Book Chapter 239
Book Chapter 239
Book Chapter 239
B. N. Singh
Arnab Roy
Dipak Kumar Maiti Editors
Recent Advances
in Theoretical,
Applied,
Computational
and Experimental
Mechanics
Proceedings of ICTACEM 2017
Lecture Notes in Mechanical Engineering
Lecture Notes in Mechanical Engineering (LNME) publishes the latest develop-
ments in Mechanical Engineering - quickly, informally and with high quality.
Original research reported in proceedings and post-proceedings represents the core
of LNME. Volumes published in LNME embrace all aspects, subfields and new
challenges of mechanical engineering. Topics in the series include:
• Engineering Design
• Machinery and Machine Elements
• Mechanical Structures and Stress Analysis
• Automotive Engineering
• Engine Technology
• Aerospace Technology and Astronautics
• Nanotechnology and Microengineering
• Control, Robotics, Mechatronics
• MEMS
• Theoretical and Applied Mechanics
• Dynamical Systems, Control
• Fluid Mechanics
• Engineering Thermodynamics, Heat and Mass Transfer
• Manufacturing
• Precision Engineering, Instrumentation, Measurement
• Materials Engineering
• Tribology and Surface Technology
To submit a proposal or request further information, please contact the Springer
Editor in your country:
China: Li Shen at li.shen@springer.com
India: Priya Vyas at priya.vyas@springer.com
Rest of Asia, Australia, New Zealand: Swati Meherishi at
swati.meherishi@springer.com
All other countries: Dr. Leontina Di Cecco at Leontina.dicecco@springer.com
To submit a proposal for a monograph, please check our Springer Tracts in
Mechanical Engineering at http://www.springer.com/series/11693 or contact
Leontina.dicecco@springer.com
Indexed by SCOPUS. The books of the series are submitted for indexing to
Web of Science.
Recent Advances
in Theoretical, Applied,
Computational
and Experimental Mechanics
Proceedings of ICTACEM 2017
123
Editors
B. N. Singh Arnab Roy
Department of Aerospace Engineering Department of Aerospace Engineering
Indian Institute of Technology Kharagpur Indian Institute of Technology Kharagpur
Kharagpur, West Bengal, India Kharagpur, West Bengal, India
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Contents
v
vi Contents
ix
Multi-scale Simulation of Elastic Waves
Containing Higher Harmonics
1 Introduction
Wave propagation can be characterized by the localized region of the sharp gradient
of field variable which changes its locations in space with time. This gives permission
to recognize the unusual nature that could be suitable for ultrasonic nondestructive
A. Sharma (B)
Mechanical Engineering Department, VIT-AP, Amaravati, India
e-mail: sharma.ambuj@vitap.ac.in
S. Kumar · A. Tyagi
Mechanical Engineering Department, IIT (BHU), Varanasi, India
e-mail: sandeep.mec@iitbhu.ac.in
A. Tyagi
e-mail: atyagi.mec@iitbhu.ac.in
© Springer Nature Singapore Pte Ltd. 2020 1
B. N. Singh et al. (eds.), Recent Advances in Theoretical, Applied, Computational
and Experimental Mechanics, Lecture Notes in Mechanical Engineering,
https://doi.org/10.1007/978-981-15-1189-9_1
2 A. Sharma et al.
spectral finite element method (WSFEM) for simulation of elastic wave propagation
in one- and two-dimensional situations [22]. In the literature, some researchers have
used wavelets as basis function to solve PDEs but most researchers have applied the
wavelet-based adaptive technique in finite difference schemes. These papers have
presented the adaptive method for propagation of a single wave, but there is a need
for different algorithms for more than one waves propagating with different veloc-
ities. Generation of higher harmonics due to material nonlinearity is not addressed
in these papers.
Multi-scale modeling is one possible solution for higher harmonics in wave prop-
agation simulation. Wavelet-based multi-scale method leads to fast and locally adap-
tive algorithms. The compactly supported refinable basis functions are main potential
advantage of the wavelet [10, 11]. However, these methods are unable to compete
with conventional finite element method. In this paper, proposed technique is inspired
by the interesting paper by Krysl et al. [23].
This paper presents multi-scale adaptive approach for solving the wave propaga-
tion problem. In the proposed wavelet-based technique, FEM is preferred due to its
capability to handle complex boundary and loading conditions instead of any other
methods. This multi-scale transformation hierarchically filters out the less significant
frequencies and offers an operative framework to retain the necessary frequencies of
the wave. In this procedure, the finest level of the coefficient matrix is calculated once
for the whole domain while the adaptively compressed coefficient matrix, which is
very small compared to complete coefficient matrix, is used in every marching step of
the solution. This paper is presenting wavelet-based nonstandard operator to improve
finite element simulation of linear and nonlinear wave propagation in a large struc-
ture. We use nonstandard operator because it is more efficient than standard operator
[24]. This will not only be useful to the structural health monitoring, but it can also
be used where waves with higher harmonics move at different group velocities. A
simple description of the nonstandard operator along with necessary algorithm and
mathematical comments is provided to remove an execution headache connected
with adaptive grid techniques. The algorithm is applied to 2D plane strain problem,
but it is general and independent of domain dimensions.
2 Mathematical Formulation
∂ 2u 2 ∂ v
2
2∂ u
2
∂ 2u
C L2 + (C 2
− C ) + C + f x = , (2(a))
∂x2 L T
∂ x∂ y T
∂ y2 ∂t 2
∂ 2v 2 ∂ u
2
2 ∂ v
2
∂ 2v
C L2 + (C 2
− C ) + C + f y = . (2(b))
∂ y2 L T
∂ x∂ y T
∂x2 ∂t 2
C L2 = λ+2μ
ρ
and C T2 = μρ are longitudinal velocity and shear velocity, respectively,
where λ = (1+ν)(1−2ν)
Eν
and μ = 2(1+ν)
E
are Lamé constants, E is Young’s modulus,
and ν is Poisson ratio. The 2D plane strain problem is discretized into the set of finite
element equations as
where [u] and [ü] are unknown coefficient vectors. [K ] and [M] are global stiffness
and mass matrix, respectively.
The matrix [T j ] is used to achieve next higher level by transforming scaling and detail
coefficients of V j−1 and W j−1 spaces, respectively. In this paper, B-spline wavelet
and Daubechies (D4) wavelet [25] are used for wave propagation.
Two observations can be made while solving some PDEs using the wavelet bases: (i)
In theoretical terms, most of the available wavelet methods have stable Riesz basis
and better condition number than FEM or FDM. (ii) But in practical applications,
wavelet methods are not yet ready to compete with the traditional FEM approach.
One important reason is while the FEM can always produce a sparse matrix with
more regular sparsity patterns, use of wavelet bases does not produce such sparse
matrices. But the combination of wavelets with other methods, such as FDM, FEM,
and recently SEM [22], show good results. Here, we have used FEM discretization
to derive a sparse matrix, as the FEM remains the most versatile tool to solve PDEs.
Let us consider a continuous wave field u(x, y) and v(x, y) for a source of excita-
tion over 2D homogeneous medium. The approximation of the continuous wave field
on the discrete domain is denoted by u j and v j . It represents the discrete wave field
that is obtained with a classic time–space finite element method for a sufficiently fine
discretization of V j ⊂ R 2 . The 2D wavelet transform cascades projections of the dis-
crete wave field over different approximation grids V1 , V2 , V3 , . . . , V j of increasing
resolution.
In this multi-scale algorithm, we used NS operator proposed by Beylkin [26]. To
the best of authors’ knowledge, no one researcher has used NS operator in wavelet–
FEM coupling or wavelet–FDM coupling. It has been proved by Beylkin [26, 27] that
NS operator is more efficient than the standard form of operator used by most of the
researchers. In this paper, we have used NS operator in two-dimensional wavelet–
finite element coupling technique. The finite element equations for the transient
problem, Eq. 3, can be expressed in the expanded form as
⎡ ⎤ ⎡ ⎤
j j j j
kuu kuv fu
⎣ ⎦ u = ⎣ ⎦. (6)
j j
kvu kvv vj fv
j
We can apply the wavelet transformation on the field variables of both the
directions:
6 A. Sharma et al.
⎡
1 ⎤
⎡ ⎤ d ⎡ ⎤
j j
1
T j
T
[0] ⎣ kuu kuv ⎦ [T ] [0] ⎢
T
u
⎢
⎥ =
⎥ T [0]
⎣
fu
⎦.
[0] T T j
kvu kvv
j [0] [T ] ⎣
e1 ⎦ [0] T T fv
j
v1
(7)
(8)
Elastic waves have been employed for identification of damage in the thin wall struc-
tures such as plates and pipes [5]. Guided Lamb waves are excited in the structures
through narrowband burst signals. In order to evaluate the performance of wavelet-
based multi-scale method. We considered an example in which a 50 × 50 mm2
homogeneous, isotropic aluminum plate with a density of 2700 kg/m3 . The simula-
tion of Lamb wave in this plate with 400 kHz central frequency is presented in Fig. 1.
Contour plots of the displacement in the x-direction at three different time instants
are well depicted in this figure.
Nonlinear Lamb wave is more sensitive to small-scale damage identification.
However, the investigations of the higher harmonics in propagating Lamb waves
Multi-scale Simulation of Elastic Waves Containing … 7
0.05 0.05
8.5E-12 8.5E-12
0.04 8E-12 0.04 8E-12
7.5E-12 7.5E-12
7E-12 7E-12
6.5E-12 6.5E-12
0.03 6E-12 0.03 6E-12
5.5E-12 5.5E-12
5E-12 5E-12
Y
Y
4.5E-12 4.5E-12
4E-12 4E-12
0.02 3.5E-12 0.02 3.5E-12
3E-12 3E-12
2.5E-12 2.5E-12
2E-12 2E-12
1.5E-12 1.5E-12
0.01 1E-12 0.01 1E-12
5E-13 5E-13
0 0
0 0.01 0.02 0.03 0.04 0.05 0 0.01 0.02 0.03 0.04 0.05
X X
(a) 8 µ s (b) 16 µ s
0.05
8.5E-12
0.04 8E-12
7.5E-12
7E-12
6.5E-12
0.03 6E-12
5.5E-12
5E-12
4.5E-12
Y
4E-12
0.02 3.5E-12
3E-12
2.5E-12
2E-12
1.5E-12
0.01 1E-12
5E-13
0
0 0.01 0.02 0.03 0.04 0.05
X
(c) 24 µ s
Fig. 1 Contour plots of the displacement in the x-direction at three different time frames for
isotropic plate
were ignored due to high computational cost. To see the efficiency of the wavelet-
based method, higher harmonics are added in the Lamb wave and propagation of
waves is observed. This study uses the following actuation function with 400 kHz
central frequency:
f o sin(t) ∗ (sin(0.1t))2 + 0.1 f o sin(2 ∗ 5t) ∗ (sin(t))2 , t < 10π
E(t) =
0, otherwise,
0.8
0.6
0.4
0.2
Amplitude
-0.2
-0.4
-0.6
-0.8
-1
0 5 10 15 20 25 30 35
Time ( S)
0.3, density = 2700 kg/m3 , and Young’s modulus E = 69 GPa are assumed as material
properties. The Lamb wave in this material has longitudinal velocity C L = 5299 m/s
and transverse velocity C T = 3135 m/s. The waves are actuated by employing pin
forces applied to the left boundary of the plate. The excitation forces are parallel to
the longitudinal (propagating) direction. In-phase pin forces are applied to the top
and bottom edge nodes of the plate for excitation of fundamental symmetric (S 0 )
modes, and the antisymmetric modes are propagated by imposing out-of-phase pin
forces. In this paper, we considered the cases in which the pure S 0 mode is excited.
Ten cycles Hanning-window actuation is given through excitation function to deliver
a limited cycle sinusoidal tone burst.
Higher frequency wave propagation problems demand enormous computer
resources because of very large number of time integration steps and highly dense
mesh. Generally, in the case of Lamb wave, 20 elements per wavelength are required
but this is not sufficient for higher harmonic simulation. Figure 3 depicts the measured
nodal displacement response of time-domain signals obtained using FEM simulation
of the plate with 40, 80, and 120 elements per wavelength. It can be observed that
higher harmonics are not properly visible in the response of the plate with 40 ele-
ments per wavelet. On the other hand, as shown in the same figure, higher harmonics
are visible for 80 elements per wavelength.
Multi-scale Simulation of Elastic Waves Containing … 9
10-3
2.5
1.5
0.5
Amplitude
-0.5
-1
-1.5
-2
-2.5
0 10 20 30
Time ( S)
Fig. 3 Comparison of response of plate for 40, 80, and 120 elements per wavelength
In the present analysis, B-spline and Daubechies (D4) wavelet are used to estab-
lish robustness and sensitivity of wavelet-based wave propagation method. To capture
higher harmonics in the plates, FEM uses 17,080 uniformly distributed nodes while
half of the FEM nodes are required after application of one level of wavelet trans-
form. Nodal displacement response of plate received from B-spline and D4 wavelet
transform at level 1 along with FEM results is demonstrated in Fig. 4. It establishes
good agreement between conventional finite element and proposed wavelet-based
method. It can be observed that B-spline wavelet produces response close to FEM
results, while there is some deviation in the results of D4 wavelet. Further, we exam-
ined wavelet-based method at various levels of wavelet transform to find the level
up to which this method can work efficiently. These results show some attenuation
but wavelets are not eliminating higher frequency components of waves which are
important in many analyses.
10 A. Sharma et al.
-3
10
2.5
FEM
2 B-spline
D4
1.5
0.5
Amplitude
-0.5
-1
-1.5
-2
-2.5
0 10 20 30 40
Time ( S)
Fig. 4 Comparison of higher harmonic Lamb waves plates response at wavelet transform level 1
4 Conclusion
The spatial derivative operators in the wave equations are handled using multi-
resolution transforms in a physical domain. We presented a wavelet-based frame-
work to reduce the size of global stiffness matrix of finite element analysis which
is becoming too large in the case of nonlinear wave propagation problem. Wavelet-
based method is not only to develop the compressed stiffness matrix, but also to
propagate higher harmonics of waves using least number of nodes and able to reduce
the computational cost significantly. Without disturbing the programming advantages
of FE regarding the implementation of boundary conditions and efficient numerical
integration of interpolation functions, wavelet-based methods are able to reduce the
size of matrix as much as one by a sixteenth of original FE matrix. These fundamental
characteristics show that the wavelet-based method can be utilized for more complex
wave propagation problems.
Multi-scale Simulation of Elastic Waves Containing … 11
References
1. Hassan W, Veronesi W (2003) Finite element analysis of Rayleigh wave interaction with finite-
size, surface-breaking cracks. Ultrasonics 41:41–52
2. Ham S, Bathe KJ (2012) A finite element method enriched for wave propagation problems.
Comput Struct 94–95:1–12
3. Per W, Bertil G (2005) High-order difference methods for waves in discontinuous media. J
Comput Appl Math 192(1):142–147
4. Tadeu A, Godinhoa L, Antónioa J, Mendesa PA (2007) Wave propagation in cracked elas-
tic slabs and half-space domains—TBEM and MFS approaches. Eng Anal Bound Elements
31(7):819–835
5. Nieuwenhuis JH, Neumann JJ, Greve DM, Oppenheim IJ (2005) Generation and detection
of guided waves using PZT wafer transducers. IEEE Trans Ultrason Ferroelectr Freq Control
52(11):2103–2111
6. Delsanto PP, Whitcombe T, Chaskelis HH, Mignogna RB (1992) Connection machine simu-
lation of ultrasonic wave propagation in materials. I: the one-dimensional case. Wave Motion
16:65–80
7. Delsanto PP, Schechter RS, Migogna RB (1997) Connection machine simulation of ultrasonic
wave propagation in materials. III: the three-dimensional case. Wave Motion 26(4):329–339
8. Casadei F (2012) Multiscale analysis of wave propagation in heterogeneous structures.
Dissertation, School of Aerospace Engineering, Georgia Institute of Technology
9. Beylkin G, Coifman R, Rokhlin V (1991) Fast wavelet transforms and numerical algorithms
I. Commun Pure Appl Math XLIV:141–183
10. Amaratunga K, Williams JR (1993) Wavelet based Green’s function approach to 2D PDEs.
Eng Comput 10(4):349–367
11. Qian S, Weiss JJ (1993) Wavelets and the numerical solution of partial differential equations.
J Comput Phys 106:155–175
12. Holmstrom M (1999) Solving hyperbolic PDEs using interpolating wavelets. SIAM J Sci
Comput 21(2):405–420
13. Lippert RA, Arias TA, Edelman A (1998) Multiscale computation with interpolating wavelets.
J Comput Phys 140:278–310
14. Liandrat J, Tchamitchian PH (1990) Resolution of the 1D regularized burgers equation using
a spatial wavelet approximation. Report NASA Langley Research Centre, Report No: NASA
CR – 187480, Hampton VA
15. Vasilyev OV, Bowman C (2000) Second generation wavelet collocation method for the solution
of partial differential equations. J Comput Phys 165:660–693
16. Rathish Kumar BV, Mehra M (2005) Wavelet-Taylor Galerkin method for the burgers equation.
BIT Numer Math 45:543–560
17. Swelden W (1998) The lifting scheme: a construction of second generation wavelets. SIAM J
Math Anal 29(2):511–546
18. Carnicer JM, Dahmen W, Peña JM (1996) Local decomposition of refinable spaces and
wavelets. Appl Comput Harmon Anal 3:127–153
19. Dahmen W, Stevenson R, Siam S (2015) Element-by-element construction of wavelets
satisfying stability and moment conditions. SIAM Numer Anal 37:319–352
20. Hong T-K, Kennett BLN (2002) A wavelet-based method for simulation of 2-D elastic wave
propagation. Geophys J Int 150(3):610–638
21. Chen XF, Yang ZB, Zhang XW, He ZJ (2012) Modeling of wave propagation in one-dimension
structures using B-spline wavelet on interval finite element. Finite Elements Anal Des 51:1–9
22. Mitra M, Gopalakrishnan S (2006) Wavelet based 2-D spectral finite element formulation for
wave propagation analysis in isotropic plates. Comput Model Eng Sci 15(1):49–67
23. Krysl P, Grinspun E, Schroder P (2002) Natural hierarchical refinement for finite element
methods. Int J Numer Methods Eng 56(8):1109–1124
24. Gines D, Beylkin G, Dunn J (1998) LU Factorization of non-standard forms and direct
multiresolution solvers. Appl Comput Harmon Anal 5:156–201
12 A. Sharma et al.
25. Stollnitz EJ, Derose TD, Salesin DH (1996) Wavelets for computer graphics. Morgan Kaufmann
Publishers Inc., California
26. Beylkin G (1992) On the representation of operator in bases of compactly supported wavelets.
SIAM J Numer Anal 6(6):1716–1740
27. Beylkin G, Keiser JM (1997) On the adaptive numerical solution of nonlinear partial differential
equations in wavelet bases. J Comput Phys 132:233–259
Effect of Skewness on Random
Frequency Responses of Sandwich Plates
Abstract This study presents the effect of skewness in natural frequency responses
of sandwich plates. The free vibration analysis is carried out by using higher order
zigzag theory (HOZT) considering random input parameters. It satisfies the trans-
verse shear stress continuity condition and the transverse flexibility effect. The in-
plane displacement throughout the thickness is assumed to vary cubically while
transverse displacement is considered to vary quadratically within the core and con-
stant at top and bottom plates. An efficient C0 stochastic finite element approach is
developed for the implementation of proposed plate theory in the random variable
surrounding. Compound stochastic effect of all input parameters is presented for the
different degrees of skewness in sandwich plates. Intensive Monte Carlo simulation
(MCS) is employed for solving the stochastic-free vibration equations and statis-
tical analysis is conducted for illustration of the results. The present algorithm for
sandwich plate is validated with previous literatures and it is found to be in good
agreement.
1 Introduction
A sandwich plate is a multilayered plate having two face sheets and a core embed-
ded in between them through adhesive. The face sheets are relatively thin but of high
strength and stiffness material, whereas the core is made up of relatively thick and
lower density material. The high specific strength and stiffness of sandwich struc-
tures make them suitable for crucial engineering applications like automobile, civil
construction, aerospace, and marine industries. Sandwich plates are widely used in
design and construction of aerospace craft. In such application, these materials are
subjected to wide environmental changes such as pressure, temperature, density, and
et al. [13]. Recently, stochastic analysis is carried out by Kumar et al. [14–19], Karsh
et al. [20–26], and Mukhopadhyay et al. [27, 28]. Most of the research is carried
out by using deterministic approach, whereas few researchers focused on stochastic
approach.
Here, the effect of skewness (Fig. 1) on natural frequency response, having taken
into consideration the compound variation of all input parameters, is studied. There-
after, this paper is presented as: Theoretical formulation is described in Sect. 2,
result and discussion are illustrated in Sect. 3, whereas conclusion and future scope
are presented in Sect. 4.
2 Theoretical Formulation
where [a] is unit step function. The equation for generalized displacement vector is
given as
n
{s( )} = ζi ( )si ( ), (2)
k=1
where [r ( )] is the random natural frequency. The global mass matrix [m( )] is
u +n l
n ˚ ¨
[m( )] = ρk ( )[n]T [ j]T [n][ j]dxdydz = [n]T [k( )][n]dxdy, (5)
k=1
whereρk ( ) is stochastic mass density of kth order, [j] is of the order of 3X11, and
[n] is the shape function matrix. The equation for stiffness matrix [k( )] is given as
u +n l
n
[k( )] = ρk ( )[ j]T [ j]dz. (6)
kl
For storing the global stiffness in one array, we have used the skyline technique.
For getting static solution, Gaussian decomposition scheme is used and for free
vibration analysis simultaneous iteration technique is used.
almost same which lies in between φ = 15° and φ = 30°. This corroborates the fact
obtained vide probability density function (PDF) plots.
4 Conclusions
Based on higher order zigzag theory (HOZT), the accuracy and applicability of
the proposed finite element model for free vibration analysis of sandwich plates
are studied. The novelty of the present study includes the skewness effect on free
vibration of sandwich plates. The natural frequency of sandwich plate is compared
with that of skewed sandwich plate by means of probability density function (PDF)
plots. The first, second, and third natural frequencies of unskewed sandwich plates are
compared with plates having skewness of 15°, 30°, 45°, and 60°. It is observed that
the unavoidable source uncertainties cause significant deviation of natural frequency
from the mean deterministic value. Therefore, it is of utmost importance to consider
the effect of skewness and source uncertainty in design and analysis of sandwich
plate and other complex structures for safe and realistic design. Based on these
observations, the present work can be extended to deal with more complex structures.
18 R. R. Kumar et al.
Fig. 2 Random natural frequency (rad/s) of sandwich plates for a first, b second, and c third natural
frequencies with skew angle, (ϕ) = 0°, 15°, 30°, 45°, and 60°
Effect of Skewness on Random Frequency Responses … 19
Acknowledgements The first and second authors would like to acknowledge the financial support
received from MHRD GOI during this research work.
References
19. Kumar RR, Pandey KM, Dey S (2020) Stochastic free vibration analysis of sandwich plates: a
radial basis function approach. In: Reliability, Safety and Hazard Assessment for Risk-Based
Technologies. Springer, Singapore, pp 449–458
20. Karsh PK, Mukhopadhyay T, Dey S (2018) Spatial vulnerability analysis for the first ply failure
strength of composite laminates including effect of delamination. Compos Struct 15(184):554–
567
21. Karsh PK, Mukhopadhyay T, Dey S (2018) Stochastic dynamic analysis of twisted functionally
graded plates. Compos B Eng 15(147):259–278
22. Karsh PK, Mukhopadhyay T, Dey S (2019) Stochastic low-velocity impact on functionally
graded plates: probabilistic and non-probabilistic uncertainty quantification. Compos B Eng
15(159):461–480
23. Karsh PK, Mukhopadhyay T, Dey S (2018) Stochastic investigation of natural frequency for
functionally graded plates. In: IOP conference series: materials science and engineering 2018
Mar, vol 326, no 1. IOP Publishing, p 012003
24. Karsh PK, Kumar RR, Dey S (2019) Stochastic impact responses analysis of functionally
graded plates. J Braz Soc Mech Sci Eng 41(11):501
25. Karsh PK, Kumar RR, Dey S (2019) Radial basis function based stochastic natural frequencies
anaysis of functionally graded plates. Int J Comput Methods
26. Karsh PK, Mukhopadhyay T, Chakraborty S, Naskar S, Dey S (2019) A hybrid stochastic
sensitivity analysis for low-frequency vibration and low-velocity impact of functionally graded
plates. Compos B Eng 176:107221
27. Mukhopadhyay T, Naskar S, Karsh PK, Dey S, You Z (2018) Effect of delamination on the
stochastic natural frequencies of composite laminates. Compos B Eng 1(154):242–256
28. Mukhopadhyay T, Naskar S, Karsh PK, Dey S, You Z (2018) Effect of delamination on the
stochastic natural frequencies of composite laminates. Compos B Eng 154:242–256
29. Dey S, Mukhopadhyay T, Naskar S, Dey TK, Chalak HD, Adhikari S (2017) Probabilistic
characterisation for dynamics and stability of laminated soft core sandwich plates. J Sandw
Struct Mater 1–32
30. Wang CM, Ang KK, Yang L (2000) Free vibration of skew sandwich plates with laminated
facings. J Sound Vib 235:317–340
31. Kulkarni SD, Kapuria S (2008) Free vibration analysis of composite and sandwich plates
using an improved discrete Kirchhoff quadrilateral element based on third order zigzag theory.
Comput Mech 42:803–824
Design and Simulation of 3-DoF Strain
Gauge Force Transducer
A. Jaiswal (B)
Department of Mechatronics Engineering, Manipal Institute of Technology,
Manipal Academy of Higher education (MAHE), Manipal 576104, Karnataka, India
e-mail: ankurmech13@gmail.com
H. P. Jawale
Department of Mechanical Engineering, Visvesvaraya National
Institute of Technology, Nagpur 440010, Maharashtra, India
e-mail: j.hemanth@rediffmail.com
K. Shrivastava
Department of Ocean Engineering and Naval Architecture, Indian
Institute of Technology, Kharagpur, India
e-mail: krshrivastava@gmail.com
1 Introduction
Force transducers are extensively used to measure the applied force on structural
members. It generally consists of an elastic load-bearing element which undergoes
strain deformation upon application of force or torque. This deformation could be
measured using strain gauges, thus converting the applied force into a proportional
electric signal.
Traditionally, force transducers are uniaxial and multiple such transducers could
be used to contrive multiaxial transducers. For instance, Kaneko [1] demonstrated
that a combination of two similar three-axis force sensors could form one six-axis
force sensor and developed a characteristic matrix, connecting the load and sensor
output vectors. Svinin and Uchiyama [2] proposed a generalized theoretical model
for multidimensional force sensor designed using elastic components connected in
parallel.
Kim et al. [3] and Kim [4] described a design and development procedure of
a six-component force and moment sensor. They derived equations for calculating
rated strains on the components of the sensor and performed a finite element (FEM)
analysis for confirming the strains obtained from the theoretical analysis. Liu and
Tzo [5] proposed identical T-shaped bar-type six-axis force sensor. They calculated
the sensitivity of force-sensing member with respect to design parameters using finite
element analysis in conjunction with a design optimization. Liu et al. [6] designed
a six-dimensional piezoelectric heavy force sensor based on load-sharing principle.
They performed a finite element analysis using ANSYS software for calculating
the static and dynamic response of the designed sensor under heavy loading. The
experiments on the fabricated sensor demonstrated very less nonlinearity errors and
verified the numerical results. Liang et al. [7] and Sunand et al. [8] derived the
equations of force and moment for the six-component force sensor and determined
the theoretical value of strains. This result was validated using finite element method
(FEM) and optimization method by trial and error, employing the response surface
methodology (RSM).
Bayo and Stubbe [9] proposed a frame/truss-type six-axis sensor for robotic appli-
cations. They employed the axial deformation behavior of the six-axis sensor to
eliminate disadvantages of the crossbar and obtained better measure of isotropy and
decoupling. They also compared the performance parameters (condition number,
overall static and dynamic stiffness, strain gauge sensitivity) and found significant
improvement over the earlier design of six-axis wrist force sensor. Chao and Chen
[10] applied penalty method to optimize the geometrical design parameters of a
wrist-type elastic force sensor. They used calibration matrix along with strain gauge
sensitivity as a performance index for design optimization. Based on these design cri-
teria they devised a novel decoupled wrist force sensor with enhanced force sensitivity
and minimum stiffness.
Kang [11] formulated closed-form solution of forward kinematics for 6-DoF Stew-
art platform based force transducer using linearization approach. Later, Dwarakanath
Design and Simulation of 3-DoF Strain Gauge Force Transducer 23
et al. [12] performed design parameter sensitivity analysis for optimal size of Stewart
platform based force transducer and fabricated a prototype of the sensor.
Present study is an extension to the work done by Deshpande et al. [13]. They
did a mathematical analysis of a novel three-axis force transducer and validated the
analytical results with experimental results. This paper incorporates finite element
analysis to study the behavior of the initial design of force transducer (FT1) and
determines the optimal locations for mounting the strain gauges. A modified design
of the force transducer (FT2) is also proposed in order to enhance the performance
of the cantilever-beam-type force-sensing elements. In the new design, the principal
stress and shear stress are also analyzed using finite element analysis for various
loading conditions. The comparison of analysis results is presented herewith. This
sensor could find its application in robotic manipulators and multiaxial load cells.
Force transducers are deployed for knowing the numerical value of the applied force
and torque on the body, which is ideally equal to the true value of the applied load.
The output of the transducer depends on the characterized relationship between the
output value and the property under measurement. Force transducers are designed
so as to have elastic member for loading elements in any of the six loading axes.
The input loading creates deflection in the elastic member which in turn creates
proportional elongation also known as strain. The strain is measured as a quantifiable
change in resistance of the coiled conductor in the strain gauge. Thus, strain gauges
proportionally convert the strain developed into change in the electric current at
output terminals [14].
There are various methods used to determine torque and force applied on any
elastic member among which the most common method is using strain gauges. The
strain gauge can be configured to distinguish tensile and compressive strains, to give
conversions in the form of positive or negative signals, thus making it possible to
distinguish expansion as well as contractions of the members having strain gauges
on it. The gauge factor is fundamental term correlating the sensitivity of the strain
gauge. It is defined as the ratio of change in electrical resistance (R) to the mechanical
strain ε captured by gauge conductor coil, and expressed as
ρL
R= (1)
A
R/R R/R
GF = = (2)
L/L ε
where
ρ = Resistivity/conductivity constant,
L = Length of the sensor,
24 A. Jaiswal et al.
This paper presents an innovative design for a force transducer capable of sensing
triaxial loads and moments. The transducer is designed in a frame-like structure by
stacking three cantilever beams upon each as depicted in Fig. 1. Each frame carries
its individual strain rosettes which are sensitive to a specific orthogonal loading
direction. Stacking all three frames into a single structure produces a 3-DoF force-
sensing transducer.
This paper also discusses variations to the original design of the transducer in
order to enhance its sensing capabilities. The alterations were proposed to attain
uniform stress distribution and better strain dispersion among the structural members.
Figure 2a depicts the original and Fig. 2b depicts modified designs of the force
transducer. The two designs are further evaluated for stress distribution patterns
using FE analysis. For the sake of simplicity, the original design of the transducer
will be denoted as FT1 and the modified new design would be denoted as FT2, further
along this paper.
Design and Simulation of 3-DoF Strain Gauge Force Transducer 25
Fig. 1 a CAD model depicting the force transducer (green) also schematics of pulleys and
deadweight for loading (brown). b Experimental setup [14]
Fig. 2 CAD model depicting transducer designs a FT1 (original design) and b FT2 (modified
design)
26 A. Jaiswal et al.
The material used for constructing the transducer was considered to be isotropic
and homogeneous. Standard structural steel properties were assumed as shown in
Table 1.
The meshing for FE analysis was performed in ABAQUS using a ten-noded quadratic
tetrahedron element (C3D10). Size of the mesh was program controlled. Figure 3
illustrates the meshing pattern for FT1 and FT2 transducer designs.
The transducer was constrained in all directions at its base to depict a fixed–fixed
boundary condition. This was done to replicate a real-life scenario when the trans-
ducer is firmly placed on a solid surface. The load was applied on the topmost surface
of the transducer. The load was applied for uniaxial and triaxial loading conditions,
with varying magnitudes, in order to achieve a detailed analysis. The boundary condi-
tions and loading patterns were kept identical for FT1 and FT2 transducers as shown
in Fig. 4.
Fig. 3 Meshed FE model of force transducer a FT1 (original design) and b FT2 (modified design)
The static analysis was performed by applying various discrete load levels having
uniform increment from 0.2 to 2 N. Loads were applied in all three uniaxial directions
(x, y, and z) independently. A triaxial loading was also applied for the same load levels.
The load was applied on the cube-shaped block at the top of the transducer.
Maximum principal strain contour on different elements of transducers FT1 and FT2
at 0.2 N, 1 N, and 2 N loads is depicted in Fig. 5a, c, e and Fig. 5b, d, f, respectively.
Table 2 enumerates the maximum principal strain for different magnitudes of load
in FT1 and FT2.
Figure 6 represents the linear relationship between magnitude of load and induced
strain exhibited by the FT1 and FT2 transducers. It is observed that maximum
28 A. Jaiswal et al.
Fig. 4 Boundary conditions and loading patterns for a FT1 and b FT2
strain occurs on the element which is sensitive to the direction of applied force.
The experimental results of FT1 transducer [14] confirm this linear relationship.
Figure 7 shows the strain difference between FT1- and FT2-type transducers when
the applied load is same. It is observed that the strain difference also increases linearly
with load.
The modified design, FT2, has additional blocks at the end of each sensing ele-
ment, which provide rigidity to the cantilever-sensing element. It is observed that in
single loading conditions, the induced strain values in FT2 transducers are higher
than FT1 transducers. It is also observed that in triaxial loading conditions, the strain
difference between FT1 and FT2 is higher compared to single loading conditions.
This infers that the modified design is more sensitive to smaller variations in the load.
The advantage of the modified design is that it detects the strain when the applied
load is very small and when there is a small variation in the applied load.
Design and Simulation of 3-DoF Strain Gauge Force Transducer 29
Table 2 Value of strain obtained from ABAQUS for maximum principal elastic strain in FT1 and FT2
FT1 FT2
Load (N) Maximum principal strain (* e−5 ) Load (N) Maximum principal strain (* e−5 )
x y z xyz x y z xyz
0.2 1.513 0.459 0.594 0.829 0.2 1.551 0.59 0.619 1.194
0.4 3.026 0.918 1.189 1.645 0.4 3.101 1.181 1.238 2.371
0.6 4.551 1.381 1.788 2.474 0.6 4.664 1.777 1.862 3.565
0.8 6.068 1.844 2.38 3.29 0.8 6.215 2.368 2.481 4.745
1 7.587 2.32 2.983 4.113 1 7.766 2.958 3.1 5.93
1.2 9.106 2.767 3.584 4.954 1.2 9.316 3.549 3.719 7.121
1.4 10.625 3.229 4.171 5.755 1.4 10.87 4.14 4.338 8.299
1.6 12.144 3.681 4.77 6.569 1.6 12.43 4.735 4.962 9.485
1.8 13.663 4.143 5.366 7.41 1.8 13.98 5.326 5.581 10.6
2 15.182 4.61 5.959 8.218 2 15.53 5.917 6.2 11.84
A. Jaiswal et al.
Design and Simulation of 3-DoF Strain Gauge Force Transducer 31
3.5
Principal Strain (FT2 - FT1)
X- Direction
Y- Direction
3 Z- Direction
XYZ- Direction
2.5
1.5
0.5
0
0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Load (N)
The transducer FE model was analyzed for shear strain by applying biaxial com-
binational loading of varying magnitude. The load increment was similar to that
used in principal strain analysis. The moment generated in the sensing element of
the transducer was neglected in this analysis. The strain gauge deployed for direct
strain estimation measures both the normal and shear components of the strains. The
analysis carried out gives a better visualization of shear patterns as shown in Fig. 8a,
c, e for FT1 and Fig. 8b, d, f for FT2. The maximum shear elastic strain in FT1 and
FT2 is given in Table 3.
32 A. Jaiswal et al.
Table 3 Value of strain obtained from ABAQUS for maximum shear elastic strain in FT1 and FT2
FT1 FT2
Load (N) Maximum shear strain (* e−5 ) Load (N) Maximum shear strain (* e−5 )
x-y y-z z-x x-y y-z z-x
0.2 4.977 2.274 4.586 0.2 6.757 2.113 7.126
0.4 9.937 4.541 9.157 0.4 13.41 4.191 14.14
0.6 14.95 6.83 13.77 0.6 20.17 6.304 21.26
0.8 19.88 9.084 18.32 0.8 26.85 8.383 28.28
1 24.89 11.37 22.93 1 33.6 10.5 35.4
1.2 29.89 13.66 27.55 1.2 40.34 12.61 42.53
1.4 34.82 15.91 32.09 1.4 47.02 14.69 49.54
1.6 39.83 18.2 36.71 1.6 53.77 16.8 56.67
1.8 44.76 20.46 41.25 1.8 60.43 18.88 63.68
2 49.77 22.74 45.86 2 67.18 20.99 70.8
The variation in maximum shear strain with respect to applied load is represented
in Fig. 9. The transducer exhibits linear relationship between load and induced shear
strain in both FT1- and FT2-type transducers. The FT2 transducer exhibits higher
strain values in x-y and z-x loading directions, whereas FT1 has higher strain magni-
tude in y–z direction. Figure 10 represents the shear strain difference between FT1-
and FT2-type transducers when the applied load is same.
It is evident from the data enumerated in Tables 2 and 3 that FT2 shows enhanced
stress concentration levels for similar loading conditions as FT1 for majority of
the cases. This leads to more localized strain and provides an advantage of better
stress sensitivity. One exception to the case could be noted in Y –Z principal shear
stress loading, where FT2 shows the reduced sensitivity than its predecessor. It is
also observed that change in maximum principal strain is least in X-direction. The
member most sensitive to the X-directional loading is at the bottom and attached to
the base. The rigidity provided by the fixed–fixed boundary condition and increased
mass in the modified design could be attributed for this deficit. It warrants further
research and possible alterations to the design of FT2 in order to rectify this problem.
6 Conclusions
This paper describes a detailed analysis of strain estimation in a triaxial force trans-
ducer which has engineering applications in platform-type sensor. The experimental
observations reflect that the strain concentration at specific locations modifies the
characteristics of the transducer. A comparative study of the strain estimation for
two different transducer designs is also presented, which is necessary for the safe
design and manufacturing of the transducers. The analysis of the strain patterns helps
in determining the location of maximum stress concentration, which provides an esti-
mate location to mount the strain gauges. Some of the notable observations from the
FE analysis are summarized below:
1. The experimental results for transducer FT1 show that a linear increase in load
causes a linear increase in strain on the sensing element [13]. The transducer FT1
has been tested and validated in simulation environment.
2. The estimation of strain for FT2 has been done by simulation and results suggest
that the strain in the sensing elements increases linearly with the applied load, as
in the case of FT1.
Design and Simulation of 3-DoF Strain Gauge Force Transducer 35
3. In transducer FT2, the blocks added at the end of each sensing element confine
the principal strain to a smaller region as compared to FT1. The advantage of the
modified design is that it detects the strain when the applied load or the variation
in the applied load is very small.
The strain-sensing capability of the proposed design (FT2) is better than that of the
FT1 transducer. This kind of transducer finds applications in serial- and spatial-type
parallel manipulators used for payload sensing.
References
1. Kaneko M (1993) A new design of six-axis force sensors. In: Proceedings of the IEEE
international conference on robotics and automation, pp 961–967
2. Svinin MM, Uchiyama M (1995) Optimal geometric structures of force/torque sensors. Int J
Robot Res 14:560–573
3. Kim GS, Kang DI, Rhee SH (1999) Design and fabrication of a six-component force/moment
sensor. Sens Actuators A 77:209–220
4. Kim GS (2001) The design of a six-component force/moment sensor and evaluation of its
uncertainty. Meas Sci Technol 12:1445
5. Liu SA, Tzo HL (2002) A novel six-component force sensor of good measurement isotropy
and sensitivities. Sens Actuators A 100:223–230
6. Liu W, Li Y et al (2011) Research on parallel load sharing principle of piezoelectric six-
dimensional heavy force/torque sensor. Mech Syst Signal Process 25:331–343
7. Liang Q, Zhang D et al (2013) Design and analysis of a novel six-component F/T sensor based
on CPM for passive compliant assembly. Meas Sci Rev 13:253–264
8. Sunand Y, Liu Y et al (2015) Design and optimization of a novel six-axis force/torque sensor
for space robot. Measurement 65:135–148
9. Bayo E, Stubbe J (1989) Six-axis force sensor evaluation and a new type of optimal frame truss
design for robotic applications. J Field Robot 6:191–208
10. Chao LP, Chen KT (1997) Shape optimal design and force sensitivity evaluation of six-axis
force sensors. Sens Actuators A Phys 63:105–112
11. Kang CG (2001) Closed-form force sensing of a 6-axis force transducer based on the Stewart
platform. Sens Actuators A 90:31–37
12. Dwarakanath T, Dasgupta B, Mruthyunjaya T (2001) Design and development of a Stewart
platform based force–torque sensor. Mechatronics 11:793–809
13. Deshpande M, Jawale HP, Thorat HT (2016) Development, calibration and testing of three axis
force sensor. In: 2016 7th international conference on mechanical and aerospace engineering
(ICMAE), IEEE, pp 285–289
14. Stefanescu DM (2011) Handbook of force transducers: principles and components. Springer
Science & Business Media
15. Yurish S (2014) Modern sensors, transducers and sensor networks. Lulu.com
16. Chen D, Song A, Li A (2015) Design and calibration of a six-axis force/torque sensor with
large measurement range used for the space manipulator. Procedia Eng 99:1164–1170
A Micromechanical Study of
Fibre-Reinforced Composites with
Uncertainty Quantification and Statically
Equivalent Random Fibre Distribution
1 Introduction
Composite materials have become an advisable choice in aerospace and other indus-
tries due to their high strength to weight ratio and high modulus [1]. In fibre rein-
forced composites, fibres are the main load-carrying members. Therefore, fibre
volume fraction and fibre distribution morphology have a crucial influence on the
strength and stiffness properties of composite material. To study these effects, many
researchers addressed micromechanics of composite material. The issue concerning
the micromechanical study is the generation of RVE with the desired dimensions
and spatial distribution of reinforcements in RVE, which highly depends upon the
manufacturing process. Many methods have been developed for the generation of
RVE model with a random distribution of fibres using Poisson point distribution [2].
In this microscopic scale approach, the constituents are employed in conjunction
with homogenization to predict the composite behaviour. Here, the average mechan-
ical characteristics of a lamina have to be estimated from the known characteristics
of the fibre and matrix materials taking into account the fibre volume fraction and
fibre packing arrangement. The current study in micromechanics is to estimate the
effective elastic properties of the RVE of the material with an equivalent random dis-
tribution of fibres [3, 4]. Effective elastic properties of composite material depend on
the shape, properties and spatial distribution of the fibre. Among the various uncer-
tainties present, the following uncertainties have been considered for the analysis:
(a) effects of volume fraction (b) effect of the randomness in the fibre arrangements
(c) effect of the uncertainty in the geometry of the fibre cross section.
In the current study, a micromechanical study [5] is proposed to calculate mechani-
cal properties of composite material with regular fibre array and random fibre arrange-
ment in a unit cell model [6]. In this case, an image of the cross section of a certain
unidirectional ply is taken and transformed into a computer recognizable format by
using binary image processing technique. From the binary image, RVEs are gen-
erated with single fibre and random fibre distribution by defining different scatter
conditions and also to incorporate randomness in the RVE. The evaluation of the
effect of fibre arrangements on the mechanical properties should be both qualita-
tive and quantitative [7]. To achieve this goal, a comprehensive study is carried out
with RVEs having 1, 12, 20 and 50 fibres. The overall properties are calculated as
an average over the respective volumes of the constituents. Mathematical theory of
homogenization [8] is used to obtain a suitable constitutive model at the macroscopic
level. In the next study, we considered the RVEs with fibres randomly distributed in
a periodic unit cell but with different fibre cross sections, i.e. circular, distorted and
elliptical cross sections. Then, we studied how these different fibre cross sections
affect the mechanical properties.
Fig. 2 Random fibre distribution. a Moving window technique used to calculate RVE, b Deviation
of fibre from their ideal distribution
Initially, scatter has been estimated in the fibre distribution from digital image
analysis. Using these data, information about the location of centre of the fibres, dis-
tribution of fibre radius and distance between neighbouring fibres are obtained. Then,
the obtained statistical parameters are utilized to construct a statistically equivalent
RVE based on certain numerical algorithms [9]. Figure 1 is a micrograph image of
composite microstructure of size 706 × 678 pixels (Fig. 2).
From the image, for an ideal distribution of fibre, four random fibres are chosen
along the breadth. Then, ideal positions are calculated from the deviation of centre
of fibres in the original micrograph. Figure 3 also shows centres of ideal fibre distri-
bution. After extracting the actual coordinates of the fibre centres, we estimate the
40 S. Koley et al.
Fig. 3 RVE with regular and random fibre distribution. a Regular fibre arrangements in a square
domain, b Modelling of RVE with random fibre distribution
ideal coordinates and then, the deviation from the real to the ideal coordinates are
calculated. By doing this process, scatter has been estimated and with reference to
this, an RVE is modelled with the random distribution of fibres.
3 Generation of RVE
This theory establishes mathematical relations between micro and macro-fields [8],
using a multi-scale perturbation method, the effective properties emerge as a conse-
quence of these relations. The local displacement field in the cell is given as
where x is the actual coordinate, y is the scaled unit cell coordinate, u0 (x) is the
macro response and u1 (y) is the periodic micro correction, ε denotes the ration of the
RVE size to the global structural dimension. u1 (y) can be obtained from each of the
ij
six fundamental macro-strains ex , by solving the periodic cell problem with applied
unit macro strain.
∂ ε y ∂
− (Cijkl ekl (χ rs )) = Cijrs (y) (2)
∂yj ∂yj
where χ rs is y periodic. From the periodic solution, the total strain eij (u) ≈ eijx (u0 ) +
y
eij (u1 ) ≈ (1 + Mijkl )eklx (u0 ), where Mijkl are the pointwise influence functions and
they depend on vf at the level of the cell.
In the present study, an in-house finite element code is developed to determine the
homogenized properties of composite materials using the mathematical theory of
homogenization. Polymeric matrix such as 3501-6 epoxy and AS4 carbon fibre mate-
rial is considered in this study. These properties are given in Soden et al. [12]. The
focus of the analysis is to study both the longitudinal, transverse and shear behaviour
of the material at the microscale.
RVE is modelled with single fibre keeping the fibre volume fraction 0.6. The RVE
model is meshed with three-dimensional four-node tetrahedral elements. Effective
properties for this RVE are estimated and the results are tabulated in Table 1.
Mechanical properties for the composite laminae were considered from experi-
ment studies carried out in Soden et al. [12].
Predicted effective elastic constants are given in Table 1, which are close to the
experimental results (see Table 2). The typical percentage differences of the effective
42 S. Koley et al.
elastic constants are –1.14% for E1 , 13.21% for E2 and E3 , 14.45% for G 12 and G 13
and 21.87% for G 23 with respect to the experimental results.
RVEs are generated with 12, 20 and 50 fibres maintaining the volume fraction of
0.6. Here, the fibres are randomly distributed as shown in Fig. 4. Three cases are
considered: Case 1: RVE with 12 fibres, Case 2: RVE with 20 fibres and Case 3:
RVE with 50 fibres. For Case 1, three RVE models are generated, for Case 2, four
RVE models are generated and for Case 3, eight RVE models are generated. In Fig. 4,
sample RVEs of each case are shown but the results are given for all the models.
Observations are made for the prediction of the effective elastic constants of the
RVEs with randomly distributed fibres (Table 3).
Case 1: Three RVE models are considered, i.e. RVE 1, RVE 2 and RVE 3. Axial
modulus, E1 predicted for all the three RVEs with different fibre arrangements has
percentage difference less than 1% with respect to the effective properties of RVE
Table 4 Percentage change in the effective properties of RVEs with 12 fibres with respect to RVE
model with single fibre
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
RVE 1 0.98 8.13 6.95 4.01 –2.28 –16.36 –1.31 0.67 –19.91
RVE 2 0.62 6.11 6.88 –4.44 0.72 –14.74 0.52 –0.71 –18.71
RVE 3 0.74 5.06 5.59 –7.71 –3.46 –14.24 0.44 –0.40 –15.51
with single fibre given in Table 4. The standard deviation for all the properties is
less than 1 but the percentage error is more for G 12 and ν23 , i.e. 5.8% and 10.34%,
respectively and for the other properties, it is less than 2%.
But for the transverse shear modulus, G 23 the percentage difference is much higher
compared to axial shear modulus G 12 and G 13 and it is about 15%.
Case 2: Four RVE models are considered and observations are made for the
predicted effective elastic constants. Axial modulus, E1 for all the four RVEs are
very close to the RVE with single fibre and also with the experimental value. The
standard deviation for all the properties is less than 1 (Table 5).
Axial modulus, E1 predicted for all the four RVEs with different fibre arrange-
ments has a percentage difference less than 1% as given in Table 6. But for RVE 1,
44 S. Koley et al.
Table 6 Percentage change in the effective properties of RVEs with 20 fibres with respect to RVE
model with single fibre
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
RVE 1 0.25 7.49 7.32 –6.74 –8.54 –20.39 0.40 0.63 –21.57
RVE 2 0.12 4.28 3.47 –5.68 –11.07 –14.74 –0.32 0.91 –11.30
RVE 3 0.09 4.86 4.38 –5.16 –7.27 –15.49 –0.16 0.59 –13.43
RVE 4 0.09 4.47 3.29 –4.34 –12.26 –14.39 –0.63 1.19 –11.11
Table 7 Effective properties of RVE model with 50 fibre (circular cross section)
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
(GPa) (GPa) (GPa) (GPa) (GPa) (GPa)
RVE 1 136.19 9.20 9.28 4.93 5.20 3.53 0.25 0.25 0.28
RVE 2 136.01 9.10 9.14 5.01 5.16 3.61 0.25 0.25 0.29
RVE 3 136.04 9.20 9.24 4.24 5.05 3.57 0.25 0.25 0.28
RVE 4 136.73 9.27 9.26 5.45 5.41 3.60 0.25 0.25 0.28
RVE 5 136.64 9.25 9.28 5.41 5.44 3.59 0.25 0.25 0.28
RVE 6 136.47 9.22 9.21 5.43 5.40 3.59 0.25 0.25 0.29
RVE 7 136.44 9.21 9.16 5.36 5.29 3.60 0.25 0.25 0.29
RVE 8 136.60 9.31 9.26 5.45 5.29 3.56 0.25 0.25 0.28
Average 136.39 9.22 9.23 5.17 5.28 3.59 0.25 0.25 0.28
SD 0.28 0.06 0.05 0.43 0.14 0.02 0.00 0.00 0.01
% error 0.21 0.65 0.54 8.32 2.65 0.56 0.00 0.00 3.52
the percentage difference is more compared to the other three RVE models. This is
because the number of fibres at the edges of this RVE is more.
The percentage difference for transverse shear modulus, G 23 is much more for
RVE 1, i.e 20% compared to other three RVE models. It is observed that the average
values of the elastic properties of RVE models with 20 fibres is higher comparable
to RVE with 12 fibres.
Case 3: Eight RVE models are considered, i.e. RVE 1 to RVE 8 and it is observed
that there is a reduction in effective axial modulus E1 compared with the RVEs with
12 and 20 fibres. The standard deviation is less than 1 for all the properties but the
percentage error is also less than 1% except for G 12 , G 13 and ν23 . The percentage error
for G 12 , G 13 and ν23 are 8, 2 and 3%, respectively. But if we notice the percentage
change with respect to RVE with single fibre, for the shear, i.e the axial shear modulus
G 13 and transverse shear modulus G 23 are more than 9% (Tables 7,8).
In the next study, RVEs with uncertainty in fibre cross section are considered.
Here, three cases are considered, Case 1: RVE with circular cross section, Case 2:
RVE with a distorted cross section and Case 3: RVE with an elliptical cross section.
For all the three cases, eight RVE models are generated.
A Micromechanical Study of Fibre-Reinforced Composites with Uncertainty … 45
Table 8 Percentage change in the effective properties of RVEs with 50 fibres with respect to RVE
model with single fibre
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
RVE 1 2.42 3.60 2.77 –4.93 –10.52 –15.34 –0.44 0.79 –9.03
RVE 2 2.56 4.66 4.22 –6.68 –9.66 –17.75 –0.16 0.52 –12.81
RVE 3 2.53 3.60 3.16 9.78 –7.46 –16.58 –0.16 0.52 –9.76
RVE 4 2.04 2.86 2.93 –16.01 –15.03 –17.46 0.52 0.39 –8.76
RVE 5 2.10 3.02 2.74 –15.10 –15.78 –17.19 0.36 0.52 –8.45
RVE 6 2.22 3.32 3.43 –15.57 –14.79 –17.03 0.44 0.28 –10.19
RVE 7 2.25 3.51 4.01 –14.03 –12.43 –17.31 0.59 –0.04 –11.11
RVE 8 2.13 2.40 2.96 –16.01 –12.49 –16.02 0.91 –0.24 –8.37
Table 9 Effective properties of RVE model with 50 fibre (distorted cross section)
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
(GPa) (GPa) (GPa) (GPa) (GPa) (GPa)
RVE 1 132.18 9.12 9.07 5.41 5.28 3.50 0.25 0.25 0.29
RVE 2 131.63 8.96 8.89 5.24 5.11 3.53 0.25 0.26 0.30
RVE 3 131.48 8.97 9.00 5.02 5.07 3.47 0.25 0.25 0.29
RVE 4 130.91 8.99 8.98 5.22 5.17 3.46 0.25 0.25 0.29
RVE 5 130.11 8.95 8.95 5.21 5.15 3.43 0.25 0.26 0.29
RVE 6 131.48 8.99 8.99 5.33 5.33 3.47 0.25 0.25 0.29
RVE 7 130.65 8.95 8.91 5.30 5.21 3.47 0.25 0.26 0.30
RVE 8 131.79 9.13 9.08 5.55 5.41 3.49 0.25 0.25 0.29
Average 131.28 9.01 8.99 5.29 5.22 3.48 0.25 0.25 0.29
SD 0.68 0.07 0.07 0.16 0.12 0.03 0.00 0.00 0.00
% error 0.52 0.78 0.78 3.02 2.30 0.86 0.00 0.00 0.00
RVE with circular fibre cross section has been discussed in the previous section.
Now, RVE with distorted fibre cross section are considered. Observations are made
for the distorted fibre cross section, where the effective properties for all the models
are much less than the values compared with the RVE with single fibre. This is mainly
due to reduction of fibre volume fraction (Table 9).
The standard deviation is less than 1 for all the properties but the percentage error
is also less than 1% except for G 12 and G 13 (Table 10).
From these tables, it can be seen that the values of E1 , E2 and E3 show about 7%
change. Similarly, this change for G 12 is about 18% and for G 13 and G 23 is 15%
when compared to that of RVE with single fibre. Further, there is a change in fibre
volume fraction for each RVE as they contain fibres with distorted cross section. This
distortion also affects the percentage change in the effective properties.
For the last case, RVEs are modelled with elliptical fibre cross section with
+ 0.5% of the radius of the circular fibre cross section as major axis and –0.5%
46 S. Koley et al.
Table 10 Percentage change in the effective properties of RVEs with 50 fibres with respect to RVE
model with single fibre
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
RVE 1 5.29 4.44 4.96 –15.06 –12.37 –14.09 –0.08 –0.83 –11.19
RVE 2 5.69 6.14 6.79 –11.47 –8.77 –15.27 –0.24 –1.07 –15.93
RVE 3 5.79 5.97 5.70 –6.75 –7.89 –13.38 –1.03 –0.67 –13.12
RVE 4 6.21 5.79 5.91 –10.99 –9.94 –13.03 –0.87 –0.99 –13.19
RVE 5 6.78 6.18 6.21 –10.85 –9.55 –12.06 –0.99 –1.19 –13.54
RVE 6 5.79 5.76 5.82 –13.44 –13.29 –13.26 –0.87 –0.87 –13.62
RVE 7 6.39 6.23 6.65 –12.77 –10.84 –13.33 –0.67 –1.35 –15.01
RVE 8 5.58 4.35 4.83 –18.08 –15.12 –13.86 0.12 –0.91 –11.19
of the circular fibre cross section as minor axis as shown in Fig. 5. The fibres are
distributed throughout the window with random orientations, i.e. the major axis of
the elliptical fibre cross section is placed randomly in the y–z plane.
The effective properties are shown in Table 11. The effective properties for E1 , E2
and E3 are less than the experimental values but the shear values are almost the same
or more than the experimental values. The standard deviation for all the properties
are less than 1 but the percentage change is greater than 1 for G 12 and G 13 (Table 12).
Observations are obtained for the fibres with elliptical fibre cross section, the
percentage change for effective axial modulus E1 is about 3% whereas for E2 and E3
it is about 5%. The percentage change for shear moduli is more than 18%.
6 Conclusion
Table 11 Effective properties of RVE model with 50 fibres (elliptical cross section)
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
(GPa) (GPa) (GPa) (GPa) (GPa) (GPa)
RVE 1 135.94 9.26 9.19 5.52 5.31 3.57 0.25 0.25 0.28
RVE 2 135.94 9.15 9.09 5.48 5.29 3.62 0.25 0.25 0.29
RVE 3 136.31 9.21 9.23 5.27 5.23 3.58 0.25 0.25 0.28
RVE 4 136.39 9.26 9.22 5.50 5.29 3.59 0.25 0.25 0.28
RVE 5 136.48 9.27 9.27 5.41 5.34 3.58 0.25 0.25 0.28
RVE 6 135.91 9.16 9.19 5.27 5.42 3.58 0.25 0.25 0.29
RVE 7 135.93 9.18 9.28 5.12 5.58 3.55 0.25 0.25 0.28
RVE 8 135.93 9.26 9.25 5.35 5.31 3.54 0.25 0.25 0.28
Average 136.11 9.22 9.22 5.37 5.35 3.58 0.25 0.25 0.28
SD 0.24 0.05 0.06 0.14 0.11 0.03 0.00 0.00 0.00
% error 0.18 0.54 0.65 2.61 2.06 0.84 0.00 0.00 0.00
Table 12 Percentage change in the effective properties of RVEs with 50 fibres with respect to RVE
model with single fibre
Model E1 E2 E3 G 12 G 13 G 23 ν12 ν13 ν23
RVE 1 2.60 2.91 3.69 –17.45 –12.88 –16.41 0.83 –0.48 –9.72
RVE 2 2.60 4.13 4.74 –16.59 –12.62 –18.13 0.63 –0.20 –12.92
RVE 3 2.33 3.46 3.26 –12.13 –11.17 –16.90 0.28 0.28 –9.57
RVE 4 2.28 2.89 3.41 –17.04 –12.51 –17.25 0.75 –0.04 –9.57
RVE 5 2.21 2.81 2.87 –15.11 –13.60 –16.77 0.52 0.28 –8.64
RVE 6 2.62 4.02 3.63 –12.15 –15.27 –16.86 –0.12 0.48 –10.73
RVE 7 2.61 3.79 2.78 –8.87 –18.69 –15.69 –0.59 0.99 –8.68
RVE 8 2.61 2.97 3.07 –13.72 –12.84 –15.49 0.36 –0.08 –8.60
1. For RVE with 12 fibres, the percentage error is more for G 12 and ν23 , i.e. 5.80%
and 10.34%, respectively and for the other properties, it is less than 2%.
2. In particular, the percentage change of predicated transverse shear moduli G 23
from UD composites with fibre distributed at random over transverse cross section
has higher values comparable to the experimental data. For the RVE with single
fibre, the percentage change is 21.87% with respect to experimental result and for
the other RVEs with more number of fibres, not only the transverse shear moduli
48 S. Koley et al.
has higher value but there is also increase in percentage change values for axial
shear between 6–17%.
3. The percentage change with respect to single fibre RVE, E1 is about 2% for
circular and elliptical fibre cross section whereas for distorted fibre cross section,
it is about (5–6)%. The uncertainty of fibre cross section is studied and the effective
properties are estimated. It is observed that E1 is about 136 GPa for the models
with circular fibre cross section but it is reduced to (130–131) GPa for distorted
fibre cross section which is mainly due to reduction in fibre volume fraction.
References
1. Daniel IM, Isaac O (1994) Engineering mechanics of composite materials, USA: Oxford Uni-
versity Press
2. Pyrz R (1994) Quantitative description of the microstructure of composites, part i: morphology
of unidirectional composite systems. Compos Sci Technol 50(2):197–208
3. Pyrz R (1994) Correlation of microstructure variability and local stress field in two-phase
materials. Mater. Sci. Eng.: A 177(1–2):253–259
4. Ghosh S, Nowak Z, Lee K (1997) Quantitative characterization and modeling of composite
microstructures by voronoi cells. Acta Mater 45(6):2215–2234
5. Ju J, Zhang X (1998) Micromechanics and effective transverse elastic moduli of composites
with randomly located aligned circular fibers. Int J Solids Struct 35(9–10):941–960
6. Tyrus J, Gosz M, DeSantiago E (2007) A local finite element implementation for impos-
ing periodic boundary conditions on composite micromechanical models. Int J Solids Struct
44(9):2972–2989
7. Huang Y, Jin KK, Ha SK (2008) Effects of fiber arrangement on mechanical behavior of
unidirectional composites. J Compos Mater
8. Terada K, Hori M, Kyoya T, Kikuchi N (2000) Simulation of the multi-scale convergence in
computational homogenization approaches. Int J Solids Struct 37(16):2285–2311
9. Aboudi J (2013) Mechanics of composite materials: a unified micromechanical approach,
vol 29, Elsevier
10. Hill R (1964) Theory of mechanical properties of fibre-strengthened materials: I. elastic
behaviour. J Mech Phys Solids 12(4):199–212
11. Kanit T, Forest S, Galliet I, Mounoury V, Jeulin D (2003) Determination of the size of the
representative volume element for random composites: statistical and numerical approach. Int
J Solids Struct 40(13):3647–3679
12. Soden P, Hinton M, Kaddour A (1998) Lamina properties, lay-up configurations and load-
ing conditions for a range of fibre-reinforced composite laminates. Compos Sci Technol
58(7):1011–1022
Free Vibration Analysis of Laminated
Composite Plates and Shells Subjected
to Concentrated Mass at the Centre
Abstract In this present paper, we deal with free vibration analysis of laminated
composite plates and shell panels with concentrated mass at the centre of the shell.
Two types of mass lumping schemes (RWOERI and RWERI) have been proposed
in the present study. In one of the mass lumping schemes, effect of rotary inertia
has been included. First-order shear deformation theory has been incorporated in
the formulation. The entire finite element programme has been written using FOR-
TRAN language. A nine-noded isoparametric shallow shell element has been used.
Cylindrical composite shell panels having different boundary conditions with various
concentrated masses at the centre have been analysed.
1 Introduction
Due to the curved configuration, shell structures offer better performance in carrying
load and moment for the combined bending and membrane action. The multilayered
laminated composite material contributes extra advantage due to lightweight and
tailorable properties. The analysis of laminated composite structures has attracted
several researchers since few decades. A significant number of literature are available
on free vibration of normal shell panels. However, investigation on shell panels with
A. Mandal
Narula Institute of Technology, Kolkata, India
e-mail: mandal.arpita6@gmail.com
C. Ray (B)
Department of Civil Engineering, IIEST, Shibpur, India
e-mail: chaitali@civil.iiests.ac.in
S. Haldar
Department of Aerospace Engineering & Applied Mechanics, IIEST, Shibpur, India
e-mail: salilhaldar@gmail.com
concentrated mass is very rare. Very few studies on free vibration of composite plates
with concentrated mass are reported in the available literature till date.
Bhimaraddi [1] worked on free vibration analysis of doubly curved shallow shell
panels on rectangular planform using three-dimensional elasticity theory. Boay [2]
made a study on natural frequencies of plates without and with a concentrated mass
using Rayleigh-energy method. Low et al. [3] made a comparative study of frequen-
cies for plates carrying concentrated mass. A free vibration analysis of rectangular
isotropic plate carrying a concentrated mass at different positions is analysed and
presented by Boay [4]. Low et al. [5] made a comparison between the analytical
and experimental results in terms of the frequencies for plates carrying concentrated
mass. Sheikh et al. [6] carried out a study on vibration of plates in different situations
using a high-precision shear deformable element.
It is found from the extensive literature review that most of the researchers
have paid attention towards the numerical analysis of isotropic plates with con-
centrated mass mostly. The literature review also reveals that the study on free
vibration response of shell panels with concentrated mass and, in particular, studies
on laminated composite shell panels with concentrated mass are very rare. There-
fore, the present investigation has attempted to conduct a numerical modal analy-
sis on composite plates and shell panels with concentrated mass to obtain natural
frequencies.
2 Mathematical Formulation
The finite element model has been developed in the present study and a computer
code is prepared using FORTRAN. The nine-noded laminated shell element with
five degrees of freedom (u, v, w, θ x and θ y ) at each node has been considered for the
present formulation.
The nodal displacement components at any node ‘r’ of the element can be written
as
{δr }T = u r vr wr θxr θxr (1)
where
9
9
9
9
9
u= Nr u r , v = Nr vr , w = Nr wr , θx = Nr θxr , θ y = Nr θ yr
r =1 r =1 r =1 r =1 r =1
and u, v and w are the translational displacements and θ x and θ y are the rotations
about y- and x-axis, respectively. The shape functions N r are developed by applying
Free Vibration Analysis of Laminated Composite Plates … 51
and Øx and Øy are the average shear rotation throughout the thickness of laminated
shell as shown in Fig. 1.
The constitutive relationship with respect to its reference plane may be expressed
as
{σ } = [D]{ε} (3)
{σ }T = N x N y N x y Mx M y Mx y Q x Q y (4)
where (Aij ), (Bij ), (Dij ) and (Alk ) are extensional, extension–bending coupling,
bending and transverse shear stiffness matrices, respectively, and defined as
n
n
Ai j = Q i j (z k − z k−1 ), Bi j = Q i j z k2 − z k−1
2
k=1 k=1
n
n
Di j = Q i j z k3 − Z k−1
3
, Alk = K S (Q lk )(z k − z k−1 )
k=1 k=1
where K S = 56 and zk is the distance of kth layer from the reference plane.
The stiffness matrix [K]e of an element derived from the concept of virtual work
method may be written as
+1 +1
[K ]e = ∫ ∫ [B]T [D][B]|J |dεdη (7)
−1 −1
+1 +1 h 2
T
+ Nθ y Nθ y |J |dεdη (8)
12
where
[Nu ] = [[Nr ][N0 ][N0 ][N0 ][N0 ]] [Nv ] = [[N0 ][Nr ][N0 ][N0 ][N0 ]]
[Nw ] = [[N0 ][N0 ][Nr ][N0 ][N0 ]] Nθx = [[N0 ][N0 ][N0 ][Nr ][N0 ]]
The rotary inertia in the mass matrix has significant contribution to the dynamic
behaviour of thicker laminates. However, the rotary inertia cannot be included in
the consistent mass matrix formulation. The lumped model of mass matrix has been
derived to include rotary inertia and expressed in the last two terms in Eq. (8). The
in-plane movement of mass is defined by first two terms in the mass matrix given
in Eq. (8). Transverse movement of mass contributes the major inertia and defined
by the third term. The effect of rotary inertia has been examined by developing two
types of proportionate mass lumping schemes. The effect of in-plane and transverse
movements of mass has been considered in the first lumping schemes. This lumping
scheme has been defined as RWOERI (mass lumping without rotary inertia) and may
be expressed as
m ii
m iiwl = me
m ii
(i = 1, 2, 3, 6, 7, 8, 11, 12, 13, 16, 17, 18, 21, 22, 23, 26,
27, 28, 31, 32, 33, 36, 37, 38, 41, 42, 43) (9)
where m iiwl are the ith diagonal elements corresponding to u, v and w in the proposed
model of lumped mass matrix, m ii is the ith diagonal element of the consistent mass
matrix and me is the actual mass of element.
The mass lumping scheme considering the effect of rotary inertia (RWERI) along
with in-plane and transverse movement is expressed as
m ii
m iiwl = me
m ii
(i = 1, 2, 3, 6, 7, 8, 11, 12, 13, 16, 17, 18, 21, 22, 23, 26,
27, 28, 31, 32, 33, 36, 37, 38, 41, 42, 43) (10)
h 2 m ii
m iiθ xl = m e (i = 4, 9, 14, 19, 24, 29, 34, 39, 44) (11)
12 m ii
h 2 m ii
m iiθ xl = m e (i = 5, 10, 15, 20, 25, 30, 35, 40, 45) (12)
12 m ii
The element stiffness and the mass matrices have been generated and assembled
together to form the global stiffness matrix [K 0 ] and the global mass matrix [M 0 ].
The equation of motion for the free vibration analysis is expressed as
of rotary inertia in mass matrix has been studied (RWERI) and the convergence has
also been verified.
ω
Table 2 Fundamental frequency (λ = 2π ) (in Hz) of a rectangular plate having a concentrated
mass at the plate centre
Concentrated mass (kg) Present result (RWERI) Boay [4] Sheikh et al. [6]
3 18.053 18.03 17.93
4 15.703 15.73 15.65
5 14.082 14.12 14.04
A cross-ply [0°/90°] laminated shell panel having lumped mass at the centre with
thickness ratio of a/h = 10 and different boundary conditions has been analysed
using the present formulation. The relative material properties are E 1 /E 2 = 25, G12
= G13 = 0.5E 2 , G23 = 0.2E 2 , γ 12 = γ 13 = γ 23 = 0.25. First three natural frequencies
obtained from the present formulation are shown in Table 3. Here, a and b are the
curve length of the edges of the shell panels and a = b.
56 A. Mandal et al.
4 Conclusions
It is found from the present numerical analysis that there is a significant effect of rotary
inertia on the free vibration of laminated plates and shell panels. It is recommended
that the mass lumping scheme with rotary inertia is more suitable for both plates
and shell panels. In this present investigation, both simply supported and fixed shell
panels with different R/a ratio have been taken under consideration. For all the cases,
results show the same pattern with the increasing value of applied lumped mass. The
present study reveals that the fundamental frequency of plates and shell panels with
lumped mass at the centre decreases with the increment of the applied lumped mass
value. Another observation from this study is that with the increment of lumped mass
value, the rate of change of the frequency increases the clamped shells.
References
1. Bhimaraddi A (1991) Free vibration analysis of doubly curved shallow shells on rectangular
planform using three-dimensional elasticity theory. Int J Solids Struct 27(7):897–913
2. Boay CG (1993) Free vibration on rectangular isotropic plates with and without a concentrated
mass. Comput Struct 48:529–533
Free Vibration Analysis of Laminated Composite Plates … 57
3. Low KH, Ng CK, Ong YK (1993) Comparative study of frequencies for plates carrying
concentrated mass. J Eng Mech 119:917
4. Boay CG (1995) Frequency analysis of rectangular isotropic plates carrying a concentrated mass.
Comput Struct 56(1):39–48
5. Low KH, Chai GB, TanA GS (1997) comparative study of vibrating loaded plates between the
Rayleigh-Ritz and experimental methods. J Sound Vib 199(2):285–297
6. Sheikh AH, Haldar S, Sengupta D (2002) Vibration of plates in different situations using a
high-precision shear deformable element. J Sound Vib 253(2):329–345
7. Corr RB, Jennings A (1976) A simultaneous iteration algorithm for symmetric Eigen value
problems. Int J Numer Eng 10:647–663
Buckling Analysis of Thick Isotropic
Shear Deformable Beams
Abstract This paper presents buckling analysis of thick isotropic shear deformable
beams using single variable new first-order shear deformation theory (SVNFSDT )
for beams. As in the case of Timoshenko beam theory, SVNFSDT is a displacement-
based first-order shear deformation beam theory and assumes constant transverse
shear strain through the thickness of the beam. SVNFSDT has only one unknown
function and has striking resemblance to that of Bernoulli–Euler beam theory as
far as expressions for governing differential equation, moment, and shear force are
concerned. Hence, efforts involved in obtaining buckling solutions using SVNFSDT
are only marginally higher as compared to those involved in the case of Bernoulli–
Euler beam theory. Numerical results are presented for rectangular isotropic beams
with several values of beam thickness-to-length ratio and with physically meaningful
boundary conditions. To demonstrate the efficacy of numerical results of buckling
analysis obtained using SVNFSDT, these results are compared with corresponding
results available in the literature.
1 Introduction
beams in which shear deformation effects are significant, the use of BEBT overesti-
mates the buckling loads. Hence, it is necessary to have beam theories which account
for effects of transverse shear strain which are prominent in case of thick beams. The
beam transverse shear strain is assumed to have a constant value through the beam
thickness in the case of first-order shear deformation beam theories (FSDT ). In order
to approximately satisfy the constitutive relation between the beam transverse shear
stress and the beam transverse shear strain, FSDT utilize a shear correction coef-
ficient. Timoshenko beam theory (TBT ) is the first FSDT and has two unknown
functions. On the other hand, the beam transverse shear strain is assumed to have a
polynomial variation through the beam thickness in the case of higher-order shear
deformation beam theories (HSDT ). Although FSDT and HSDT are adequate for
analyzing shear deformable beams, they generally have coupled governing differen-
tial equations, increased number of independent unknowns, and require specification
of increased number of beam boundary conditions. More details regarding FSDT and
HSDT are presented in Ghugal and Shimpi [1]. Higher accuracy offered by HSDT
comes with a price of longer computation time and effort [2]. Pakhare et al. [3] have
developed single variable new first-order shear deformation theory (SVNFSDT ) for
isotropic beams based on new first-order shear deformation theories (NFSDT ) by
Shimpi et al. [4]. Unlike TBT, SVNFSDT has single governing differential equation
involving single unknown function and requires specification of only two physically
meaningful boundary conditions at each end of the beam.
Buckling of beam is a functional failure phenomenon wherein beam loses its
ability to carry lateral loads when compressive force acting along the beam length
reaches a critical value. Beam thickness-to-length ratio and hence shear deformation
effects have direct implications on critical buckling loads wherein increase in beam
thickness-to-length ratio causes reduction in non-dimensional critical buckling load.
Thai [5] has performed stability analysis of nanobeams using nonlocal shear defor-
mation beam theory. Shimpi et al. [6] have utilized single variable HSDT to perform
stability analysis of thick beams.
In this paper, SVNFSDT is used to perform buckling analysis of thick isotropic
shear deformable beams. Numerical results obtained using SVNFSDT are compared
with existing results in order to demonstrate the effectiveness of the presented method.
2 Theoretical Formulation
In accordance with the recent work by Pakhare et al. [3] and Shimpi et al. [4] on
NFSDT, single valued displacement fields are assumed for the beam, which are on
similar lines to the displacement fields given by Shimpi el al. [4]. Lateral displacement
“w” is split into a bending component “wb ” and a shearing component “ws ”. These
two components of lateral displacement are functions of longitudinal coordinate “x”
alone. Longitudinal displacement “u” is a function of longitudinal coordinate “x”
and lateral coordinate “z”. Assumed displacement field is given below:
Buckling Analysis of Thick Isotropic Shear Deformable Beams 61
dwb
u(x, z) = −z (1)
dx
d 2 wb
Mx = −E I (3)
dx2
whereas shear force “Qx ” of the proposed theory is given below:
d 3 wb
Q x = −E I (4)
dx3
As stated by Shimpi et al. [6], conditions of gross equilibrium for the beam acted
upon by an axial tensile load “N x ” are given below:
dw d Mx
Nx + − Qx = 0 (5)
dx dx
d Qx
=0 (6)
dx
Beam gross equilibrium equations are utilized to obtain the relation between wb
and ws . Equations (3), (5), (6) along with the relation between wb and ws are utilized
to derive governing differential equation of SVNFSDT for beam under the influence
of an axial compressive load which is given below:
h 2 (1 + μ)No d 4 wb d 2 wb
EI − + N o =0 (7)
5 dx4 dx2
In Eq. (7), “μ” is Poisson’s ratio and “E” is modulus of elasticity of the beam
material, “h” is the beam thickness, “I” is the beam cross-sectional area moment of
inertia, and “N o ” is axial compressive load acting on the beam.
General solution of Eq. (7) can be given as below:
N0
β2 = (9)
h 2 (1+μ)N0
EI − 5
Depending upon the type of end supports, physically meaningful boundary con-
ditions, viz, free, clamped, and simply-supported boundary conditions can be pre-
scribed at each end of the beam in terms of the displacements. Furthermore, as in
Timoshenko and Goodier [7], it is possible to represent two distinct beam clamped
boundary conditions, viz, “clamp type 1” and “clamp type 2” in the present beam
theory.
Four linear algebraic equations in C 1 , C 2 , C 3, and C 4 can be obtained by sub-
stituting general solution given by Eq. (8) in four boundary conditions imposed on
the beam, two at each beam end. Then, critical buckling load for the beam can be
obtained by performing eigenvalue buckling analysis on these four linear algebraic
equations.
Fig. 2 Effects of “clamp type 1” and “clamp type 2” conditions on the variation of non-dimensional
critical buckling loads (N cr = (N o L 2 )/(EI)) with respect to beam thickness-to-length ratio for CS
beam
(Fig. 1). Also for any beam fixity conditions, increase in beam thickness-to-length
ratio causes N cr to decrease.
Figure 2 depicts the effects of “clamp type 1” and “clamp type 2” conditions on the
variation of N cr obtained using SVNFSDT with respect to beam thickness-to-length
ratio for CS beam.
For CS beam, N cr is higher for the beam with “clamp type 2” boundary conditions
as compared to the beam with “clamp type 1” boundary conditions for a given beam
thickness-to-length ratio (Fig. 2). Hence “clamp type 2” boundary conditions make
the beam stiffer as compared to “clamp type 1” boundary conditions. With reduction
in beam thickness-to-length ratio, effects of clamping conditions attenuate.
4 Concluding Remarks
In this paper, buckling analysis of thick isotropic shear deformable beams is per-
formed by using single variable new first-order shear deformation beam theory. For
thick isotropic rectangular beams with various beam fixity conditions, influence of
Buckling Analysis of Thick Isotropic Shear Deformable Beams 65
References
1. Ghugal YM, Shimpi RP (2001) A review of refined shear deformation theories for isotropic and
anisotropic laminated beams. J Reinf Plast Compos 20(3):255–272
2. Lee KH, Lim GT, Wang CM (2002) Thick Lévy plates re-visited. Int J Solids Struct 39:127–144
3. Pakhare KS, Mitra M, Shimpi RP (2016) Development of single variable new first-order shear
deformation theories for plates and beams. Master of Technology Dissertation, Department of
Aerospace Engineering, Indian Institute of Technology Bombay
4. Shimpi RP, Patel HG, Arya H (2007) New first-order shear deformation plate theories. ASME
J Appl Mech 74:523–533
5. Thai HT (2012) A nonlocal beam theory for bending, buckling, and vibration of nanobeams. Int
J Eng Sci 52:56–64
6. Shimpi RP, Shetty RA, Guha A (2017) A simple single variable shear deformation theory for a
rectangular beam. J Mech Eng Sci 231:4576–4591
7. Timoshenko SP, Goodier JN (1951) Theory of elasticity. McGraw-Hill Book Company, New
York, pp 35–39
Spectral Finite Element for Dynamic
Analysis of Piezoelectric Laminated
Composite Beams
Namita Nanda
1 Introduction
N. Nanda (B)
Department of Applied Mechanics, Indian Institute
of Technology Delhi, New Delhi 110 016, India
e-mail: namita_nanda@rediffmail.com
laminated composite beams. Many researchers used the conventional finite element
method to study the dynamic behavior of structures. However, the conventional FEM
requires very fine mesh to obtain reliable dynamic solutions of such structures. This
creates very large problem size which costs more computational effort and time [1].
A more efficient method is the frequency-domain spectral finite element method [2].
In this method, the discretization of the time-domain solution to frequency domain
in terms of spectral amplitudes (Fourier coefficients) is done using the fast Fourier
transform (FFT). The fundamental difference between FEM and SFEM is that the
spectral element stiffness matrix is exact and frequency dependent. The governing
equations are solved in the frequency domain, which are transformed from time-
domain equations using the discrete Fourier transform. These responses are then
transformed back to the time domain using the inverse FFT. Due to the exact for-
mulation of the system, only one spectral element is sufficient for solution in most
cases where thousands of finite elements are required.
The frequency-domain SFEM considered in the present study is different from
the time-domain SFEM proposed by Patera [3]. In time-domain SFEM, Legendre
polynomials or Chebyshev polynomials are used as shape functions to formulate the
finite element matrices in the time domain. The element mass matrix formulated is
diagonal hence the computational cost is less expensive than conventional FEM. This
method has been used for wave propagation analysis in laminated composite beams,
plates, and panels [4–7]. Though the method is more robust and has high accuracy,
the problem size remains large. The time-domain SFEM analysis is inadequate to
furnish most of the useful frequency domain properties such as wave dispersion and
nature of each wave modes.
The frequency-domain SFEM proposed by Doyle [2] has been widely used by
researchers to study wave propagation in structures. Wave propagation analysis in
isotropic beams has been studied [8–11] using this frequency-domain SFEM. Lee
et al. [11] presented a brief review of the spectral element method in structural
dynamics. Spectral element models for wave propagation and dynamic analysis of
composite beams have been developed [12–15]. A spectral finite element based on
an efficient layerwise theory has been developed by Nanda et al. [16] for wave
propagation analysis of anisotropic beams. Wavelet spectral finite element method
has also been used to study wave propagation in laminated composite plates with
transverse cracks [17].
Many theories and models have been developed for analyzing piezoelectric lam-
inated composite structures [18–21] with different displacement field and electric
potential approximations. However, the studies on smart piezoelectric composite
structures using SFEM are limited [22, 23] and have not been well addressed. More-
over, these spectral element formulations used linear variation of the electric potential
through the thickness. However, the electric potential distribution is actually nonlin-
ear due to the induced potential effects. In the present work, a spectral finite element
model in the frequency domain is presented for the free vibration and high-frequency
wave propagation analyses of piezoelectric laminated composite beams. The dis-
placement field of the beam is represented by first-order shear deformation theory
Spectral Finite Element for Dynamic Analysis … 69
with both linear and layerwise through thickness distribution of electric potential
[21] in the piezoelectric layer.
where σx , τzx , εx , γzx , Dz , and E z denote the axial stress, shear stress, normal
strain, shear strain, electric displacement, and electric field, respectively. Q̂ 11 , Q̂ 55 ,
ê31 , and η33 are the reduced elastic, piezoelectric, and dielectric coefficients. The
displacement field of the beam is represented by the first-order shear deformation
theory. The in-plane and transverse displacements u and w are represented as
∂u ∂u ∂w
εx = , γzx = + (3)
∂x ∂z ∂x
The electric potential (φ) for the piezoelectric layer (Fig. 1) is expressed in two
ways:
(i) Linear distribution through the thickness.
The electric potential and the electric field of the ith piezoelectric layer is expressed
as
Piezoelectric layer
where φ̄i = (φi + φi−1 )/2, φ̃i = φi − φi−1 , · · · z̄ i = (z i + z i−1 )/2 and φi , φi−1 are
the electric potential at the top and bottom faces of the ith piezoelectric layer and ψ0
is the first derivative of ψ0 . From Eq. (5a), it is observed that the electric potential
is quadratic in thickness direction. The first two terms describe the conventional
linear part. The quadratic term represents the bending deformation contribution to
the potential.
The resulting electric field is
E zi = −∂φi /∂z = −(φ̃i (x)/ h i ) − (ê31 (z − z̄ i )ψ0 /η33 ) (5b)
The spectral elements developed using the above representations for φ using
Eqs. (4) and (5a, 5b) are denoted as FSDT-A and FSDT-M, respectively.
The equations of motion and the corresponding variationally consistent bound-
ary conditions for the beam with piezoelectric layers are derived from Hamilton’s
principle, which states that
t2
(δT − δU + δW )dt = 0 (6)
t1
where t 1 and t 2 are the starting and finish time, respectively; δT and δU are the
first-order variations of total kinetic and total strain energies of the piezoelectric
composite beam, respectively, and δW denotes the total virtual work done by the
external mechanical and electrical forces. These variations are obtained as
δT = ρ(u̇δ u̇ + ẇδ ẇ)bdz (7)
z
δU = (σx ∂εx + τzx ∂γzx − Dz ∂ E z )bdz (8)
z
The variation of the virtual work done by external surface force q and the applied
surface charge density q0 is expressed as
Spectral Finite Element for Dynamic Analysis … 71
δW = q∂w0 bdx − q0 ∂φbdx (9)
Substituting Eqs. (7–9) into Eq. (6) and using Eqs. (1–3), the governing differential
equations of the smart composite beams are expressed as
∂φ : G kz + q0 = 0 (10d)
L
where F 2 is the applied distributed force along the z-direction and G kz = Dz h1p dz.
k=1
The coefficient G kz is obtained for FSDT-A and FSDT-M using the expressions for Dz
and E z from Eqs. (1), (4), and (5a, 5b), and subsequently the electric-field potential
function in terms of primary displacement variables is found out from Eq. (10d).
The stress resultants are defined as
L z k
L z k
L z k
Nx = σx dz , Mx = σx zdz , Qx = τzx dz (11)
k=1 z k=1 z k=1 z
k−1 k−1 k−1
Thus,
⎧ ⎫ ⎡ ⎤⎧ ⎫ ⎧ p⎫
⎨ Nx ⎬ A11 B11 0 ⎨ ∂u 0 /∂ x ⎬ ⎨ N x ⎬
M = ⎣ B11 D11 0 ⎦ ∂ψ0 /∂ x − Mp (12)
⎩ x⎭ ⎩ ⎭ ⎩ x ⎭
Qx 0 0 A55 ψ0 + ∂w0 /∂ x 0
p p zk
where N x , Mx = ê31 E z (1, z)dz
z k−1
Equation (9) is subjected to the boundary conditions at x = 0, and x = l as given
below:
u 0 = u ∗0 , or N x = N x∗
w0 = w0∗ , or Q x = Q ∗x
ψ0 = ψ0∗ , or Mx = Mx∗ (13)
The general solution of the homogeneous form of Eqs. (10a, 10b, 10c, 10d) is
written in the spectral form as
72 N. Nanda
N
N
u 0 (x, t) = û(x, ωn )eiωn t = (ũ j e−ik j x )eiωn t
n=1 n=1
N
N
w0 (x, t) = ŵ(x, ωn )eiωn t = (w̃ j e−ik j x )eiωn t
n=1 n=1
N
N
ψ0 (x, t) = ψ̂(x, ωn )eiωn t = (ψ̃ j e−ik j x )eiωn t (14)
n=1 n=1
where û, ŵ, and ψ̂ are the Fourier coefficients (or spectral components) of the
axial, flexural, and shear deformations, respectively, ũ j , w̃ j , and ψ̃ j are the wave
amplitudes, which are evaluated from the essential and natural boundary conditions,
ωn is the discrete frequency at the nth sampling point, and k j is the wavenumber.
The summation is performed up to the Nyquist frequency in FFT ω N , where N is the
number of samples.
Substituting Eq. (14) into the governing differential Eqs. (10a, 10b, 10c, 10d), we
get
⎡ ⎤⎧ ⎫
( Ã11 k 2 − I0 ωn2 ) 0 ( B̃11 k 2 − I1 ωn2 ) ⎨ ũ j ⎬
⎣ 0 ( Ã55 k 2 − I0 ωn2 ) ik A55 ⎦ w̃ j = 0 (15)
⎩ ⎭
( B̃11 k 2 − I1 ωn2 ) −ik A55 ( D̃11 k 2 + A55 − I2 ωn2 ) ψ̃ j
where Ã11 , B̃11 , and D̃11 are the modified stiffness coefficients due to ê31 and η33
terms. Equation (15) will have a nontrivial solution by setting the determinant of the
coefficient matrix of ũ j , w̃ j , and ψ̃ j to zero. This gives a sixth-order characteristic
equation in terms of the wavenumber k j ,
aw k 6j + bw k 4j + cw k 2j + dw = 0 (16)
The coefficients aw , bw , cw , and dw are not listed here due to their lengthy appear-
ance. Since Eq. (16) is a cubic equation in k 2j , there are three pairs of wavenumbers,
±k1 , ±k2 , and ±k3 : three for forward modes and three for backward modes. These
three modes constitute the axial mode (mode 1), flexural mode (mode 2), and the
shear mode (mode 3).
The solution for the spectral components of the displacement field (Eq. 14) is
obtained as
⎧ ⎫
⎪
⎪ ũ 1 e−ik1 x ⎪ ⎪
⎪ ⎪
⎧ ⎫ ⎡ ⎤⎪⎪
⎪ ũ e −ik1 (l−x) ⎪
⎪
⎪
⎨ û(x, ωn ) ⎬ R11 R12 R13 R14 R15 R16 ⎨ ⎪ 2
−ik2 x ⎬
⎪
ũ e
ŵ(x, ωn ) = ⎣ R21 R22 R23 R24 R25 R26 ⎦ 3
(17)
⎩ ⎭ ⎪
⎪ ũ 4 e−ik2 (l−x) ⎪⎪
ψ̂(x, ωn ) R31 R32 R33 R34 R35 R36 ⎪ ⎪ ⎪
⎪
⎪
⎪ ũ 5 e−ik3 x ⎪ ⎪
⎪
⎪
⎩ ⎭
ũ 6 e−ik3 (l−x)
Spectral Finite Element for Dynamic Analysis … 73
where R1 j , R2 j , and R3 j are the amplitude ratios for the three modes of propagation.
At nodes 1 (x = 0) and 2 (x = l), the spectral amplitudes of the nodal displacements
û 1 = û(0, ωn ), ŵ1 = ŵ(0, ωn ), ψ̂1 = ψ̂(0, ωn ), û 2 = û(l, ωn ), ŵ2 = ŵ(l, ωn ), and
ψ̂2 = ψ̂(l, ωn ) of the element of length l are expressed as
where
T T
ûe = û 1 ŵ1 ψ̂1 û 2 ŵ2 ψ̂2 and ũ = ũ 1 ũ 2 ũ 3 ũ 4 ũ 5 ũ 6 .
Here K̂(ωn ) is the exact dynamic stiffness matrix of the piezoelectric laminated
beam.
The SFEM developed using Eqs. (4) and (5a, 5b) for φ (FSDT-A and FSDT-M,
respectively) are employed to perform dispersion, free vibration, and wave propa-
gation analyses of smart composite beams. The results for natural frequencies of a
piezoelectric cantilever beam using the present SFEM are compared with the FEM
74 N. Nanda
results [21]. The beam has length L = 100 mm and thickness h = 5 mm. The material
properties are for steel: E 1 = E 2 = E 3 = 210 GPa, G12 = G23 = G13 = 80.77 GPa, ν 12
= ν 13 = ν 23 = 0.3, ρ = 7850 kg/m3 and for PZT G1195 N: E 1 = E 2 = E 3 = 63 GPa,
G12 = G23 = G13 = 24.2 GPa, ν 12 = ν 13 = ν 23 = 0.3, d 31 = d 32 = 254× 10−12 m/V,
η11 = η22 = 1.53 × 10−8 , η33 = 1.5 × 10−8 F/m, ρ = 7600 kg/m3 . The performance
of both the FSDT-based spectral elements is evaluated over a wide range of thickness
ratios (h p / h), which is the ratio of the thickness of the piezoelectric layer to the total
beam thickness. Tables 1 and 2 present the natural frequencies in Hertz for different
thickness ratios for both open and closed-circuit electrical boundary conditions. The
closed-circuit condition is obtained by grounding both the lower and upper surfaces
of piezoelectric layer. In open-circuit condition, only the lower surface of the piezo-
electric layer is grounded. The comparison shows good agreement with the published
results. The free vibration results reveal that the natural frequencies predicted by the
conventional FSDT-A show significant deviation from those of the coupled FSDT-M
when the thickness ratio is more. This may be because of the predominant induced
potential effect. The accuracy of the coupled FSDT-M based SFEM is established
from the close agreement of the results with the ANSYS 2D solutions.
To study the effect of material properties of PZT layers, a cantilever composite
beam (L = 600 mm, h = 30 mm) with a surface-bonded piezoelectric layer of PZT
5A and G1195 N is considered (Fig. 2). The material properties of graphite–epoxy
composite [0/90/90/0] and piezoelectric layers are
Composite: E 1 = 181 GPa, E 2 = E 3 = 10.3 GPa, G12 = G13 = 7.17 GPa, G23 =
2.87 GPa, ν 12 = ν 13 = 0.28, ν 23 = 0.33, ρ = 1578 kg/m3.
PZT 5A : E 1 = E 2 = 61 GPa, E 3 = 53.2 GPa, G12 = 22.6 GPa, G23 = G13 = 21.1
GPa, ν 12 = 0.35, ν 13 = ν 23 = 0.38, d 31 = d 32 = –171×10−12 m/V, d 33 = 374×10−12
m/V, d 15 = d 24 = 584×10−12 m/V, η11 = η22 = 1.53 × 10−8 F/m, η33 = 1.5 × 10−8
F/m, ρ = 7600 kg/m3.
Table 3 compares the first four natural frequencies of cantilever composite beam
(hp /h = 0.1, 0.5, 1.0) with two different piezoelectric layers, viz., PZT-5A and
G1195 N at the top using both FSDT-A- and FSDT-M-based SFEM. The study
of natural frequencies in Table 3 shows that, with increase of thickness ratio, FSDT-
A results deviate significantly from those of the coupled FSDT-M for both PZT-5A
and G1195 N. These deviations become more significant in case of a beam with
PZT G1195 N layer than those for the beam with PZT-5A layer at higher thickness
ratios, i.e., 0.5 and 1.0. This is because the modulus induced in PZT G1195 N is
more than PZT-5A. At higher modes, the differences in frequencies obtained using
FSDT-M and FSDT-A spectral elements are noteworthy. Coupled FSDT-M-based
SFEM is used to study free vibration, dispersion, and wave propagation analysis
of cantilever composite beams. A four-layered cantilever composite beam with two
PZT-5A patches at the top as shown in Fig. 3 is considered. The geometric properties
of the beam are L 1 = 200 mm, L 2 = 150 mm, L 3 = 200 mm, L p = 25 mm, hc =
10 mm, and hp = 1 mm. Table 4 compares the first four natural frequencies of the
beam for three lamination sequences [0]4 , [0/90/90/0] and [0/90/0/90].
Next, the dispersion relation, i.e., the variations of the nondimensional wavenum-
ber (k i h) with the frequency is studied. For this, a cantilever steel beam with a
piezoelectric layer of G1195 N at the top is considered. The material and geometric
76 N. Nanda
Table 3 Comparison of the first four natural frequencies (Hz) of cantilever composite beam
[0/90/90/0] with a piezoelectric layer at the top
Thickness Mode no. PZT-5A PZT G1195 N
ratio FSDT-A FSDT-M FSDT-A FSDT-M
0.1 1 104.37 104.37 106.20 106.20
2 614.62 614.62 623.78 623.78
3 1577.1 1577.1 1603.4 1603.4
4 2706.3 2706.3 2729.5 2729.5
0.5 1 57.983 58.594 60.425 61.035
2 355.84 357.67 370.48 374.76
3 963.75 966.80 1004.6 1015.0
4 1738.3 1738.9 1813.9 1815.8
1.0 1 37.842 39.673 38.452 43.335
2 236.21 245.36 239.87 269.77
3 649.41 674.44 661.62 741.58
4 1242.7 1289.1 1267.7 1416.0
properties are the same as described in the previous problem. Figure 4 shows the dis-
persion relations for the asymmetric beam with and without a surface-bonded PZT
layer using the FSDT-M. It is noticed that the wavenumbers corresponding to axial
(k 1 h), flexural (k 2 h), and shear rotation (k 3 h) modes with piezoelectric layer being
more than that of the beam without the piezoelectric layer. Moreover, the wavenum-
bers of all these modes increase with the increase of thickness of the piezoelectric
layer. The cutoff frequency, i.e., the frequency at which the evanescent wave mode
(imaginary wavenumbers) changes to propagating wave mode (real wavenumbers),
decreases with an increase of thickness ratio.
The FSDT-M-based spectral element is then used for wave propagation in a smart
composite beam subjected to modulated tone burst excitations. A graphite–epoxy
Spectral Finite Element for Dynamic Analysis … 77
Non-dimensional wavenumber
at the top thickness ratio = 0.1 k2h
6 thickness ratio = 0.5
4
k1h
2
Real
0
k3h
Imag
-2
0 100 200 300 400 500
Frquency (kHz)
composite [0/90/90/0] beam with PZT-5A actuator and sensor (Fig. 3) is considered.
The Lamb wave is generated using five-cycle sinusoidal tone burst modulated by
Hanning window with center frequency of 100 kHz (Fig. 5). The wave signal of
100 kHz frequency with peak voltage of 200 is applied to the PZT actuator and the
response measured at the PZT sensor is shown in Fig. 6. The geometric properties
of the beam for this example are L 1 = 300 mm, L 2 = 200 mm, L 3 = 300 mm, L p =
5 mm, hc = 5 mm, and hp = 0.5 mm. The first peak in Fig. 6 is the symmetric S 0 mode
and the second peak is the antisymmetric A0 mode. Symmetric modes are described
as extensional or axial modes whereas the antisymmetric modes are described as the
flexural modes. The group speed (C g = Re [dωn /dk j ]) for the S 0 and A0 modes at
100 kHz frequency are obtained as 7785.5 m/s and 1702 m/s. The time of arrival of
the S 0 and A0 modes at the sensor found from group speed relation (L 2 /C g ) are 26
µs and 118 µs, respectively, which are in good agreement with the times of flight of
S 0 and A0 modes in the wave propagation response (Fig. 6).
200 0.3
Load (V)
100 0.0
0 50 100 150 200 250 300
Frequency (kHz)
0
-100
-200
0 50 100 150 200
-6
Time (s) ×10
78 N. Nanda
Normalized Voltage
0.5 A0 mode
S0 mode
0.0
-0.5
-1.0
0 50 100 150 200
-6
Time (s) × 10
The axial and flexural velocity responses of the same symmetric cross-ply
[0/90/90/0] cantilever beam (Fig. 3) due to axial and transverse excitation of 100 kHz
frequency with a peak voltage of 200 applied to the PZT actuator are investigated.
The beam has L 1 = 300 mm, L 2 = 200 mm, L 3 = 500 mm, L p = 5 mm, hc =
5 mm, and hp = 0.5 mm. The response at a distance of 200 mm from the actuator is
plotted. Figures 7 and 8 are the axial and flexural velocity response of the cantilever
beam obtained using the developed FSDT-M-based SFEM. The group speed C g for
the axial and flexural modes at 100 kHz frequency are obtained as 7785.5 m/s and
1682 m/s, respectively. The time of arrival of the first peak of the excitation cor-
responding to axial mode from group speed relation is 26 µs (= 0.2/7785.5). The
second and third peaks correspond to the reflection from the left and right boundaries
and the time of arrival is obtained as 103 µs and 155 µs, respectively. These values
of the time of arrival of axial mode are in perfect agreement with the times of flight
in the wave propagation response (Fig. 7). Similarly, the time of flight of the flexural
mode in the wave propagation response in Fig. 8 is matching well with the time of
arrival obtained from group speed (= 0.2/1682).
0.0
-0.5
-1.0
0 50 100 150 200
-6
Time (s) ×10
Spectral Finite Element for Dynamic Analysis … 79
w/wmax
0.0
-0.5
-1.0
0 50 100 150 200 250 300
-6
Time (s) ×10
4 Conclusions
This paper presents the frequency-domain spectral finite element method for disper-
sion, free vibration, and wave propagation analyses of piezoelectric composite beams
based on the first-order shear deformation theory. The electric potential distribution
through the thickness of the piezoelectric layer is expressed in two ways, i.e., linear
and coupled layerwise through thickness potential distribution consistent with the
FSDT. The results for natural frequencies for a piezoelectric cantilever beam using
the SFEM developed here are compared with the FEM results. The SFEM results
using FSDT-A and FSDT-M show large deviation from each other when the piezo-
electric layer thickness is increased. This shows the predominant induced potential
effect when the thickness ratio is increased. The comparison proved the superiority of
the spectral finite element formulation based on coupled layerwise through thickness
potential distribution over the conventional linear potential distribution in the piezo-
electric layer. Furthermore, the developed spectral element has been used to study
the free vibration in composite beams fully and partially covered with PZT layer
at the top. The dispersion studies show that at a given frequency, the wavenumbers
corresponding to axial, flexural, and shear rotation modes with piezoelectric layer
are more than that of the beam without the piezoelectric layer. The cutoff frequency
decreases with the increase of thickness ratio. A detailed analysis of wave propaga-
tion of composite beams with PZT actuator and sensors is also provided. The times of
flight of axial and flexural modes in the wave propagation response are corroborated
from the group speeds.
Acknowledgements The author gratefully acknowledges the financial assistance by the Science
and Engineering Research Board, Department of Science and Technology, New Delhi, under Start-
up grant for Young Scientists.
80 N. Nanda
References
1 Introduction
mately one laminate thickness from the free edge of width”. It is often known as the
“boundary layer effect” or “free edge effect”. It was also showed that the mismatch of
shear coupling coefficients between the two adjacent plies is responsible for soaring
in an interlaminar shear stress in the angle-ply laminate. Later, several researchers
have used these results as a reference result for validation. Further, Pagano [18]
solved for cross-ply [0/90]s laminate by using the analytical solution. He showed
that the interlaminar normal stress σ33 is increasing near the free edge and it can
cause delamination for cross-ply laminate. Similarly, Wang et al. [2] also used to
solve cross-ply [0/90]s and [90/0]s laminates by using finite element method. In the
same paper, they used [90/0/45/ − 45]s laminate which is a combination of angle-
ply and cross-ply laminate for understanding the behavior of σ33 near free edges. For
same geometric and material properties, they observed that the stacking sequence
[45/ − 45/0/90]s and [90/0/45/ − 45]s gives tension and compression through the
thickness, respectively, for the axial extension. Further, through the experiments, they
observed that the laminate [45/ − 45/0/90]s stacking sequence undergoes delami-
nation in the midplane, while laminate [90/0/45/ − 45]s stacking sequence did not
delaminate. Next, Makeev et al. [19] developed an iterative method for the one-term
approximate solution of partial differential equations. This approach can be applied
to several boundary value problems.
From literature, it is observed that some of the researchers solved angle-ply lam-
inate ([45/ − 45]s ), some of them used cross-ply laminate ([90/0]s or [0/90]s ), and
some of them solved quasi-isotropic ([90/0/45/ − 45]s ) laminate. Further, some
of the researchers used analytical solution and some of them used finite element
method while some of them used finite difference method. However, which method
is a best fit for the assessment of a given mechanical problem including free edge
effects in the laminate structures cannot be answered universally and priorly. It fur-
ther appears through literature that the results obtained from various approaches have
shown some of the similarity, but in some cases, discrepancies were also observed.
These discrepancies do exist in sign as well as in the magnitude of the interlam-
inar stresses. As a requirement, an analysis method should be in good agreement
between the possible correctness and computational expense. There are very few
methods available which gives closed-form solution by using recovery relation to
get interlaminar stress components and satisfy the above requirements. Nonetheless,
there are challenging opportunities, for predicting the interlaminar stress components
accurately and efficiently.
Following the above motive, a new analytical methodology is used by the author.
The present method gives a closed-form solution by using recovery relation. The
proposed approach is a combination of the VAM and the approximation method. We
know that VAM is the powerful mathematical tool to address the linear as well as
nonlinear challenges in beam modeling. Berdichevskii [20] was the first researcher
who used VAM for modeling shell structure, by simplifying the original 3-D problem
to a 1-D problem by sustaining strain energy. The final results from this approach
84 S. B. Salunkhe and P. J. Guruprasad
include linear as well as nonlinear stiffness terms. Further, this theory was developed
and much work has been done by Hodges and his co-workers [21–24]. Next, the
approximation approach used in this paper is based on an assumed stress distribution
derived from more sophisticated solution and on the equilibrium equations of elas-
ticity. This method involves the use of polynomial, and it is an extension of Pagano
and Pipes approach. Further, the precision of the method is examined by correlating
the present results, wherever possible, with the data available in the literature and the
3-D FEM. Finally, the method is used to study the thickness-wise 3-D stress behavior
of pretwisted, antisymmetric, symmetric, quasi-isotropic, and crossed ply laminated
strip. In particular, the proposed approach can provide a correct 3-D elasticity solu-
tion for any laminated composite strip through 1-D analysis in a computationally
efficient manner.
2 Mathematical Formulation
In the first part of this section, VAM is employed to develop reduced-order 1D strip-
like model. In the latter part of the section, the approximation method is employed
to obtain out of plane stresses.
δt
O δh
x 2 κ2
11 = γ11 − x2 κ3 + x3 κ2 + k1 x22 κ1 + 2 1 + w3(1) κ2 +O δb , δt , δh
2
2
O()
2
O
δh2
2
O
δh2
(1) (1) 1 (1) 2
22 = w2,2 − x3 w3,22+ w3,2 +O(δh ) (1)
2
O()
δt
O δh
212 = w1(1) − 2x3 κ1 + k1 x2 w3,2
(1)
− w3(1) + κ1 x2 w3,2
(1)
− w3(1)
O()
2
O
δh2
+O δb , δh , δt , δh
2
The 2D strain measures can be extracted from the 3D strain measures by using
the relation αβ = αβ + x3 ραβ , where αβ are the middle surface membrane strains
and ραβ are the middle surface. The expressions for membrane strains, which are
obtained by inspection of Eq. (1) are bending curvatures.
86 S. B. Salunkhe and P. J. Guruprasad
x22 κ21
11 ≈ γ11 − x2 κ3 + k1 x22 κ1 + 2
+ w3 κ2
2
22 ≈ w2,2 + 21 w3,2
2 12 ≈ w1,2 + k1 x2 w3,2 − w3 + κ1 x2 w3,2 − w3
To receive the zeroth-order 2D strain energy density through VAM, use the non-
underlined terms from the above equation in energy functional and integrate over the
thickness. Thus, one can get zeroth-order approximation. Similarly, to get first-order
approximation consider the complete above equation, i.e., admit underlined term.
The 2D strain energy density can be expressed as
⎧ ⎫
⎪
⎪ 11 ⎪
⎪
⎪ 22 ⎪
⎪ ⎪
⎪⎪ ⎪
⎪
1
A B ⎨ 212 ⎬
U2D = 11 22 212 ρ11 ρ22 2ρ12 . . (3)
2 B D ⎪ ⎪ ρ11 ⎪
⎪
⎪ ρ22 ⎪
⎪ ⎪
⎪
⎪ ⎪
⎪
⎩ ⎭
2ρ12
where the A, B, and D terms in the strain energy represents membrane, bending and
coupling stiffness matrices, respectively. Following the procedure of minimization
by using variational minimization principle and after integrating U2D along the width
direction, the 1-D strain energy, U1D , can be expressed in terms of the linear and
nonlinear 1-D strain measures as,
1 T 1
U1D = L [SL ] L + TL [SL N ] N + TN [S N ] N (4)
2 2
where [SL ] is linear, [SL N ] linear–nonlinear, and [S N ] nonlinear stiffness matrix. The
terms of these stiffnesses are given in Appendices and L is linear and N nonlinear
strain measures which are defined as follows:
L = {γ11 , κ1 , κ2 , κ3 }T
N = {κ21 , κ22 , κ2 γ11 , κ2 κ3 , κ2 κ1 }T
VAM is used to develop reduced-order 1-D strip-like model. Further, based on the
theory, a finite element formulation is generated for the structural analysis, where
the linear and nonlinear cross-sectional stiffnesses derived from VAM serves as the
input. After solving the differential equations by finite element method, we have a
solution for 1-D displacement. From this, one can obtain the 1-D generalized strain
measures through the nonlinear beam constitutive law. Knowing the 1-D strains and
curvatures from which one can obtain the warping solutions. Further, 2-D strain field
can be obtained using the already established kinematics. Now, the plane stress field
can be obtained using the plane stress reduced transformed stiffness matrix which
depends on the ply orientation. Expressions for the in-plane stress components are
given below:
The warping terms wi , wi,2 , w3,22 used above are for healthy case available in paper
of [24]. The superscript 0 and I above the warping terms represents the zeroth and
the first-order approximation, respectively. For a case of in-plane loading, however,
the present theory produces only in-plane stress components. These values from the
variational asymptotic method are shown in Fig. 2. Results are compared with the
3-D finite element result. It shows that results are accurate for interior regions
removed from free edges. However, discrepancies at the free edge are due to the
following issues:
• In strain energy, the strain terms are restricted to the zeroth and the first-order
approximation.
• Procedure of the dimensional reduction adopted in this work is based on the 2-D
strain energy functional as a starting point. It will be reduced to 1-D strain energy
functional through VAM. It makes ad hoc assumption of the line originally normal
to the midplane remains normal even after deformation.
(a) (b)
Fig. 2 a Variation of (a) τ12 and b σ22 in the interface along the width of laminate and comparison
between the results of present approach by using VAM and the 3-D finite element method
Determination of Interlaminar Stress Components … 89
These issues can be resolved by considering the 3-D strain energy functional as a
starting point and through asymptotically correct dimensional reduction procedure
using VAM. By this process, a 3-D problem is reduced to a 1-D beam problem and
the constitutive law associated with the warping fields is obtained. For primary study,
the goal was set to develop an analytical solution based on 1-D, which can produce
accurate 3-D elasticity solution with less computational time. By using present VAM
approach, it is not possible to get interlaminar shear stress components. To estimate
these components near the free edges, approximation method which is based on
global 3-D equilibrium equations is used, which is mentioned in next subsection.
This method is based on assumed stress distribution from more sophisticated solu-
tions and equilibrium equation of elasticity, and it is an extension of the approach of
Pagano and Pipes [29]. In this method involve the use of polynomial for one lami-
nate thickness from free edge, and constants of the polynomials are determined from
interface continuity or by traction free conditions. A midplane coordinate system is
considered at the free edge. Due to symmetry and antisymmetry of the stress com-
ponents, only half portion of the width of laminate is considered. Now, we assume
[29] the shear stress τ12 has following distribution along the width:
τ12
k
= A1 x2 + A2 x22 σ0 Q k0 , 0 ≤ x2 ≤ h
(8)
τ12
k
= σ0 Q k0 , h ≤ x2 ≤ b
In Eq. 8, σ0 Q k0 denotes the value of τ12 in the kth ply as determined from the VAM-
based 1-D theory. Here, k represents the layer number; A1 and A2 represent constants.
To solve the stress components in free edge boundary layer, we use the equilibrium
equation without considering the body forces.
∂σ11
k
∂τ k ∂τ k
+ 12 + 13 = 0 (9)
∂x1 ∂x2 ∂x3
For uniaxial loading, the axial stress is independent of x1 , then equilibrium equation
becomes
∂τ12
k
∂τ k
+ 13 = 0 (10)
∂x2 ∂x3
∂τ13
k
= −σ0 (A1 + 2 A2 x2 ) Q k0 (11)
∂x3
90 S. B. Salunkhe and P. J. Guruprasad
After integration
τ13
k
= −σ0 (A1 + 2 A2 x2 ) h Q k1 + x3 Q k0 (12)
since τ12 is constant within each ply. The value of h Q k1 is determined by the require-
ment that τ13 vanish on the top or bottom of the surface. By following the boundary
conditions, the value of constants A1 and A2 are determined.
τ12
k
(h, x3 ) = σ0 Q k0
(13)
τ13 (h, x3 ) = 0
k
The value of constants are A1 = 2/ h and A2 = −1/ h 2 . Now the complete equation
can be written is as
x2 x2
τ12
k
= 2− σ0 Q k0 , 0 ≤ x2 ≤ h
h h (14)
τ12 = σ0 Q 0 ,
k k
h ≤ x2 ≤ b
x
2
τ13
k
= − 1 13
k
, 0 ≤ x2 ≤ h
h (15)
τ13
k
= 0, h ≤ x2 ≤ b
σ22
k
= σ0 P0k F(x2 ), 0 ≤ x2 ≤ h
(17)
σ22 = σ0 P0k ,
k
h ≤ x2 ≤ b
where σ0 P0k is determined from the VAM-based 1-D theory, and F is variable along
the free edge. To solve these stresses in the free edge boundary layer, we use the
second equilibrium equation.
∂τ12
k
∂σ22
k
∂τ k
+ + 23 = 0 (18)
∂x1 ∂x2 ∂x3
∂σ22
k
∂τ k
+ 23 = 0 (19)
∂x2 ∂x3
∂τ23
k
dF
= −σ0 P0k (20)
∂x3 d x2
Determination of Interlaminar Stress Components … 91
dF
τ23
k
= −σ0 P1k h + x3 P0k (21)
d x2
In Eq. 21, value of P1k h is determined by interface continuity conditions or the require-
ment that τ23 vanish on the laminate surface.
The third equilibrium equation is as
∂τ13
k
∂τ k ∂σ33
k
+ 23 + =0 (22)
∂x1 ∂x2 ∂x3
∂τ23
k
∂σ33
k
+ = 0 (23)
∂x2 ∂x3
∂σ33
k
d2 F
= σ0 (P1k h + x3 P0k ) 2 (24)
∂x3 d x2
The value of P2k h 2 is determined by the requirements that σ33 vanish on the top or
bottom of the surfaces. Similarly, we can find σ33
x2 k h
σ33
k
= 1 − 3(1 + C) σ̃ , 0 ≤ x2 ≤
h 33 3
h (26)
σ33
k
= −C σ̃33
k
, ≤ x2 ≤ h
3
σ33
k
= 0, h ≤ x2 ≤ b
and C and K are constants. The σ33 produces the moment and the stress distribution
of Eq. 26 must be self-equilibrating,
h
0 σ33
k
d x2 = 0 (28)
92 S. B. Salunkhe and P. J. Guruprasad
Substituting Eq. 26 into this integral yields and performing this integration gives the
value of C = 15 and Eq. 26 becomes
1 x2 k h
σ33
k
= 5 − 18 σ̃33 , 0 ≤ x2 ≤
5 h 3
σ̃33
k
h (29)
σ33
k
= , ≤ x2 ≤ h
5 3
σ33
k
= 0, h ≤ x2 ≤ b
d2 F K x2 h
= 5 − 18 0 ≤ x2 ≤
d x22 5h 2 h 3
(30)
d2 F −K h
2
= ≤ x2 ≤ h
d x2 5h 2 3
Integration yields
dF K x2 x2 h
= 5 − 9 + K a0 0 ≤ x2 ≤
d x2 5h 2 h 3
(31)
dF −K x2 h
= + K a1 ≤ x2 ≤ h
d x2 5h 2 3
Since τ23
k
is a free edge stress, it must vanish at x2 = 0. Thus, a0 = 0. The constant
a1 is determined from continuity at x2 = h/3, with the result a1 = 5h 1
and Eq. 32
becomes
36 9x2 k x2 h
τ23
k
= 5− 23 , 0 ≤ x2 ≤
25 h h 3
36 x2 k h (33)
τ23 =
k
1− 23 , ≤ x2 ≤ h
25 h 3
τ23
k
= 0, h ≤ x2 ≤ b
Determination of Interlaminar Stress Components … 93
Since σ22
k
is a free edge stress, it must vanish at x2 = 0. Thus, b0 = 0. The constants
b1 is determined from continuity condition at y = h3 , with the result
1
b1 = (37)
45
Now, K is determined from continuity condition at y = h, with the result
90
K = (38)
7
Now, Eq. 19 becomes
9 x2 k x22 h
σ22
k
= 5−6 P0 2 , 0 ≤ x2 ≤
7 2h h 3
1 x2 x2 h (39)
σ22
k
=− 9 2 − 18 + 2 σ0 P0k , ≤ x2 ≤ h
7 h h 3
σ22
k
= σ0 P0k , h ≤ x2 ≤ b
94 S. B. Salunkhe and P. J. Guruprasad
25 k x3 k
23
k
=− P + P0 σ0 (41)
14 1 h
This section begins with the validation of the proposed modeling approach for inter-
laminar stress components. The geometric and material properties corresponding to
each case against which the present method has been compared are given in Table 1.
The geometric representation of laminate structure has been shown in Fig. 3. The
method is then extended for the pretwisted, antisymmetric, or symmetric laminated
composite strip of solid cross section. All the results are for thin laminated compos-
ite strip subjected to uniform axial extension. Moreover, these results are an average
value calculated from the gauss points located near to the interface. The discussion
is for all the stress components through the thickness of few laminates. At the end
of section, the results are summarized.
In this section, many interlaminar stress components and in-plane stress components
results have been compared with the literature to evaluate the performance of the
present method. The results are for through the thickness near to the free edge and at
the interface of the composite laminate. Specifically, the results are compared with
Pipes et al. [16] for the symmetric angle-ply laminate, Wang et al. [2] for cross-ply
and quasi-isotropic laminate under the axial extension.
In Fig. 4a, the behavior of all 3-D stress components along the width at interface
between 45◦ and −45◦ for a graphite–epoxy laminate are shown. The material and
geometric properties are given in the first row of Table 1. These geometrical and
material properties are same as that of Ref. [16]. It is observed from the figure that
the values and the trend of stress components of the present approach are almost
the same with the corresponding results. Though small differences observed are due
to the difference in the finite difference method used by Pipes et al. [16] and the
finite element method used in the current work. Furthermore, in the present case all
3-D stress components has been predicted from the 1-D approach. However, in the
reference all 3-D stress components results are predicted from the 2-D approach. It
can be observed from Fig. 4a that the values of stress components start to decrease or
increase near the free edge, i.e., at x2 /b = 0.8. This portion of the width of the free
edge is a function of geometry, ply orientation and the properties of the laminate.
However, a simple thumb rule proposed by Pipes et al. [16] is that the width of the
boundary layer is equal to the laminate thickness. In Fig. 4a, the value of τ13 increases
(a) (b)
Fig. 4 a Result of stress components in the interface between the +45◦ and −45◦ under axial
extension along the width. b Interlaminar shear stress τ13 distribution through the laminate thickness
96 S. B. Salunkhe and P. J. Guruprasad
and becomes high near the free edge and at the same time σ11 , τ12 decreases. This
can be explained by the interlaminar stress transfer for the angle-ply laminate which
occurs due to the mismatch in the shear coupling coefficient between θ and −θ layers.
The mechanism of interlaminar stress transfer for angle-ply laminate is also called as
the first mode mechanism. The interlaminar shear stress can be one of the reasons for
delamination at the free edge in case of angle-ply laminates. The interlaminar normal
stress σ33 disappears at all interfaces as well as on free edge. It can be understood that
this stress may not do any damage to the angle-ply laminate. The present method is
further extended to evaluate the performance of through the thickness of the laminate.
In Fig. 4b shows the distribution of interlaminar shear stress τ13 across the laminate
thickness at the free edge. It is witnessed from the figure that shear stress τ13 vanishes
at free surface and laminate midplane, while it attains the maximum value in the
interface between the 45◦ and −45◦ layers. Figure 5 shows the distribution of τ13
stress along thickness for [45◦ / − 45◦ ]s laminate. Stress distribution through the
thickness, distribution of stress component near to the free edge region is compared
with Pipes et al. [16]. All the above results and their comparison with Pipes et al. [16]
result shows that there is not much variation in the two results.
In Fig. 6, the comparison has been made between the present and Ref. [2] results
for [90◦ /0◦ ]s and [0◦ /90◦ ]s laminate. The result shows a good correlation at the
free edge. However, the results are inaccurate at the interior regions removed from
free edge. This is because they found out the results of interior regions by using
CLT and the present results are by using VAM. It is noted that when the cross-
ply lamina is subjected to tensile loading in one direction; it shrinks in the other
direction perpendicular to the applied load. If the laminae with different Poisson’s
ratios are bonded together, then the interlaminar stress is being produced to force all
laminae to deform uniformly at the interfaces. This mechanism of interlaminar stress
transfer due to mismatch of Poisson’s ratio between cross-ply laminate is termed the
second mode mechanism. In such a situation, the values of σ11 and τ12 are same for
Determination of Interlaminar Stress Components … 97
(a) (b)
Fig. 6 a Variation of interlaminar normal stress σ33 in the interface of a [90◦ /0◦ ] and [0◦ /90◦ ]
laminate along the normalized width and its comparison between the reference and present result.
b Interlaminar shear stress τ23 result at interface (along the normalized width) of [90◦ /0◦ ] and
[0◦ /90◦ ] in a [90◦ /0◦ ]s and [0◦ /90◦ ]s graphite/epoxy laminate, respectively, and its comparison
between the present and the reference result
Fig. 7 Variation of
interlaminar stress τ23 along
the thickness at free edge of
[90◦ /0◦ ]s laminate and its
comparison between present
and reference results
all over width. However, the values of interlaminar normal stress σ33 for [0◦ /90◦ ]s
layup drastically increases from negative to positive. Whereas in [90◦ /0◦ ]s laminate,
contrary behavior has been observed. From this one can predict that the interlaminar
normal stress σ33 is responsible for delamination in the case of cross-ply laminate.
Further, it can be observed that the interlaminar shear stress τ23 value starts from zero
and increases but near the end of an interface ( x2 = 0.99*h), it tends to become zero.
From this, it can be concluded that the τ23 could not be responsible for delamination
in the case of cross-ply laminate. Similarly, through the thickness at x2 = h/3, results
of interlaminar shear stress τ23 is presented in Fig. 7. The results are compared with
the most favorable literature results. Again, it demonstrates a good agreement. From
the above studies, a hostile behavior in angle-ply and cross-ply laminates is noticed.
98 S. B. Salunkhe and P. J. Guruprasad
Fig. 8 Variation of
interlaminar normal stress
σ33 in the interface of a
[90◦ /0◦ /45◦ / − 45◦ ]
laminate along the width and
its comparison of present,
3-D FEM and reference
result
(a) (b)
Fig. 10 Variation of
interlaminar stress τ13 along
the thickness at free edge of
a [102◦ / − 102◦ ]s laminate and
its comparison between
present and reference results
It can be concluded that the trend and the magnitude of the interlaminar stress
components along the thickness and width matches well within the limits of error.
Thus, this approach has been proved to deliver results efficiently. In the next section,
all the six stress components behavior along the thickness of the different ply stacking
sequences have been computed and plotted.
100 S. B. Salunkhe and P. J. Guruprasad
In this section, the approach is applied to the additional laminates to check the utility
of the method. In addition to this, it will be helpful to check the effect of out of
plane deformation on interlaminar stress components between the asymmetric and
the symmetric laminates. This section gives the behavior of the stress components
through the thickness for the different layup orientations. The 3-D FEM simulation
are run for the present case of the cantilevered beam. In the simulation, a strategy
was formed such that the top and the bottom surface yield traction. However, in
real practice, the top and the bottom surfaces are traction free. For simplicity, these
conditions are relaxed in the present case. Excluding the top and the bottom surface of
geometry, all the results between the proposed scheme and the 3-D FEM are showing
a good agreement. These discussions are restricted to the transverse stress-related
studies. These stress results are shown for symmetric or asymmetric, cross-ply, angle-
ply and quasi-isotropic laminate subjected to an axial loading of 1000N. For the stress
analyses, the following boundary conditions have been enforced:
1. The stress boundary conditions on the top and bottom surfaces are given by
h
σ33 (x1 , x2 , ± ) = 0
2
h
τ13 (x1 , x2 , ± ) = 0 (42)
2
h
τ23 (x1 , x2 , ± ) = 0
2
Determination of Interlaminar Stress Components … 101
The parametric study has been are carried out due to the following things:
1. The interlaminar stress components changes due to the stacking sequence under
a uniform axial load.
2. It will help to understand the behavior of stress components along the thickness.
3. To check the computational efficiency and the accuracy of the present method.
The implementation of the developed approach is demonstrated first for the
Winckler’s type ([α2 /(90◦ − α)4 /α2 /α2 /(α − 90◦ )4 /α2 ]T ) antisymmetric laminate.
This layup exhibits a strong nonlinear extensional-twist coupling. This coupling is
nonlinear in nature, and generally, it is called as trapeze effect. Before the 3-D stress
analyses, the author validated static, dynamic, and the stability results by using the
present approach for the same laminate. The static validation for healthy and dam-
aged laminate is provided in the author’s paper [27]. Therefore, from the previous
study, it is observed that the method is capable of capturing nonlinearity. For this
analysis also, the cantilevered strips made of graphite/cyanate and glass/epoxy are
considered with Winckler’s type layup corresponding to α = 20◦ . The material and
the geometrical properties corresponding to AS4/3501-6 are shown in the second
row of Table 1.
The distribution of the stress components through the laminate thickness is shown
in Fig. 12. Similarly, the results are also validated for the symmetric layup with all the
geometric and the material properties remain the same to the antisymmetric layup
orientation. The 3-D stress components results for the symmetric Winckler’s type
layup are shown in Fig. 13. The in-plane stress results are exactly matching with the
3-D elasticity solution. Hence, by using the present formulation (which makes no ad
hoc kinematic assumption) the 3-D analyses (capturing nonlinearity) can be carried
out without depending on the computationally expensive 3-D FE analyses. These
two figures show that the thickness-wise stress distribution for the same physical
condition. By comparing these two figures following salient observation can be noted:
• Axial stress σ11 of the symmetric layup is more as compared to the antisymmetric
and less fluctuation along the thickness is observed. In both cases, the stress pattern
is symmetric about its midplane of the laminate.
• Both the transverse shear stresses and the normal transverse stresses are found to
be more in the antisymmetric layup.
• Smooth transitions are observed and are represented by least quadratic functions
for σ33 for both cases between the plies. The linear variations of τ13 and τ23 near
the free edge get influenced due to the alternating stacking sequence of angle-ply
laminates. This resulted in sharp transitions for τ13 and τ23 between the plies.
• These significant variations of interlaminar stress components may be due to con-
straining of the axial twist of general laminates under extension. It is to be noted
102 S. B. Salunkhe and P. J. Guruprasad
(a) (b)
(c) (d)
(e) (f)
Fig. 12 Stress distribution through the thickness of the antisymmetric Winckler’s type of laminate
at the free edge due to the axial loading when α = 20◦
that the same physical layers are used in the two laminates. It is clearly seen that by
changing the stacking sequences in the laminate (from symmetric to Antisymmet-
ric), the numerical values of both stress components are considerably increased,
almost throughout the entire thickness of the laminate. Thus, it is envisioned here
that by changing the fiber orientations from symmetric to antisymmetric, the state
of interlaminar stresses can be significantly changed. One can easily see the vari-
ation of the plot between in these two type of layup in Figs. 12 and 13.
Determination of Interlaminar Stress Components … 103
(a) (b)
(c) (d)
(e) (f)
Fig. 13 Stress distribution through the thickness of the symmetric Winckler’s type of laminate at
the free edge due to axial loading when α = 20◦
• The nonlinear analyses of the same antisymmetric layup were carried out. The
comparison for the magnitudes of the stress components shows more difference
between them.
• By altering the stacking sequences of the individual layers and the fiber orienta-
tions, the interlaminar stress components can be controlled to avoid the free edge
effect.
A comparison is further extended for the layup orientation [45◦ /0◦ /45◦ /0◦ /45◦ /0◦ ]s .
The material and the geometric properties have been given in the first row of Table 1.
The stress results for the angle-ply laminate under the extensional loading are given
in Fig. 14. The comparisons are shown for through the thickness profile in the figure,
and a good agreement between them can be observed.
To understand the effect of stacking sequence, the quasi-isotropic laminate with
ply properties shown in the first row of Table 1 under the uniform axial extension
has been considered. A good correlation can be observed for the stress distribution
across the thickness with 3-D FEM results in Fig. 15. Some deviations observed are
due to the higher thickness to width ratio of the laminate. This drawback can be
overcome by using the higher order terms in asymptotic series. All the interlami-
nar stress components approaches zero near the top and the bottom surface of the
laminate. As shown in Fig. 15, the interlaminar shear is symmetric concerning the
midplane interface. The maximum interlaminar shear and the normal stress occur at
the midplane of laminates.
In all the above cases, in the composite structure under the mechanical load, the
results achieved from the present 1-D formulation matched almost with its corre-
sponding elasticity and the 3-D FE solutions. The present results in Table 2 show
that the 1-D VAM and the approximation method-based formulation reduces the
computational time without much dishonor to the exactness. Such a simplified and
a computationally efficient approach is most useful in the analyses of laboratory
coupon type specimens to minimize the free edge effects.
4 Conclusion
The key feature of any successful structural design is to predict the structural response
at a given condition accurately. By keeping this thing in mind, here in the work, our
goal was set to develop an analytical solution based on 1-D, which can produce accu-
rate 3-D elasticity solution with less computational time. To achieve the goal, the
1-D analytical approach based on the variational asymptotic method and the approx-
imate method were used to recover an accurate 3-D elasticity solution. This approach
also satisfies the interlaminar displacement continuity and the transverse equilibrium
requirements as demanded by the exact 3-D formulation. Based on the derived theory,
the numerical codes were developed in-house. The 3-D displacements, interlaminar
normal, and the shear stress distributions were obtained. To check the correctness of
it, results were compared to the available literature for the different layup orienta-
Determination of Interlaminar Stress Components … 105
(a) (b)
(c) (d)
(e) (f)
Fig. 14 Stress distribution through the thickness of [45/0/45/0/45/0]s laminate at free edge due
to axial loading
tions. They show a good agreement especially with the numerical solution of Pipes
[16, 17], Pagano [7, 18, 29], Sen [3], and Wang [32, 33]. Further, the deviation in
the results was less than 1% when compared to the commercial 3-D finite element
packages. Subsequent to the analyses results obtained from the recovery relation, the
following things are confirmed and noted:
106 S. B. Salunkhe and P. J. Guruprasad
(a) (b)
(c) (d)
(e) (f)
Fig. 15 Stress distribution through the thickness of [90/0/45/ − 45]s laminate at free edge due
to axial loading
• Mechanism of the free edge effect in angle-ply laminates is due to the mismatch
in the shear coupling coefficients between the adjacent layers.
• In-plane shear stress components vanish at the free edge in angle-ply laminates.
• In cross-ply laminates, the interlaminar normal stress appears to be singular at the
free edge.
• In cross-ply laminates, the mechanism of the free edge effect is due to the mismatch
in Poisson’s ratio between 0◦ and 90◦ plies.
Determination of Interlaminar Stress Components … 107
• In quasi-isotropic laminate, the stacking sequence plays a vital role in the devel-
opment of interlaminar stresses.
• The effect of interlaminar stress was found to be restricted near the laminate free
edge.
• The approach presented in this paper is possibly the simplest and the efficient of
all the currently available methods.
• It is attractive and can be extended during the design phase in order to study the free
edge stress components and their effect under the static and the fatigue loading.
• The computational time in this case is also very less as compared to the 3-D finite
element solver, i.e., ANSYS® .
• The development of an analytical solution for the interlaminar 3-D stress compo-
nents by using 1-D approach is important from the point of saving the time as well
as can be built into machine intelligence.
References
Abstract This work details wrinkling for hyperelastic materials with the Ogden
material model. First, generalized expressions are derived for stresses and tangent
stiffness terms. Material constants obtained through testing work have been used to
determine stiffness terms as well as second Piola–Kirchhoff’s stress for a range of
stretch ratios. This is then followed with a new model based upon modified defor-
mation gradient tensor and hence Green tensor. Wrinkling parameter is monitored
with material constants and stretch ratios. Analytically performance of material has
been accessed for different conditions of uniaxial tension, equi-biaxial loading and
plane strain condition. It has been observed that over a range of stretches; uniaxial
condition and plane strain condition results in a situation which is consistent in the
context of wrinkling, i.e. possibility of wrinkling gets boosted. On the other hand,
when equi-biaxial loading or stretch ratio approaches unity for a region, chances of
wrinkling being triggered reduce substantially. Developed numerical models are val-
idated against tested work on planar hyperelastic structures for the cases of uniaxial,
biaxial as well as shear loading. It has been observed that numerical results obtained
hint possible scenarios of wrinkling.
1 Introduction
hydraulic hoses, seals, gaskets, tank liners, aircraft applications. One of the dominant
application is in the medical section. Doctors use to have NBR gloves for surgical pro-
cedures. Such gloves are susceptible to wrinkling. The wrinkle is fold or bifurcation
or generation of the wavy surface due to the inability of material to resist compressive
loads. This class of problem has been tackled in the literature by two approaches.
First being tension field theory, in which secondary principal stresses are assumed to
be zero and analytical formulations are built [2–4]. But it is not easy to determine the
wrinkling details such as wavelength and amplitude from this approach. The second
approach is based on the bifurcation theory. This treats the problem on the lines of
buckling of plates. Wrinkling of NBR is considered in this paper. The material under
study is incompressible hyperelastic. Further, Ogden’s constitutive rule is represent-
ing the characteristics of the same. The formulation accounts for wrinkling features
by introducing a wrinkling tensor and modified stretches in the kinematics. Material
constants are obtained from experiments with the in-house developed test rig. Test
prototype has been developed to take care of large deformations with large strains
under uniaxial, biaxial as well as shear conditions. It also generates stress relaxation
data. Wrinkling measurements are done using non-contact method with a brand new
technique using the Arduino controller.
2 Methodology
There are three different criteria, which determine the existence of wrinkling in a
membrane, rubber or soft tissue.
1. Principal stress criterion,
2. Principal strain criterion, and
3. Combined criterion, including both principal stresses and principal strains.
For the wrinkling criterion based on principal stresses/strains, the membrane is
said to be in a taut condition when both the principal stresses/strains are positive. If
both principal stresses/strains are zero, then the slack condition of membrane results.
Obviously, wrinkling will not be there for the taut and slack states. Thus, wrinkling
occurs in the membrane when the minor principal stress/ strain is zero and the major
principal stress/strain is tensile. Tension field theory is depending upon fictitious
wrinkled surface and everything is required to be derived based upon following
or similar geometry as shown in Fig. 1. As shown, in a direction perpendicular to
uniaxial stress direction, the material is not able to resist compression as compressive
stiffness or bending stiffness of the material is very small. Based upon the deformed
configuration of the figure, wrinkling strain is defined as
β −1
Wrinkling strain = (1)
β
A Study on Wrinkling Characteristics of NBR Material 111
where β is wrinkling parameter, which decides the amount of wrinkling and defined
with respect to pseudo-deformed surface or fictitious non-wrinkled average surface.
Deformed length is L = β ∗ L
For wrinkling based on the combined criterion, the membrane is in a taut condition
when the minor principal stress is positive and is in the slack condition when the
major principal stress is zero. When the minor principal stress is negative and the
maximum principal strain is positive, the membrane gets wrinkled. Here, σ1 and
σ2 are the major and minor principal stresses, respectively, e1 and e2 are major and
minor strains, respectively, ν is Poisson’s ratio.
The wrinkling criteria are summarized below:
a. Principal stress criterion
i. σ1 > 0 and σ2 > 0 taut
ii. σ2 ≤ 0 and σ1 > 0 wrinkled
iii. σ2 ≤ 0 and σ1 ≤ 0 slack
b. Principal strain criterion
i. e1 > 0 and e2 > −ve1 taut
ii. e2 ≤ −ve1 and e1 > 0 wrinkled
iii. e2 ≤ 0 and e1 ≤ 0 slack
c. Combined criterion
i. σ1 > 0 and σ2 > 0 taut
ii. σ2 ≤ 0 and e1 > 0 wrinkled
iii. σ1 ≤ 0 and e1 ≤ 0 slack
Principle strain criterion is used in user subroutine developed for predicting the
occurrence of wrinkling. The procedure used in this subroutine for finite element
implementation of wrinkling is summarized in Fig. 2. At the start of every increment,
112 V. S. Pawar et al.
ABAQUS calls the subroutine with trial strain increment and trial strain obtained from
the previous increment except for the first increment. At the start of the first increment,
strain increment executed based on tangent stiffness owing to linear elasticity is
obtained. At each increment, strain increment is rotated in the principal direction
within the framework. Based upon the framework used in the implementation, the
state of the membrane is accessed. Accordingly, the scheme picks suitable tangent
stiffness available and updates stress at the end of increment. Next trial increment is
evaluated based upon stress updates of the previous increment. This loop continues
until the convergence criteria get satisfied. Tolerance (TOLR) is set of the order of
10−12 . Since the problem is of large deformation, Jaumann’s rate of Kirchoff’s stress
is used for tangent stiffness updates. This being incompressible material, Jaumann’s
rate of Kirchoff’s stress is the same as Jaumann’s rate of Cauchy’s stress.
NBR is tested for uniaxial testing and is calibrated. It has been observed that the
Ogden model, reduced-order model of order 2 are best for representing the material
and both the models are found to be stable.
A Study on Wrinkling Characteristics of NBR Material 113
The stored energy function for rubber-like material is given by Ogden [2] in terms
of principal stretches as
N
μp α α α
∅(λ1 , λ2 , λ3 ) = (λ1 p + λ2 p + λ3 p − 3) (2)
p=1
αp
N
μp α α −α p −α p
∅(λ1 , λ2 , λ3 ) = (λ1 p + λ2 p + λ1 λ2 − 3) (3)
p=1
αp
N
2μ = μpαp (4)
p=1
dφ
σj = p + λj (5)
dλ j
where σ j = λ j P j
Now considering an incompressible material under uniaxial tension, with the
stretch ratio given as λ = ll0 , the principal stresses are given by
N
α
σj = p + μpλ j p (6)
p=1
σ2 = σ3 = 0, yielding
114 V. S. Pawar et al.
N
α − 21 α p
σj = (μ p λ j p − μ p λ j ) (7)
p=1
It can be rightly said that, for rubber and biological materials, more sophisticated
models are necessary. Such materials may exhibit non-linear stress–strain behaviour
at modest strains, or are elastic up to large strains. These non-linear stress–strain
behaviours need to be accommodated using specifically tailored strain-energy density
functions.
Typical wrinkling configuration and parameters involved in the model are high-
lighted in Fig. 1. Ogden model captured material response for both cases of with
wrinkling parameter and without wrinkling parameter. On the onset of wrinkling,
when secondary principal stress is approaching zero, the maximum principal stress
determined are shown in Fig. 3. In realtime conditions, situations result such that
uniaxial tension condition may prevail which forces other two stretches to be equal
or sometime plane strain assumptions may sustain while it may be possible that
equi-biaxial stretch involves in some of the applications. Second Piola–Kirchhoff’s
stress is used as a stress measure. Incompressibility condition is accommodated by
imposing constraints. The associated multiplying parameter is obtained to ensure the
plane stress problem. Wrinkling depth, orientation and associated loads are captured
experimentally.
Figure 4 evaluates the performance of NBR under the conditions of uniaxial,
equi-biaxial and plane strain conditions analytically. Since the grip of the test rig
developed is such that samples can be fixed over a wide range of width, uniaxial and
plane strain situations can be immediately achieved. Since the prototype is developed
by keeping a wide range of stretch ratios in mind, equi-biaxial condition can be
maintained as well. Secondary principle stress variations with almost all possibilities
are shown in Fig. 4. All the cases are depicted in terms of true stresses as the problem
is of large deformation. The developed framework has considered Jaumann rate of
Kirchoff’s stress as already mentioned for quick convergence. As the material is
incompressible, it can be termed as Jaumann’s rate of Cauchy’s stress as well, which
is required in the Eulerian description. Over a range of stretch, it is possible to have
negative compressive stress for plane strain and for uniaxial condition it is zero. As
opposed to this, if equi-biaxial case is to be dealt with, most zones are turning out
to be wrinkle free. For unequal stretch ratios, the intermediate scenario appears as
shown in Fig. 4.
Figure 5 depicts maximum Cauchy’s principal stresses against corresponding true
strain over a range for which material was tested. No significant difference resulted in
stress for the different cases that have been considered in this study. The behaviour
has been captured perfectly through the Ogden model, reduced-order polynomial
A Study on Wrinkling Characteristics of NBR Material 115
Fig. 3 Maximum principal stress at the onset of wrinkling using Ogden material model
model of order 2 and order 3. Most of the applications such as large shear, pure
shear and biaxial loading conditions can be approximated as plane strain problem.
For wide grip of the specimen with comparable width, this case is different from
the uniaxial loading case; and it follows the plane strain assumptions. With a slight
stretch in the perpendicular direction to loading as is the case of typical NBR tubes
or gloves, secondary principal true stress is approximately zero. Hence such NBR
gloves or tubes wrinkle. This is illustrated in Fig. 5 for low true strains. For low
strength NBR, these effects get more pronounced.
In general, the use of tension field theory captured wrinkling but only orientation
can be obtained with this theory. Implicit solution schemes for user subroutines in
the finite element package require the calculation of the tangent stiffness matrix.
Primarily this is done by resolving the stiffness matrix in principal directions and
then back into the original coordinate system; thus rendering FE program to have
co-rotational part to be dealt in subroutine through transformation tensor R. There
is an additional shear stiffness that gets developed due to changes in the shear stress
which will change the direction of the principal stresses. These are related to the
current principal stresses and stretches (Cauchy’s stresses and stretches).
116 V. S. Pawar et al.
Fig. 4 Secondary principal stress against true strain for different cases of NBR
4 Summary
This work presents the analytical development of wrinkling using the Ogden model
and the same has been numerically treated using user subroutines. NBR of moderate
strength has been considered for the analysis and the developed framework is applied
to it. Merits and demerits of the type of test used in the context of the developed
prototypes have been discussed. Conditions for wrinkling have been elaborated and
possible situations are addressed. For small strains, it can be concluded that as the
strength of NBR gets reduced with the reduction in polymer percentage, chances of
wrinkling get triggered. This is true for larger stretch ratios for biaxial loading and
in general for plane strain and uniaxial conditions where the width of such structure
is small compared to length.
A Study on Wrinkling Characteristics of NBR Material 117
Fig. 5 Maximum principal true stress against true strain for different cases of NBR
References
1. Khan AS, Farrokh B (2006) Thermo-mechanical response of nylon 101 under uniaxial and
multi-axial loadings: part I, experimental results over wide ranges of temperatures and strain
rates. Int J Plasticity 1506–1529
2. Ogden RW (1972) Large deformation isotropic elasticity—on the correlation of theory and
experiment for incompressible rubberlike solids. Proc R Soc London Ser A, Math Phys Sci
326(1567):565–584
3. Steigmann DJ, Pipkin AC (1989) Finite deformation of wrinkled membranes. Q J Mech Appl
Math 42(3):427–440
4. Deng X, Pellegrino S (2012) Wrinkling of orthotropic viscoelastic membranes. AIAA J 50:668–
681
First Ply Failure Study of Laminated
Composite Conoidal Shells Using
Geometrically Nonlinear Formulation
Abstract The literature clearly indicates that the failure of laminated composite
plates have been studied in detail while composite shell structures have received lim-
ited attention. Though few recent papers reported first ply failure of conoidal shells
in order to apply these shell forms confidently in industrial applications, the failure
investigations need to be continued for different parametric variations keeping the
industrial requirements in mind. This study attempts to fill the lacuna. An isopara-
metric finite element code considering geometrically nonlinear strains and constant
shear deformation along the thickness direction of the shell is proposed to study fail-
ure initiation in uniformly loaded laminated conoidal shell roofs having two adjacent
boundaries simply supported and other two adjacent edges clamped. Moreover, fail-
ure locations and modes or tendencies of failure are also reported. The results are
studied for varying stacking orders and laminations. This study recommends that the
practicing engineers must adopt 30°/−30°/30° laminate to maximize load-carrying
capacity of the shell for the edge condition taken up here.
1 Introduction
The conoidal shells are stiffer and easy to cast compared to singly curved cylindrical
and doubly curved spherical shell surfaces. Additionally, the conoidal shell form
allows natural ventilation and entry of diffused sunlight from both ends which is pre-
ferred in medicinal and chemical plants. Thus the practicing civil engineers choose
K. Bakshi (B)
Indian Institute of Technology Indore, Indore, India
e-mail: Bakshi.kaustav@gmail.com; kaustav.bakshi@iiti.ac.in
Discipline of Civil Engineering, Khandwa Road, Simrol, Madhya Pradesh 453552, India
D. Chakravorty
Jadavpur University, Kolkata, India
e-mail: prof.dipankar@gmail.com
© Springer Nature Singapore Pte Ltd. 2020 119
B. N. Singh et al. (eds.), Recent Advances in Theoretical, Applied, Computational
and Experimental Mechanics, Lecture Notes in Mechanical Engineering,
https://doi.org/10.1007/978-981-15-1189-9_10
120 K. Bakshi and D. Chakravorty
conoidal shells as roofs to cover large unsupported areas one finds in aircraft hangers,
stadiums and shopping malls. The high specific strength/stiffness of the laminated
composites render lesser foundation force and seismic force compared to conven-
tional reinforced cement concrete shell roofs. Moreover, the material offers designer
flexibility, where the relatively stronger direction of a lamina can be oriented along
the major direction of load transfer. Keeping these advantages in mind the practic-
ing civil engineers prefer the laminated conoids over the isotropic ones. A group of
researchers started working on bending and dynamic responses of laminated conoids.
Dey et al. [1] studied static bending responses of laminated composite conoids. Ghosh
and Bandyopadhyay [2] applied an isoparametric finite element formulation to study
static responses of isotropic conoidal shells with cut-outs. Dynamic analysis of stiff-
ened conoids was conducted by Nayak and Bandyopadhyay [3] using eight noded
doubly curved elements. Das and Chakravorty [4] studied the free vibration of lam-
inated conoids for complicated edge conditions. Bending responses of delaminated
conoids were reported by Kumari and Chakravorty [5]. But the authors did not study
the damage initiation and propagation in the composite material. Dynamic instability
of laminated conoids was reported by Pradyumna and Bandyopadhyay [6].
A shell roof can be confidently used in practical applications when the allowable
load before failure is known to the practicing engineers. The failure of the laminated
composite is progressive as indicated by Singh and Kumar [7]. The failure initiates at
the weakest ply in the laminate. The failure initiation is termed as the first ply failure.
If the latent damage remains undetected and unprotected then the cracks propagate
within the laminate and culminate to total collapse under service condition. Such a
sudden catastrophe can be avoided by studying failure initiation in composite shell
roofs. The failure of composite plates was studied by many researchers in detail.
Reddy and Reddy [8] reported the failure of composite plates using geometrically
linear and nonlinear strains. Kam et al. [9] conducted an experiment to study failure
initiation and propagation in laminated plates. Sathish Kumar and Srivastava [10]
worked on the failure of stiffened plates. The failure study of composite plates was
continued by Lal et al. [11] and Coelho et al. [12] where the authors studied pro-
gressive failure of the composite material using geometrically nonlinear formulation.
The research reports on laminated composite shells are really scanty compared to
the study carried out on composite plates. Prusty et al. [13] and Adali and Cagdas
[14] worked on the failure of singly curved and doubly curved shell panels using
geometrically linear strains. The present authors worked on failure initiation in lam-
inated conoids using geometrically linear [15] and nonlinear strains [16]. The failure
initiation in clamped conoids [15] and a conoid with a set of parallel edges clamped
and other edges free [16] were studied by the present authors. The failure study on
composite conoidal shells still requires attention from the researchers for industrially
important parametric variations to apply these shell form confidently in industrial
applications. Thus the present study focuses on failure initiation in conoids with two
adjacent boundaries simply supported and other two adjacent edges clamped. This
paper also reports the failure location on the shell surface and modes of failure.
First Ply Failure Study of Laminated Composite … 121
2 Mathematical Formulation
A conoidal shell having radii of curvatures Ryy and Rxy is taken up here where the total
laminate thickness ‘h’ is consisting of thin arbitrarily oriented layers (h = h1 + h2 +
…) (see Fig. 1). The reference axes (x and y) are taken up at the shell mid-surface.
The ‘z’ axis is oriented along the thickness direction of the shell.
The displacement at any arbitrary point is expressed in Eq. (1) assuming a constant
shear deformation acting along the z-axis.
u, v, w, α, β are shown in Fig. 1. u0 , v0, and w0 are the degrees of freedom at the
mid-surface.
{ε}L and {ε}NL are linear and nonlinear mid-surface strains, respectively. The
linear strains are adopted from Kumari and Chakravorty [5] for an undamaged shell.
The nonlinear strains are the same as those were reported in [16]. The laminate
constitutive relationship is given as the following:
The stress resultants {F} and laminate stiffness matrix [D] are adopted from
Kumari and Chakravorty [5] for an undamaged shell.
Fig. 1 The conoidal shell with global reference axes and degrees of freedom
122 K. Bakshi and D. Chakravorty
The isoparametric finite element code is generated using the same element reported
by Bakshi and Chakravorty [15]. The strain–displacement relation is expressed in
Eq. (4).
1
{ε} = ([B] L + [B] N L ){u i } (4)
2
[B]L is taken up from Kumari and Chakravorty [5] for an undamaged shell. The
nonlinear matrix [B]NL is the same as it was reported in [16].
The total potential energy (π ) of the conoid [refer to Eq. (5)] is minimized with
respect to its displacements to derive the governing equation of shell bending.
¨ ¨
1
π= {ε}T [D]{ε}d A − {u}T {q}d A (5)
2
A A
The external load intensity on the shell is {q}, shell displacements are {u}, and
laminate strains are {ε}.
The governing equation is given as the following:
∂π
=ψ (6)
∂{u}
Displacement of the shell for the nth iteration is calculated following Fig. 2.
Table 1 Nondimensional central displacements (ŵ × 103 ) of simply supported composite spherical
shell under uniformly distributed load
Lamination 0°/90° 0°/90°/0° 0°/90°/90°/0°
Reddy [18] 16.980 6.697 6.833
Present FEM 17.009 6.707 6.835
a/b = 1, a/h = 100, E 11 = 25E 22 , G12 = G13 = 0.5E 22 , G23 = 0.2E 22 , ν = 0.25, E 22 = 106
N/cm2 , R/a = 1030
3 Numerical Problems
Table 1 confirms accuracy of the proposed static formulation as the central displace-
ments of the spherical shell obtained using the proposed finite element code are in
a close match with the results reported by Reddy [18]. Figure 3 confirms that the
First Ply Failure Study of Laminated Composite … 125
Table 2 Nonlinear first ply failure loads in Newton for a (02◦ /90◦ )s plate
Failure theories Side/thickness Failure loads [9] Experimental Failure loads
failure load [9] (proposed code)
Maximum stress 105.26 147.61 157.34 135.94
Maximum strain 185.31 218.10
Hoffman 143.15 133.21
Tsai–Wu 144.42 134.50
Tsai–Hill 157.58 134.91
Length = 100 mm, ply thickness = 0.155 mm
proposed code accurately formulated the conoidal shell geometry as good agreement
is noted between the proposed results with the data reported by Hadid [19]. The first
ply failure and geometrically nonlinear formulations are correctly formulated in the
proposed code as it is found in Table 2 that the experimental and numerical failure
loads reported by Kam et al. [9] show a close match with present results.
The additional problems concentrate on failure initiation in uniformly loaded
conoidal shells. The failure initiation is studied for three different fiber orienta-
tions. The fibers run along 0°, 45°, and 90° with global x-axis of the shell. These
126 K. Bakshi and D. Chakravorty
laminations are stacked in symmetric and antisymmetric orders to form two- and
three-layered laminates by keeping the total material thickness as fixed. The nondi-
mensional first ply failure loads of the shell for different laminations and stacking
orders are reported in Table 4. The dimensions of the shell are furnished with the
table. The nondimensional failure pressure (NFL) is obtained by using the expres-
sion, NFL = (F/E 22 )(a/ h)4 , where ‘F’ is the failure load in MPa. Apart from the
failure pressure, failure initiation location on the shell surface, first failed ply number
and failure modes or tendencies are also reported in that table. The first lamina is the
topmost one (see Fig. 1). The highest number is assigned to the bottommost lamina.
It is obvious that the working load of a laminate can be obtained by assigning the
factor of safety to the minimum failure pressure for a given laminate.
An in-depth study of Table 4 shows that the practicing civil engineers must adopt
the angle-ply laminates for the conoids as the nondimensional failure pressure values
are greater in magnitude for angle-ply laminations than the cross-ply ones. The
minimum failure pressure is yielded by Hoffman failure theory for the cross-ply
laminations and by maximum strain theory for the angle-ply ones. It is important
to note from Table 4 that all the cross and angle-ply laminates taken up here fail
through transverse tension acting perpendicular to the fiber and they fail at points
lying on the simply supported boundaries of the conoid. It can be concluded from
these observations that diagonal tension caused by a shear force acting at the simply
supported edges of the shell is the cause of failure for all the laminates considered
in Table 4.
In cross-ply laminates, the fibers run along the beam direction of the conoid in 0°
lamina and along the arch direction in 90° lamina. The elastic modulus of a lamina
is relatively higher along its fiber direction. It is noted from Table 4 that the 90°
lamina fails first for both the 0°/90° and 0°/90°/0° laminates. The diagonal tension
has almost equal components along x and y directions of the shell. The arch direction
of the conoid is stiffer than the beam direction by virtue of the curved geometry of
the shell. The 90° fiber, being aligned along the arch direction of the conoid makes
the beam direction a relatively weaker side of the shell. Thus, for both the 0°/90°
and 0°/90°/0° laminates, the component of diagonal tension acting along the arch
First Ply Failure Study of Laminated Composite … 127
Table 4 Nondimensional first ply failure pressure, failed ply number, failure location, and failure
modes/tendencies of laminated conoid
Lamination Failure Nondimensional Failed Failure Failure
criteria failure pressure ply location mode/failure
(NFL) (x, y) (m, tendency
m)
0°/90° Maximum 1510.73 2 (0, 0.75) 2
stress theory
Maximum 1502.55 1 (0.25, 1) 2
strain theory
Hoffman 1496.42L 2 (0, 0.75) 2
failure
theory
Tsai–Hill 1500.51 2 (0, 0.75) 2
failure
theory
Tsai–Wu 1497.45 2 (0, 0.75) 2
failure
theory
0°/90°/0° Maximum 1927.48 2 (0, 0.75) 2
stress theory
Maximum 1922.37 2 (0, 0.75) 2
strain theory
Hoffman 1865.17L 2 (0, 0.75) 2
failure
theory
Tsai–Hill 1891.73 2 (0, 0.75) 2
failure
theory
Tsai–Wu 1866.19 2 (0, 0.75) 2
failure
theory
45°/−45° Maximum 1991.83 2 (0, 0.88) 2
stress theory
Maximum 1621.04L 2 (0, 1) 2
strain theory
Hoffman 1971.40 2 (0, 0.88) 2
failure
theory
Tsai–Hill 1979.57 2 (0, 0.88) 2
failure
theory
Tsai–Wu 1973.44 2 (0, 0.88) 2
failure
theory
(continued)
128 K. Bakshi and D. Chakravorty
Table 4 (continued)
Lamination Failure Nondimensional Failed Failure Failure
criteria failure pressure ply location mode/failure
(NFL) (x, y) (m, tendency
m)
45°/−45°/45° Maximum 3266.60 3 (0, 0.13) 2
stress theory
Maximum 2939.73L 3 (0, 0.13) 2
strain theory
Hoffman 3250.26 3 (0, 0.13) 2
failure
theory
Tsai–Hill 3245.15 3 (0, 0.13) 2
failure
theory
Tsai–Wu 3161.39 3 (0, 0.13) 2
failure
theory
Note 1 Failure mode/tendencies as 2 indicates tensile matrix cracking [15]
Note 2 a/b = 1, a = 1000 mm, h = 10 mm, hh = 200 mm, hl = 50 mm
Note 3 ‘L’ indicates minimum failure pressure
Fig. 4 Nondimensional first ply failure pressure for varying ‘θ°’ in θ°/−θ°/θ° laminate
5 Conclusion
(3) The symmetric stacking orders found to fail in higher magnitudes of externally
superimposed pressure compared to the antisymmetric orders. Thus, the practic-
ing civil engineers must adopt the three-layered symmetric angle-ply laminate
to fabricate the conoidal shell option studied here.
(4) Among the angle-ply laminations studied in this paper, it is found that for a
constant quantity of material use, the 30°/−30°/30° laminate shows the highest
failure pressure. Thus the 30°/−30°/30° laminate is recommended for practi-
cal use. The bottommost lamina of the 30°/−30°/30° laminate fails first. The
failure initiates from a point lying on the simply supported boundary of the
30°/−30°/30° shell.
References
1. Dey A, Bandyopadhyay JN, Sinha PK (1992) Finite element analysis of laminated composite
conoidal shell structures. Comput Struct 43:469–476
2. Ghosh B, Bandyopadhyay JN (1994) Bending analysis of conoidal shells with cut-outs. Comput
Struct 53:9–18
3. Nayak AN, Bandyopadhyay JN (2006) Dynamic response analysis of stiffened conoidal shells.
J Sound Vib 291:1288–1297
4. Das HS, Chakravorty D (2008) Natural frequencies and mode shapes of composite conoids
with complicated boundary conditions. J Reinf Plast Compos 27:1397–1415
5. Kumari S, Chakravorty D (2010) On the bending characteristics of damaged composite conoidal
shells—a finite element approach. J Reinf Plast Compos 29:3287–3296
6. Pradyumna S, Bandyopadhyay JN (2011) Dynamic instability behavior of laminated hypar and
conoid shells using a higher-order shear deformation theory. Thin Walled Struct 49:77–84
7. Singh SB, Kumar A (1998) Postbuckling response and failure of symmetric laminates under
in-plane shear. Compos Sci Technol 58:1949–1960
8. Reddy YSN, Reddy JN (1992) Linear and nonlinear failure analysis of composite laminates
with transverse shear. Compos Sci Technol 44:227–255
9. Kam TY, Sher HF, Chao TN, Chang RR (1996) Predictions of deflection and first-ply failure
load of thin laminated composite plates via the finite element approach. Int J Solids Struct
33:375–398
10. Satish Kumar YV, Srivastava A (2003) First ply failure analysis of laminated stiffened plates.
Compos Struct 60:307–315
11. Lal A, Singh BN, Patel D (2012) Stochastic nonlinear failure analysis of laminated composite
plates under compressive transverse loading. Compos Struct 94:1211–1223
12. Coelho AMG, Mottram JT, Harries KA (2015) Finite element guidelines for simulation of
fiber-tension dominated failures in composite materials validated by case studies. Compos
Struct 126:299–313
13. Prusty BG, Ray C, Satsangi SK (2001) First ply failure analysis of stiffened panels—a finite
element approach. Compos Struct 51:73–81
14. Adali S, Cagdas IU (2011) Failure analysis of curved composite panels based on first-ply and
buckling failures. Procedia Eng 10:1591–1596
15. Bakshi K, Chakravorty D (2013) First ply failure study of composite conoidal shells used as
roofing units in civil engineering. J Fail Anal Prev 13:624–633
16. Bakshi K, Chakravorty D (2017) Geometrically nonlinear first ply failure loads of laminated
composite conoidal shells. Procedia Eng 173:1619–1626
First Ply Failure Study of Laminated Composite … 131
Abstract While designing the wind tunnel components, extreme care is taken for the
designing of the contraction wall profile. Contraction plays an important role in flow
uniformity and turbulence reduction inside the test section. Due to this, researchers
are having a greater interest in the designing of a contraction wall profile. The sixth-
order polynomial curve proposed by Sargison et al. is a well-acknowledged design
for the contraction wall profile. Almost all design criteria were considered for the
designing of the sixth-order polynomial curve except the proportionality criteria of
inlet and outlet radii. According to the design criteria, the inlet radius must be larger
compared to the outer radius of the contraction but in the sixth-order polynomial
design, both the radii are almost similar. A transformed polynomial design was
proposed by Daniel Brassard to meet this design criterion by changing the value of
polynomial according to a proposed variable α. In this paper, the open-source CFD
tool OpenFOAM was used to analyse the transformed sixth-order polynomial. The
adverse pressure gradient, uniformity of flow, and turbulence intensity inside the test
section as well as at the contraction outlet was taken into account for the analysis. The
comparative results from the analysis show that the transformed polynomial model
is giving better outcomes than the conventional sixth-order polynomial model.
1 Introduction
A wind tunnel is an aerodynamic tool that can simulate the fluid flow around a body
as well as the forces produced at the fluid–body interface. Wind tunnels are basically
categorized with respect to the tunnel geometry (open-circuit wind tunnel or closed-
circuit wind tunnel) or by the wind speed in the test section relative to the speed of
sound [1]. With the help of instrumentation measurement of local as well as the total
velocity along with the temperature and pressure can be done using a wind tunnel.
Flow visualization can also be done inside the test section by using smoke generators.
The main parts of a conventional wind tunnel are the settling chamber, contraction,
diffuser, test section and drive section [2]. Due to the wide real-world applications,
plenty of efforts has been done in designing wind tunnel [3, 4]. Lots of researches
are going on in the field of optimized designing of wind tunnel components. The
creation of uniform flow inside the test section is the key objective of wind tunnel
design.
The contraction plays a crucial role in producing a uniform flow with minimal
turbulence inside the test section. Many research works have been done on the wind
tunnel contraction design [5–11]. The prime objectives for the design of contraction
are to improve the uniformity of flow, minimization of turbulence inside the working
section, minimization of the boundary layer thickness at the entry of the test section
and to evade boundary layer separation inside the contraction. The design of the con-
traction wall profile using a higher order polynomial equation got more significance
in due time. The present work investigates the transformed sixth-order polynomial
profile for the contraction wall profile of the wind tunnel.
Flow uniformity inside the test section of a wind tunnel is directly affected by the
contraction design. The criteria that should be satisfied with the design of the contrac-
tion wall profile ensures uniform flow at the contraction outlet. So that the velocity
or pressure gradients should not create a progression of boundary layer thickness
or generation of boundary layer separation. Sargison et al. proposed a sixth-order
polynomial curve that fits in with all the following seven conditions that define the
contraction profile [12].
y(x = 0) = H ; y (x = 0) = 0; y (x = 0) = 0;
(1)
y(x = l) = h; y (x = l) = 0; y (x = l) = 0; y (x = i) = 0;
where ‘i’ is the position of inversion point from the inlet, l refers to the total length
of contraction, H and h represent height at the inlet and the outlet of the contraction,
respectively. The location of the inversion point has been found to be one of the
main factors in the optimization of the design. For finding the optimal position of
Analysis of Transformed Sixth-Order Polynomial … 135
inversion point, the degree of curvature at contraction inlet is varied. In their paper,
0.6 was found to be the optimal position of inversion point. Solving the following
matrix equation provides the values of unknown coefficients.
Aw = B; (2)
where
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
30i 4 20i 3 12i 2 6i 0 a
⎢ l6 l5 l4 l3 ⎥ ⎢ −h ⎥ ⎢b⎥
A=⎢
⎣ 6l 5
⎥, B = ⎢ ⎥ ⎢ ⎥
⎣ 0 ⎦, w = ⎣ c ⎦ (3)
5l 4 4l 3l 2 ⎦
30l 4 20l 3 12l 2 6l 0 d
The design for the contraction must ensure a larger radius at inlet compared to the
outlet radius of the contraction profile. But in most of the higher order polynomial
designs, both inlet and outlet radii of the profiles are nearly similar. For providing
better results, transformed polynomial curves were proposed by Daniel Brassard by
making the inlet and outlet radius of the contraction proportional to their area [13].
The transformed polynomial profiles were obtained by varying the value of poly-
nomial according to a proposed variable ‘α’. An analysis of transformed fifth-order
polynomial profile showed better results compared to untransformed polynomial
curve design [14]. In this paper, the analysis of transformed sixth-order polynomial
is evaluated by providing various values for ‘α’ in the following equation:
1 1 1 α
y = η Hi α − Heα + Heα (4)
where, η is the sixth-order polynomial designed by Sargison et al. for the contraction;
‘α’ is some function of ζ (where ζ = x/l); ζ is the dimensionless distance from the
contraction inlet, with ζ = 1 at the contraction outlet and ζ = 0 at the inlet of the
contraction.
By judicious selection of ‘α’ (either as a constant or as a function of ζ ), the radius
at the inlet and radius ant the outlet of the profile can be made proportional to their
area. The values of α considered for the present analysis are 2, 1.5, 0.9, 0.7, 0.5, sine
function of ζ , the quadratic function of ζ . Figure 1 shows the profiles generated by
giving different values of ‘α’.
136 R. Lakshman and R. Basak
4 Numerical Setup
δu δv
+ =0 (5)
δx δy
δu δu δP δ δ
u +v =− + μ∇ 2 u + −ρu 2 + (−ρu v ) (6)
δx δy δx δx δy
δv δv δP δ δ
u +v =− + μ∇ 2 v + (−ρv2 ) + (−ρu v ) (7)
δx δy δx δx δy
After creating structural grids for the geometry, areas adjacent to the walls are pro-
vided with very fine grids. SnappyHexMesh utility in OpenFOAM is used for mesh-
ing the models. The model has meshed into hex-dominant meshes having 202,140
elements with 100,483 nodes (Fig. 2). A grid convergence study was also conducted
to eliminate spatial error.
Analysis of Transformed Sixth-Order Polynomial … 137
Fig. 2 a Geometrical model of wind tunnel contraction along with the test section. b Hex-dominant
mesh created using SnappyHexMesh utility. c Detailed mesh view along the corner of the geometry
The geometrical model involves the three-dimensional contraction along with the test
section of 0.52 length. The dimension details of the geometry are given in Table 1.
The extension of the geometrical model up to the test section improves the stability
of the computation as well as enhances the flow uniformity. Figure 2 shows the
structure and the mesh grid generated for the geometrical model. The minimum y+
Table 2 Details of the grids selected for the grid convergence study
Grid Type Refinement level No of cells No of faces No of points
Coarse 1 20,034 61,980 22,038
Medium 2 51,091 168,516 66,459
Fine 3 202,140 676,034 272,040
Very fine 4 700,077 2,393,935 995,022
value for the geometry created for the computation was below 5. The pressure at the
outlet and velocity at the inlet are used as boundary conditions. For the numerical
simulation, the residual value of 10−5 was taken as the convergence criterion. A
grid independence study was done to estimate discretization errors and to finalize
the satisfactory mesh size. With the help of SnappyHexMesh utility in OpenFOAM,
four sets of grids were generated for the grid independence study.
Four grid levels were selected (coarse, medium, fine, and very fine) for the grid con-
vergence study. The grid level was changed using SnappyHexMesh utility available
on OpenFOAM by changing the refinement level from 1 to 4. The details about the
number of cells, faces, and points in each refined grids are shown in Table 2. Velocity
U = (2.1, 0, 0) m/s, the turbulent length scale l = 1 mm and the turbulence intensity
I = 2% was given at the domain inlet as the boundary condition. A constant zero-
pressure condition is given as the outlet boundary condition. Non-slip wall conditions
(|U| = 0 m/s and δP/δn = 0) is given to the side and upper walls. The spatial error
can be eliminated by selecting the grid with the least difference in result compared
to the subsequent level. By doing so the solution will be depending on the physics
and boundary condition but not on the type of grid applied [16].
In this study, the velocity (U) and the coefficient of the pressure value (CP) along
the contraction were selected as the criterion for the study. The results obtained were
plotted and was shown in Figs. 3 and 4.
From the result, it was found out that the difference in results between very fine and
fine grids is much inferior compared to the difference in results between medium and
the fine grid. Therefore, the fine grid with 202,140 cells is selected for the numerical
simulation in the current study.
Fig. 3 a Velocity variation along the axis of contraction for various grids selected. b Velocity
variation along the outlet zone of contraction
Fig. 4 a Coefficient of pressure variation along the axis of contraction for various grids selected.
b Coefficient of pressure variation along the outlet zone of contraction
Fig. 5 Comparison of the velocity profile obtained at x = 0.5 m from the test section inlet
In this paper, the dimensions of contraction and test section used for the validation
is taken for the computation. In order to obtain steady condition flow, contraction
along with the test section was considered for the geometry. The SnappyHexMesh
utility is used for meshing the geometry. Figure 2 shows the mesh grid generated.
The same boundary condition which is used for validation is applied for the analysis
(velocity at inlet = 2.1 m/s and outlet pressure is equated to atmospheric pressure).
The boundary layer separation can occur if there is a reversal in flow direction by
the section of the boundary layer nearer to the wall causes. Wall shear stress is an
important parameter to find the presence of separation. The points where the value
of wall shear stress becomes zero or negative suggests a possibility of separation at
that point. Shear stress distribution along the wall was plotted for each profile and
found that none of them are subjected to separation. The wall shear stress distribution
for all the contraction profiles is shown in Fig. 6. It is found that all the models are
having a positive value for wall shear stress which proves the absence of separation.
The wall pressure coefficients along the contraction for all the transformed and
untransformed profiles are shown in Fig. 7. This dimensionless parameter coefficient
of pressure, C was used to normalize the data. There is a risk of separation if the
pressure gradient is high enough. If there is a constant pressure region on the CP
Analysis of Transformed Sixth-Order Polynomial … 141
Fig. 6 Wall shear stress contour plot for the profiles when a α = 0.5, b α = 0.7, c α = 1.5, d α =
2, e α = quadratic function of ζ , f α = sine function of ζ , g untransformed
Fig. 7 a Pressure coefficient along the contraction, b magnified view at the outlet of contraction
graph in this adverse pressure gradient region, it indicates the presence of separation.
The expression for finding Cp is as shown below.
(P − P∞ )
Cp =
0.5ρU 2
where P is the static pressure measured along the wall, P∞ is the static pressure at
the outlet of contraction and U is the mean value of the velocity measured at the
contraction outlet.
From the figure, the adverse pressure gradient is found at the inlet and outlet
portion of the contraction. The outlet adverse pressure gradient is found to be more
significant compared to the inlet. From the comparison, it is clear that the adverse
pressure is more when the alpha value is less and a lesser adverse pressure gradient
142 R. Lakshman and R. Basak
is seen for the profile having higher alpha value (α = 1.5 and α = 2). On comparing
Quadratic function and sine function, the quadratic function is having superior results.
The prime objective of the design of contraction is to improve the uniformity of
flow inside the test section. Figure 8 shows the velocity profile at the mid vertical
plane of the working section. The same graph was used to analyse the flow uniformity
along the test section. Due to the growth in turbulence intensity, flow uniformity is
found to be disturbed adjacent to the walls. It is seen that the flow uniformity is getting
better with the increase in the value of ‘α’. It is also detected that the contraction with
‘α = quadratic function of ζ ’ hold better uniformity than other contraction profiles.
Figure 9 shows the turbulence intensity profile for the various profiles (both trans-
formed and untransformed). The turbulence intensity value is very much depended
on the contraction profile and contraction ratio. If the turbulence intensity in the test
section is very high, it may generate unfavorable conditions and measurements. So
the value of turbulence intensity inside the test section must be kept as low as possi-
ble. From the figure, it is clear that the value of turbulence is falling with an increase
Fig. 8 a Velocity profile at the contraction outlet along the vertical plane, b magnified the view of
the velocity profile at the contraction outlet along the vertical plane
Fig. 9 a Turbulence intensity at the contraction outlet. b Turbulence intensity at the mid-vertical
plane of the test section
Analysis of Transformed Sixth-Order Polynomial … 143
Table 3 The average turbulence intensities and uniformity at the outlet of contraction and test
section midplane for various values of α
Value of α Average turbulence intensity (%) Flow uniformity (%)
At contraction At test section At contraction At test section
outlet midplane outlet midplane
0.5 6.735 6.368 86.210 88.403
0.7 6.077 5.958 87.332 88.555
1.5 5.441 5.352 87.224 88.246
2 5.255 5.225 87.287 88.163
The quadratic 5.716 5.628 87.454 88.428
function of ζ
Sine function 5.652 5.580 87.177 88.321
of ζ
Untransformed 5.426 5.439 87.227 88.254
6 Conclusion
The proportionality of inlet and outlet radius with respect to their cross-sectional
area is one of the key design criteria for the contraction of the wind tunnel. While
analysing the sixth-order polynomial contraction proposed by Sargison et al., it was
found that the inlet radius and outlet radius are not having much difference. This
can be resolved by using a transformed polynomial curve using the Eq. (2). Both
the normal and transformed curves were analysed based on uniformity of flow, tur-
bulence intensity, and adverse pressure gradient by using open source CFD tool
OpenFOAM. Experimental data were taken for validation of results. The results
show the transformed contraction is having better results compared to the untrans-
formed polynomial profile. As the value of alpha is increasing it was found that the
uniformity and turbulence intensity is getting better. This analytical study thus proves
the significance of using transformed curves for the design of the contraction of the
wind tunnel.
Although this analysis (using the k-ε model turbulence model) found that the
transformed polynomial curves are performing better compared to untransformed
144 R. Lakshman and R. Basak
References
1. Mehta RD, Bradshaw P (1979) Design rules for small low speed wind tunnels. Aeronaut J
83(827):443–453
2. Dolan DS, Zepeda D, Taufik T (2011) Development of wind tunnel for laboratory wind turbine
testing. In: North American power symposium (NAPS), pp 1–5. IEEE
3. Panda MK, Samanta AK (2016) Design of low cost open circuit wind tunnel—a case study.
Indian J Sci Technol 9(30)
4. Cattafesta L, Bahr C, Mathew J (2010) Fundamentals of wind-tunnel design. Encycl Aerosp
Eng
5. Rouse H, Hassan MM (1949) Cavitation-free inlets and contractions. Mech Eng 71(3):213–216
6. Bell JH, Mehta RD (1988) Contraction design for small low-speed wind tunnels
7. Mikhail MN (1979) Optimum design of wind tunnel contractions. AIAA J 17(5):471–477
8. Wolf T (1995) Design of a variable contraction for a full-scale automotive wind tunnel. J Wind
Eng Ind Aerodyn 56(1):1–21
9. Lastra MR, Oro JMF, Vega MG, Marigorta EB, Morros CS (2013) Novel design and experi-
mental validation of a contraction nozzle for aerodynamic measurements in a subsonic wind
tunnel. J Wind Eng Ind Aerodyn 118:35–43
10. Hoghooghi H, Ahmadabadi MN, Manshadi MD (2016) Optimization of a subsonic wind tunnel
nozzle with low contraction ratio via ball-spine inverse design method. J Mech Sci Technol
30(5):2059–2067
11. Abdelhamed AS, Yassen YES, ElSakka MM (2015) Design optimization of three dimensional
geometry of wind tunnel contraction. Ain Shams Eng J 6(1):281–288
12. Sargison JE, Walker GJ, Rossi R (2004) Design and calibration of a wind tunnel with a
two dimensional contraction. In: Australasian fluid mechanics conference, The University of
Sydney, Australia
13. Brassard D, Ferchichi M (2005) Transformation of a polynomial for a contraction wall profile.
Trans ASME I J Fluids Eng 127(1):183–185
14. Lakshman R, Basak R (2018) Analysis of transformed fifth order polynomial curve for the
contraction of wind tunnel by using OpenFOAM. In: IOP conference series: materials science
and engineering, vol 377, no 1, p 012048. IOP Publishing
15. OpenFOAM (2015) The open source CFD toolbox, User guide, Version 3.0.1
16. Anderson JD, Wendt J (1995) Computational fluid dynamics, vol 206. McGraw-Hill, New York
Fatigue Life Assessment of an Existing
Railway Bridge in India Incorporating
Uncertainty
Abstract Fatigue Life Assessment of an existing railway steel truss bridge near
Dumdum, Kolkata is presented. The deterministic approach using the Palmgren–
Miner rule and probabilistic approach by (i) available probabilistic formulations and
(ii) direct Monte Carlo Simulation (MCS) have been adopted in the study. The bridge
is unique of its type since the movement of traffic is transverse to the main span of
the truss. The truss bridge is modeled in STAADPro.V8i SS5 with a moving load of
(i) coal rakes, (ii) Passenger rakes, and (iii) EMU cars. The train running schedule,
composition of train and loading is obtained from the Indian Railway. After analysis,
the stress range time spectra is obtained and the reservoir method is applied to evaluate
the damage. Also, the probability of fatigue failure of the bridge is estimated which
indicates a risk of 77% damage of the critical members after 80 years. It is observed
that fatigue failure is supposed to occur in between 80 and 90 years of service. The
probabilistic approach with direct MCS yields the most conservative prediction.
1 Introduction
The Indian Railway (IR) in its vast network is having over one lakh bridges and most
of them are steel bridges. The assessment of fatigue life of such existing bridges is
a major issue in the perspective of ever-increasing traffic load (e.g., fatigue failure
of Jubilee bridge in Bandel, West Bengal). There have been lots of studies on the
assessment of fatigue life in deterministic [1] and probabilistic format [2] in the past
few decades. Solimon et al. [3] predicted the life of existing fatigue prone steel bridges
by integrating Structural Health Monitoring (SHM) data by probabilistic bilinear S-
N approach. The effect of changing the value slope of AASTHO S-N lines below
the constant amplitude fatigue threshold on the fatigue life is investigated through a
parametric study. Krejsa [4] described methods for probabilistic assessment of the
reliability of steel structures and bridges that are exposed to cyclic loads taking into
account the propagation of fatigue cracks from surface. Ye et al. [5] presented a
critical review of the fatigue life assessment of railway bridges. Lee and Cho [6]
presented a new approach for probabilistic fatigue life prediction for bridges using
finite element model updating based on SHM data. Ding and Chen [7] evaluated
fatigue damage using closed-form spectral methods considering narrowband and
Gaussian response characteristics. After a thorough literature survey, scarcity is noted
regarding the fatigue life assessment of existing railway bridges in India. Thus, in this
study, the fatigue life assessment of an existing railway trussed bridge near Dumdum,
Kolkata is accomplished by (i) deterministic approach using Palmgren-Miner rule
[8] using IS:800-2007 [9] (Approach I), (ii) Monte Carlo Simulation (MCS) on
Approach I to incorporate uncertainty in load and parameters (Approach II), (iii)
Probabilistic approach as per formulation of Crandall and Mark [10] (Approach III),
and (iv) Probabilistic approach as per Wirscheng [2]. The bridge is unique of its type
since the movement of traffic is transverse to the main span of the truss. The truss
bridge is modeled in STAAD.Pro.v8iSS5 [11] with moving load of (i) coal rakes,
(ii) Passenger rakes, and (iii) EMU cars. The train running schedule, composition of
train and loading is obtained from the IR. After analysis, the stress range time spectra
is obtained and the reservoir method is applied to evaluate damage by approach I.
The uncertainty in loading and strength parameters are incorporated by the MCS in
approach II. In approach III, the expectation of damage is estimated through zero-
crossing rate of random stress assuming Rayleigh distribution for stress range history.
In approach IV the probabilistic fatigue life is assessed assuming the stress range
history to be lognormal. Also, the probability of failure of the bridge is evaluated
following [12] (approach V) considering the endurance strength of the material. In
Sect. 2, these four approaches are presented in brief followed by the modeling of the
bridge in Sect. 3 and results and discussion in Sect. 4.
2.1 Approach I
This approach is based on the Palmgren–Miner’s rule and uses the S-N curve of IS:
800-2007 [9]. The calculation of the remaining fatigue life is generally manifested
through the calculation of the damage accumulation. The task is to find the year of
operation when Damage (D) exceeds unity. This method, based on the S-N approach,
Fatigue Life Assessment of an Existing … 147
is called the linear Palmgren–Miner’s rule for damage accumulation and is expressed
by the form
ni
D= (1)
Ni
where n i is the number of cycles occurring at the stress range spectrum for ith type
train. Ni is the number of cycles corresponding to a particular fatigue strength for the
induced stress range magnitude, σi by the ith type train and obtained from the S-N
curve. The standard S-N curve can be expressed in the form of Basquin equation as
N Sb = C (2)
where N is the number of stress cycles to failure at a constant amplitude stress range
S, and C and b are the material parameters, termed as the fatigue strength coefficient
and the fatigue strength exponent, respectively. The ni is obtained from the stress
range spectrum using Reservoir method following the approach of Lee and Noori
[13]. The traffic flow data as in 2017 through the IR is obtained as in Table 1. The
extrapolated cumulative number of train run in upcoming years is presented in Fig. 1,
which is based on the trend and perspective of growth in the IR in the past few years.
700000
600000
y = 7300x - 1E+07
500000
number of train
400000
y = 4663.5x - 9E+06
300000
200000
y = 1812x - 4E+06
100000
year
Goods Train Passenger Train
Emu Cars Linear (Goods Train)
Linear (Passenger Train) Linear (Emu Cars)
2.2 Approach II
In this approach, the involved parameters, viz. wheel loading of different trains,
yearly number of trains, dynamic magnification factor, S-N curve parameters b and
C are assumed to be uncertain with the lognormal distribution. The Coefficient of
Variation (COV) for wheel load for goods, passenger, and EMU train are assumed
as 36%, 30% and 25% based on statistical analysis. The COV for the number of
trains, dynamic magnification factor, b and C are taken as 15%, 10%, 10%, and
10%, respectively. The uncertainty is incorporated through a direct Monte Carlo
Simulation (MCS). The MCS converges at around 1.2 lakhs simulation.
Based on Crandall and Marks [10] approach, if the stress history follows the normal
distribution, the peak follows Rayleigh’s distribution as
S S2
f (S) = 2 exp − 2 (3)
σs 2σs
ωT √ α b
E(D) = 2σ S 1 + (4)
2πC 2
where ω the frequency of load is cycle variation and T is expected fatigue damage
time.
2.4 Approach IV
This approach [2] is similar to approach I. But, in place of the MCS, second-moment
reliability method is used. The approach assumes that the involved parameters are
lognormal. This approach attempts to find after how many years of operation the
fatigue reliability (or probability of failure) becomes less than the acceptable limit
(or more than the safety threshold). The probability of failure is estimated by p f =
φ(−β), where, φ is the CDF of the standard normal distribution function and β is
safety index. Then, following [2]
ln Ñ N s
β= (5)
σlnN
Fatigue Life Assessment of an Existing … 149
where Ns is the intended service life and the tilde denote median values.
C̃ D
Ñ = σlnN = ln 1 + VC2 1 + VD2 1 + VB2 (6)
B b Sb
e
2.5 Approach V
If the induced repeated stress is below the endurance strength of the material, the-
oretically the component will not fail no matter how many cycles are repeated. In
this approach, the endurance strength of a material in fatigue is estimated following
Mischke [12]. The endurance strength is given by S e as
Se = K a K b K c K d K e K f ∅Sut (7)
The present study focuses on the fatigue life assessment of the Dumdum bridge in
Kolkata. The plan view of the bridge is shown in Fig. 2. The length of the bridge
along train movement is 6 m and the bridge span is 30 m. The height of the truss from
the deck is 6 m. The direction of the traffic is transverse to the bridge span direction.
One typical bridge is considered for analysis and is modeled in STAADPro Select
Series 6. The moving load as per the IR specification of the train configuration is
applied on the bridge (Fig. 3). The load data of three kinds of railway goods rakes is
considered to act as a moving load over the span of the bridge. The specification of
a load of each rake are as follows:
(i) Taken over coal rakes (Composition − engine + 55 BOXN, Rake Length =
624.15 m, Total Weight = 5140 t);
150 M. Chanda et al.
A typical stress spectrum obtained after the STAADPro analysis is shown in Fig. 4.
It will provide ni by the reservoir method. The maximum axial stress in the truss
member, effective stress (including dynamic magnification factor) and the number
Fatigue Life Assessment of an Existing … 151
300
250
Stress Range, MPa
200
150
100
50
0
1
7
13
19
25
31
37
43
49
55
61
67
73
79
85
91
97
103
109
115
121
127
133
139
145
151
Number of moving load generation cases
of cycles to failure for each category of train (N i ) is shown in Table 2. Then, by using
Eqs. (1) and (2) the fatigue life estimated through Approach I, deterministically. The
bridge is in operation from the year 2017. It is observed that the fatigue life of the
bridge is 85 years.
Now, the uncertainty information on the involved parameters (see Sect. 2.2) is
incorporated by the direct MCS and the probabilistic fatigue life is estimated as
80 years by Approach II. In approach III, assuming S-N curve parameters as of
approach I, σ S as of approach II the probabilistic fatigue life is estimated by Eq. (4)
as 90 years. The trend of accumulated damage over operation year by approach III is
shown in Fig. 5. Since the extrapolated growth rate model of the train is considered
as linear in Fig. 1, the trend of accumulated damage over the operation year is also
linear. The probabilistic fatigue life is further assessed following Approach IV. The
variation of Probability of failure by Eqs. (5) and (6) over the operation year is
presented in Fig. 6. It can be observed that the truss attains 77% failure after 80 years
of operation.
The endurance strength of the member material is estimated in Approach V
by Eq. (7). The mean values of the parameters are K a = 0.09106, K b = 1,
K c = 0.774 (1, 0.163), K d = 1, K e = 0.5722, K f = 1.0, ∅ = 0.505 (1, 0.146)
152 M. Chanda et al.
Damage
Operation year
Fig. 5 Accumulated damage over operation year by Approach III
100
80
Probability of failure
60
40
20
0
0 20 40 60 80 100 120
-20
year of operation
Fig. 6 Probability of failure by Approach IV
following equations mentioned in Mischke [12] and based on the fabrication of the
considered bridge. The uncertainty is incorporated in the parameters as mentioned
in Sect. 2.5 by the direct MCS. The MCS converges at 10,000 iterations yielding a
probability of failure of most critical as 13%. In a nutshell, the fatigue life obtained
by various approaches is presented in Table 3. It may be observed that the MCS
based probabilistic fatigue life approach yields the most conservative result. How-
ever, predictions by all the approaches are in close conformity yielding the range of
80–90 years, which further indicate acceptability of all the approaches and accuracy
of the assumptions. In the most conservative sense the bridge is expected to exhaust
its fatigue life in the year 2097 (Approach II) and in optimistic sense the same is in
the year 2107 (Approach III).
Fatigue Life Assessment of an Existing … 153
However, the main limitation of the study is its assumed linear extrapolated traffic
growth rate data, which is supposed to vary in the future. Though uncertainty on this
extrapolated data is duly incorporated in the study, there is always a chance of higher
deviation than the assumed COV.
5 Conclusions
Fatigue Life Assessment of an existing railway bridge of India, viz. Dumdum bridge,
Kolkata is presented. The fatigue Life is investigated both by the deterministic and
probabilistic approaches. The bridge is modeled in STAAD Pro and analyzed for
moving load. Uncertainty in various parameters is considered through available prob-
ability formulation and by the direct MCS. All the approaches yield similar results,
with approach II (by the MCS) predicting the most conservative assessment. The
expected fatigue life of the trussed bridge is in between 80 and 90 years. As per
approach II, the fatigue life of the bridge is supposed to be exhausted in 2097. It is
observed from the Figure that there is a 77% chance of fatigue failure of the critical
members at 80 years. The approach V estimates a 13% probability of failure of the
most critical member in consideration to the endurance strength.
Acknowledgements The authors gratefully acknowledge the Indian Railway for sharing the infor-
mation on the considered bridge, loading on the bridge, train schedule, and other uncertainty
information.
References
1. Wang X, Sayed AM, Wu Z (2015) Modeling of the flexural fatigue capacity of RC beams
strengthened with FRP sheets based on finite-element simulation. J Struct Eng ASCE
141(8):04014189-1-14
2. Wirsching PH (2006) Application of reliability methods to fatigue analysis and design. In:
Halder A (ed) Recent developments in reliability based civil engineering. World Scientific,
USA, pp 125–139
3. Solimon M, Frangpool DM, Kown K (2013) Fatigue assessment and service life prediction of
existing steel bridges by integrating SHM into a probabilistic bilinear S-N approach. ASCE J
Struc Eng 139(10):1728–1740
154 M. Chanda et al.
4. Krejsa N (2014) Probabilistic reliability assessment of steel structures exposed to fatigue VSB,
Technical University Ostrava, Czech Republic
5. Ye XW, Su YH, Han JP (2014) A state-of-the-art review on fatigue life assessment of steel
bridges. Math Problems Eng. Article iD 956473
6. Lee YJ, Cho S (2015) Probabilistic fatigue life prediction for bridges using system reliability
analysis and SHM-based finite element model updating. In: 12th international conference on
applications of statistics and probability in civil engineering, ICASP12 Vancouver, Canada,
12–15 July 2015
7. Ding J, Chen X (2015) Fatigue damage evaluation of broad-band Gaussian and non-Gaussian
wind load effects by a spectral method. Probab Eng Mech 41:139–154
8. Ciaverella M, Antuono PD, Demelio GP (2017) A simple finding on variable amplitude
(Gassner) fatigue SN curves obtained using Miner’s rule for unnotched or notched specimen.
Eng Fract Mech 176:178–185
9. IS: 800-2007. Indian Standard. General Construction In. Steel—code of practice (Third
Revision), BIS, India
10. Crandall SH, Mark WD (1963) Random vibrations in mechanical systems. Elsevier
11. STAAD.Pro, Select Series 6. Bentley Inc
12. Mishke CR (1987) Prediction of stochastic endurance strength. ASME J Vib Acous Stress Rel
Des 109:113–122
13. Lee NW, Noori M (2016) A framework for fatigue reliability assessment of orthotropic bridge
decks under stochastic fatigue truck loads. In: Wang HW, Li J, Zhang J, Chen JB (eds) 6th
Asian-Pacific symposium on structural reliability and its applications (APSSRA 2016), 28–30
May 2016, Shanghai, China
Numerical Simulation of Acoustic
Emission Waveforms Generated
by Tension and Shear Cracks in RCC
Beams
Abstract The objective of this paper is to model acoustic emission (AE) sources
and examine the propagation characteristics of AE waveforms generated by these
sources in—concrete structures using finite element (FE) simulation. In order to
model AE sources in the form of tensile and shear cracks, a new simulation technique
is developed, where the sudden release of nodes is introduced at the crack face on a
stressed concrete FE model. In this context, two different concrete models are used
for the study, namely the concrete half-space model and the concrete beam model.
The resulting AE waveforms are analysed using the wave speed and scalogram to
classify various types of AE sources. It is shown that due to a tensile crack at the
surface, S-waves and surface (or Rayleigh) waves are formed, whereas, a shear crack
at the surface, predominantly produces S-waves. On the other hand, body waves (P-
and S- waves) are predominantly formed for internal tensile as well as shear cracks.
The developed simulation technique and signal processing tools are expected to be
helpful for the direct correlation of AE waveforms with the recorded signal and
identification of various AE sources in an experiment.
1 Introduction
This method demands the ability to record and store as many signals as possible along
their waveform so that they can be converted from analog to digital (A/D) signals.
These requirements can be easily met by newly developed high capability sensors as
well as multichannel acquisition systems. With the advent of increased computational
power and greater availability of software based on finite element methods, several
studies have been done on this area, which proves that finite element (FE) modeling
has a major role in simulating elastic wave propagation associated with structural
vibrations, acoustic phenomena and ultrasound problems [5]. FEM has been used
to generate elastic waves from fatigue cracks in a large isotropic plate-like structure
to produce the quantitative analysis. The simulations on the two-dimensional (2D)
model showed that fatigue cracks can be introduced in plates in a realistic way by
the sudden release of nodes in finite element mesh [6]. However, there are limited
literature presenting quantitative results for materials like concrete [7, 8].
In this study, the AE sources are modelled in the form of tensile and shear cracks,
with the sudden release of nodes at the crack face on a stressed concrete FE model in
an effort to examine the propagation characteristics of the waveforms generated by
these sources in the concrete structures using finite element (FE) simulation. The aim
is to develop a direct correlation of AE waveforms with the recorded signal and the
identification of various AE sources in an experiment conducted. Towards this, two
different concrete models are used for the study, namely concrete half-space model
and concrete beam model. The resulting AE waveforms are analysed using the wave
speed and scalogram to classify various types of AE sources.
2 Methodology
σ = σm + σn
For a crack that occurs suddenly, configuration m is a static problem while con-
figuration n is a dynamic one as shown in Fig. 1. Since we are interested in the events
after the formation of a crack, configuration n is our main concern. So we could use
a coarser mesh for the static part (configuration m) and finer mesh for the dynamic
part (configuration n). This method is also effective for studying different modes
158 A. Roy et al.
of fracture separately, by applying the required stress at the crack face as shown in
configuration n.
This method can be easily adapted for modelling different modes of fracture. In
this study, two modes of fracture are modelled using this technique. Mode-I (Tensile)
and Mode-II (Shear). In Mode-I, pure tensile stress is applied at the surface, and for
Mode-II, pure shear is applied. For tensile crack, the approximate flexural strength
of concrete (IS 456) is applied at the crack surface.
σ c = 0.7 f ck
For shear crack, the maximum shear strength for M20 concrete (IS 456) was
applied at the crack face.
σ c = 2.8 N/mm2
3 FE Analysis
Fig. 2 Schematic representation of concrete half-space model and concrete beam model with sensor
positions and crack source
the area of concern is near to the crack, so a different type of meshing was adopted
which provides a transition from fine to coarse mesh as we move away from the
centre. It also has a minimal effect of reflection caused by mesh transitions. Fine
mesh is having a mesh size of 0.5 mm and coarse mesh 10 mm. For concrete beam
model, the same quadratic structured mesh with minimized mesh transitions is used.
Loading and support conditions are also varied for every model. Concrete beam
model is created as per the experimental setup. The size and loading conditions are
identical to the flexural strength test specified in IS 516 (1959) [10].
For M20 grade of concrete, static Young’s modulus (E) is 22.36 GPa, Poisson
ratio (υ) is 0.15 and density (ρ) is 2400 kg/m3 . The longitudinal wave velocity is
3136.4 m/s, whereas the transverse wave velocity is 2012.6 m/s.
E(1 − υ)
VL =
ρ(1 + υ)(1 − 2υ)
E
VT =
2ρ(1 + υ)
160 A. Roy et al.
Pencil-Lead breaks are widely used as a reproducible source for test signals in acous-
tic emission applications. This type of source is also referred to as Hsu–Nielsen
source, based on the works of Hsu and Nielsen. A lead pencil is used to apply pres-
sure on the surface under investigation until the lead breaks. The sudden release of
pressure, at the point of breakage, creates a microscopic displacement and an acoustic
wave associated with that. This is one of the most common artificial sources because
of its simplicity in implementation, in the laboratory as well as the field. The lead
break source was modelled in ABAQUS using the Eqs. 1 and 2 given in [5] and
applied to the half-space model.
where
t(t−τ )
, 0<t<τ
f (x) = τ2
−2.1(t−τ ) (2)
1−e ,t≥τ
From Fig. 4, it is clear that the response from the lead break source is not able
to simulate an actual crack. The amplitude of the lead break signal is having large
variations when compared with that of crack.
Though the pattern is almost the same, it cannot reproduce the higher frequency
components. These are justifiable considering the fact that these waveforms vary
significantly with respect to the type of crack, crack length, thickness, the total
length of plate etc., which can’t be included in a single lead break source. There
are several functions that can resemble a lead break source. In order to obtain the
most appropriate function resembling the crack, we have to adopt a trial and error
procedure, which is not considered in the present work.
45 0 LINE
Sensors
Fig. 6 a Recorded and filtered signal at output nodes, b magnitude scalogram for a tensile crack
received at a distance of 150, 300, 450 mm at 0° angle with the bottom edge of the half-space
and frequency filtered signal along with the respective scalogram at the output node
located at 0° with the bottom edge at 150, 300 and 450 mm due to a tensile crack.
From the scalogram shown in Fig. 6b, it is clear that the major frequency content of
the signal occurs at the time of around 84.5 µs for the sensor located at a distance
of 150 mm from the crack and the velocity is found out to be 1776 m/s. This can
be due to an S-Wave or Rayleigh (R) Wave because they both have almost similar
theoretical velocity. S-Wave has a velocity of 1820 m/s and R-wave has a velocity
of 1779 m/s. The corresponding frequency content is 40 kHz.
Figure 7 shows the variation of recorded and frequency filtered signals, and the
corresponding scalogram for a tensile crack at the output node located at 45° with
the bottom edge at 150, 300 and 450 mm.
As shown in Fig. 7, two distinct peaks are observed in signals. Better resolution
of these peak signals is observed with an increase in distance between crack and
the sensor. In the filtered signal for the first plot, we can notice that the first peak
occurs at 71 µs. Since the crack is located at a distance of 150 mm, the velocity
Numerical Simulation of Acoustic Emission … 163
Fig. 7 a Recorded and filtered signal at output nodes, b magnitude scalogram for a tensile crack
received at a distance of 150, 300, 450 mm at 45° angle with the bottom edge of the half-space
equals to 2987 m/s. The theoretical velocity of the P wave is 3116 m/s. The second
peak occurs at a time of 117 µs, which corresponds to a velocity of 1928 m/s. The
respected frequency contents are 58 kHz for the first peak and 25 kHz for the second
peak. The scalogram of the above waveforms presented in Fig. 7b shows that the
separation of peaks is visible as the distance of the sensors increases. So the first
peak corresponds to P-Wave and the second peak corresponds to S or R-Waves.
Figure 8 shows the variation of recorded and filtered signals for a shear crack for
sensor position of 150 mm, 300 mm and 450 mm, respectively at 0° angle with the
Fig. 8 a Recorded and filtered signal at output nodes, b magnitude scalogram for a shear crack
received at a distance of 150, 300, 450 mm at 0° angle with the bottom edge of the half-space
164 A. Roy et al.
Fig. 9 a Recorded and filtered signal at output nodes, b magnitude scalogram for a shear crack
received at a distance of 150, 300, 450 mm at 45° angle with the bottom edge of the half-space
bottom edge of the half-space along with magnitude scalogram. From Fig. 8a, we
can observe that the first peak occurs at a time of 82.72 µs at a 150 mm sensor, with
a velocity of 1813 m/s. The theoretical surface (R) wave velocities in concrete are
around 1800 m/s. So the first peak corresponds to an S-Wave.
From the scalogram shown in Fig. 8b, it is clear that the maximum value of
frequency is 50 kHz. So from the frequency content its evident that the wave is
S-Wave because Rayleigh wave will have less frequency.
Figure 9a shows the recorded and filtered signals at the output node located at 45°
with the bottom edge at 150, 300 and 450 mm due to a shear crack. The corresponding
scalogram is also presented in Fig. 9a. In the filtered signal, 2 different peaks are
noticeable, which occurs at a time of 69 µs and 123 µs, respectively. Corresponding
velocities are 3074 and 1710 m/s. So we can deduce that the first peak is that of
a P-Wave. We can also notice the separation of peaks as the distance of sensors
increase.
From Fig. 9b, we can see that for the first plot the major frequency content lies
around 65 kHz, which occurs at 90 µs. It is evident that in all the three figures the
maximum frequency value is around 50–65 kHz. So we can infer that for a shear
crack measured at the same surface it will have predominantly S-Wave. But for the
second and third plot we can see a separation of frequencies, which is represented by
a circle. So by comparing waveforms and the scalogram from Fig. 9, we can deduce
that it is the shear wave that is separating from the P-Wave. So, in this case, S-Wave
and P-Wave are having comparable frequencies.
From the above results for half-space model for tensile and shear crack, we can
conclude the following:
1. Waves received at sensors places at 0° angle with the bottom edge, shows that
surface waves are predominant for both tensile and shear cracks.
Numerical Simulation of Acoustic Emission … 165
2. For Surface cracks, waves received at sensors placed at 45° angle with the bottom
edge, both P- and S-waves are obtained with distinct frequency peaks separated
as longer sensor distance.
The displacement time history of nodes at six different locations are recorded under
tensile and shear cracks. The nodes are located at a distance of 50, 100 and 150 mm
from the middle of the beam. Three of them are at the bottom and the other three
at the top. Similar to half-space two different modes of cracks are modelled: tensile
and shear.
Figure 10 shows the recorded signal for a surface tensile crack received at the bottom
sensor located at a distance of 100 mm from the middle of the beam. The recorded
signal is similar to that of the half-space in the initial portion, and later on, the
reflections from boundaries affect the signal. Hence only the first 100 µs of the
recorded signals are considered in each case. Recorded signals are filtered with an
elliptical high pass filter above 20 kHz. The waveform is similar to that of half-
space with tensile crack signals received at sensors placed at 0° angle with the
bottom edge. The high-frequency peak at 60 µs with a velocity of 1666 m/s with the
major frequency contents at 40 kHz, indicates that the signal contains predominantly
S-Wave and R-Wave.
Fig. 10 a Recorded and filtered signal at a bottom output node located at a distance of 100 m from
a surface tensile crack, b magnitude scalogram of the signal
166 A. Roy et al.
Fig. 11 a Recorded and filtered signal at a top output node located at a distance of 100 m from a
surface tensile crack, b magnitude scalogram of the signal
Figure 11 shows the recorded signal for a surface tensile crack received at the top
sensor located at a distance of 100 mm from the middle of the beam. The recorded
waveform received at the top sensor, placed at 141.42 mm from crack has 2 peaks,
which are similar to the recorded waveform for tensile crack signals, received at
the sensors placed at 45° angle with the bottom edge in half-space. The initial peak
occurs at 44 µs, corresponding to a velocity of 3214 m/s and 2nd peak at 60 µs
with a velocity of 2357 m/s with the frequency content of 120 kHz and 60 kHz
respectively. From the scalogram, the Higher frequency content of nearly 120 kHz
corresponding to the first peak is observed along with a lower frequency content of
60 kHz corresponding to the later part of the waveform. So we can infer that the first
peak corresponds to P-Wave and the second peak can be that of an S-Wave or surface
wave.
Figure 12 shows the recorded signal for a surface shear crack received at the bottom
sensor located at a distance of 100 mm from the middle of the beam. The high-
frequency peak at 59 µs with a velocity of 1694 m/s with major frequency content
at 50 kHz indicates that the signal contains predominantly S-Wave and R-Wave.
Figure 13 shows the recorded signal for surface shear crack received at a distance
100 mm from the middle of the beam. Similar to the recorded signal obtained in half-
space, it has 2 peaks. The higher frequency corresponding to the first peak and lower
frequency for the later part of the waveform is observed. The initial peak occurs at
45 µs, corresponding to a velocity of 3142 m/s and 2nd peak at 60 µs with a velocity
of 2357 m/s with the frequency content of 110 kHz and 70–90 kHz respectively. So
we can infer that the first peak corresponds to P-Wave and second can be S-Wave or
surface wave.
Numerical Simulation of Acoustic Emission … 167
Fig. 12 a Recorded and filtered signal at the bottom output node located at a distance of 100 m
from a surface shear crack, b magnitude scalogram of the signal
Fig. 13 a Recorded and filtered signal at the top output node located at a distance of 100 m from
a surface shear crack, b magnitude scalogram of the signal
Internal crack is also modelled in ABAQUS. The crack with a length of 10 mm,
placed in the middle of the beam as shown in Fig. 14. The nodes above and below
the crack surface are given symmetric boundary conditions (horizontal symmetry).
The sensor is at the same location as that of surface cracks as shown in Fig. 14.
Figure 15 shows the recorded signal of internal tensile crack at the bottom sensor
located at a distance of 100 mm from the middle of the beam. Two distinct peaks are
observed with high-frequency contents. For sensor placed at 111.8 mm, the initial
peak occurs at 37 µs, corresponding to a velocity of 3021 m/s and 2nd peak at 80 µs
168 A. Roy et al.
Fig. 15 a Recorded and filtered signal at the bottom output node located at a distance of 100 m
from an internal tensile crack, b magnitude scalogram of the signal
with a velocity of 1863 m/s with the frequency content of 110 kHz and 80 kHz
respectively. So we can infer that the first peak corresponds to P-Wave and second
can be S-Wave or surface wave.
Similar behaviour of the waves is observed in case of internal shear cracks. It
also shows a similar pattern as the half-space shear crack signal recorded at internal
sensors. Figure 16 shows the recorded signal of internal shear crack at the bottom
sensor located at a distance of 100 mm from the middle of the beam.
Two distinct peaks are observed with high-frequency contents. The initial peak
occurs at 38 µs, corresponding to a velocity of 2942.19 m/s and second peak at
62 µs with a velocity of 1803 m/s with the frequency content of 100 kHz and 80 kHz
respectively. So we can infer that in this case also, the first peak corresponds to
P-Wave and second can be S-Wave or surface wave.
From the above results for 2D concrete beam model for tensile and shear crack,
we can conclude the following,
Numerical Simulation of Acoustic Emission … 169
Fig. 16 a Recorded and filtered signal at the bottom output node located at a distance of 100 m
from an internal shear crack, b magnitude scalogram of the signal
1. For surface cracks, surface waves are dominated if the signals are recorded on
the same surface. For these surface waves, the frequency content of 60–90 kHz
was observed.
2. For surface cracks, P-waves become prominent with higher frequency content
100–120 kHz for the signals received on the other surface of the beam.
3. For Internal cracks, both P-waves and S/R-waves are seen for recorded signals
on the surface.
5 Conclusion
Two different models were used for the study, concrete half-space model and beam
model. The major conclusions gathered from the study are explained below.
1. Due to a tensile crack at the surface, P-Waves, S-Waves, and surface (R) waves
are formed in the beam and half-space. P-Wave is having less amplitude when
compared to S-Waves so they are not clearly visible in the scalogram. Since the
S-Waves and surface waves are having almost the same velocity they can’t be
distinguished in the unfiltered signal.
2. Shear crack at surface predominantly produces S-Waves. Surface waves and
P-Waves are not generated.
3. Body waves (P and S waves) are predominantly formed for internal tensile as
well as shear crack. But the interaction of body waves with the surface causes
the formation of surface waves. Body waves also have significant amplitude and
energy when compared to surface waves.
170 A. Roy et al.
The study has to be extended by analysing different models and conducting several
experimental studies. The model has to be analysed three dimensionally in order to
include the effect of side reflections also.
Acknowledgements The authors want to acknowledge the funding received from the Board of
Research in Nuclear Sciences, Department of Atomic Energy, Government of India, with grant
number 36(2)/15/05/2016-BRNS/36058 (16BRNS0003) with NRFCC, BRNS, for carrying out the
research work presented in this paper.
References
1. Behnia A, Chai HK, Shiotani T (2014) Advanced structural health monitoring of concrete
structures with the aid of acoustic emission. Constr Build Mater 65:282–302
2. Berkovits A, Fang D (1995) Study of fatigue crack characteristics by acoustic emission. Eng
Fracture Mech 51(3):401–416
3. Ohtsu M (1996) The history and development of acoustic emission in concrete engineering.
Mag Concr Res 48(177):321–330
4. Ohno K, Ohtsu M (2010) Crack classification in concrete based on acoustic emission. Constr
Build Mater 24(12):2339–2346
5. Hill R, Forsyth S, Macey P (2004) Finite element modelling of ultrasound, with reference to
transducers and AE waves. Ultrasonics 42(19):253–258
6. Lee C, Drinkwater B, Friswell M, Scholey J, Wilcox P, Wisnom M (2006) Acoustic emission
during fatigue crack growth in aluminium plates. In: 9th European conference on NDT, ECNDT
2006, Berlin
7. Sause M (2011) Investigation of pencil-lead breaks as acoustic emission sources. J Acoust
Emiss 29:184–196
8. Banerjee S, Mal AK (2005) Acoustic emission waveform simulation in multilayered compos-
ites. J Strain Anal Eng Des 40(1):25–32
9. Kumar P (2009) Elements of fracture mechanics. McGraw Hill Education (India) Pvt. Ltd
10. Indian Standard. Method of tests for strength of concrete. IS 516: 1959, Edition 1.2 (07-1991)
Applicability of Tricycle Modelling
in the Simulation of Aircraft Steering
System
Nomenclature
1 Introduction
Landing gear systems are used in aircraft for energy absorption during landing and
for aircraft ground manoeuvring. Tricycle arrangement of landing gear is commonly
used in most of the aircraft because of its stability and less expensive design. Steering
system is one of the major subsystems associated with the aircraft landing gear
systems for steering operation on the ground using landing gear along with the rudder
steering. The rudder steering is used at higher speeds and landing gear steering is
used during taxiing or any other manoeuvring at lower speeds when the aircraft is on
the ground. Aircraft Steering is done by using Nose Wheel Steering (Nose Gear) in
most of the aircrafts. All-wheel steering (Nose + Main Gear) is used in large aircraft
during sharp turns. In this paper, simulation studies are carried out using the different
ground dynamic models to understand the steering response of the aircraft steering
system for the given pilot inputs.
2 Literature Survey
In the available literature [1–6], more emphasis is given on bicycle modelling for
the aircraft ground dynamics model. Biannic et al. [1] developed a simplified Linear
Fractional Transformation Model for predicting the aircraft-on-ground forces to be
used for the development of robust ground control systems. Pouly [2], Pouly et al.
[3] used bicycle methodology to simulate aircraft steering system. Ross et al. [4]
developed high-fidelity non-linear model for lateral control of aircraft-on-ground
using bicycle methodology. Chen et al. [5] proposed a non-linear control model with
kinematic constraints of nose wheel cornering and the tyre side-slip effects, to predict
the steering response of the aircraft. Pavan et al. [6] predicted the steering response
of the given aircraft using bicycle methodology and analysed the hydraulic system
based on the torque output from the aircraft steering simulations.
Though the tricycle modelling is more accurate, very few authors have reported
this method [7, 8]. Duprez et al. [7] developed a feedback linearizing control model
using tricycle method to predict the yaw rate control during aircraft on ground. Chen
et al. [8] developed a dynamical adaptive backstepping controller to address the path
following the control problem of the aircraft on ground. Also, the comparison and
applicability of these modelling methods have not been discussed in the available
literature.
In this paper, mathematical models for tricycle modelling are formulated and
their mathematical relationships are compared with the bicycle model formulations.
Steering response results are obtained from both the models and are compared in
detail for similar inputs.
174 S. Sathish et al.
Aircraft nose wheel steering system includes Nose Landing gear (NLG) model,
Aircraft Ground Dynamics model and Tyre model. Aircraft ground dynamics model
is commonly modelled as bicycle/single-track model as shown in Fig. 3. This two
degree of freedom model simplifies the aircraft model to study the effects of steering
Applicability of Tricycle Modelling … 175
input on the attitude and path of the aircraft. Major assumptions are rigid landing
gear, low slip angles for linearization of tyre model, constant low aircraft velocity and
no roll or pitch movement on the ground. In the bicycle model, both the main landing
gears are assumed as single gear with four tyres (for twin wheel arrangement) and
one nose gear with two tyres.
From Fig. 3, Force balance along the lateral direction and Moment balance about
C.G of aircraft results in equations of motion as [6],
176 S. Sathish et al.
. −1 Vy + r × LNW Vy − r × LMW
Vy = × CSNW × + 2 × CSMW ×
Ma/c Vx Vx
CSNW × δf
+ − [r × Vx ] (1)
Ma/c
⎡ ⎛ ⎞⎤
Vy + r × LNW
CS × L ×
⎢ −1 ⎜ ⎟⎥
⎟⎥ + CSNW × LNW × δf .
NW NW
. Vx
r=⎢
⎣ Izz × ⎜
⎝ Vy − r × LMW ⎠⎦ Izz
−2 × CSMW × LMW ×
Vx
(2)
Equations (1) and (2) are solved to get lateral velocity (vy ) and yaw rate (r) to
predict the lateral resisting force and sideslip on the tyres which are given by
These are linear and coupled equations that can be solved using state-space meth-
ods in MATLAB–Simulink for the unknowns, lateral velocity, Vy and yaw rate, r. Vy
and r are used in calculating side slip using Eqs. (5) and (6). Using sideslip, resistive
torque is calculated as shown in Eq. (20).
.
Izz r = FYNW LNW − FYMWL LMW − FYMWR LMW (8)
Applicability of Tricycle Modelling … 177
Substituting Eqs. (9) to (14) above on Eqs. (7) and (8) and defined in terms of
unknowns as
⎡ ⎛ ⎞⎤
Vy + r × LNW Vy − r × LMW
⎢ −1 ⎜ CS × + CS ×
Vx + r × 2b ⎟ ⎥
NW MWL
. ⎢ ⎜ Vx ⎟⎥
Vy = ⎢ ×⎜ ⎟⎥
⎣ Ma/c ⎝ Vy − r × LMW ⎠⎦
+ CSMWR ×
Vx − r × 2b
CSNW × δf
+ − [r × Vx ] (15)
Ma/c
⎡ ⎛ Vy + r × LNW ⎞⎤
CSNW × LNW × − CSMWL × LMW
. ⎢ −1 ⎜ Vx ⎟⎥
r =⎢ ⎜
⎣ Izz × ⎝ Vy − r × LMW Vy − r × LMW ⎠⎦
⎟⎥
× − CSMWR × LMW ×
Vx + r × 2b Vx − r × 2b
CSNW × LNW × δf
+ (16)
Izz
As seen above, Eqs. (15) and (16) are coupled, non-linear and hence they cannot be
solved using linear state-space methods. Non-linearity arises because of the presence
of unknown term, r in the denominator of sideslip terms. The function block approach
in MATLAB-Simulink is first applied to the bicycle model to validate the approach
with state-space methods and the same is applied to tricycle model.
Torque balance equation [6] of Landing Gear with reference to Fig. 1 is given as
Tnet = Ta − Tr (18)
Ta = A PL L L (19)
As seen from Eq. (20), resistive torque model is depending only on nose wheel
parameters as steering operation in most of the aircraft are done using nose wheel
alone. Hence the controller system is going to depend on this resistive torque model
apart from bicycle or tricycle method for aircraft ground dynamics model.
Applicability of Tricycle Modelling … 179
7 Simulations
Fig. 7 Simulink function blocks method to solve aircraft ground dynamics model
Applicability of Tricycle Modelling … 181
Both the bicycle and tricycle models are simulated with typical aircraft data [6] with
the longitudinal velocity of 6 m/s for a typical input steering profile. Comparison
of various parameters between the bicycle and tricycle model is shown in Table 2.
As expected, there is no difference in nose wheel parameters because of the same
equations prevail for both the models. There is an increase of 5% in slip angle on
the main wheel because of the additional track parameter in the tricycle equations.
In addition, it is found that there is a difference of 3% in lateral force on the main
wheel between the models.
Variation of for lateral force and sideslip on main gear for bicycle and tricycle
modelling method is shown in Fig. 8.
In the sensitivity study 1, both the models are simulated with different steer rates
to understand its influence on the modelling methods as shown in Table 3. Parameters
such as slip angle, lateral force, yaw rate, etc., increase slightly with an increase in
the steering rate of aircraft. For 50% increase in steering rate, the slip angle and
lateral force at the nose wheel increase by 12%.
In the sensitivity study 2, both the models are simulated with different longitudinal
velocities of aircraft to understand its influence on the modelling methods as shown
in Table 3. Also, it is observed that with an increase in the linear velocity, slip angles
Table 2 Comparison
Parameters Bicycle model Tricycle model
between bicycle and tricycle
model Max yaw rate (rad/s) 0.168 0.168
Max lateral velocity (m/s) 0.390 0.390
Slip angle at MWL (deg) 4.60 4.39
Max lateral force at MWL 3872 3700
(N)
Max slip angle at MWR NA 4.82
(deg)
Max lateral force at MWR NA 4000
(N)
Fig. 8 Comparison of lateral force and sideslip between tricycle and bicycle method on MLG
182 S. Sathish et al.
Table 3 Comparison between Bicycle and tricycle model for different steering rate (Study 1)
Parameters Bicycle model Tricycle model Bicycle model Tricycle model
Steering rate (deg/s) 2.5 5
Max yaw rate (rad/s) 0.1685 0.1685 0.1775 0.1775
Max lateral velocity −0.39 −0.39 −0.42 −0.42
(m/s)
Slip angle at MWL 4.6 4.3 4.9 4.7
(deg)
Max lateral force at 3872 3700 4137 3948
MWL (N)
Max slip angle at MWR NA 4.8 NA 5.1
(deg)
Max lateral force at NA 4000 NA 4319
MWR (N)
become more than 5 deg and hence the linear tyre model considered in the simulations
is not applicable. For a 33.3% increase in longitudinal velocity, a 32% increase in
slip angle and lateral force is observed (Table 4).
Table 4 Comparison between bicycle and tricycle model for different longitudinal velocity (Study
2)
Parameters Bicycle Tricycle Bicycle Tricycle Bicycle Tricycle Bicycle Tricycle
model model model model model model model model
Longitudinal 4 6 8 10
velocity
(m/s)
Max yaw 0.094 0.094 0.168 0.168 0.275 0.275 0.442 0.442
rate (rad/s)
Max lateral −0.078 −0.078 −0.390 −0.390 −1.22 −1.22 −3.18 −3.18
velocity
(m/s)
Slip angle at 1.80 1.72 4.60 4.39 9.85 9.28 20.00 18.35
MWL (deg)
Max lateral 1503 1450 3872 3700 8380 7820 18,000 15,400
force at
MWL (N)
Max slip NA 1.85 NA 4.82 NA 10.35 NA 21.25
angle at
MWR (deg)
Max lateral NA 1560 NA 4000 NA 8720 NA 17,900
force at
MWR (N)
Applicability of Tricycle Modelling … 183
9 Conclusions
In the above study, the lateral forces and slip angles on the main wheel are predicted
from bicycle and tricycle modelling methods. It is observed that the bicycle modelling
methodology is 5% less accurate when compared to tricycle method. The steering
actuator response is based on the resistive torque generated on the tyres. When the
nose wheel steering is used, resistive torque is calculated using lateral forces and slip
angles from the nose gear alone. Hence, the bicycle or tricycle modelling method
will not change the steering actuator response. But in tri-wheel steering, resistive
torque is calculated using lateral forces and slip angles from both the gears. If the
bicycle model is considered in the all-wheel steering system, the predicted steering
actuator response will be different from actual because of the difference in the lateral
forces and slip angles of the main gear. Hence, it is concluded that the bicycle model
is sufficient in the simulation of the nose wheel steering system as in many aircraft
but tricycle modelling is required in the simulation of the tri-wheel steering system
as in some wide-body aircraft to predict the right steering response.
Acknowledgements The authors would like to thank “Council of Scientific and Industrial Research
(CSIR)”, Government of India for funding this project under the 12th Five Year Plan. The
authors also like to thank Director, CSIR-National Aerospace Laboratories (NAL), India and Head,
Structural Technologies Division, CSIR-NAL, for their kind support.
References
1. Biannic J-M, Marcos A, Jeanneau M, Roos C (2006) Nonlinear simplified LFT modelling of
an aircraft on ground. In: Proceedings of the 2006 IEEE international conference on control
applications, Munich, pp 2213–2218
2. Pouly G (2009) Analysis and synthesis of advanced control laws for vehicle ground guidance.
PhD thesis, University de Haute Alsace
3. Pouly G, Lauffenburger J-P, Basset M (2009) Reduced order H∞ control design of a nose landing
gear steering system. In: Proceedings of the 12th IFAC symposium on transportation systems,
pp 478–483
4. Roos C, Biannic J-M, Tarbouriech S, Prieur C, Jeanneau M (2010) On-ground aircraft control
design using a parameter-varying anti-windup approach. Aerosp Sci Technol 14:459–471
5. Chen B, Jiao Z, Ge SS (2011) Nonlinear control of aircraft on ground runway keeping. IEEE
Xplore, pp 576–581
6. Pavan MS, Jaidev Vyas J, Balamurugan G (2016) Modelling and simulation of aircraft nose
wheel steering system. IEEE Xplore
7. Duprez J, Mora-Camino F, Villaume F (2004) Control of the aircraft-on-ground lateral motion
during low speed roll and manoeuvers. In: IEEE aerospace conference proceedings, pp 2656–
2666
8. Chen B, Jiao Z, Ge SS (2013) Aircraft-on-ground path following control by dynamical adaptive
backstepping. Chinese Journal of Aeronautics, 26:668–675
Free Vibration and Stress Analysis
of Laminated Box Beam
with and Without Cut-Off
Abstract This paper presents the free vibration and static analysis of composite box
beam using refined beam theory. The structural model based on one-dimensional
(1D) is derived in the Carrera Unified Formulation (CUF) framework. The principle
of virtual displacement has been used along with CUF to formulate the finite ele-
ment arrays in the terms of fundamental nuclei, which either do not depend on the
expansion order or on the class of the beam model. In the present study, the various
composite box beam models with and without cut-off model using different aspect
ratio has been analyzed the free vibration and static analysis. The results of free
vibration analysis are compared to published literature. The present study indicates
the high-level accuracy reached by refined beam models with lower computational
costs than 3D solid elements.
1 Introduction
In modern times, laminated composite beams, plates, and shells are extensively used
in high-speed aircraft, rocket, launch vehicle, aerospace, or civil structures due to
their high specific strength and stiffness, excellent fatigue and corrosion resistance.
In engineering application, many structures are required stiffness-to-weight ratio,
thus the structure made of laminated composite materials represents efficient con-
structive solutions. To analyze the mechanical behavior of such type of structures
are extremely complex process. Although a considerable number of analytical and
numerical methods have been proposed over the years. To analyze the structural
behavior of this type of structure, the beam theories are extensively used. In the case of
modeling of weight-sensitive structures, the one-dimensional (1D) approach is more
significant due to its simplicity and low computational cost and more efficient than
two-dimensional (2D) (plate and shell) and three-dimensional (3D) (solid) elements.
The classical theories that are most employed are proposed by Euler–Bernoulli [1]
and Timoshenko [2], which are also known as first-order shear deformation theories
(FSDTs).
For modeling the composite structure these theories are inadequate. Several higher
order shear deformation theories (HSDTs) are purposed for the composite structure,
which can be classified into two main categories: equivalent single layer (ESL)
and layer-wise (LW) theories. The ESL enables to produce piecewise continuous
displacement and transverse stress in the thickness direction of the laminate structures
[3, 4]. In this approach, the number of unknowns is independent of the number of
layers, for example, first-and second-order shear deformation theories [5, 6]. Third-
order shear deformation theories were purposed for the analysis of beams [7] and
plates [8]. Arya et al. [9] presented a HSDT for the static analysis of laminated
composite beams. Later, Li et al. [10] extended this refined model to study the free
vibration of angle-ply laminated beams. And recently, Sahoo and Singh [11] studied
new inverse zig-zag shear deformation theories for the static analysis of sandwich
laminated plates. Carrera [12] compared different 2D theories to investigate the
effects of the curvature and shear deformation on the buckling and vibrations of
cross-ply laminated shells. Although ESL exhibits many applications in static and
dynamic analysis of composite beams, it results in efficient theories for laminate
structures. The main drawback of the ESL approach is that the continuity of transverse
shear and normal stresses is not always assured. In the domain of LW, a continuous
displacement assumption is considered in each separate layer. In LW approach, the
number of degrees of freedom (DOFs) depends directly on the number of layers. In
the LW approach the required computational cost is more than the ESL approach.
Robin and Reddy [13] used LW theory to develop modeling of thick composites.
Shimpi and Ainapure [14] presented a trigonometric shear deformation theory to
develop modeling of two-layered cross-ply beams. Later, the same theory extended
and utilized [15] for free vibration analysis of cross-ply laminated beams.
The present work has focused on refined theories with generalized displacement
variables for the free vibration and stress analysis of the laminated box beam. For
preliminaries model, the Carrera unified formulations (CUF) are used. CUF formu-
lation has been developed over the last decade for plate/shell models [12, 16–18]
and it has recently been extended for beam modeling [19]. In this study the cross
section of laminated box beam made of orthotropic material with discretized with 16
nine-node elements (16L9). Two types of cross-sectional distribution 16L9 (a) and
16L9 (b) are used for the analysis as shown in Fig. 2. A combination of orthotropic
fiber composite layers is used to be a construction of walls. The objective of this
Free Vibration and Stress Analysis of Laminated Box … 187
study is to develop a laminated box beam model with cut-off and without cut-off
and this model is used for analysis in the form of free vibrations, displacements, and
stresses with the help of refined one-dimensional beam theories. For finite element
formulation, CUF framework is used to prepare a model, which is a hierarchical for-
mulation leading to very accurate and computationally efficient finite element (FE)
models. The laminated box made of two orthotropic layers with same thickness. In
order to demonstrate the effectiveness of the proposed refined elements, the results
in terms of natural frequencies, and displacements, are computed and compared with
the available research literature.
where T stands for the terms in expansion, in Einstein’s generalized notation it stands
for summation, u τ is the displacement vector, and Fτ represents expansion function
to approximate the behavior of the cross section of box beam.
In this work Eq. (1) consists of Lagrange polynomials, which are used to build the
1D higher order models. In this paper, the nine points (L9) cross-sectional polynomial
(Fig. 2) set was adopted and the interpolation functions are given as:
1 2
Fτ = r + rrτ s 2 + ssτ , τ = 1, 3, 5, 7
4
1 1
Fτ = sτ2 s 2 − ssτ 1 − r 2 + rτ2 r 2 + rrτ 1 − s 2 , τ = 2, 4, 6, 8
2 2
Fτ = 1 − r 2 1 − s 2 , τ = 9 (2)
188 R. B. Bharati et al.
where r and s vary from −1 to +1, whereas rτ and sτ are the coordinates of the nine
points whose locations in the natural coordinate frame. The displacement of a L9
element therefore
u x = F1 u x1 + F2 u x2 + F3 u x3 + · · · + F9 u x9
u y = F1 u y1 + F2 u y2 + F3 u y3 + · · · + F9 u y9
u z = F1 u z1 + F2 u z2 + F3 u z3 + · · · + F9 u z9 (3)
σ p = {σzz σx x σx z }T , ε p = {εzz εx x εx z }T
T T
σn = σzy σx y σ yy , εn = εzy εx y ε yy (4)
where the subscripts p and n stands for the terms lying cross section and planes,
respectively.
Hooke’s law and strain–displacement relations are, respectively,
pp ε p + C
σp = C np εn
np ε p + C
σn = C nn εn
εp = Dpu
εn = Dny + Dnp (5)
where
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
∂
0 0 ∂z ∂
0 ∂z 0 0 0 ∂∂y
⎢ ⎥ ⎢ ⎥
D p = ⎣ ∂∂x 0 0 ⎦, Dn A = ⎣ 0 ∂∂x 0 ⎦, Dny = ⎣ ∂∂y 0 0 ⎦ (6)
∂ ∂
∂z
0 ∂x 0 0 0 0 ∂∂y 0
Free Vibration and Stress Analysis of Laminated Box … 189
Box beam is a complex laminated structure, for such type of structure can be
considered constituted by a certain number of straight orthotropic layers, material
coordinate system (1; 2; 3) generally do not coincide with the physical coordinate
system (x; y; z) as shown in Fig. 1. The matrices of the material coefficient of the
generic material k based on the above approach are
⎡ ⎤ ⎡ k ⎤ ⎡ k ⎤
11
C k 12
C k 14
C k
C 16
C k 13
C k
C k
C k
C
15 55 56 35
kpp
C = ⎣C
12
k 22
C k
C24
k ⎦, C k
pn = ⎣ k
C25 26
C k
C23
k ⎦, C k
nn = ⎣ k
C56 66
C k
C k ⎦ (7)
36
k
C14 k
C24 k
C44 k
C45 k
C46 k
C43 k
C35 k
C36 k
C33
For the sake of brevity, shape function and explicit form of the coefficients are
shown in [12]. To handle for arbitrary shaped cross section, classical finite element
technique is adopted here and generalized displacement vector becomes
where Ni (y) is the shape function and qτ i is the nodal displacement vector:
T
qτ i = qu xτ i qu yτ i qu zτ i (9)
The equation of motion can be directly derived from the Principle of Virtual
Displacement (PVD), which states
where δL int , δL ext , δL ine , and δ stands for internal work, external work, inertial work,
and virtual variation, respectively.
With the help of Eqs. (1), (3), (4), and (5), the virtual variation can be written as
where K i jτ s is in the form of a fundamental nucleus for the stiffness matrix and can
be written as follows:
T k
K i jτ s = Il
ij np D p (Fs I ) + C
Dnp (Fτ I ) C nn
k
Dnp (Fs I )
k
+D Tp (Fτ I ) C pp D p (Fs I ) + C
kpn Dnp (Fs I ) dA
T
+ +Il
i j,y kpn Fs dA + I Ay
Dnp (Fτ I ) + D Tp (Fτ I )C
190 R. B. Bharati et al.
k
+ Il
i,y j T
I Ay np D p (Fs I ) + C
Fτ C nn
k
Dnp (Fs I ) dA
+ Il
i,y j,y T
I Ay I Ay Fτ C nn
k
Fs dA (12)
where q̈ stands for nodal acceleration and M i jτ s stands for mass matrix in the form
of fundamental nucleus:
ij
M i jτ s
= Il Fτ ρ k I Fτ dA (16)
T
P = Pu x Pu y Pu z (19)
M q̈ + K q = 0 (20)
In this section, the free vibration analysis of composite box beam is discussed. The
cantilever box beam is prismatic with length L = 762 mm, width b = 24:21 mm,
and height h = 13:46 mm. Each wall of the box beam has a total thickness equal to t
= 0.762 mm was consider in the first numerical example. The length to height ratio,
L/h is assumed equal to 10 and each layer of the structure is made of an orthotropic
material, whose density and mechanical properties along the fiber (L) and transverse
(T ) directions are ρ = 1601 kg/m3 , E L = 142 GPa, E T = 9.8 GPa, GLT = 6 GPa, GTT =
4.83 GPa, ν = 0.5. The box beam made of single and double layers and lamination
schemes are reported in Table 1. The Lagrange Element (LE) consists of 9 four-
noded (9B4) beam elements along the longitudinal axis and various approximations
of the cross-sectional kinematics are assumed shows in Fig. 2 where each rectangle
represents one L9 (Fig. 3) polynomial set. The two 16L9 models are considered
for the free vibration analysis. Natural frequencies of the box beam reported in
Table 2 and the computed results were compared with the finite element solution
provided by [16]. The compared results have good agreement with references. The
natural frequencies related to the mode number for the various aspect ratio and
stacking sequences of 16L9(a) and 16L9(b) models were shown in Fig. 4 and Fig. 5
respectively. From the results given, it should be clear that aspect ratio and stacking
sequences can significantly influence the natural frequencies of the model.
In this section the static analysis of a composite box beam has been discussed, the
box beam was clamped at the one end while a point load applied at the other end.
The magnitude of the applied load was F = −5000 N. The geometrical data of the
structure were as the previous and length to height ratio was equal to the 10. 9B4
elements were used to describe the box beam and the number of beam elements was
192 R. B. Bharati et al.
7.00E+03
6.00E+03
5.00E+03
FREQUENCY (HZ)
4.00E+03
3.00E+03
2.00E+03
1.00E+03
0.00E+00
0 2 4 6 8 10 12
MODE NUMBER
7.00E+03
6.00E+03
5.00E+03
FREQUENCY (HZ)
4.00E+03
3.00E+03
2.00E+03
1.00E+03
0.00E+00
0 2 4 6 8 10 12
MODE NUMBER
derived from the convergence analysis. The vertical displacement reported in Table 3
for the various cases and computed results were compared with shell element of the
commercial software. The vertical displacement was computed at two points ‘a’ and
‘b’ those projected on the middle of top and bottom surface at the free end of the
beam. The computed results were compared with commercial software and have a
Table 3 Vertical displacement of box beam with and without cut-off model
Without cut-off With cut-off
CUF model Shell CUF model Shell
16L9(a) 16L9(b) element 16L9(a) 16L9(b) element
Table 4 Natural frequencies (Hz) of box beam with and without cut-off model
Without cut-off With cut-off
CUF model Shell element CUF model Shell element
16L9(a) 16L9(b) 16L9(a) 16L9(b)
Case 1 514.70 513.56 510.64 489.52 501.55 503.42
Case 2 741.72 738.7 735.26 737.78 741.33 737.25
Case 3 1251.73 1222.93 1193.36 1067.37 1169.48 1167.86
Case 4 518.87 498.84 495.61 498.19 485.67 482.38
good agreement. The study shows that the different stacking sequence can influence
the vertical displacement of the box beam.
In this section, the static and free vibration analysis of a composite box beam with
cut-off has been discussed. The cut-off located on the center of the bottom surface
of the box beam, the dimension of the cut-off was width (bc ) = 11:343 and length
(lc ) = 44:866 mm. The vertical displacement and natural frequencies of box beam
(L/h = 10) without and with cut-off have been reported in Table 3 and Table 4
respectively. The computed results are in good agreement with shell elements model
of commercial software. For the static analysis, the magnitude of the applied force was
F = −5000 N. Vertical displacement and natural frequencies have been analyzed for
various cases. The computed results show that the cut-off model was more displaced
than the without cut-off model for static analysis and natural frequencies of cut-off
model were lower than the without cut-off model for free vibration analysis. The
natural frequencies related to the mode number for case 1 and case 3 of 16L9(b)
models have been shown in Figs. 6 and 7. The graph shows that the low variation
trends of frequencies at lower modes and higher modes but more variation in between
lower and higher modes.
5 Conclusion
In this present work, free vibration and static analyses of composite box beam have
been carried out. The analyses were performed by means of a refined beam model
based on the Lagrange Expansion (LE). The principle of virtual displacement has
been used along with CUF to formulate the finite element arrays in the terms of
fundamental nuclei, which neither depend on the expansion order nor the class of the
beam model. The present methodology can deal with full material anisotropy and the
cut-off also can be easily implemented in the box structure. Various composite box
Free Vibration and Stress Analysis of Laminated Box … 195
6.00E+03
5.00E+03
FREQUENCY (HZ)
4.00E+03
3.00E+03
2.00E+03
1.00E+03
0.00E+00
0 2 4 6 8 10 12
MODE NUMBER
Fig. 6 Natural frequencies for 16L9(b) model of box beam for case 1
5.00E+03
4.50E+03
4.00E+03
3.50E+03
FREQUENCY (HZ)
3.00E+03
2.50E+03
2.00E+03
1.50E+03
1.00E+03
5.00E+02
0.00E+00
0 2 4 6 8 10 12
MODE NUMBER
Fig. 7 Natural frequencies for 16L9(b) model of box beam for case 3
beams have been analyzed and in the domain to focus on the parametric studies have
been performed to see the effects of cut-off versus free vibration and static analysis.
The computed results show that the different stacking sequence and aspect ratio
can influence the results of both free vibration and static analysis. The results were
compared with published literature obtained in LE expansion of CUF models. The
provided comparison shows that with lower computation cost, the 1D-CUF approach
yields eventually the same results of the 3D solid element solution.
Acknowledgements The authors are grateful to the Director of Indian Institute of Technology
(Indian School of Mines), Dhanbad, India and Politecnico di Torino, Italy, for research collaboration
and facilities.
196 R. B. Bharati et al.
References
Abstract In the present study, free vibration analysis of porous functionally graded
material (FGM) circular arches has been performed using finite element methodol-
ogy. The present mathematical model is formed on the higher-order shear deforma-
tion theory (HSDT). The mechanical properties of the functionally graded material
arches are varying along the thickness direction. The effective mechanical proper-
ties of FGM are defined by a modified power law in terms of volume fractions of
its constituents. Two distributions of porosity are considered, viz., even and uneven
porosity distributions. Convergence and a comparative study have been executed to
verify the present formulation. The effect of various influencing parameters, viz.,
volume fraction index, temperature change, porosity index, and thickness to length
ratio, with both distributions of porosity (even and uneven porosity distributions), on
the frequency parameter of the arch have been studied.
1 Introduction
properties of FGM, like high specific strength, ability to withstand ultrahigh temper-
ature gradients and high specific stiffness, their use in spacecraft, nuclear reactors,
and space structures has been increased [2]. The curved beams and circular arches
are being widely used as structural members in various engineering applications
such as in civil, mechanical and aerospace industries [3]. Therefore, it is important
to analyze and find out the vibrational behavior of these structures in order to avoid
catastrophic failure.
The FGMs are difficult to manufacture, microstructural voids, or porosities may
arise in the FGMs because of the substantial difference in the solidification tempera-
ture of the material constituents, during the process of sintering [4, 5]. The porosity
also affects the response of the FGM structure, for example, Ebrahimi et al. [6]
scrutinized the vibration response of the FGM porous beam subjected to different
types of thermal loadings using the semi-analytical method. Ebrahimi and Jafari
[7] obtained Navier solution for the analysis of thermomechanical vibrations of the
porous FGM beams under the various thermal loading conditions, viz., uniform,
the linear, sinusoidal temperature rises, and nonlinear. They modified power law to
calculate the mechanical properties of the FGM with the even and uneven porosity
distributions. Gupta and Talha [8] examined the effect of the porosity on the vibration
characteristics of the FGM plates and found that the porosity noticeably affected the
vibration response of a thin plate. Sometimes the porous FGMs can be very useful
in biomaterial applications, such as bone replacements and dental implants [9].
The FGM structures like arches are frequently used in various engineering appli-
cations. Malekzadeh [10] scrutinized the vibration analysis of the FG thick arches
under thermal environment condition. He found that the material properties depen-
dent on temperature have a notable effect on the natural frequency. The in-plane
thermomechanical vibrations of FGM circular beams using beam theory approach
have been studied by Eroglu [11]. Filipich and Piovan [12] discussed the analyti-
cal solution for dynamics of the thick FGM curved beams using the power series
method. They also implemented the neutral axis shifting concept to minimize the
algebraic calculations. Lim et al. [13] scrutinized the free vibration response of func-
tionally gradient arches under the thermal environment condition analytically, using
state-space method.
It is carrying out from the literature that the thermos-elastic vibration analysis
of the FGM porous circular arches has not been proclaimed to the best of authors’
knowledge. So in this paper, the effect of porosity on frequency parameters of FGM
arches under the thermal environment is analyzed. It is assumed that mechanical
properties are temperature dependent and varying along with the thickness. To incor-
porate even and uneven porosity distributions in the formulation the modified power
laws have been chosen. To obtain the final governing equation of the vibrations for
the arch, a C0 continuous finite element method has been used. The validation and
convergence studies have been executed to establish the efficacy of the current finite
element model. The effect of various influencing parameter, viz., porosity index (α),
temperature change (T ), porosity distribution type, volume fraction index (n), and
the thickness to length ratio (h/L) on the natural frequency of the arches have been
investigated.
Free Vibration Analysis of the Functionally Graded … 199
Consider an FGM circular arch with the thickness h, length of the arch (along A-
A ) L = Rθ0 , and the mid-surface radius of the arch R as displayed in Fig. 1. The
curvilinear coordinate axes (ξ, z) are used to obtain the present formulation, where
ξ = r θ and z = r − R.
The circular arch made of porous FGM and having continuously varying mechanical
properties along the thickness is considered. The bottom surface of FGM circular
arch (z = −h/2) is metal-rich, while the top surface (z = h/2) is assumed to be
ceramic-rich. In this paper, even and uneven distributions of the porosity throughout
the thickness of FG circular arches have been considered. In the even distribution,
the porosity disseminated uniformly over the cross section, whereas in uneven dis-
tribution, the porosity occurred frequently in the neighboring of the middle region
of the cross section as displayed in Fig. 2.
Fig. 2 Cross section of the FGM arch with the even and uneven porosity distributions
200 M. Amir and M. Talha
For the even distribution, the effective mechanical properties are obtained by a
power law as
z n
H (z) = (Hc − Hm ) 0.5 + + Hm − α(Hc + Hm ) (1)
h
The effective mechanical properties of the FGM circular arch for the uneven
porosity distribution can be declared as in [7]:
z n α 2|z|
H (z) = (Hc − Hm ) 0.5 + + Hm − (Hc + Hm ) 1 − (2)
h 2 h
Figure 3a, b shows the alteration of effective mechanical properties of porous FGM
with z/h for the even and uneven porosity distributions. The effective property E(z)
is calculated for even (Eq. 1) and uneven (Eq. 2) porosity distributions of the FGM
at E c = 390 Gpa, E m = 214 Gpa, and α = 0.1. The effective elastic constant for
the even porosity distribution at the porosity index (α = 0.1) reduces throughout the
thickness as shown in Fig. 3a. But for the uneven porosity distribution, the effective
elastic constant reduces utmost at z/h = 0 as given in Fig. 3b, because, at the middle
plane the porosity concentration is maximum. While in case of even distribution the
pores are spread uniformly throughout the thickness.
Where α is the porosity index, Hc and Hm are the mechanical properties of
the ceramics and metal, respectively. The mechanical properties of the FGM are
considered as temperature dependent and declared as a function of temperature:
H = H0 (H−1 T −1 + 1 + H1 T + H2 T 2 + H3 T 3 ) (3)
Fig. 3 The variation of the effective elastic modulus of porous FGM with z/h at α = 0.1 for the
various n; a even porosity b uneven porosity
Free Vibration Analysis of the Functionally Graded … 201
Using the curvilinear coordinate axes (ξ , z) the displacement field is defined in the
terms of midplane displacements and the higher-order rotation terms. By incorpora-
tion of the traction-free conditions on the top and the bottom surface of the arch and
accommodating C0 continuity the modified displacement field becomes
4z 3
ū(ξ, z) = u(ξ ) + zφξ (ξ ) − 2
φξ (ξ ) + βξ (ξ )
3h
w̄(ξ, z) = w(ξ ) (4)
∂w
where βξ = ∂ξ
; The primary field variables are represented as
202 M. Amir and M. Talha
{} = u, w, φξ , βξ (5)
The strain terms for the circular arch are linearly defined as [14, 15]
1 ∂ ū w̄
εξ ξ = +
(1 + z/R) ∂ξ R
1 ∂ w̄ ū ∂ ū
γξ Z = − + (7)
(1 + z/R) ∂ξ R ∂z
The linear stress–strain relations of the FGM circular arch under thermal
environment conditions are defined as
σξ ξ Q11 0 εξ ξ δ
= − T (9)
τξ z 0 Q55 γξ z 0
where Qi j and δ are the material elastic coefficients and thermal expansion coefficient,
respectively. T is a change in the temperature.
The strain energy and the kinetic energy of the FGM circular arch are given as
L
1 1 e T
Us = {ε} {σ }dV = b
T
Λ [B]T [D][B] Λe dξ
2 2
V 0
Free Vibration Analysis of the Functionally Graded … 203
1 e T e
e
= Λ K Λ (11)
2
T L
1 ˙ ˙ 1 eT e
Tk = ρ Ū Ū dV = b ρ ˙ ˙ dξ
[N ]T [N ]
2 2
V 0
1 eT e
= ˙ ˙
[M] (12)
2
The variational principle can be used to derive the governing equation for the
free vibration of the FGM circular arch, which is a generalized principle of virtual
displacement. The equilibrium equation for free vibration analysis can be obtained
as
¨ + [K ]{} = 0
[M] (13)
or
[K ]{} = λ[M]{} (14)
where λ is the eigenvalue, which can be obtained by incorporating the end conditions
of the arch.
3 Numerical Results
To determine the efficacy and precision of the present finite element model, con-
vergence and validation studies are carried out. The validation is performed for free
vibration analysis of straight FGM beam with even and uneven porosity distributions.
The simply supported (S-S) straight beam made of FGM (SUS304/Si3 N4 ) is consid-
ered whose mechanical properties are assumed to be temperature dependent and are
given in Table 1. The frequency parameters of the FGM straight beam are evaluated
at T = 20 K and also compared with those results presented by Ebrahimi and
Jafari [7]. It is seen from Table 2 that the present results are in reasonable agreement
with those results presented by Ebrahimi and Jafari [7].
Table 3 shows the first three frequency parameters of the FGM circular arch
for the h/L = 0.2 are compared with those results given by Malekzadeh [10] and
Eroglu [11]. The mechanical properties of FGM (Ti–6Al–4V/ZrO2 ) are considered
as temperature dependent as provided in Table 1. The frequency parameters of FGM
circular arch have been obtained for clamped–clamped boundary conditions with
temperature change T = 400 K, θ0 = 60◦ , and n = 2. It is noticeable that the
current results are in the excellent agreement with those results presented by Eroglu
[11] and Malekzadeh [10]. It is also evident from this study that performance and
rate of convergence of the present model are very good in terms of solution accuracy.
204 M. Amir and M. Talha
Table 2 Comparison of the frequency parameters λ̄ of S-S FGM straight beam with the porosity
at L/h = 20 and T = 20K
n Even porosity Uneven porosity
α=0 α = 0.1 α = 0.2 α=0 α = 0.1 α = 0.2
0 Present 6.38672 6.94562 7.84147 6.38672 6.70292 7.0885
Ref. [7] 6.30389 6.88939 7.82479 6.30389 6.64111 7.04845
0.5 Present 4.41545 4.48376 4.57254 4.41545 4.50182 4.59810
Ref. [7] 4.27875 4.36908 4.47876 4.27875 4.38047 4.49154
1 Present 3.88201 3.88076 3.87884 3.88201 3.93159 3.98587
Ref. [7] 3.72764 3.74463 3.75865 3.72764 3.79112 3.85843
2 Present 3.49029 3.45061 3.40190 3.49029 3.51835 3.54861
Ref. [7] 3.33104 3.30844 3.27411 3.33104 3.37261 3.41549
5 Present 3.16545 3.10088 3.02325 3.16545 3.17928 3.19398
Ref. [7] 3.01417 2.96886 2.90872 3.01417 3.04242 3.0708
√
The dimensionless frequency parameter is defined as λ̄ = λL 2 / h ρm /E m , where
L = Rθ .
Table 4 displays the frequency parameter of the FGM (Ti–6Al–4V/ZrO2 ) porous
circular arch for different values of h/L (h/L = 0.2, 0.1, 0.05), gradient indices (n =
0, 1, 2, 5), porosity indices (α = 0, 0.1, 0.2) for both even and uneven distributions at
T = 400 K and θ0 = 60◦ . On observing Table 4, it is concluded that the frequency
parameter (λ̄) reduces with the increase in the value of n. This is because of the
metallic content in FGM increases with an increase in the value of n, consequently
stiffness decreases and hence frequency decreases. In case of even distribution of
porosity, on increasing the value of α, the frequency parameter increases up to n <
1 but decreases after n ≥ 1. In case of uneven distribution of porosity the frequency
parameter ever increases with increasing in α. This is due to the modulus term in
the expression for the uneven distribution of porosity. The influence of thickness to
Free Vibration Analysis of the Functionally Graded … 205
Table 4 Influence of porosity on the frequency parameter of the fully clamped FG circular arch
with the volume fraction indices (n) at T = 400 K, and θ0 = 60◦
h/L n Even porosity Uneven porosity
α=0 α = 0.1 α = 0.2 α=0 α = 0.1 α = 0.2
0.2 0 5.6522 5.7033 5.7681 5.6522 5.7006 5.7536
1 5.2119 5. 2108 5.2091 5.2119 5.2356 5.2609
2 5.0896 5.0752 5.0570 5.0896 5.1067 5.1245
5 4.9740 4.9476 4.9147 4.9740 4.9850 4.9961
0.1 0 8.7536 8.8330 8.9335 8.7536 8.8215 8.8965
1 8.0799 8.0793 8.0781 8.0799 8.1096 8.1418
2 7.8869 7.8653 7.8382 7.8869 7.9063 7.9272
5 7.7030 7.6624 7.6118 7.7030 7.7131 7.7236
0.05 0 14.0980 14.2256 14.3873 14.0980 14.3027 14.5276
1 13.0111 13.0092 13.0061 13.0111 13.1518 13.3045
2 12.7284 12.6958 12.6546 12.7284 12.8528 12.9873
5 12.4553 12.3948 12.3192 12.4553 12.5649 12.6829
length ratio (h/L) is noticed that the frequency parameter increases with a decrease
in the value of h/L.
Table 5 presents the dimensionless frequency parameter of the porous FGM (Ti–
6Al–4V/ZrO2 ) circular arch. In this table, the effect of temperature rise on the fre-
quency parameter of the porous circular arch with the porosity is studied. The fre-
quency parameter (λ̄) of the FGM porous circular arch have been determined for the
various values of the porosity index (α = 0, 0.1, 0.2) and volume fraction index (n
= 0, 1, 2, 5) at T = 0, 100, 200, 300 K, h/L = 0.2 and θ0 = 60◦ for the both even
and uneven distributions of porosity as given in Table 5. It is noticed that for the both
even and uneven distributions of porosity, the frequency parameters start declining
with the rise in the temperature. This is because of the value of elastic moduli reduces
with rising in the temperature.
4 Conclusions
A C0 finite element method has been adopted to scrutinize the frequency response
of the porous FGM circular arch with clamped–clamped boundary conditions in the
thermal environment. The effect of porosity (microstructural defects) on the free
vibrations of the FGM circular arch is examined. It is concluded that the frequency
parameters reduce with the rise in the temperature for both even and uneven poros-
ity distributions. The frequency parameters reduce with the increase in the volume
fraction index. It is also observed that the frequency parameters always increase by
decreasing the value of thickness to length ratio.
206 M. Amir and M. Talha
Table 5 The change in the frequency parameter of the C–C FGM circular arch with distinct values
of n, α, and T, at θ0 = 60◦ , and h/L = 0.2
T (K) n Even porosity Uneven porosity
α=0 α = 0.1 α = 0.2 α=0 α = 0.1 α = 0.2
0 0 6.3288 6.3826 6.4510 6.3288 6.3817 6.4396
1 5.8625 5.8607 5.8584 5.8625 5.8890 5.9172
2 5.7322 5.7162 5.6963 5.7322 5.7515 5.7717
5 5.6089 5.5801 5.5445 5.6089 5.6217 5.6347
100 0 6.1727 6.2263 6.2943 6.1727 6.2247 6.2817
1 5.7095 5.7078 5.7056 5.7095 5.7353 5.7628
2 5.5802 5.5645 5.5449 5.5802 5.5989 5.6186
5 5.4580 5.4296 5.3944 5.4580 5.4703 5.4828
200 0 6.0086 6.0616 6.1289 6.0086 6.0596 6.1154
1 5.5505 5.5489 5.5469 5.5505 5.5756 5.6024
2 5.4228 5.4075 5.3882 5.4228 5.4410 5.4600
5 5.3022 5.2743 5.2397 5.3022 5.3140 5.3260
300 0 5.8354 5.8877 5.9539 5.8354 5.8852 5.9397
1 5.3848 5.3835 5.3816 5.3848 5.4093 5.4353
2 5.2595 5.2445 5.2258 5.2595 5.2771 5.2955
5 5.1410 5.1138 5.0800 5.1410 5.1524 5.1639
Acknowledgements The authors would like to acknowledge the financial support provided by the
Science and Engineering Research Board, Department of Science and Technology (SERB-DST),
Government of India to carry out this research. Project number-YSS/2015/001290, Dated-09th
NOV 2015.
References
1 Introduction
Functionally gradient materials (FGMs) are advanced materials having the progres-
sive variation of their constituent materials along the predetermined direction [1].
Since the past three decades, these materials have proven their worth in numer-
ous applications such as aerospace, civil, biomechanical, and mechanical industries
[2]. The ability to retain the structural integrity under extreme thermal–-mechanical
loading makes theses material suitable for structural application as well.
Considerable investigations have been perpetrated to inspect the vibrational
attributes of gradient structures. Early research efforts for dynamic analysis of FGM
plate have been presented by Praveen and Reddy [3] in which nonlinear time-
dependent thermoelastic examination of FGM plate has been studied using a finite
A. Gupta (B)
School of Engineering, Shiv Nadar University, Noida, UP, India
e-mail: ankit.gupta1@snu.edu.in
M. Talha
School of Engineering, I.I.T Mandi, Mandi, India
e-mail: talha@iitmandi.ac.in
element method. The buckling and vibration response of the FGM polygonal plate
using TSDT has been studied by Cheng and Batra [4]. Ferreira et al. [5] employed
FSDT and TSDT with the global collocation method to examine the vibration behav-
ior of gradient plates. Neves et al. [6] explored the structural characteristics of sand-
wich gradient plates using a meshless technique and HSDT. Talha and Singh [7]
employed a modified displacement field with thirteen DOFs/node to demonstrate
the displacement and vibration attributes of a gradient plate. Gupta and Talha [8, 9]
developed non-polynomial HOSNT to analyze the flexural and vibration response
of the gradient plate.
In this article, the vibration behavior of the gradient plate with initial geometric
imperfection has been investigated using non-polynomial HOSNT. The effective
material properties of the gradient plate have been computed using sigmoid rule.
Initial geometric imperfection in the plate has been incorporated using a generic
function that is capable to model various imperfection modes such as sine, global,
and local. Comparative studies are provided to illustrate the correctness of the present
formulation.
2 Mathematical Idealization
A hybrid HOSNT developed by the authors [10] is used in this study for the vibration
analysis of the gradient plate.
H −1 z
U = u − z Ax + Bx + Nsinh Bx
h h
H z
V = v − z Ay + B y + Nsinh−1 By
h h
z
W = w + cosh2 Bz (1)
h
T
{} = u v w A x A y Bx B y Bz . (2)
In this study, the Sigmoid law has been employed to compute the effective Young’s
modulus and density of the gradient plate and is written as
V ol 1f r (z) = 1 − 0.5(1 − 2z/t)n
V ol 2f r (z) = 0.5(1 + 2z/t)n
Employing the rule of mixture, the effective Young’s modulus and density of the
gradient material can be computed by
E(z) Ec Em
= + 1 − V ol 1f r (z) for 0 ≤ z ≤ t/2
V ol 1f r (z)
ρ(z) ρc ρm
E(z) Ec Em
= V ol 2f r (z) + 1 − V ol 2f r (z) for − t/2 ≤ z < 0
ρ(z) ρc ρm
[K ]{
} = λ[M]{
} (5)
In this section, first validation examination has been accomplished to exhibit the
exactness of the present solution, then after some new results have been shown for
geometrically imperfect FGM plate. The Young’s modulus and density of metal (Ti–
6AL–4V) is E m = 105.7 × 109 Pa and ρ m = 4429 kg/m3 and for ceramic (Si3 N4 ) is
E c = 322.27 × 109 Pa and ρ m = 2370 kg/m3 , respectively. The generic imperfection
function is employed to generate various geometric imperfection models as shown
in Eq. (6) [12].
x x y x
ζ̄ = hς cos β1 π − χ1 / cosh α1 − χ1 × cos β2 π − χ2 / cosh α2 − χ2 (6)
a a b b
Table 3 Change in frequency of (SSSS) gradient plate with a/h and geometric imperfection (n =
1, b/a = 1)
Modes a/h
5 10 20 50 100
Perfect 1.0308 1.2046 1.2420 1.2561 1.2587
Sine-type 1.0221 1.2424 1.2697 1.2803 1.2824
% difference* 0.8440 3.1380 2.2303 1.9266 1.8829
Global-type 1.0094 2.0394 2.6731 2.5640 2.5442
% difference 2.0760 69.301 115.225 104.124 102.129
Local-type 1.0259 1.2051 1.2472 1.5283 1.7239
% difference 0.4754 0.0415 0.4187 21.6702 36.9588
* % difference = 100 * (Perfect plate frequency − Im perfect plate frequency/Perfect plate frequency)
Table 4 Change in frequency of (CCCC) gradient plate with a/h and geometric imperfection (n
= 1, b/a = 1)
Modes a/h
5 10 20 50 100
Perfect 1.7039 2.1089 2.2920 2.3947 2.4199
Sine-type 1.7409 2.1381 2.3102 2.4088 2.4333
% difference 2.1715 1.3846 0.7941 0.5888 0.5537
Global-type 1.8577 3.3921 3.4341 3.3444 3.3330
% difference 9.0264 60.8469 49.8298 39.6584 37.7330
Local-type 1.7230 2.1094 2.2952 2.5619 3.3570
% difference 1.1210 0.0237 0.1396 6.9821 38.7247
the results obtained from the perfect plate and the imperfect plate is 3.1380, 102.129
and 36% for S-type, G–type and L-type imperfection, respectively, for CCCC bound-
ary condition. It is noteworthy that the influence of imperfection is more in thin
plates. The consistent behavior is observed for SSSS boundary constraint as shown
in Table 4.
Tables 5 and 6 depict the change of frequency of gradient plate with b/a and
geometric imperfection for SSSS and CCCC boundary constraints, respectively. It is
found that the nondimensional frequency increases with b/a. Again the same obser-
vation is perceived as outlined in the previous discussion that G-type imperfection
has a substantial impact on the frequency. It is also evident that the change in nondi-
mensional frequency is more in the case of a rectangular plate compared to the square
plate.
216 A. Gupta and M. Talha
Table 5 Change in nondimensional frequency of (CCCC) gradient plate with b/a and geometric
imperfection (n = 1, a/h = 10)
Modes b/a
1 1.5 2 2.5 3
Perfect 2.1089 2.4278 2.9625 3.5697 4.2066
Sine-type 2.1381 2.4785 3.0322 3.6574 4.3118
% difference 1.3846 2.0883 2.3527 2.4568 2.5008
Global-type 3.3921 4.2595 5.1168 5.9837 6.8742
% difference 60.8469 75.4469 72.7190 67.6247 63.4146
Local-type 2.1094 2.4287 2.9637 3.5712 4.2084
% difference 0.0237 0.0371 0.0405 0.0420 0.0428
Table 6 Change in nondimensional frequency of (SSSS) FGM plate with b/a and geometric
imperfection (n = 1, a/h = 10)
Modes b/a
1 1.5 2 2.5 3
Perfect 1.2046 1.3192 1.5289 1.7777 2.0460
Sine-type 1.2424 1.3896 1.6320 1.9128 2.2126
% difference 3.1380 5.3366 6.7434 7.5997 8.1427
Global-type 2.0394 2.9937 3.8567 4.1483 4.1438
% difference 69.3010 126.9330 152.2533 133.3521 102.5318
Local-type 1.2051 1.3203 1.5306 1.7800 2.0489
% difference 0.0415 0.0834 0.1112 0.1294 0.1417
The change in frequency with “n” and geometric imperfection with fully fixed and
simply supported boundary constraints is given in Table 7 and 8. It is clear that
the frequency decreases as “n” increases. This is because that as “n” increases, the
metallic fraction in the gradient plate increases, therefore the stiffness decreases. It
is noteworthy that the effect of imperfection is more in gradient plate having more
metallic contents.
Table 7 Variation of frequency of (CCCC) gradient plate with “n” and geometric imperfection
(b/a = 1, a/h = 10)
Modes “n”
0 1 2 5 10 100
Perfect 2.1916 2.1089 2.0606 2.0181 2.0053 1.9992
Sine-type 2.2217 2.1381 2.0895 2.0467 2.0338 2.0277
% difference 1.3734 1.3846 1.4025 1.4172 1.4212 1.4256
Global-type 3.6804 3.3921 3.2877 3.2074 3.1847 3.1742
% difference 67.9321 60.8469 59.5506 58.9317 58.8141 58.7735
Local-type 2.1921 2.1094 2.0611 2.0186 2.0058 1.9997
% difference 0.0228 0.0237 0.0243 0.0248 0.0249 0.0250
Table 8 Variation of frequency of (SSSS) gradient plate with “n” and geometric imperfection (b/a
= 1, a/h = 10)
Modes “n”
0 1 2 5 10 100
Perfect 1.2571 1.2046 1.1738 1.1467 1.1385 1.1345
Sine-type 1.2951 1.2424 1.2117 1.1846 1.1764 1.1725
% difference 3.0228 3.1379 3.2288 3.3051 3.3289 3.3494
Global-type 2.1332 2.0394 2.0138 1.9954 1.9906 1.9885
% difference 69.6921 69.301 71.5624 74.0124 74.8441 75.2755
Local-type 1.2576 1.2051 1.1744 1.1472 1.1390 1.1351
% difference 0.0397 0.0415 0.0511 0.0436 0.0439 0.0528
in the stiffness. It is notable that the impact of imperfection is maximum in the case
of FFFF whereas least in CCCC boundary constraint.
4 Conclusions
In this article, the vibration behavior of the geometrically imperfect gradient plate
has been investigated using hybrid HOSNT. It is concluded that the geometric imper-
fection has a prominent effect on thin plates. The L-type and G-type imperfection
have minimum and maximum effect on the frequency parameter of the FGM plate.
It is also concluded that the influence of geometric imperfection decreases as the
constraint at the boundaries of the gradient plate increases.
218 A. Gupta and M. Talha
Table 9 Change in nondimensional frequency of gradient plate with various boundary constraints
and geometric imperfection (b/a = 1, a/h = 10)
Model n Boundary conditions
CCCC CFCF SSSS CFSF FFFF
Perfect 0 2.1916 1.3983 1.2571 0.9784 0.8806
5 2.0181 1.2835 1.1467 0.8928 0.8007
10 2.0053 1.2750 1.1385 0.8865 0.7950
Sine-type 0 2.2217 1.4214 1.2951 1.0673 0.1930
% difference 1.3734 1.6520 3.0228 9.0862 78.0831
5 2.0467 1.3060 1.1846 0.9764 0.1922
% difference 1.4171 1.7530 3.3051 9.3638 75.9960
10 2.0338 1.2975 1.1764 0.9697 0.1921
% difference 1.42123 1.76471 3.32894 9.38522 75.8364
Global-type 0 3.6804 2.5870 2.1332 1.7695 1.9474
% difference 67.9321 85.0104 69.6921 80.8565 121.145
5 3.2074 2.3013 1.9954 1.6683 1.8548
% difference 58.9317 79.2988 74.0124 86.8616 131.647
10 3.1847 2.2852 1.9906 1.6654 1.8483
% difference 58.8141 79.2314 74.8441 87.8624 132.491
Local-type 0 2.1921 1.3988 1.2576 0.9874 0.0254
% difference 0.02281 0.03576 0.03977 0.91987 97.1156
5 2.0186 1.2839 1.1472 0.9011 0.0253
% difference 0.0247 0.0311 0.0436 0.9296 96.8402
10 2.0058 1.2754 1.1390 0.8947 0.0253
% difference 0.0249 0.0313 0.0439 0.9249 96.8176
References
8. Gupta A, Talha M (2016) An assessment of a non-polynomial based higher order shear and
normal deformation theory for vibration response of gradient plates with initial geometric
imperfections. Compos Part B 107:141–161
9. Gupta A, Talha M (2017) Large amplitude free flexural vibration analysis of finite element
modeled FGM plates using new hyperbolic shear and normal deformation theory 287–308
10. Gupta A, Talha M (2017) Influence of porosity on the flexural and vibration response of gradient
plate using nonpolynomial higher-order shear and normal deformation theory. Int J Mech Mater
Des
11. Gupta A, Talha M, Singh BN (2016) Vibration characteristics of functionally graded material
plate with various boundary constraints using higher order shear deformation theory. Compos
Part B Eng 94:64–74
12. Yang J, Huang X-L (2007) Nonlinear transient response of functionally graded plates with
general imperfections in thermal environments. Comput Methods Appl Mech Eng 196:2619–
2630
13. Akavci SS (2014) An efficient shear deformation theory for free vibration of functionally
graded thick rectangular plates on elastic foundation. Compos Struct 108:667–676
Characterization of 2D Nanomaterials
for Energy Storage
1 Introduction
During the last couple of decades, graphene, and h-BN have emerged as the most
promising and fascinating two-dimensional (2D) nanomaterials [1, 2]. Their excep-
tional mechanical and thermal properties are attributed to its space frame honey-
comb lattice structure and quantum confinement. Researchers are exploring diver-
sifying areas of applications that include microelectronics, biomedical, reinforcing
nanocomposites, desalination membrane, clean energy devices, and solar cells for
utilizing these aforementioned 2D nanomaterials [3].
Hydrogen chemical group is a highly combustible and potential futuristic energy
source for an uncontaminated and sustainable economy; but, its room temperature
storage has raised a few questions to the materials scientific community. Ozturk
et al. [4] proposed a 3D carbon-based nanoporous material (containing fullerene
compound sandwiched between graphene sheets) for the hydrogen storage applica-
tion. In addition to the hydrogen storage dilemma, water desalination, and separation
of ions also make these 2D nanomaterials prone to hydrogenation effect [5, 6].
In the present scrutiny, molecular models were designed in the background
of molecular dynamics (MD) based simulations technique. Accomplishment of
MD computational study depends on the interatomic chemical potential that is
mainly used for perceiving the bonded and nonbonded atomistic interaction. Herein,
the adaptive intermolecular reactive empirical bond order (AIREBO) potential is
employed in all the simulations for calculating and analyzing atomistic interactions
between carbon–hydrogen chemical systems in graphene [7]; whereas, reactive force
field (ReaxFF) is utilized for graphene, h-BN and hydrogen chemical systems [8,
9]. For the per atom stress tensor, virial stresses [10, 11] were predicted using the
mathematical formula (refer Eq. 1):
⎛ ⎞
1 1 j
σiαj = α ⎝ m α viα v αj + i ⎠
rαβ f αβ (1)
ϕ 2 β=1,n
The brittle behavior of graphene and h-BN leads to characterizing them from
the fracture point of view, that is quantified by the fracture toughness/critical stress
intensity factor (K IC ) as described in the Eq. 2:
√
K IC = Y σ πa (2)
here, Y is the dimensionless geometry parameter (Y ≈ 1), σ is the normal virial stress
value at the first bond rupture instant and a is half of the central crack length present
in graphene and h-BN sheet. To avoid thermal disturbances and maintain a realistic
room temperature, all the computations were executed at room temperature (300 K)
under NPT ensemble [12]. Open source code software large-scale Atomic/Molecular
Massively Parallel Simulator (LAMMPS) was employed for this simulations [13, 14];
whereas, open visualization tool OVITO was executed for post handling of dump
files produced by LAMMPS [15, 16]. The authors have taken an integration time
step of 0.5 fs and the nanomaterials were subjected to a tensile deformation (strain
rate being equal to 0.0005 ps−1 ). The nanosheet size was taken to be 270 Å (in length
and breadth) and the periodic boundary conditions were taken.
Preliminary computations were carried out to validate the AIREBO and ReaxFF
potential factors for graphene and h-BN, respectively. After validation, tensile test
was conducted for predicting the mechanical behavior of these nanomaterials, as
shown in Figs. 1 and 2. It was concluded from Fig. 1 that failure stress and strain
considerably reduced for the fully and partially hydrogenated graphene sheets as
Characterization of 2D Nanomaterials for Energy Storage 223
160
Pristine
Fully Hydrogenated
Partially Hydrogenated (50%)
120
Stress (GPa)
80
40
0
0.00 0.10 0.20 0.30
Strain [/]
Fig. 1 Stress–strain response of pristine, fully and partially hydrogenated sheets of graphene at
300 K
80
HBN
40
0
0.00 0.05 0.10 0.15 0.20
Strain [/]
compared to pristine one. Moreover, the partially hydrogenated graphene (with 50%
coverage) has comparatively higher fracture strength than the fully hydrogenated
one. Out-of-plane displacement in the hydrogenated graphene and the change in its
hybridization state from sp2 to sp3 were attributed as the main reasons behind diluting
the properties of graphene (see Fig. 3). The out-of-plane displacement leads to an
increase in carbon–carbon bond length, thus, decreasing its bond strength. Besides
this, decrement in contribution of s-orbital and increment of p-orbital (directional
dependent) lead to bond strength deterioration in the in-plane direction. Hence, we
observed an overall deteriorating tensile strength for the hydrogenated graphene
224 A. Verma and A. Parashar
structure. Similar kind of declining effect was predicted for hydrogenated h-BN as
shown in Fig. 2. Despite the reduction in fracture stress of fully hydrogenated (100%
coverage) as well as partially hydrogenated (50% coverage with H atom attached only
to B atoms) h-BN nanosheets, a noteworthy enlargement in fracture toughness of H-
BN (partially hydrogenated) was detected. The improvement in fracture toughness
of H-BN was accredited to shifting of the polarization charge state of hydrogen with
increase in the strain values and also to the crack tip blunting phenomenon. Similar
kind of effort was implemented to witness the influence of hydrogen passivation
of crack tip atoms on the fracture toughness of aforementioned 2D nanosheets. The
fracture toughness in terms of stress intensity factor (K IC ) was estimated and has been
showcased in Table 1. It may be predicted from this table that the hydrogenation has
deteriorating effect on K IC of graphene, but it helps in improving the K IC of h-BN
nanosheets. This enhanced fracture toughness would enable these 2D nanomaterials
to soak large energy erstwhile to failure and augmented toughness too, leading to
amended energy storage and structural applications.
Thus, we have shown that the molecular dynamics-based simulation helps in
predicting positive as well as negative impact of hydrogenation on h-BN and graphene
nanosheets, respectively.
3 Conclusion
Acknowledgements Monetary funding from the Nano mission unit of Government of India is
appreciatively recognized (Project number: DST-952-MID).
References
1. Novoselov KS, Fal’ko VI, Colombo L, Gellert PR, Schwab MG, Kim K (2012) A roadmap for
graphene. Nature 490(7419):192–200
2. Rajasekaran G, Narayanan P, Parashar A (2016) Effect of point and line defects on mechanical
and thermal properties of graphene: a review. Crit Rev Solid State Mater Sci 41(1):47–71
3. Verma A, Parashar A, Packirisamy M (2018) Atomistic modeling of graphene/hexagonal boron
nitride polymer nanocomposites: a review. Wiley Interdiscip Rev: Comput Mol Sci 8(3):e1346
4. Ozturk Z, Baykasoglu C, Kirca M (2016) Sandwiched graphene-fullerene composite: a novel
3-D nanostructured material for hydrogen storage. Int J Hydrog Energy 41(15):6403–6411
5. Verma A, Parashar A (2017) The effect of STW defects on the mechanical properties and
fracture toughness of pristine and hydrogenated graphene. Phys Chem Chem Phys 19:16023–
16037
6. Kumar R, Parashar A (2017) Fracture toughness enhancement of h-BN monolayers via
hydrogen passivation of a crack edge. Nanotechnology 28(16):165702
7. Rajasekaran G, Parashar A (2017) Enhancement of fracture toughness of graphene via crack
bridging with stone-thrower-wales defects. Diam Relat Mater 74:90–99
8. Verma A, Parashar A (2018) Reactive force field based atomistic simulations to study frac-
ture toughness of bicrystalline graphene functionalised with oxide groups. Diam Relat Mater
88:193–203
9. Verma A, Parashar A (2018) Molecular dynamics based simulations to study the fracture
strength of monolayer graphene oxide. Nanotechnology 29(11):115706
10. Zhou M (2003) A new look at the atomic level virial stress: on continuum-molecular system
equivalence. Proc R Soc Lond A: Math Phys Eng Sci 459(2037):2347–2392
11. Verma A, Parashar A (2018) Molecular dynamics based simulations to study failure morphol-
ogy of hydroxyl and epoxide functionalised graphene. Comput Mater Sci 143:15–26
12. Hoover WG (1985) Canonical dynamics: equilibrium phase-space distributions. Phys Rev A
31(3):1695
13. Plimpton S (1995) Fast parallel algorithms for short-range molecular dynamics. J Comput Phys
117(1):1–19
14. Verma A, Parashar A (2018) Structural and chemical insights into thermal transport for strained
functionalised graphene: a molecular dynamics study. Mater Res Express 5(11):115605
15. Stukowski A (2009) Visualization and analysis of atomistic simulation data with OVITO—the
open visualization tool. Modell Simul Mater Sci Eng 18(1):015012
16. Verma A, Parashar A, Packirisamy M (2018) Tailoring the failure morphology of 2D
bicrystalline graphene oxide. J Appl Phys 124(1):015102
Cold Expansion of Elongated Hole:
A Realistic Finite Element Simulation
Abstract Elongated holes are found in many aerospace structural components and
are prone to fatigue failure due to the stress concentration effect at the edge of the
hole. Different approaches like shape reworking, shape optimization, and cold expan-
sion combined with interference fitting are currently used as a repair/life extension
option, to overcome fatigue cracking problems around elongated holes. In practice,
these approaches lead to either addition of material or modification of geometry. A
novel and economical method derived from renowned hole cold expansion process
is proposed in the literature to enhance the fatigue life of elongated hole without
adding material and modifying hole geometries. Benefit of implementing proposed
novel cold expansion method for elongated hole is investigated in this work through
a simplified three-dimensional nonlinear Finite Element simulation. Distributions of
induced beneficial residual stress around and along the thickness direction of elon-
gated hole are predicted from the FE simulation. The results indicate introduction of
significant beneficial residual stresses throughout the thickness surface of elongated
hole. These beneficial residual stresses are responsible for fatigue life extension and
can be further used to quantify the achieved fatigue life enhancement.
1 Introduction
wing pivot fittings in F-111 aircarfts [1–3]. These fuel flow vent holes are machined
elongated holes with semi-circular ends (i.e., noncircular holes) in wing pivot fitting
components for the purpose of allowing fuel flow in the wings. Among several
such functional elongated holes in wing pivot fittings, some of the elongated holes
are extremely vulnerable for premature fatigue cracking owing to flight operational
loads [4–6]. To repair the fatigue-damaged elongated holes, different life extension
approaches, viz., shape reworking, shape optimization, and cold expansion combined
with interference fitting approaches are used in practice [1–6].
In shape reworking approach, the material layers on the elongated hole bound-
aries are progressively machined to slightly higher sizes so that the material layers
containing fatigue cracks are completely eliminated [1, 4]. Although shape rework-
ing eliminates fatigue cracking problem during repair stage, it is not possible to
avoid fatigue cracking problems which arise during later stages of service life of the
component. The shape optimization approach involves optimizing fatigue-damaged
elongated hole shapes with an objective of minimizing stress concentrations and
maximizing fatigue life/damage tolerance by employing optimization algorithms [1,
6]. Although optimization of elongated hole shapes significantly reduces the peak
stresses at stress concentration locations, the production of optimized hole shapes
often poses difficulties due to complexities in hole geometries. Hence, to overcome
the limitations of shape reworking and shape optimization approaches, cold expan-
sion combined with interference fitting approach is developed. In this approach, a spe-
cially designed oversized mandrel/sleeve combination made of hardened material is
permanently inserted into elongated hole opening causing cold expansion combined
with interference fitting. As a result, compressive residual stresses are developed
around damaged elongated hole impeding the fatigue crack initiation/propagation
under in-service operational loads [1–6]. Though the cold expansion combined with
interference fitting approach is found to be capable of enhancing the fatigue life
of elongated holes to larger extents, it leads to addition of extra material on the
wing pivot fitting assemblies. Application of proceeding approaches in repairing
fatigue-damaged elongated holes lead to either shape modification or material addi-
tion thereby posing a difficulty in meeting one of the mandatory air force durability
and damage tolerance certification requirements. This requirement insists to pre-
vent or repair fatigue-damaged structural holes without adding extra material and
imposing costly structural replacements [7, 8].
To overcome the aforementioned limitations of different life extension
approaches, a novel cold expansion method which is derived from the renowned
hole cold expansion process is proposed in Ref. [9] for enhancing the fatigue life of
elongated holes. Although the novel method of cold expansion for elongated holes
is proposed, none of the researchers have previously attempted to investigate the
method in detail and quantify resulting fatigue life enhancement benefits. Hence, an
attempt is made in the present work to realistically simulate the novel cold expan-
sion method for typical elongated hole in Al 7075-T651 plate and adequately predict
the cold expansion-induced beneficial residual around the elongated hole and its
thickness directions by developing a simplified FE framework.
Cold Expansion of Elongated Hole: A Realistic Finite … 229
Oversized Portion
Insert
Elongated
Hole
Step 1 Step 2
Fig. 2 Dimensions (mm) of the plate with elongated hole (R—radius of the semi-circular end,
C—spacing between semi-circular ends)
nonlinear material behavior is modeled using tangent modulus of 1000 MPa under
Mises plasticity option with isotropic hardening rule.
The situation of implementing proposed cold expansion method to elongated holes
can be considered as equivalent to sequential cold expansion of two closely spaced
adjacent circular holes (Fig. 1). Taking this consideration into account, the mechanics
of cold expansion process around closely spaced adjacent holes have been reviewed
from the published literature [12–16]. It is identified that either simultaneous or
sequential cold expansion of closely spaced adjacent holes (hole center-to-center
spacing is equal to two times the hole diameter) induces approximately same level
of compressive residual stresses around the holes. Hence, during simulation of cold
expansion for elongated hole, it can be considered that both circular hole regions
(at either semi-circular ends) are simultaneously cold expanded for the purpose of
simplifying FE modeling and reducing computational time.
The complete cold expansion method for elongated hole is simulated in two stages,
viz., gradual expansion of the material layer by layer on semi-circular end during first
stage and release of expanded material layer by layer in the same sequence during
second stage. For accounting the effects of layer by layer material expansion and
relaxation, the total thickness (3 mm) of the plate is divided into 8 equal elemental
divisions as shown in Fig. 3. These eight divisions are further grouped as three layers
having three elemental divisions in each layer. This grouping of layers is only for
the purpose of studying through-thickness effects, whereas, in the actual plate, the
material along the thickness is continuous. In the first stage of simulation, each layer
along the thickness direction is sequentially expanded one after the other starting
from mandrel entry side (top plane) and moving towards the exit side (bottom plane)
of the plate. The expansion of each layer is simulated by applying the displacements
that cause circular hole region to expand by 2% of its diameter (expansion level).
This indicates the gradual engagement of mandrel into circular hole region from
the entry side (top plane), expansion of material along hole thickness direction and
removal of mandrel from the exit side (bottom plane).
In the second stage of simulation, all the displacements which are applied in the
first stage of simulation are successively removed layer by layer starting from mandrel
entry side towards the exit side of the plate. This indicates the gradual elastic-plastic
recovery (spring-back) of expanded material on the circular hole region starting from
mandrel entry side to exit side. Throughout the simulation, the effect of insert in
elongated hole slot is considered by constraining the displacements on straight edge
portion of elongated hole boundary. Whereas, in actual cold expansion situations,
after cold expansion of circular hole regions at either end of elongated hole, the insert
is permanently ejected-out thereby leaving the semi-circular ends in cold expanded
state. Thus, after the complete FE simulation, the semi-circular ends of elongated
hole are in cold expanded state.
232 S. Anil Kumar and N.C. Mahendra Babu
To validate the FE simulation framework developed for simulating the cold expansion
of elongated hole in the present work, either experimental or numerical results are not
available in literature. Hence, for validating the present FE simulation framework, the
case of two adjacent circular holes whose radius equal to end radii of elongated hole
is considered in the configuration identical to elongated hole configuration shown in
Fig. 2. For this case having two adjacent circular holes in the plate of dimensions
shown in Fig. 2, the results of sequential cold expansion process are available [16].
Therefore, for the purpose of validation, a separate half-symmetry FE model having
two adjacent circular holes is developed as shown in Fig. 4 using 14,896 numbers of
8-noded solid 185 element type after testing mesh convergence.
To capture through-thickness variation effects, the thickness (3 mm) of the plate
is discretized into 8 equal elemental divisions as shown in Fig. 4. The complete simu-
lation of cold expansion around adjacent circular holes is carried out through several
loading steps by following the FE simulation framework employed for elongated
hole. From the FE simulation on cold expansion of adjacent circular holes for 2%
expansion level, the beneficial residual stress distribution is predicted and presented
in Fig. 5. It is evident from Fig. 5 that cold expansion-induced beneficial residual
stresses vary throughout the thickness of holes starting from a minimum magnitude
of 228 MPa (compressive) to maximum magnitude of 606 MPa (compressive). Also,
the beneficial residual stress variations on top and bottom planes of two adjacent
holes are predicted and validated with published results [16] as shown in Fig. 6.
This validation study shows that present simplified FE simulation framework is
Fig. 4 Quarter symmetry FE model of the plate with two adjacent circular holes
Cold Expansion of Elongated Hole: A Realistic Finite … 233
Fig. 6 Variation of normalized beneficial residual stresses over the normalized distance on hole
centerline section
234 S. Anil Kumar and N.C. Mahendra Babu
The beneficial residual stress distributions predicted around cold expanded elongated
hole is as shown in Fig. 7. It is clear from Fig. 7 that significant beneficial residual
stresses are induced around the hole and along the thickness direction of critical
locations, viz., semi-circular ends of elongated hole. These beneficial residual stresses
are found to significantly vary throughout the thickness of elongated hole starting
from the top plane to the bottom plane. Under remote fluctuating loads, sections HH1
and VV1 shown in Fig. 2 are found to be critical for fatigue failures around elongated
hole. Hence, the variations of beneficial residual stresses on different planes (top, mid-
thickness, and bottom) of critical sections (HH1 and VV1 ) are predicted and presented
in Figs. 8 and 9. Also, the through-thickness variations of beneficial residual stress
at two critical locations (H and V) on elongated hole boundary is predicted as shown
in Fig. 10.
Along with the thickness of location ‘H’, the beneficial residual stresses vary
from a minimum magnitude of 171 MPa (compressive) on bottom plane to 613 MPa
(compressive) on a plane which is at a distance of 2.25 mm from top plane (Fig. 10).
Fig. 8 Variation of normalized beneficial residual stresses over the normalized distance on
centerline section HH1 of elongated hole
Fig. 9 Variation of normalized beneficial residual stresses over the normalized distance on section
VV1 of elongated hole
236 S. Anil Kumar and N.C. Mahendra Babu
Fig. 10 Variation of beneficial residual stresses through the thickness of critical locations H and V
of elongated hole
Similarly, along the thickness of location ‘V’, the beneficial residual stresses vary
from a minimum magnitude of 444 MPa (compressive) on bottom plane to 633 MPa
(compressive) on a plane which is at a distance of 0.75 mm from top plane (Fig. 10).
In the region between elongated hole edge and edge of the plate, the magnitudes
of beneficial residual stresses are found to be maximum at semi-circular end and
gradually decays over the distance away from the semi-circular ends as observed
from Figs. 8 and 9. On top, mid-thickness and bottom planes, the beneficial residual
stresses remain compressive up to 1.25 mm distance from the elongated hole edge (H
and V) as shown in Figs. 8 and 9. Beyond this region, equilibrating tensile residual
stresses of small magnitudes are induced up to certain distance and further decays
to negligible magnitude as observed in Figs. 7 and 8. Similar trend is observed
for all the planes along the thickness direction. Due to the minimum magnitude of
beneficial residual stress on bottom plane (Fig. 10), the locations on this bottom plane
are the most probable locations for fatigue crack initiation/propagation. As observed
from Figs. 7, 8, 9 and 10, the through-thickness variation of cold expansion—induced
beneficial residual stresses are due to nonuniform expansion/recovery of the material
and difference in material support conditions along the thickness direction.
Thus, the developed FE simulation framework is simple to implement and capable
of adequately predict the cold expansion-induced beneficial residual stresses around
Cold Expansion of Elongated Hole: A Realistic Finite … 237
the hole and along with the thickness of elongated holes. The present predictions con-
firm the presence of significant beneficial residual stresses around the semi-circular
ends and its thickness directions. As a result, the possibility of fatigue crack initia-
tion/propagation around the semi-circular end locations is reduced thereby leading to
appreciable fatigue life enhancement of elongated hole without adding extra material
and modifying hole geometry.
5 Conclusions
References
1. Heller M, Evans RL, Allan RB (1999) Cold expansion tests for plates containing elongated
holes. Defense Science and Technology Organization Report No. DSTO-TN-0233
2. Weller S, McDonald M (2000) Stress analysis of the F-111 wing pivot fitting. Defense Science
and Technology Organization Report No. DSTO-TN-0271
3. Das R, Jones R (2012) Extending the fatigue life of a fuel flow vent hole in an aircraft com-
ponent using shape optimization. In: Proceedings of international conference on engineering
optimization, Brazil
4. Kevin WC (1997) Strain surveys of FFVH#13 and stiffener run outs #2 in the F-111 WPF for
a range of rework shapes. Defense Science and Technology Organization Report No. DSTO-
TR-0567
5. Krishnapillai K, Jones R, Peng D (2009) Three-dimensional fatigue-based structural design
optimization of fuel flow vent holes. J Eng Fail Anal 16:361–390
6. Allan RB, Heller M (1997) Stress analysis of an interference fit life extension option for a cold
expanded elongated FFVH on the F-111 aircraft. Defense Science and Technology Organization
Report No. DSTO-TR-0549
238 S. Anil Kumar and N.C. Mahendra Babu
7. Wood HA, Engle Jr RM (1979) USAF damage tolerance design handbook: guidelines for the
analysis and design of damage tolerant aircraft. Technical Report No. AFFDL-TR-79-3021
8. (2002) Aircraft structural integrity program general guidelines. Department of Defense
Handbook, MIL-HDBK-1530A (USAF)
9. Ruffin AC (1993) Extending the fatigue life of aircraft engine components by hole cold
expansion technology. J Eng Gas Turbines Power 115:165–171
10. Anil Kumar S, Mahendra Babu NC (2014) Numerical simulation of cold expansion of cut-out
with semi-circular ends. In: Proceedings of international conference on advances in design and
manufacturing 2014 (ICAD & M’14) organized by NIT-Trichy, pp 219–223. ISBN 978-93-
84743-12-3
11. Babu NCM, Jagadish T, Ramachandra K, Sridhara SN (2008) A simplified 3-D finite element
simulation of cold expansion of a circular hole to capture through thickness variation of residual
stresses. J Eng Fail Anal 15:339–348
12. Karuppanan S, Hashim MH, Waheb AA (2013) Finite element simulation of residual stresses
in cold-expanded plate. Asian J Sci Res. ISSN 1992-1454
13. Anil Kumar S, Mahendra Babu NC (2017) Effect of proximity hole on induced residual stresses
during cold expansion of adjacent holes. J Mater Today Proc
14. Kim C, Kim D-J, Seok C-S, Yang W-H (2004) Finite element analysis of the residual stress
by cold expansion method under the influence of adjacent holes. J Mater Process Technol
153–154:986–991
15. Papanikos P, Maguid SA (1999) Elasto-plastic finite-element analysis of the cold expansion of
adjacent fastener holes. J Mater Technol 92–93:424–428
16. Papanikos P, Maguid SA (1998) Three-dimensional finite-element analysis of the cold
expansion of adjacent holes. Int J Mech Sci 40:1019–1028
Effect of Module on Wear Reduction
in High Contact Ratio Spur Gears Drive
Through Optimized Fillet Stress
Abstract The current study intends to predict the wear obstruction of high contact
ratio spur gears over optimized fillet stress between the wheel and pinion. As the
fillet region of the wheel is considered to be the critical region with high stress
concentration, it becomes essential to study the wear depth at this enhanced bending
strength. The optimization of fillet stress is carried out through profile modification
technique and the tooth wear of the gear drive is examined using contact analysis
and analytical method by taking in the tooth load distribution at the contact points,
fillet stress and strength during contact for various work cycles. This examination
additionally investigates the impact of module over depletion of material from the
tooth surface of the transmission system and lastly the outcomes are conferred that
the reduced stress for the high contact ratio gear drive has enhanced wear resistance.
Keywords High contact ratio · Optimized · Module · Spur gear · Wear reduction
Nomenclature
h Wear depth in mm
ha Addendum coefficient in mm
i Gear ratio
k Tooth thickness factor
m Gear module in mm
p Base pitch in mm
s Sliding distance in mm
v Sliding velocity in m/s
z Teeth number
Symbols
Subscripts
Abbreviations
FE Finite element
FHPDTC First highest point of double tooth contact
FLPDTC First lowest point of double tooth contact
HPTC Highest point of tooth contact
LPTC Lowest point of tooth contact
SHPDTC Second highest point of double tooth contact
SLPDTC Second lowest point of double tooth contact
Effect of Module on Wear Reduction in High Contact … 241
1 Introduction
In the course of power transmission, the full load (F N ) is shared by two consecutive
teeth of the meshing spur gear for an instant of time (double pair contact regions—
BC and DE) and for the remaining instant three teeth takes the full load (triple pair
contacts regions—AB, CD and EF). Hence, it is important to get the critical contact
point (Fig. 1) and its radial distance [10] which is given by Eqs. (1)–(6).
B A
C
PD
E
F
SLPDTC FHPDTC
HPTC
2
rSHPDTC = 2 − r 2 − AF + 2 p
rap bp b + rbp
2
(2)
2
rSLPDTC = 2 − r2 − p
rap bp b + rbp
2
(3)
2
rFHPDTC = 2 − r 2 − AF + p
rap bp b + rbp
2
(4)
2
rFLPDTC = 2 − r2 − 2p
rap bp b + rbp
2
(5)
2
rLPTC = 2 − r 2 − AF
rap bp + rbp
2
(6)
AF = 2 − r 2 + r 2 − r 2 − a sin α
rap bp ag bg 0 0 (7)
3 Methodology
The fillet stress and contact pressure are assessed using FE analysis for which a code
has been developed in Ansys Parametric Design Language (APDL). Bi-dimensional
Effect of Module on Wear Reduction in High Contact … 243
five teeth model during contact (Fig. 2 and Table 1) with plain strain condition
and four-noded quadrilateral elements (PLANE 42) with two degrees of freedom is
assumed. CONTA172 (Contact) and TARGE169 (Target) is utilized for study from
ANSYS. For this examination, the face width is taken uniform throughout the plane.
Optimized edge length is found by performing a convergence study at the critical
(a) Bi-Dimensional Model teeth during contact (b) Fine meshed region
(b) Pressure distribution in contact point (b) Von Mises stress in contact point
regions of the gears. Rim portion of the gears is arrested in both the axis and the
rim of the pinion is arrested only in the radial axis whereas in the other axis, the
tangential force is applied in the rim.
Generalized equation by Anderson and Erikson [11] for calculating the tooth wear
depth is specified as
h i,n = h i,(n−1) + Jw (σH )i sp i (8)
where
Jw = HJ , wear coefficient—5 × 10−16 m2 /N [12]
h—wear depth
Sliding distance at any instant engagement point is by [12]
vp i − vg i
sp i = 2ai (9)
vp i
Semi contact width (ai ) established during engagement along the contact path is
specified as
(1−νp2 ) (1−νg2 )
4Fi Ep + Eg
ai = (10)
π b R1 + R1
( p )i ( g )i
vg i = ωg Rg i (12)
Rg i
= rog sin α0 + X i (14)
Tooth profile layout with the coordinates is presented in Fig. 3. Equation (8)
reveals the amount of wear depth after each working cycle. Totally 2000 working
Effect of Module on Wear Reduction in High Contact … 245
hi
hi (xi,yi)
F
E hi sin θ′
θ′
D hi cos θ′
kpπm C
B
A
ri
cycles is considered for the study in which the profile is updated periodically. Initial
coordinates
are specified as (x i , yi ) and the worn out coordinates are mentioned as
xi , yi ,
where
xi = Ri cos θi
yi = Ri sin θi
In profile modification technique, the tooth thickness is modified along the reference
line which is considered to be an effective technique in gear design as suggested
by Prabhu Sekar and Muthuveerapan [13–15]. By varying the tooth thickness factor
(k p ), the fillet stress of the pinion is optimized with respect to the gear. The tooth
layout utilized for analysis and the critical points for the loading [10] in the wheel
and pinion of high contact ratio gears engaged for finite element analysis are shown
in Fig. 2. In this examination, the optimized fillet stress is accomplished between the
gear drives by varying the thickness of the tooth at the pitch line. It is inferred that
246 R. Ravivarman et al.
18
16
15.69275
k p =0.4854
15
15.01889
σt maxg = σt maxp = 15.2643 14.76379
14
0.42 0.44 0.46 0.48 0.50 0.52
Tooth thickness factor
the unbalanced maximum fillet stresses begin to decrease with reduction in tooth
thickness factor (k p = 0.42) by which the required kp for achieving the balanced
fillet stress (σ t max ) decreases eventually. The optimum estimation point of tooth
thickness factor (k p ) for attaining the balanced σ t max between the pinion and wheel
is determined (Fig. 4). Maximum balanced stress at the fillet is achieved (σ t maxp =
σ t maxg = 15.269 MPa) at k p = 0.4854. Henceforth, the tooth load-carrying capacity
of the spur gear drive is improved through the design modification suggested than
the unbalanced one.
The effect of module (m) on high contact ratio spur gear drive is examined for the
increase of gear module (m = 1, 2 and 3) in optimized event. Along the contact
path of instantaneous engagement points, the predicted values of load distribution
of the gear drive, stress at the fillet region, contact pressure and wear depth are
studied. While balancing the maximum bending stress for different module, the tooth
thickness factor (k p ) is found to be constant because the ratio in which the size of the
gear changes is also constant for increasing module. From Fig. 5b, it is evidenced a
certain amount of decline in maximum balanced fillet stress for increasing module
(Table 2) is observed as stated in Lewis equation where gear module is inversely
proportional to fillet stress.
It is observed that the load share (Fig. 5a) is identical during the complete course of
mesh cycle that is primarily because of the joint impact in constant bending moment
arm and critical tooth thickness factor (k p ). As m increases, the contact pressure
decreases during the total course of the work cycle due to the influence of the radius
Effect of Module on Wear Reduction in High Contact … 247
of curvature (Fig. 5c). The accumulated wear depth reduces considerably along the
contact path because of the reduced contact pressure and sliding distance. Thus, the
material worn out from the pinion tooth surface is lower in the high contact ratio
gear drive when the module increases (Fig. 5d).
248 R. Ravivarman et al.
Fig. 5 (continued)
6 Conclusions
The maximum amount of material worn from the tooth surface on the high contact
ratio spur gear drive (for optimized fillet stress) is assessed through the contact
analysis and the subsequent conclusions have been brought down.
• In a transmission drive with a higher gear ratio, the bending strength of the wheel
and pinion is not same which is optimized through profile modification.
• In increasing module, the balanced bending stress decreases since the critical tooth
thickness of the teeth increases m number of times.
• There is a decrease in contact pressure observed during the complete course of the
mesh cycle because of the increase in the radius of curvature.
• A high contact ratio gear drive with optimized stress in the fillet region has
improved wear resistance for increasing gear module due to the influence of
reduced contact pressure and fillet stress.
References
1. Tunalioglu MS (2011) A research of tooth profile damages in internal gears. Gazi University
Institute of Science and Technology, Ankara, pp 1–5
2. Tunalioglu MS, Tuc B (2014) Theoretical and experimental investigation of wear in internal
gears. Wear 309(1):208–215
3. Archard JF (1953) Contact of rubbing flat surfaces. J Appl Phys 24:981–988
4. Andrews JD (1991) A finite element analysis of bending stresses induced in external and
internal involute spur gears. J Strain Anal 26(3):154–163
5. Dhanasekaran S, Gnanamoorthy R (2008) Gear tooth wear in sintered spur gears under dry
running conditions. Wear 265(1):81–87
6. Imrek H, Düzcükoglu H (2007) Relation between wears and tooth width modification in spur
gears. Wear 262:390–394
7. Walton D, Goodwin AJ (1998) The wear of unlubricated metallic spur gears. Wear 222:103–113
8. Sivakumar P, Gopinath K, Sundaresh S (2009) Performance evaluation of high-contact-ratio
gearing for combat tracked vehicles—a case study. Proc Inst Mech Eng Part D J Automob Eng
224:631–643
9. Prabhu Sekar R, Geo VE, Jesu Martin L (2017) A mixed finite element and analytical method to
predict load, mechanical power loss and improved efficiency in non-standard spur gear drives.
J Eng Tribol Part J IMECHE (in press)
10. Thirumurugan R, Muthuveerappan G (2011) Critical loading points for maximum fillet and
contact stresses in normal and high contact ratio spur gears based on load sharing ratio. Mech
Based Des Struct Mach 39(1):118–141
11. Andersson S, Eriksson B (1990) Prediction of the sliding wear of spur gears. Proc NORDTRIB
90
12. Flodin A, Andersson S (1997) Simulation of mild wear in spur gears. Wear 207:16–23
250 R. Ravivarman et al.
13. Prabhu Sekar R, Muthuveerappan G (2014) A balanced maximum fillet stresses on normal
contact ratio spur gears to improve the load carrying capacity through non-standard gears.
Mech Based Des Struct Mach 43:150–163
14. Ravivarman R, Palaniradja K, Prabhu Sekar R (2018) Evolution of balanced root stress and
tribological properties in high contact ratio spur gear drive. Mech Mach Theory 126:491–513
15. Ravivarman R, Palaniradja K, Prabhu Sekar R (2018) Influence of gear ratio on wear depth of
nonstandard HCR spur gear drive with balanced fillet stress. Mater Today Proc 5:17350–17359
Force Estimation on a Clamped Plate
Using a Deterministic–Stochastic
Approach
1 Introduction
The forward and inverse problems are two important areas of research in structural
dynamics. In the forward problem, the governing differential equations are solved
to obtain structural responses for known forces and constraints. In the inverse prob-
lem approach, the excitation forces or structural parameters are obtained, which are
difficult to measure directly. The regularization and statistical inversion [1] are two
types of approaches to solving an inverse problem.
The most common engineering structures are plates, and their analysis is important
in the fields of civil, mechanical, and aerospace engineering. The plate structures are
often exposed to different loading environments. There are plenty of techniques
2 Formulations
The present technique uses state-space modeling of the system. The equation of
motion of a linear dynamic system can be described by Eq. (1)
where M, C, and K are the mass, damping, and stiffness matrices, respectively.
Matrix S f is the force selection matrix, and f (t) is the external force vector. The
physical displacement, velocity, and acceleration are represented by z(t), ż(t) and
z̈(t), respectively.
Using coordinate transformation z(t) = Φp(t) and premultiplying Φ T in Eq. (1)
we obtain [6]
where matrix Φ contains modal vectors and p(t) is the modal coordinates.
For mass-normalized eigenvectors Φ, following expression holds,
Φ T MΦ = I, Φ T K Φ = Ω 2 , Φ T CΦ = Γ
where Ac , Bc , and G c are state transition matrix, input influence matrix, and output
matrix, respectively, and these matrices are defined as,
0 I 0
Ac = , Bc = , G c = Sd Φ Sv Φ
−Ω 2 −Γ M −1 S f
xn+1 = Axn + B f n
254 A. Shrivastava and A. R. Mohanty
yn = Gxn (4)
where n represents time-index. Matrices A, B, and G are the discrete time equivalent
of matrices Ac , Bc , and G c , respectively.
Use of the reduced-order model provides a two-fold advantage: it allows the
use of available measurements for limited degrees-of-freedom and it reduces the
computational time. A reduced-order model of the system is obtained by selecting
the dominant modes that adequately describe the system’s behavior. In this case, only
a few mode shape vectors (Φr ) and corresponding natural frequencies (Ωr ) are used
in Eq. (3).
The linear dynamic system with modeling and measurement errors can be described
by following set of equations,
xn+1 = Axn + B f n + wn
yn = Gxn + vn (5)
where wn and vn are the random Gaussian noise vectors correspond to process and
measurement noise with known covariance Q and R, respectively.
The KF-RLS based input force estimation technique is presented in Table 1, for
detailed derivation refer the reader to Ref. [2]. The initial state and error estimates
are assumed to be known. The innovation covariance S(n), Kalman gain K a (n), and
innovation Z (n) are identified in the first step and used in the second step for the
3 Numerical Example
A clamped plate is considered here for the verification of the present approach.
Finite element model of the plate is shown in Fig. 1, and Table 2 summarizes plate’s
properties. The Kirchhoff plate theory is used in the finite element modeling with a
4-noded quadrilateral finite element.
The following steps are followed in numerical simulations:
1. The displacement response is obtained using Eq. (4) for known transverse force
acting at a particular location.
2. Noisy response is simulated by adding Gaussian random noise (yrand ), the added
noise can be expressed as, yrand = N L × ystd × r, where N L is the percentage
Force Estimation on a Clamped Plate Using … 255
Recursive least-square
Bs (n) = G[AMs (n − 1) + I ]B
Ms (n) = [I − K a (n)C][AMs (n − 1) + I ]
K b (n) = γ −1 Pb (n − 1)BsT (n)[Bs (n)γ −1 Pb (n − 1)BsT (n) + S(n)]−1
Pb (n) = [I − K b (n)Bs (n)γ −1 Pb (n − 1)]
29 31 33 35
15 17 19 21
b
z y 1 3 5 7
x clamped
noise level, ystd is the standard deviation of the exact response, and r is random
sequences with zero mean and unit variance.
3. Force identification is performed using proposed approach.
The following parameters are selected in all the cases presented in next section:
null initial conditions, sampling frequency is 20,000 Hz, fading factor γ = 0.1.
Process noise is not considered and the measurement noise covariance R can be
calculated as (N L × ystd )2 .
The normalized mean square error between the true and estimated force has been
used to quantify the estimation accuracy,
N
( f (i) − F (i))2
i=1
Error (%) = 100 ×
N
f (i)2
i=1
In the first example, three different measurement sets are used to estimate a low-
velocity impact applied at node 17. First two modes are used to form reduced-order
model, and the displacement responses with 5% measurements noise level are used.
It is clear from Fig. 2 that the accuracy of estimated force is increased when more
number of simulated responses is included in the output vector. Here, the simulated
response means the response obtained by solving full-order finite element model,
and the predicted response means the response obtained by expanding simulated
response using SEREP method.
(a) (b)
80 80
Actual Actual
Measurement set 1 Measurement set 1
60 75 Measurement set 2
Measurement set 2
Measurement set 3 Measurement set 3
70
Force [N]
Force [N]
40
65
20
60
0
0.98 1 1.02 1.04 1.06 1.08 1.1 1.02 1.025 1.03 1.035 1.04
Time [s] Time [s]
Fig. 2 a The plot of true and estimate impact force at node 17, b zoomed portion of (a) (details of
measurement sets are given in Table 3)
Force Estimation on a Clamped Plate Using … 257
Table 3 Normalized mean square error for three different measurement sets used to estimate
low-velocity impact at node 17 (measurement noise level N L is 5%)
Measurement sets Node numbers NMSE (%)
Simulated responses Predicted responses using
SEREPa
1 17,25 4.68 3.28
2 17, 25, 27, 33 3.82 6.95
3 17, 25, 27, 33, 24, 31 2.05 7.62
a Responses are predicted using simulated responses at node 17, 25, and first two modes
In Table 3, normalized mean square errors are presented for different cases where
simulated, and predicted responses are used for force estimation. It can be observed
that the error decreases when the number of simulated response is increased. But,
when predicted responses from SEREP are used, the error is increased. This loss of
accuracy is due to the error present in predicted responses. However, the percentage
NMSE is below 10%, and only two simulated responses (at node 17 and 25) are used
to form all the measurement sets. For the sampling frequency of 20,000 Hz, the esti-
mated force convergences to the actual force within 450 time-steps. Effect of reduced-
order modeling on converges and CPU time for different sampling frequencies is
presented in Ref. [9].
In another example, random load acting at node 17 is estimated from predicted
responses at nodes 17 and 25 using simulated responses at nodes 9 and 19. The
simulated and predicted response at node 25 is shown in Fig. 3. It can be observed
that the predicted response matches well with the simulated response. The actual and
estimated force is shown in Fig. 4 for different measurement noise levels.
(a) (b)
30
Simulated Simulated
8
20 Predicted Predicted
6
Displacement [m]
Displacement [m]
10
4
0 2
-10 0
-2
-20
-4
-30
0 0.5 1 1.5 2 0.988 0.989 0.99 0.991
Time [s] Time [s]
Fig. 3 a Time history and b details of simulated and predicted (using SEREP) displacement at
node 25 due to a random load applied at node 17 (for prediction, response at nodes 9, 19, and first
two modes are considered)
258 A. Shrivastava and A. R. Mohanty
(a) (b)
40
30
Actual Actual
=0.01 =0.01
Estimated, N L Estimated, N L
=0.05 =0.05
Estimated, N L 20 Estimated, N L
20
=0.1 =0.1
Estimated, N L Estimated, N L
10
Force [N]
Force [N]
0
-20
-10
-40
0 0.5 1 1.5 2 0.818 0.82 0.822 0.824 0.826
Time [s] Time [s]
Fig. 4 a The plot of true and estimated random force at node 17, b zoomed portion of (a)
(a) (b)
Simulated Simulated
40 40
Predicted Predicted
Displacement [m]
Displacement [m]
20 20
0 0
-20 -20
0.1 0.12 0.14 0.16 0.18 0.2 0.139 0.14 0.141 0.142 0.143 0.144
Time [s] Time [s]
Fig. 5 a Time history and b details of simulated and predicted (using SEREP) displacement at
node 25 due to a sinusoidal load applied at node 17 (for prediction, response at nodes 9, 19, and
first two modes are considered)
Force Estimation on a Clamped Plate Using … 259
(a) (b)
150 60
Actual Actual
=0.01
Estimated, N L =0.01 40 Estimated, N L
100
Estimated, N L =0.05 =0.05
Estimated, N L
20
Estimated, N L =0.1 =0.1
Estimated, N L
50
Force [N]
Force [N]
0 -20
-40
-50
-60
0 0.02 0.04 0.06 0.08 0.1 0.7105 0.711 0.7115 0.712
Time [s] Time [s]
Fig. 6 a The plot of true and estimated sinusoidal force at node 17, b zoomed portion of (a)
5 Conclusions
In this paper, a Kalman filter-based input estimation technique is used for force iden-
tification on a clamped plate. Full-order finite element model is used in numerical
simulations for response generation and the SEREP technique is used for response
prediction. The proposed approach uses a reduced-order model of the dynamic
system. The proposed approach has been verified for different types of loads and
measurement noise levels.
The following conclusions can be drawn from the present study:
1. The Kalman filter and recursive least-square technique can be used to estimated
external transverse forces acting on plate type structures. Based on the obtained
results, it can be concluded that all kinds of loads can be estimated using the
present approach.
2. The SEREP technique can be used to predict responses prior to force estimation,
which can further reduce the number of response measurements. However, it is
found that the use of predicted response in force estimation causes an increase
in estimation error.
A force location identification technique is expected to combine with the present
approach to make it more effective. As future work, the proposed approach can be
verified experimentally.
References
1. Jari K, Somersalo E (2006) Statistical and computational inverse problems. Springer Science
& Business Media
2. Tuan PC, Ji CC, Fong LW, Huang WT (1996) An input estimation approach to on-line two-
dimensional inverse heat conduction problems. Numer Heat Tr A-Appl 29(3):345–363
260 A. Shrivastava and A. R. Mohanty
3. Ma CK, Chang JM, Lin DC (2003) Input forces estimation of beam structures by an inverse
method. J Sound Vib 259(2):387–407
4. Liu JJ, Ma CK, Kung IC, Lin DC (2000) Input force estimation of a cantilever plate by using
a system identification technique. Comput Methods Appl Mech Eng 190(11):1309–1322
5. Lee MH, Chen TC (2010) Intelligent fuzzy weighted input estimation method for the input
force on the plate structure. Struct Eng Mech 11(1):1
6. Lourens E, Papadimitriou C, Gillijns S, Reynders E, De Roeck G, Lombaert G (2012) Joint
input-response estimation for structural systems based on reduced-order models and vibration
data from a limited number of sensors. Mech Syst Signal Process 29:310–327
7. Shrivastava A, Mohanty AR (2017) Detection of unbalance in a rotor-bearing system using a
deterministic-stochastic approach. 24th ICSV, London, UK
8. Shrivastava A, Mohanty AR, Pekpe K, Cassar J (2017) Estimation of strain in a rotating shaft
using deterministic-stochastic approach. 1st ICEASSM Accra, Ghana
9. Shrivastava A, Mohanty AR (2019) Kalman filter-based force estimation in a clamped plate
using reduced order model and noisy measurements. Inverse Prob Sci Eng 27(8):1–24
10. Baqersad J, Niezrecki C, Avitabile P (2015) Extracting full-field dynamic strain on a wind
turbine rotor subjected to arbitrary excitations using 3D point tracking and a modal expansion
technique. J Sound Vib 352:16–29
Dynamic Analysis of Composite
Cylinders Using 3-D Degenerated Shell
Elements
Abstract This paper deals with the formulation of an 8-noded degenerated shell
finite element for modeling and analysis of laminated composite shell structures. A
MATLAB code has been developed based on the formulation to analyze the com-
posite shell structures. The formulation is capable of solving both plate and shell
structures. Use of degenerated shell elements allows the formulation to be used
for any type of shells with various shapes and thickness ratios. The formulation is
also capable of solving isotropic and laminated composite materials. The formula-
tion developed has been validated with the results available in open literatures and
software (ANSYS).
1 Introduction
Shells with variable thickness have extensively been used in many fields such as
aerospace, rocket, aviation, and submarine technology. Over the years much research
has been conducted in attempts to produce precise, competent, and reliable shell
elements. Various shell theories have been developed, over the years, based on the
thickness of the shells.
Three separate classes of shell elements have been widely used for analyzing shell
structures: flat elements, curved shell elements, and degenerated shell elements. Flat,
plate-like elements which approximate the curved shell by a faceted surface, hence
sometimes called facet elements, show completely uncoupled behavior between in-
plane stretching and bending. The coupling between in-plane stretching and bending
only appears indirectly by linking adjacent elements through the nodal degrees of
freedom. These elements are not preferred due to shortcomings such as the absence
of curvature of the elements within the element. Also, slope discontinuity between
neighboring plate elements can generate bending moments in the sections of struc-
ture where they do not exist. The interior of the individual elements will not have any
coupling between bending effects and membrane effects due to curvature of shell.
Curved shell elements are founded on several shell theories which are also quite
popular. These elements have various limitations due to fact that the shell theories
are not consistent with each other. Also, it is very difficult to find appropriate defor-
mation idealizations where truly strain-free rigid body movements are allowed. The
degenerated shell element is not based on any of the available shell theories and can
be applied over a wide range of thicknesses and curvatures. The degenerate solid
approach is used to develop this element, which is formulated on Reissner–Mindlin
assumptions where, the shear deformation and rotary inertia effect of the shell is
considered in the formulation and the 3D field is reduced to a 2D field in form of
mid-surface nodal variables.
In the late sixties, Ahmad et al. [1] developed a Mindlin-type, degenerated, curved
shell element which is quite competent as well as effortless. It can be used for any
arbitrary shape and does not depend upon any specific shell theory. In order to elim-
inate shear and membrane locking, Zeinkiewicz et al. [2] improved the degenerated
shell element developed by Ahmad et al. [1], by reducing the order of numerical
integration. Huang & Hinton [3] presented a new nine node degenerated shell ele-
ment formulation. To avoid locking phenomena, they proposed the assumed strain
method where an enhanced interpolation of the transverse shear strains in the natural
co-ordinate system is used. The nine-node degenerated shell element formulation
developed by Huang and Hinton [3] was later extended by Jayashankar et al. [4] to
conduct free vibration analysis of thick laminated composites. Because the degener-
ated shell element formulation works well for both thick and thin shells, nine-noded
degenerated shell element was preferred over conventional solid elements for the
modeling and analysis of laminated composite shell structures. Balamurugan and
Narayanan [5] developed a nine-noded degenerate shell finite element model for the
vibration and active vibration control of piezo-laminated composite plates and shells
bonded with piezoelectric sensor and actuator layers.
Most of the research that has been done on degenerated shells is focused on the
static analysis of structures such as dams, tanks, and dome, etc. Very less work is done
on dynamic analysis using degenerated shells and fewer on the dynamic analysis of
composite structures using degenerated shells. The main objective of this paper is to
develop a MATLAB code for dynamic analysis of composite cylindrical shell using
3-D degenerated element which can accurately solve different types of shells and
plates with both isotropic and composite materials.
Dynamic Analysis of Composite Cylinders … 263
2 Methodology
Some of the important aspects of the standard finite element scheme under plane
strain platform are outlined in the following sub-sections.
Fig. 1 A 16-noded solid shell element and equivalent 8-noded degenerated shell element
264 P. Tiwari et al.
where,
⎧ ⎫ ⎡⎧ ⎫ ⎧ ⎫ ⎤ ⎧ ⎫ ⎧ ⎫
⎨ xi ⎬ 1 ⎨ xi ⎬ ⎨ xi ⎬ ⎨ xi ⎬ ⎨ xi ⎬
⎢ ⎥
y = ⎣ y + yi ⎦ and V3i = yi − yi (3)
⎩ i⎭ 2 ⎩ i⎭ ⎩ ⎭ ⎩ ⎭ ⎩ ⎭
zi z i top z i bottom z i top z i bottom
For thin structures, it is convenient to replace V3i by a unit vector v3i . Thus equation
changes to
⎧ ⎫ ⎡⎧ ⎫ ⎤
⎨x ⎬
⎨ xi ⎬ ζ
y = Ni (ξ, η)⎣ yi + ti v3i ⎦ (4)
⎩ ⎭ ⎩ ⎭ 2
z zi
where v2i , v1i , α and β are unit vectors in y and x directions and rotations in x and y
directions respectively as shown in Fig. 2.
Strain definitions are needed to find strain displacement matrix [B]. Since the ele-
ments have different coordinate system from global coordinates, we need strains in
local coordinates by converting global strains to local strains.
⎧ ⎫ ⎧ ∂u
⎫
⎪
⎪ εx ⎪⎪ ⎪
⎪ ∂ x ⎪
⎪
⎪
⎪ ⎪ ⎪ ⎪
⎪ ε ⎪ ⎪ ⎪
⎪ ∂v ⎪
⎪
⎨ y ⎬ ⎨ ∂u ∂ y ∂v ⎬
ε = γx y = ∂ y + ∂ x (6)
⎪
⎪ ⎪
⎪ ⎪
⎪ ∂v ⎪
⎪
⎪
⎪ γ yz ⎪
⎪ ⎪
⎪ + ∂w ⎪
⎪
⎪ ⎪ ⎪ ∂z
⎩ ⎭ ⎩ ∂u ∂ y
⎪
⎭
γx z ∂z
+ ∂w ∂x
Also
∂u ∂u ∂ x ∂u ∂ y ∂u ∂z ∂u ∂u ∂u ∂v ∂u ∂w ∂u
= + + and = + +
∂x ∂x ∂x ∂y ∂x ∂z ∂ x ∂x ∂ x ∂u ∂ x ∂v ∂ x ∂w
where ∂∂xx , ∂∂xy , ∂∂zx , ∂∂yx , ∂∂yy , . . . . and ∂u , ∂u , ∂u , ∂v , ∂v , . . . . are direction cosines
∂u ∂v ∂w ∂u ∂v
between global and local coordinate system.
Solving Eq. (6) using Eq. (7), we get
⎧ ⎫
⎧ ⎫ ⎡ ⎤⎪ ε x ⎪
⎪ ε ⎪ 2 2 2 ⎪
⎪ ⎪
⎪
⎪
⎪ x ⎪ l
⎪ ⎢ x2
m x n x l m
x x m n
x x l n
x x ⎪
⎪ εy ⎪ ⎪
⎪
⎪ ⎪⎪ ⎥ ⎪
⎪ ⎪
⎪
⎨ εy ⎬ ⎢ l y my 2
ny 2
lym y myny lyn y ⎥⎨ ⎬
⎢ ⎥ ε
γx y = ⎢ 2l x l y 2m x m y 2n x n y l x m y + l y m x m x n y + m y n x l x n y + l y n x ⎥ z
⎪
⎪ ⎪
⎪ ⎢ ⎥⎪ γ x y ⎪
⎪ ⎪ ⎣ 2l y l z 2m y m z 2n y n z l y m z + l z m y m y n z + m z n y l y n z + l z n y ⎦⎪⎪ ⎪
⎪
⎪ γ yz ⎪
⎪
⎩ ⎪
⎭
⎪
⎪
⎪ γ ⎪
⎪
⎪
γx z 2l x l z 2m x m z 2n x n z l z m x + l x m z m z n x + m x n z l z n x + l x n z ⎩ ⎪ yz ⎪
⎭
[T ]
γ x z
(7)
Here l x , l y , l z , m x , . . . are the direction cosines which can be calculated using the
Jacobian matrix of the element.
d x dy dz d x dy dz
nx , n y , nz = , , × , , (8)
dξ dξ dξ dη dη dη
nx , n y , nz
l x , l y , l z = (9)
nx , n y , nz
m x , m y , m z = n x , n y , n z × lx , l y , lz (10)
where,
⎧ ∂u
⎫ ⎡ ∂ Ni ⎤
⎪
⎪
⎪ ∂ξ ⎪
⎪
⎪ ∂ξ
0 0 − ∂∂ξNi ζ t2i l2i ∂ Ni ζ ti l1i
∂ξ
⎪
⎪ ∂u ⎪
⎪ ⎢ ∂ Ni Ni ζ ti l2i
2
∂ Ni ζ ti l1i ⎥
⎪
⎪ ∂η ⎪
⎪ ⎢ 0 0 − ∂∂η ⎥⎧ ⎫
⎪
⎪ ∂u ⎪
⎪ ⎢
∂η 2 ∂η 2
⎥
⎪
⎪ ⎪
⎪ ⎢ 0 0 0 − Ni 2ti l2i Ni ti l1i
⎥⎪ u ⎪
⎪
⎪ ∂ζ ⎪
⎪ ⎢ 2 ⎥⎪⎪ i⎪ ⎪
⎪
⎪ ∂v ⎪
⎪ 8 ⎢
∂ Ni
− ∂∂ξNi ζ ti2m 2i ∂ Ni ζ ti m 1i
⎥⎪⎪ ⎪
⎪
⎨ ∂ξ ⎬
⎢
0 ∂ξ
0 ∂ξ 2 ⎥⎨ vi ⎬
∂v
= ⎢ 0 ∂ Ni
0 Ni ζ ti m 2i
− ∂∂η ∂ Ni ζ ti m 1i ⎥ wi (12)
⎪ ∂η ⎪ ⎢ ∂η 2 ∂η 2 ⎥⎪ ⎪
⎪
⎪ ⎪
⎪ i=1 ⎢ ⎥⎪ ⎪
⎥⎪⎪ αi ⎪
∂v
⎪
⎪ ⎪
⎪ ⎢ 0 0 0 − Ni t2i m 2i Ni ti m 1i
⎪
⎪
⎪
∂ζ
⎪
⎪ ⎢ 2 ⎥⎪⎩β ⎪ ⎭
⎪
⎪
∂w ⎪
⎪ ⎢ 0 0 ∂ Ni
− ∂∂ξNi ζ ti2n 2i ∂ Ni ζ ti n 1i
⎥ i
⎪
⎪ ∂ξ ⎪
⎪ ⎢ ∂ξ ∂ξ 2 ⎥
⎪
⎪ ∂w ⎪
⎪ ⎣ 0 0 ∂ Ni
− ∂∂ηNi ζ ti n 2i ∂ Ni ζ ti n 1i ⎦
⎪
⎪ ∂η ⎪
⎪ ∂η 2 ∂η 2
⎩ ∂w ⎭ 0 0 0 − Ni t2i n 2i Ni ti n 1i
∂ζ 2
[∂ N ]
Stress strain relationship matrix [D] is considered in local coordinate system and
taken as
Dynamic Analysis of Composite Cylinders … 267
⎡ ⎤
1 υ 0 0 0
⎢υ ⎥
⎢ 1 0 0 0 ⎥
E ⎢ (1−υ) ⎥
[D] = ⎢ 0 0 0 0 ⎥ (14)
1 − υ2 ⎢ 2
K s (1−υ) ⎥
⎣0 0 0 2
0 ⎦
K s (1−υ)
0 0 0 0 2
where E, υ and K s are modulus of elasticity, Poisson’s ratio and shear correction
factor for the given Isotropic material.
For orthotropic materials, the stress strain relationship matrix [D] will be taken
as
⎡ ⎤
E 1 υ21 E 1 0 0 0
⎢υ E E 0 ⎥
⎢ 12 2 2 0 0 ⎥
1 ⎢ ⎥
[D] = ⎢ 0 0 G 12 0 0 ⎥ (15)
1 − υ12 υ21 ⎢ ⎥
⎣ 0 0 0 K s G 13 0 ⎦
0 0 0 0 K s G 23
[M]ẍ + [K ]x = 0 (19)
We have considered different Gauss points for bending and shear to avoid shear
locking and to get more accurate results.
For Isotropic materials, we have taken 3 Gauss points each in ξ and η directions and
two Gauss points in ζ direction in case of bending. For shear, we have taken 2 Gauss
points each in ξ and η directions and one Gauss point in ζ direction.
tl tl
ζ = ζl + 1 + ζ and dζ = dζ (20)
t t
v = w = β = 0 at x = 0, a
u = w = α = 0 at y = 0, b
The elastic properties of the lamina with respect to the material axes has been
taken as E 1 /E 2 = 10, G12 = G13 = 0.6 E 2 , G23 = 0.5 E 2 , ν 12 = 0.25, and ρ = 1.
Thickness ratio (a/h) is assumed to be 5.
Since it is clear form Table 1 that the program converges at 10 × 10, so we have
taken mesh size as 10 × 10 for square plate.
u = v = w = α = β = 0 at z = 0
√
Table 1 Non-dimensional frequency parameter ω̄ = ωa 2 / h ρ/E 2 of a simply supported (SSSS)
square cross ply (0/90/90/0) plate. a/h = 5
Mode 5×5 6×6 7×7 8×8 9×9 10 × 10 12 × 12 15 × 15
no.
1 7.714 7.712 7.711 7.711 7.711 7.710 7.710 7.710
2 12.169 12.168 12.168 12.168 12.167 12.167 12.167 12.167
3 12.169 12.168 12.168 12.168 12.167 12.167 12.167 12.167
4 13.507 13.490 13.483 13.479 13.477 13.476 13.474 13.474
270 P. Tiwari et al.
The results obtained from MATLAB code, created for solving free vibration of
isotropic and laminated composite plates and shells, are compared with the results
available in open literature and are in good agreement with the literature available in
open source.
We have compared our result with the natural frequencies of composite plate and
isotropic circular cylinder, available in literatures and result generated by ANSYS.
The problem considered here is a cross ply (0/90/90/0) of square cross section and
having simply supported boundary conditions. The elastic properties of the lamina
with respect to the material axes has been taken as E 1 /E 2 = 10, G12 = G13 = 0.5 E 2 ,
G23 = 0.6 E 2 , ν 12 = 0.25, and ρ = 1. Thickness ratio (a/h) is assumed to be 5.
As we can see from Table 3, the results obtained for composite plates from present
formulation are in good agreement with the results available in open literature.
√
Table 3 Non-dimensional frequency parameter ω̄ = ωa 2 / h ρ/E 2 of a simply supported (SSSS)
square cross ply (0/90/90/0). a/h = 5
E 1 /E 2 ratio
10 20 30 40
Exact [6, 7] 8.2982 9.5671 10.326 10.854
Liew et al. [8] 8.4298 9.6782 10.424 10.944
FSDT (present) 8.1974 9.7612 10.6894 11.3315
Degenerated (present) 7.985 9.4399 10.3628 11.016
Dynamic Analysis of Composite Cylinders … 271
An Isotropic cylindrical shell of 0.6 m height, 0.15 m outer radius and 0.1 mm
thickness is considered. We have assumed Aluminum as material with E = 71 GPa,
υ = 0.33 and density as 2770 kg/m3 . Bottom side of shell is assumed to be fixed and
top is assumed to be free.
It is evident from Table 4 that the result obtained for isotropic cylindrical shell from
present formulation are in good agreement with results available in open literature
as well as the results obtained from software (ANSYS).
The formulation developed based on 3-D degenerated shell elements has been
validated by comparing results obtained for composite plates and isotropic shells. The
results are in good agreement with open literature and software. Thus we conclude
that the formulation is correct and is able to produce accurate results.
From parametric study, it is evident that the natural frequency of composite shell
is increasing as we increase the thickness of the shell, which is consistent with our
understanding of composite shells. Other parameters such as fiber orientation and no
of lamina are also varied to understand their impact on natural frequency.
4 Conclusion
In the present work, a finite element formulation has been created for the dynamic
analysis of laminated composite shell using 8-noded 3-D degenerated shell element.
Present formulation is capable of analysis both isotropic and laminated composite
shells of arbitrary geometry. The results generated so far are in good agreement with
the results available in open literature as well as with the software generated results.
References
1. Ahmad S, Irons BM, Zienkiewicz OC (1970) Analysis of thick and thin shell structures by
curved finite elements. Int J Numer Meth Eng 2(3):419–451
2. Zienkiewicz OC, Taylor RL, Too JM (1971) Reduced integration technique in general analysis
of plates and shells. Int J Numer Meth Eng 3(2):275–290
3. Huang HC, Hinton E (1986) A new nine node degenerated shell element with enhanced
membrane and shear interpolation. Int J Numer Meth Eng 22(1):73–92
4. Jayasankar S, Mahesh S, Narayanan S, Padmanabhan C (2007) Dynamic analysis of layered
composite shells using nine node degenerate shell elements. J Sound Vib 299(1–2):1–11
5. Balamurugan V, Narayanan S (2008) A piezolaminated composite degenerated shell finite ele-
ment for active control of structures with distributed piezosensors and actuators. Smart Mater
Struct 17(3):35031
6. Khdeir AA, Librescu L (1988) Analysis of symmetric cross-ply elastic plates using a higher-order
theory, Part II: buckling and free vibration. Compos Struct 9:259–277
7. Reddy JN (1997) Mechanics of Laminated Composite Plates. Theory and Analysis. CRC Press,
Boca Raton, FL
276 P. Tiwari et al.
8. Liew KM, Huang YQ, Reddy JN (2003) Vibration analysis of symmetrically laminated plates
based on FSDT using the moving least squares differential quadrature method. Comput Methods
Appl Mech Eng 192(19):2203–2222
9. Rawat A, Matsagar V, Nagpal AK (2016) Finite element analysis of thin circular cylindrical
shells. Proc Indian Natl Sci Acad 82(2):349–355
A New Hybrid Unified Particle Swarm
Optimization Technique for Damage
Assessment from Changes of Vibration
Responses
Abstract Unified particle swarm optimization (UPSO) and artificial bee colony
(ABC) are two effective swarm based techniques to solve inverse damage detec-
tion problems. A hybrid algorithm of ABC and UPSO named UPSO_Scout has
been proposed in the present paper. Comparisons among these three algorithms are
demonstrated using three structural problems (beam, plane truss and space truss)
considering natural frequencies and mode shapes as damage indicators. Proposed
algorithm shows improved performance than UPSO and ABC for damage detection.
Slow convergence speed of ABC reduces its effectiveness for larger problem. UPSO
is sometimes prone to fall in local minima despite of its faster convergence speed.
Efficacy of the proposed algorithm has also been examined in noisy environment.
1 Introduction
Structural health monitoring (SHM) has been turned into a significant area of study
within the civil, mechanical and aerospace engineering research community in recent
years. Damage identification at initial phases is a crucial part of structural health mon-
itoring to evaluate the safety and to ward off the cataclysmic failure of the structure.
A number of non-destructive tests are available to identify and quantify the damages.
Among all the methods, damage identification using vibrational characteristics has
become popular in recent years. Existence of damage changes the dynamic response
parameters (frequency, mode shapes, FRF, etc.) by altering the physical properties
(mass, stiffness, damping, etc.) of the structures. Identification and quantification of
damage can be carried out through investigating these changes in dynamic parame-
ters. This is called inverse method for damage detection. Adams et al. [1], Cawley
and Adams [2], Messina et al. [3] used change in natural frequency to detect damage
in beams. Yang [4] used modal residual force criteria as damage indicator. Majumder
et al. [5] used frequency change to detect damages in truss. Seyedpoor [6], Dinh-
Cong et al. [7] used modal strain energy for damage detection for beam and truss.
Mohan et al. [8] used frequency response function (FRF) to detect damage in beam
and space frame. Kaveh and Zolghadr [9], Fei Kang et al. [10], Nanda et al. [11],
Ding et al. [12], Nhamage et al. [13], Mishra et al. [14] , Barman et al. [15, 16],
Kim et al. [17] used combined frequency and mode shapes data to quantify damage
in different structures, as beam, plate, frame, truss etc. Another crucial attribute in
inverse method of damage detection is to employ efficient optimization tool to save
computational time and cost. Various soft computing tools such as particle swarm
optimization (PSO) [6, 8], genetic algorithm (GA) [18], neural network (NN) [19],
artificial bee colony (ABC) [12], ant colony optimization (ACO) [5], modified charge
system search (CSS) [9], fuzzy cognitive map (FCM) [20], unified particle swarm
optimization (UPSO) [11, 15, 16] antlion optimization (ALO) [14] etc. have been
used by the researchers to solve the inverse problem commendably.
So, above literature review has given a clear idea about how metaheuristic tech-
niques have been gaining popularity from last two–three decades in the field of
damage detection of structures. However, no optimization techniques are full proof
as each of them has their advantages and disadvantages in particular field of appli-
cation. So, the researchers have been trying to make new robust hybrid algorithm
by mixing two or more original algorithms as per the suitability of their particular
field of applications. Hybrid algorithms show improved performance compared to
the parent algorithms. So, keeping that in mind a new and effective optimization
algorithm UPSO_Scout, which is a hybrid algorithm combining the features of ABC
into UPSO has been proposed in the present study. Proposed algorithm has been used
to solve inverse problem with an aim for detecting damages in the beam and truss of
varying complexity. Natural frequencies and mode shapes are used as the diagnostic
parameters. Performance of the proposed algorithm has also been scrutinized in case
of noisy vibration data.
2 Mathematical Background
The finite element model for isotropic beam has been formulated using one-
dimensional two noded element. Each node has two degrees of freedom (δ y , θz )
(Fig. 1a). Beam rigidity E I and cross-sectional area A are considered same through-
A New Hybrid Unified Particle Swarm Optimization … 279
(a) (b)
out the length of the element. Element stiffness matrix and element mass matrix,
respectively, are given by [21].
⎡ ⎤ ⎡ ⎤
12 6l −12 6l 156 22l 54 −13l
EI ⎢ 6l 4l 2 −6l 2l 2 ⎥ ρ Al ⎢ 22l 4l 2 13l −3l 2 ⎥
[K ]e = 3 ⎢
⎣
⎥ [M]e = ⎢ ⎥
l −12 −6l 12 −6l ⎦ 420 ⎣ 54 13l 156 −22l ⎦
6l 2l 2 −6l 4l 2 −13l −3l 2 −22l 4l 2
Here, l is length of the element and ρ is mass density of the material. Finite ele-
ment formulation of truss structures have been formulated using two-dimensional
two noded truss element with two (δx , δ y ) and three (δx , δ y , δz ) degrees of free-
dom (Fig. 1b), respectively, for plane and space truss. Axial rigidity E A and cross-
sectional area A are considered same throughout the length of the element. Element
stiffness matrix and element mass matrix in element coordinate system, respectively,
are given by [21].
E A 1 −1 ρ Al 2 1
[k]e = [m]e =
l −1 1 6 12
Here, l is length of the element and ρ is mass density of the material used.
These matrices can be transformed into global coordinate system using the following
equation, respectively.
Here, c = cos α, s = sin α. α is the inclination of the element axis with global X
axis.
280 S. K. Barman et al.
Only one damage parameter per structural element is considered as all the materials
are considered isotropic. So, damaged stiffness matrix of an element can be written
as
where K dam and K undam are damaged and undamaged element stiffness matrix,
respectively. α is damage parameter.
The objective function considered here is given below, Nanda et al. [11]:
i=n m 2
1
fi
i=n
F(%damage in each element) = − 1 + (1 − MACii ) (2)
n i=1 f ia i=1
where
T
ϕ ϕai 2
MACii = T mi
T (3)
ϕmi ϕmi ϕai ϕai
Here, n stands for the number of modes, f and ϕ denote the frequency and cor-
responding mode shape, respectively. Suffix m and a stand for measured response
A New Hybrid Unified Particle Swarm Optimization … 281
Where, Hn and H are respectively with noise and noise free data; NL is noise level
(0.05 relates to 5 % noise level); is a random number between 0 and 1.
3 Optimization Algorithms
Particle swarm optimization (PSO) was developed by Kennedy and Eberhart [22] as
a stochastic optimization algorithm based on concepts and rules that govern socially
organized populations in nature, such as bird flocks, fish schools and animal herds
to search for food or to avoid predators. UPSO is a refinement of PSO and was first
proposed by Parsopoulos and Vrahatis [23]. Basic steps involved in UPSO are as
follows:
(i) Random initialization of swarm position and velocity:
xi j ∈ [xmin , xmax ], vi j ∈ [−vmax , vmax ],
vmax = 0.5 ∗ (xmax − xmin ), ∀i ∈ N , ∀ j ∈ S
(ii) Evaluate the swarm.
(iii) Cycle = 1.
(iv) Repeat.
(v) Calculate pb: the best position ever visited by individual particle; gb: the best
position ever visited by all particles; l b: best position ever visited by any of
its neighbour.
(vi) Update velocity:
Vit+1
j = μG it+1
j + (1 − μ)L i j , Vi j
t+1 t+1
∈ [−vmax , vmax ] (5)
282 S. K. Barman et al.
Here,
t
j = χ vi j + c1 r 1 pbi j − x i j + c2 r 2 gbi j − x i j
G it+1 t t
(6)
t
j = χ vi j + c1 r 3 pbi j − x i j + c2 r 4 lbi j − x i j
L it+1 t t
(7)
t log(2.0)
μ(t) = exp −1 (8)
tmax
Update position:
j = x i j + Vi j , x i j ∈ [x min , x max ]
xit+1 t t+1 t+1
(9)
Artificial bee colony (ABC) algorithm [25] simulates the behaviour pattern of a
honey bee colony. Inside a bee hive, bees share information about food source with
each other in the dancing area through waggle dance. In ABC algorithm, the position
of a food source represents a possible solution to the optimization problem and the
nectar amount of a food source corresponds to the quality (fitness) of the associated
solution. The number of the employed bees or the onlooker bees is equal to the
number of solutions in the population. The steps involved in ABC algorithm are
(i) Random initialization of population: xi j ∈ [xmin , xmax ], ∀i ∈ S N , ∀ j ∈ D
(ii) Fitness evaluation.
(iii) Cycle = 1.
(iv) Repeat.
(v) Produce new solution for employed bees (Eq. 10) and evaluate them.
t
j = x i j + rand[−1, 1] x i j − x k j
xit+1 t t
(10)
SN
pi = Fitnessi / Fitnessn
n=1
(viii) Produce new solution for onlooker bees (Eq. 10) based on the probability and
evaluate them.
(ix) Apply greedy selection process for onlooker bees.
(x) Determine the abandoned solution for the scout bee and if exists replace it
with a randomly created solution:
limit = S N ∗ D (12)
In the present study the termination criteria is when either the objective function
assumes a predefined value or maximum iteration is reached.
In this section, a hybrid UPSO algorithm has been proposed by combining the scout
bee phase of ABC algorithm with the existing UPSO algorithm to enhance its explo-
ration property, and hence to enhance its performance for damage detection problems.
The algorithm is named as scout unified particle swarm optimization (UPSO_Scout).
The steps involved in this algorithm are mentioned below:
(i) Random initialization of swarm position and velocity:
xi j ∈ [xmin , xmax ], vi j ∈ [−vmax , vmax ],
vmax = 0.5 ∗ (xmax − xmin ), ∀i ∈ N , ∀ j ∈ S
(ii) Evaluate the swarm.
(iii) Cycle = 1.
(iv) Repeat.
(v) Calculate pb, gb, l b.
284 S. K. Barman et al.
4 Numerical Results
The results obtained from the numerical study for the three structures have been
presented here.
Damage has been modelled as percentage reduction in the bending stiffness for the
particular element. Two damage cases are selected for demonstration of the algo-
rithms and mentioned in Table 4. First eight natural frequencies of damaged struc-
tures are mentioned in Table 5. 100 numerical experiments are conducted for each
damage case. Damage prediction for E1 and E2 damage cases by three algorithms
are shown in Fig. 3a and b, respectively. All three algorithms are capable of detecting
damages for both damage cases E1 and E2. A summary of all the numerical experi-
ments conducted in the cantilever beam has been presented in Table 6. Mean number
of function evaluation represents the average convergence speed. Among the three
algorithms ABC has the lowest convergence speed (higher mean function evaluation)
286
(a) (b)
0.06 0.25
UPSO UPSO
Damage severity
Damage severity
In this case of 25 member plane truss, damage has been modelled as the percentage
reduction in axial stiffness of the particular member. Two damage cases have been
considered and mentioned in Table 7. First eight natural frequencies of damaged
structures are mentioned in Table 8. 100 numerical experiments are conducted for
each damage case. Damage prediction for E3 and E4 damage cases by UPSO and
UPSO_Scout is shown in Fig. 4a and b, respectively. Comparison among the three
algorithms can be performed from the summary of the experiments, presented in
Table 9. ABC has failed to minimize the objective function and thus failed to detect
the damage in both the cases. UPSO has slightly faster average convergence rate
compared to UPSO_Scout for all the cases. However, the success rate of UPSO_Scout
(a) (b)
0.06 0.25
UPSO UPSO
0.05
Damage severity
Damage severity
UPSO_Scout 0.2 UPSO_Scout
0.04 Actual Actual
0.15
0.03
0.1
0.02
0.01 0.05
0 0
1 15 20 24 25 1 15 20 24 25
Member number Member number
is much higher (100%, 81%) for damage case (E3, E4) in comparison with the
success rate of UPSO (85%, 57%) and ABC (0%, 0%). In case of E3 UPSO has
taken least computational time, whereas, in case of E4 UPSO_Scout has taken least
computational time. So, UPSO_Scout has been found to have superior performance
in comparison with UPSO and ABC.
A New Hybrid Unified Particle Swarm Optimization … 291
In the case of 72 member space truss, damage has been modelled in similar manner as
the plane truss. Two damage cases have been considered and mentioned in Table 10.
First eight natural frequencies of damaged structures are mentioned in Table 11.
100 numerical experiments are conducted for each damage case. Damage prediction
for E5 and E6 damage cases by UPSO and UPSO_Scout are shown in Fig. 5a, b.
Performance of the three algorithms can be compared from the summary of the
experiments, presented within Table 12. ABC has failed to detect the damage in both
the cases. UPSO_scout has faster average convergence speed compared to UPSO for
damage case E5. The difference in convergence speed between these two algorithms
is not significant enough for damage case E6. However, success rate of UPSO_Scout
is much higher (100%, 77%) for damage case (E5, E6) in comparison with the success
rate of UPSO (86%, 19%) and ABC (0%, 0%). Again, UPSO_scout is found to be
least time consuming for both the damage cases. Thus, UPSO_Scout has been found
to have superior performance in comparison with UPSO and ABC.
(a) (b)
0.06 0.25
UPSO UPSO
Damage severity
0.05
Damage severity
The relative measurement error of frequency is about 1% and the mode shape is
about 10% in practical cases [29]. To incorporate this error we have considered
a range for different noise levels, where the noise for frequency is varying from
0 to 1% and the same for mode shapes is varying from 0 to 10%. The different
noise levels are mentioned in Table 13. As from the previous section, it is clear
that overall performance of UPSO_Scout is better than the other two; in this section
only capability of the same is investigated for damage detection using noisy data.
Figure 6 depicts the damage detection scenario using noisy data for six damage
cases (E1–E6) considered before. For single element damage cases (E1, E3 and E5)
where the amount of damage is very low (5%), the algorithm gives almost correct
representation of damage up to noise level N2. But, for higher level of noise (N5, N10)
the algorithm fails to differentiate between the change in vibrational parameter for
noise and damage, as damage is very low and gives false representation of damage.
Whereas the three element damage cases (E2, E4, E6) have comparatively higher
magnitude of damage in the range of 10–20% and these cases have been investigated
for noise cases starting from N5 and gradually for higher noise levels (N6, N7, N10).
For these damage cases except E6 the algorithm has been able to detect damages
almost accurately for noise level up to N7. At noise level N10 the algorithm detects
the damage location correctly but failed to quantify the damages correctly. For E6
(Fig. 6f) the algorithm has been able to detect damages for noise level N1 but failed
for noise level N2.
A New Hybrid Unified Particle Swarm Optimization … 293
(a) (b)
0.06 0.3
N1 N2 N5 N10 Actual N5 N6 N7 N10 Actual
0.05 0.25
Damage severity
Damage severity
0.04 0.2
0.03 0.15
0.02 0.1
0.01 0.05
0 0
1 2 3 4 5 7 8 1 2 4 5 8 10
Member number Member number
(c) (d)
0.06 0.3
Damage severity
N1 N2 N5 Actual N5 N6 N7 N10 Actual
Damage severity
0.05 0.25
0.04 0.2
0.03 0.15
0.02 0.1
0.01 0.05
0 0
1 9 14 15 16 24 25 1 9 14 15 17 20 24
Member number Member number
(e) (f)
0.3
N1 N2 N5 Actual 0.25 N1 N2 Actual
Damage severity
Damage severity
0.15
0.2
0.1 0.15
0.1
0.05
0.05
0 0
1 13 17 18 31 50 72 1 17 29 35 41 68 70
Member number Member number
Fig. 6 Damage prediction by the UPSO_Scout in noisy environment : a damage case E1 b damage
case E2 c damage case E3 d damage case E4 e damage case E5 f damage case E6
5 Conclusions
In the present study, the finite element formulation is developed for beam, plane
and space trusses to find out natural frequencies and mode shapes for undamaged
and damaged structures. The finite element-based computer code is developed in the
MATLAB environment. Inverse algorithms based on UPSO, ABC and UPSO_Scout
are developed to detect and quantify damages in such structures. The requirements
of number of modes in damage detection have been decided based on cumulative
effective modal mass for all three structures. The proposed algorithm is found to be
more capable to recover from local minima without losing its convergence speed in
comparison with UPSO and ABC. UPSO_Scout has produced more number of suc-
cessful runs compared to UPSO and ABC. Thus, damage can be predicted correctly
294 S. K. Barman et al.
with the help of UPSO_Scout even with a small number of numerical experiments,
and overall computational efforts can be reduced significantly. Thus UPSO_Scout
algorithm is found to be more reliable than UPSO and ABC to assess damages in
structures. UPSO_Scout algorithm also performs satisfactorily in damage assessment
in case of noisy environment.
Acknowledgements This research work is financially supported by ISRO (Indian Space Research
Organisation) IIT Kharagpur cell. The authors are grateful to ISRO cell for their financial support
to carry out the research work at Department of Aerospace Engineering, IIT, Kharagpur.
References
1. Adams RD, Cawley P, Pie CJ, Stone BJ (1978) A vibration technique for non-destructively
assessing the integrity of structures. J Mech Eng Sci 20:93–100. https://doi.org/10.1243/JMES_
JOUR_1978_020_016_02
2. Cawley P, Adams RD (1979) The location of defects in structures from measurements of natural
frequencies. J Strain Anal Eng Des 14:49–57. https://doi.org/10.1243/03093247V142049
3. Messina A, Williams EJ, Contursi T (1998) Structural damage detection by a sensitivity and
statistical-based method. J Sound Vib 216:791–808. https://doi.org/10.1006/jsvi.1998.1728
4. Yang QW (2009) A numerical technique for structural damage detection. Appl Math Comput
215:2775–2780. https://doi.org/10.1016/j.amc.2009.08.039
5. Majumdar A, Maiti DK, Maity D (2012) Damage assessment of truss structures from changes in
natural frequencies using ant colony optimization. Appl Math Comput 218:9759–9772. https://
doi.org/10.1016/j.amc.2012.03.031
6. Seyedpoor SM (2012) A two stage method for structural damage detection using a modal strain
energy based index and particle swarm optimization. Int J Non Linear Mech 47:1–8. https://
doi.org/10.1016/j.ijnonlinmec.2011.07.011
7. Dinh-Cong D, Vo-Duy T, Ho-Huu V, Nguyen-Thoi T (2019) Damage assessment in plate-like
structures using a two-stage method based on modal strain energy change and Jaya algorithm.
Inverse Probl Sci Eng 27:166–189. https://doi.org/10.1080/17415977.2018.1454445
8. Mohan SC, Maiti DK, Maity D (2013) Structural damage assessment using FRF employing
particle swarm optimization. Appl Math Comput 219:10387–10400. https://doi.org/10.1016/j.
amc.2013.04.016
9. Kaveh A, Zolghadr A (2015) An improved CSS for damage detection of truss structures using
changes in natural frequencies and mode shapes. Adv Eng Softw 80:93–100. https://doi.org/
10.1016/j.advengsoft.2014.09.010
10. Kang F, Li J, Xu Q (2012) Damage detection based on improved particle swarm optimization
using vibration data. Appl Soft Comput 12:2329–2335. https://doi.org/10.1016/j.asoc.2012.
03.050
11. Nanda B, Maity D, Maiti DK (2014) Modal parameter based inverse approach for structural joint
damage assessment using unified particle swarm optimization. Appl Math Comput 242:407–
422. https://doi.org/10.1016/j.amc.2014.05.115
12. Ding ZH, Huang M, Lu ZR (2016) Structural damage detection using artificial bee colony
algorithm with hybrid search strategy. Swarm Evol Comput 28:1–13. https://doi.org/10.1016/
j.swevo.2015.10.010
13. Nhamage IA, Lopez RH, Miguel LFF (2016) An improved hybrid optimization algorithm
for vibration based-damage detection. Adv Eng Softw 93:47–64. https://doi.org/10.1016/j.
advengsoft.2015.12.003
A New Hybrid Unified Particle Swarm Optimization … 295
14. Mishra M, Barman SK, Maity D, Maiti DK (2019) Ant lion optimisation algorithm for structural
damage detection using vibration data. J Civ Struct Heal Monit 9:117–136. https://doi.org/10.
1007/s13349-018-0318-z
15. Barman SK, Maiti DK, Maity D (2016) A simple inverse technique to assess delamination in
composite beam. In: Proceedings of the 61st congress of the indian society of theoretical and
applied mechanics, VIT University Vellore, Tamil Nadu, India, 11–14 Dec, pp 1–8
16. Barman SK, Jebieshia TR, Tiwari P, Maiti DK, Maity D (2019) Two-stage inverse method
to detect delamination in composite beam using vibration responses. AIAA J 57:1312–1322.
https://doi.org/10.2514/1.J057471
17. Kim N Il, Kim S, Lee J (2019) Vibration-based damage detection of planar and space trusses
using differential evolution algorithm. Appl Acoust 148:308–321. https://doi.org/10.1016/j.
apacoust.2018.08.032
18. Maity D, Tripathy RR (2005) Damage assessment of structures from changes in natural fre-
quencies using genetic algorithm. Struct Eng Mech 19:21–42. https://doi.org/10.12989/sem.
2005.19.1.021
19. Zapico JL, González MP, Friswell MI, Taylor CA, Crewe AJ (2003) Finite element model
updating of a small scale bridge. J Sound Vib 268:993–1012. https://doi.org/10.1016/S0022-
460X(03)00409-7
20. Beena P, Ganguli R (2011) Structural damage detection using fuzzy cognitive maps and Hebbian
learning. Appl Soft Comput 11:1014–1020. https://doi.org/10.1016/j.asoc.2010.01.023
21. Chandrupatla TR, Belegundu AD (2002) Introduction to finite elements in engineering. Prentice
Hall, Upper Saddle River, New Jersey, USA
22. Kennedy J, Eberhart R (1995) Particle swarm optimization. In: Proceedings of IEEE
international conference on neural networks, Piscataway, NJ, pp 1942–1948
23. Parsopoulos KE, Vrahatis MN (2005) Unified particle swarm optimization for solving con-
strained engineering optimization problems. In: Proceedings of the first international confer-
ence on advances in natural computation—Volume Part III. Springer, Berlin, Heidelberg, pp
582–591
24. Parsopoulos KE, Vrahatis MN (2010) Particle swarm optimization and intelligence
25. Karaboga D, Akay B (2009) A comparative study of artificial bee colony algorithm. Appl Math
Comput 214:108–132. https://doi.org/10.1016/j.amc.2009.03.090
26. Esfandiari A, Bakhtiari-Nejad F, Rahai A (2013) Theoretical and experimental structural dam-
age diagnosis method using natural frequencies through an improved sensitivity equation. Int
J Mech Sci 70:79–89. https://doi.org/10.1016/j.ijmecsci.2013.02.006
27. Priestley MJN, Seible F, Calvi GM (1996) Seismic design and retrofit of bridges. Wiley, New
York, USA
28. ASCE (2002) SEI/ASCE7-02: minimum design loads for buildings and other structures,
SEI/ASCE 7. Reston, Virginia, USA
29. Hao H, Xia Y (2002) Vibration-based damage detection of structures by genetic algorithm. J
Comput Civ Eng 16:222–229. https://doi.org/10.1061/(ASCE)0887-3801(2002)16:3(222)
Semi-active Control of a Three-Storey
Building Structure
Abstract This paper proposes two different semi-active control methods. Linear
Quadratic Gaussian (LQG) controller is used as the first method for feedback control
of the structure. The second method has used structural stabilization using pole
placement and optimizing the response by pattern search genetic algorithm approach.
Magnetorheological (MR) damper is attached to the structure to provide the required
damping force. A clipped optimal control method is used to provide the required
voltage to the MR damper to generate the desired control force to the structure.
1 Introduction
Semi-active [1] control of a system has come to acceptance more than active and
passive control systems in structural control nowadays. These systems have managed
to possess the adaptability of active control systems along with their intrinsic stability
as passive control systems. These systems can be operated using very low power.
Different control methods [2] and optimization methods can be implemented easily
with semi-active systems. This can further exceed the performance of that of active
control systems.
Magnetorheological (MR) damper [3] is one semi-active device that seems to
be particularly suitable for seismic protection. MR dampers [4, 5] are semi-active
devices. Therefore, they have very low power requirements and a controllable damp-
ing force. They can be made to achieve large force capacities as well. They also have
a low sensitivity to temperature changes.
This study proposes two different semi-active control methods. The first method
is Linear Quadratic Gaussian (LQG) controller used for semi-active feedback con-
trol of the structure. The second method involves structural stabilization using pole
placement and adding compensatory gain and optimizing the response by pattern
search genetic algorithm (GA) approach. Clipped optimal control strategy has been
used to calculate the voltage required for MR damper to produce the desired control
damping force to the structure.
2 Mathematical Formulation
A three-story shear building model has been taken up from Mahdi and Abdelhafid
[6] and is shown in Fig. 1. A linear MR damper is attached to produce optimal control
damper force from the required calculated voltage using clipped optimal controller.
State space design of the model has been done to implement the proposed control
designs. The controller K c (s) has been developed both by using LQG methods and
genetic algorithms each. Block diagrams of this semi-active control system are shown
in Figs. 1 and 2.
ż = Az + Bx (1)
y = C z + Dx (2)
where [E] and [0] are identity and zeros matrices of suitable sizes. The vectors z and
x can be written in Eq. (4)
u̇ f
z= and x = (4)
u ẍ g
The structure is controllable up to four numbers of states out of six states. The
control force required is produced by a single linear MR damper using clipped optimal
control strategy.
M ü + C u̇ + K u = f − Mẍ g (5)
where f is the controlled damper force, defined by Eqs. (6) and (7), u = [ u 1 u 2 u 3 ]
are the floor displacements of the frame structure with respect to the ground.
The state space equation and output equations are modified in Eqs. (6) and (7).
ż = Az + B f + G ẍ g (6)
y = C z + D f + vm (7)
The controller K c (s) has been developed using LQG [8] methods which has been
defined substantially. The major advantage of the LQG design is that it not only
controls the output response of the controllable states to desirable values but also
reduce the response of uncontrollable states. For a continuous-time linear structure,
defined on t [t 0 , t 1 ], described by Eqs. (6) and (7) the state feedback law u = −kx
minimizes a quadratic cost function defined in Eq. (8).
Semi-active Control of a Three-Storey Building Structure 301
t1
J = x (t1 )F(t1 )x(t1 ) +
T
(x T Qx + u T Ru + 2x T N u)dt
t0
where E is the expected value. T is the final time. Using Matlab programs, the gain
matrix K matrix has been computed by solving the algebraic Riccati equation.
x max − x1min
x1 = x1min + 1 l ×D (10)
2 −1
where
x 1 is the real value of one sub-string;
D decoded value for the binary sub-string. The objective function is expressed as
given in Eq. (11)
F(x1 , x2 , x3 ) = f i (X ) ± Pi (11)
of cycles until a termination criterion is satisfied that is either the objective function
acquire a predefined value or maximum iteration is reached. A flowchart is shown in
Fig. 3.
A type of clipped optimal controller [13] is introduced in this study. The controller
K c (s) is designed both by LQG and GA each to estimate the desired control force,
f c (Eq. 12). y is the measured responses and f is the measured damper force. L is a
Laplace transform.
Semi-active Control of a Three-Storey Building Structure 303
−1 y
fc = L K c (s)L (12)
f
v = Vmax H {( f c − f ) f } (13)
where Vmax is the voltage required for the saturation of the magnetic field of the
MR damper and H is the Heaviside step function and v is the voltage required. The
controller algorithm of linear variation of the desired control force with the required
voltage is shown in Fig. 4.
A three-storey frame model as shown in Fig. 1 spanning one bay in X direction and
Z direction each is chosen for the present work. El-Centro (CA, USA) earthquake
data (October, 1942) has been used as dynamic load to the structure. State space
design of the model has been done to implement the proposed control design. The
compensator gain has been suitably adjusted by adding a pole for a stable root locus
diagram of the open-loop system. Suitable controllers from LQG design and genetic
304 P. Chaudhuri et al.
⎡ ⎤
12 −6.84 0
⎢ ⎥
⎢ −6.84 13.7 6.84 ⎥105
⎣ ⎦
0 −6.84 6.84
N/m
⎡ ⎤
175 −50 0
⎢ ⎥
Damping matrix, [C] C =⎢⎣ −50 100 −50 ⎦
⎥
0 −50 50
Ns/m
Location matrix of MR = [ 1 0 0]T
damper force
External force distribution = [ 1 1 1]T
matrix
From figures it can be observed that LQG and GA control are both efficient in
reducing the vibration response to a large extent.
306 P. Chaudhuri et al.
Displacements (mm)
150
100
Uncontrolled
50
LQG controlled
0 GA controlled
1 2 3
Floors (No.)
20
Displacements (mm)
0
1 2 3
-20 Uncontrolled
-80
Floors (No.)
show the flexibility of both the methods. The root mean square value of displacements
in each floor for all earthquake data has been tabulated in Table 2.
The efficiency of both the methods can be clearly observed in Table 2. The LQG
controlled and GA controlled structures have shown a significant reduction in the
displacement values for the structure.
4 Conclusion
The controller is designed using both the conventional LQG method and genetic
algorithm. These methods can show much flexibility to produce desirable range of
outputs in different loading conditions. The clipped optimal controller has been used
to get the optimized voltage required by MR damper to produce damper force. Suit-
able MR dampers can be placed to the structure and the responses can be studied.
Both the methods have shown significant efficiency in reducing the seismic vibra-
tion of building. The proposed methods of controlling the structure itself can be
implemented in higher storey space frame building.
References
1. Yi F, Dyke SJ, Caicedo JM, Carlson JD (2001) Experimental verification of multi-input seismic
control strategies for smart dampers. J Eng Mech (ASCE) 127(11):1152–1164
2. Jansen LM, Dyke SJ (2000) Semiactive control strategies for MR dampers: comparative study.
J Eng Mech (ASCE) 126(8):795–803
3. Park JM, Gang HG, Sohn JW (2015) Parametric study on damping force characteristics of MR
damper with various inner magnetic core shapes. In: 22nd International congress on sound and
Vibration, Florence, Italy, pp 1–2
4. Khan SA, Suresh A, Ramaiah NS (2014) Principles characteristics and applications of magneto
rheological fluid damper in flow and shear mode. Procedia Mater Sci 6:1547–1556
5. Cha YJ, Zhang J, Agarwal AK, Dong B, Friedman A, Dyke SJ, Ricles J (2013) Comparative
studies of semi-active control strategies for MR dampers: pure simulation and real-time hybrid
tests. J Struct Eng (ASCE) 139(7):1237–1248
6. Mahdi A, Abdelhafid O (2015) Seismic response reduction using semi-active magneto-
rheological dampers. Innov Constr
7. Ogata K (2009) Modern control engineering, 5th edn. Prentice Hall, pp 29–44, 645–688
8. Dyke SJ, Spencer BF Jr, Sain MK, Carlson JD (1996) Modeling and control of magnetorheo-
logical dampers for seismic response reduction. Smart Mater Struct 5(5):565–575
9. Pratihar DK (2014) Soft computing fundamentals and applications. Narosa Publishing House,
pp 40–51, 87–89
10. Holland JH (1984) Adaptive control of ill-defined system. NATO Conf. Series (II Systems
Science) Springer 16, pp 317–333
11. Jose K, Pratihar D (2016) Task allocation and collision-free path planning of centralized
multi-robots system for industrial plant inspection using heuristic methods. Robot Auton Syst
80:2016. https://doi.org/10.1016/j.robot
12. Saxena P, Dixit A (2009) Software component retrieval using genetic algorithms. In: 2009
International Conference on Computer and Automation Engineering ICCAE, Bangkok, pp
151–155. https://doi.org/10.1109/iccae.2009.62
13. Dyke SJ, Spencer BF Jr, Sain MK, Carlson JD (1999) An experimental study of mr dampers
for seismic protection. Smart Mater Struct 7(5):1–19
A Direction-Based Exponential
Crossover Operator for Real-Coded
Genetic Algorithm
1 Introduction
Optimization is the method of searching the best feasible candidate out of several
potential solutions. In real-world applications, it deals with many nonlinear complex
problems, which have one or more than one design variables and constraints. The
purpose of optimizing a process or the design of a product is the reduction in various
costs, such as material, rejection, and labor, or to increase the overall performance and
profit [1]. Although traditional deterministic algorithms had been used for solving
several optimization problems, these are not found to be suitable in case of solv-
ing large-scale problems with ill-conditioned or non-smoothed objective function.
Therefore, researchers had copied the Mother Nature to obtain new ways to solve
the problems and consequently, several nature-inspired optimization techniques have
been designed and successfully implemented to solve a variety of problems. Among
several bioinspired algorithms, Genetic algorithm [2] (GA), Particle swarm opti-
mization [3] (PSO), Differential evolution [4] (DE), Ant colony optimization [5]
(ACO), Bat algorithm [6] (BA), etc., are the most popular evolutionary optimization
techniques available in the literature.
Genetic algorithm (GA) is one of the most robust stochastic algorithms used in
different types of applications. It is very easy and straightforward to implement, able
to handle a large variable space and very flexible in assigning the objective function
and the constraints [7]. It works based on Darwin’s principle of natural selection.
It consists of three stochastic operators, such as selection, crossover, and mutation.
In the selection scheme, better solutions (in terms of their fitness values) are given
advantages for creating mating pools. During crossover, the properties of two mating
parents are exchanged to obtain two new children solutions. After this operation, the
solutions may find sudden changes while passing through the mutation operator,
which helps the algorithm to overcome the problem of local optimum. Among these
three main operators of a GA, the crossover scheme has a great role in searching
for new solutions to reach the globally optimal point. Several attempts were made
to design an efficient crossover operator of a GA. Wright [8] proposed a heuristic
crossover operator, in which one child is created out of two parents and the child
has a bias toward the better parent. Radcliffe [9] suggested a crossover operator,
namely, flat crossover. In this type of recombination, one child solution is obtained
randomly in between a pair of parents. Michalewicz [10] introduced an arithmetical
crossover with two variants, such as uniform and nonuniform arithmetical crossover.
Eshelman and Schaffer [11] developed a blend crossover operator (BLX-α), where
the locations of the children solutions to be created by the parents, are calculated
using the parameter α. The suggested standard value for α was equal to 0.5. Deb
and Agrawal [12] proposed the simulated binary crossover (SBX), in which two
offspring are generated from a pair of mating parents. It imitates the concept of a
single-point crossover on binary strings in continuous space. Ono and Kobayashi
[13] developed a unimodal normal distribution crossover operator (UNDX). In this
scheme, two or more children solutions are generated from three parent solutions.
To demonstrate the performance of the said operator, only three problems were
solved, which might not be sufficient to conclude anything firmly. Tsutsui et al.
[14] suggested a simplex crossover (SPX), where to create the offspring, three or
more parents were used. From some numerical experiments, they inferred that SPX
could give a better performance, where the objective function is multimodal and/or
it has epistasis with a medium number of mating parents. Another operator, namely,
parent-centric crossover (PCX) was developed by Deb et al. [15] for real parameter
optimization and its performance was compared with the other popular crossover
operator like SPX, UNDX, etc. However, it was tested on three problems only, which
might not be sufficient enough to conclude anything significant. Utilizing the Laplace
distribution to evaluate the location of the offspring solution, Deep and Thakur [16]
proposed a new crossover operator, named Laplace crossover (LX). Apart from these,
A Direction-Based Exponential Crossover Operator … 313
Kuo and Lin [17] suggested a directed crossover for an RCGA. This recombination
method was designed based on the reflection and expansion search mechanism of
Nelder-Mead’s simplex approach. Chuang et al. [18] developed a parallel-structured
RCGA, in which a direction-based crossover (DBX) was proposed. From the various
simulation results, it was observed that this method could yield good performance
for multimodal and high-dimensional hybrid objective functions only.
It is to be noted that there are several crossover operators for the RCGA available
in the literature of GA. However, the novel concept of deriving the most-favorable
search direction and its application to yield new children solutions are missing in
most of the crossover schemes. In this paper, we introduce a direction-based expo-
nential crossover operator (DEX) for a real-coded genetic algorithm (RCGA), where
two offspring are generated from a pair of mating parents using exponential functions
and predetermined directional information of the decision variables. In the proposed
DEX operator, most-promising directional information is used in an efficient way for
creating the children solutions. Moreover, two probability terms, such as directional
crossover probability ( pcd ) and exploration probability ( pe ) are introduced in the pro-
posed DEX to preserve a superior balance between the exploitation and exploration
of the search process and these are found to be the novelties of the present study.
The remaining part of the text is arranged as follows: Sect. 2 describes the details of
the proposed crossover operator, whereas the results and discussion are presented in
Sect. 3. At the end, some conclusions are made in Sect. 4.
The name of this crossover scheme clearly indicates that DEX is influenced by the
directional knowledge of the problem. The collected information of the variables
actually helps to decide the search direction of the algorithm in the variable space
to move toward the globally optimal solution. Now, the task of obtaining this data is
difficult and may have several ways for the same. Here, we propose one method to
get this information during evolution.
314 A. K. Das and D. K. Pratihar
At first, we have to calculate the average numerical value of the current population
for each decision variable. This calculation is to be done for each generation. Now,
for each variable, we have to make comparisons between the average value obtained
in the current generation and that calculated in the previous generation. When the
average value of a variable in the current generation is observed to be more compared
to that of the previous generation, then the direction for that variable is considered as
the positive in that generation. In another case, when the average value of a variable
becomes less than that in the previous generation, the direction for the variable is
taken as the negative in that particular generation. An example has been shown in
Fig. 1 with a 2-D variable space. In this example, at (T + 1)th generation, the obtained
directional information is positive for both the variables.
In this way, for every generation, except for the first and second ones, we can
obtain the directional information for the design variables. For the second gener-
ation, the calculated average values of the variables are compared with the best
solution available in that generation and in a similar way, the directional information
is obtained. However, it is not required to derive the directional knowledge for the
very first generation as there is no crossover taking place in that generation.
Utilizing the obtained directional knowledge of the variables, the locations of the
children solutions are controlled to a greater extent. However, there are other aspects
which are also responsible for determining the positions of the offspring. A more
detailed description of this scheme is given below in a few steps.
First step: Let us say, the population size of the solutions is taken as N . After
the selection operation, N /2 pairs of mating solutions are created randomly from the
N number of selected members. A mating pair participates in crossover, while the
generated random value (ranging from 0 to 1) is observed to be either less than or
equal to the crossover probability ( pc ). Otherwise, the mating pair is not allowed to
join in crossover.
Second step: After the first step, the recombination operator is applied variable-
wise for the mating solutions, which are permitted to join in the crossover. Here,
another probability, say variable-wise crossover probability ( pcv ) is applied to decide
whether the crossover is going to occur or not to a certain variable position. This
concept of pcv is implemented in a similar way to that of pc as mentioned in first
step.
Third step: The mating parents, which have passed through the said two steps
and are allowed to join in the crossover, exchange their properties to yield children
solutions according to the scheme described in this step. We consider a user-defined
very small positive number, say ε. Depending on the value of ε, there may occur two
scenarios, in which two different crossover schemes are adopted as follows:
• Scenario 1: Absolute difference between the two mating parents is more than ε
Between the two mating parents, let us take the smaller one as p1 and the larger
one as p2 . Then, a parameter, say val, is calculated using the Eq. (1):
( p2 − p1 )
(yu −yl )
val = 1 − 0.5{e }
(1)
where yu and yl are the upper and lower boundaries of the decision parameters,
respectively. Now, whenever the directional knowledge for a variable is found to be
positive and a random number (varying from 0 to 1) is observed to be either less than
or equal to a parameter, say directional crossover probability ( pcd ), two children
solutions (c1 and c2 ) are generated using the Eqs. (2) and (3).
where α is the spread factor, which generally varies from 0 to 4, and r is a random
number created in the range of (0,1). However, if the directional information of the
316 A. K. Das and D. K. Pratihar
where Er is the exploration ratio and it takes a numerical value which is seen to be
very near to 1, and r is a random number varying in the range of 0–1.
If a child solution is found to be greater than the upper limit of the variable (yu )
or less than the lower limit of the variable (yl ), then its value is taken as the value of
yu or yl , respectively. These are called the variable limiting conditions.
Fourth step: This is the last step of the proposed scheme, where the newly gener-
ated offspring (c1 and c2 ) are recognized. If a random value (yielded between 0 and
1) is obtained as either less than or equal to 0.5, then c2 and c1 are considered as the
first child and second child of the generated pair of solutions, respectively. In another
case, c1 and c2 are recognized as the first child and second child of the yielded pair of
solutions, respectively, if a random value, varying from 0 to 1, is found to be greater
than 0.5. These defined conditions are denoted as children recognition conditions.
A pseudo-code of the proposed crossover operator is presented in Table 1.
A Direction-Based Exponential Crossover Operator … 317
d [−100, 100]d
f (x) = xi2
i=1
F02: Sum of
d [−100, 100]d
different f (x) = |xi |i+1
i=1
powers
F03: rotated
i [−65, 65]d
hyper-ellipsoid f (x) = x 2j
i=1 j=1
d √ [−600, 600]d
F04: Griewank 2
i=1 xi
d
f (x) = 4000 − cos(xi / i) + 1
i=1
F05: Trid
d [−d 2 , d 2 ]d
f (x) = (xi − 1)2 − (xi xi−1 )
i=1 i=2
F06:
d [−10, 10]d
Dixon-price f (x) = (x1 − 1)2 + i(2xi2 − xi−1 )2
i=2
F07: Powell
d [−10, 10]d
f (x) = |xi sin(xi ) + 0.1xi |
i=1
d [−10, 10]d
f (x) = x12 + 106 xi2
i=2
To make a fair comparison, the common controlling parameters of both the algo-
rithms, such as population size (N = 60), number of variables
(d = 30), crossover
probability ( pc = 1.0), mutation probability pm = d1 , user-defined index parame-
ter for polynomial mutation operator (ηm = 10), and maximum number of genera-
tions (max_gen = 500), have been kept the same. The user index parameter for SBX
crossover operator (ηc ) is taken as 2. Both the algorithms stop, if these are run for
the maximum number of generation. Special parameters of the DEX operator, such
as spread factor (α), variable wise crossover probability ( pcv ), directional crossover
probability ( pcd ), small positive number (ε), exploration ratio (Er ), and exploration
probability ( pe ) are determined after several trial experiments and a set of suitable
values of these parameters is selected to test the said ten benchmark functions. These
chosen parameters’ values are given in Table 3.
The obtained results are given in Table 4, where the best ones are marked in bold.
In addition, the average number of function evaluations ( f avg ) and average CPU
time (in seconds) required in 50 runs to reach a particular accuracy of the objective
function value (i.e., equal to 1 for F06 and 0.1 for others) for all the test functions
are also provided in Table 4.
From the results, it is clear that RCGA with DEX operator is performing better
than the RCGA with SBX operator for the ten benchmark functions. Moreover, it is
observed that the proposed algorithm takes less CPU time to reach a particular accu-
racy of the objective function compared to that of the other. It might have happened
due to the fact that the directional information derived from the problem itself has
been utilized for creating the approximate children solutions in the RCGA with DEX
operator. Moreover, the use of exploration probability ( pe ) and exploration ratio (Er )
enhances the exploration capability of the RCGA, and this capability decreases over
the generation maintaining a proper balance between the exploration and exploitation
phenomena of the optimization algorithm.
Table 4 Results of benchmark test functions (F01–F10) using RCGA with DEX and SBX operators
320
Function RCGA with RCGA with Function RCGA with RCGA with Function RCGA with RCGA with
DEX SBX DEX SBX DEX SBX
F01 Best 4.059E−07 9.673E−03 F02 2.423E−10 3.650E−04 F03 3.249E−06 4.474E−02
Median 3.743E−06 4.259E−02 3.674E−07 2.939E−02 2.400E−05 2.382E−01
Worst 7.163E−04 2.811E−01 9.769E−05 2.443E+00 2.557E−03 2.303E+00
Mean 2.332E−05 5.391E−02 7.351E−06 1.524E−01 1.250E−04 3.141E−01
SD 1.008E−04 4.186E−02 1.809E−05 3.835E−01 3.761E−04 3.323E−01
f avg 1.341E+04 2.486E+04 2.060E+04 2.819E+04 1.622E+04 3.902E+04
tavg 4.469E−01 7.342E−01 7.607E−01 9.722E−01 5.520E−01 1.222E+00
F04 Best 5.849E−07 1.954E−02 F05 −3.855E+03 −2.001E+03 F06 6.666E−01 6.852E−01
Median 2.017E−05 9.949E−02 −2.792E+02 8.171E+03 6.701E−01 2.461E+00
Worst 3.927E−02 1.108E+00 8.166E+03 3.892E+04 4.417E+00 5.845E+00
Mean 7.915E−03 1.678E−01 −3.219E+02 9.562E+03 1.024E+00 2.738E+00
SD 8.127E−03 2.386E−01 3.263E+03 9.041E+03 7.923E−01 1.381E+00
f avg 1.456E+04 2.638E+04 3.041E+04 1.249E+05 2.869E+04 2.590E+05
tavg 4.910E−01 8.184E−01 9.598E−01 3.653E+00 9.488E−01 7.053E+00
F07 Best 1.535E−02 4.278E−01 F08 5.001E−06 1.081E−03 F09 3.077E−03 4.954E+01
Median 9.045E−02 1.746E+00 1.309E−05 4.885E−03 2.232E−02 2.545E+02
Worst 1.931E−01 6.572E+00 2.063E−04 2.284E−02 1.185E+00 1.457E+03
Mean 9.065E−02 2.088E+00 2.468E−05 6.394E−03 7.151E−02 3.584E+02
SD 3.923E−02 1.214E+00 3.327E−05 5.084E−03 1.858E−01 2.993E+02
f avg 2.813E+04 9.369E+04 6.773E+03 1.182E+04 2.809E+04 5.393E+05
tavg 9.445E−01 2.939E+00 2.319E−01 3.628E−01 8.834E−01 1.314E+01
(continued)
A. K. Das and D. K. Pratihar
Table 4 (continued)
Function RCGA with RCGA with Function RCGA with RCGA with Function RCGA with RCGA with
DEX SBX DEX SBX DEX SBX
F10 Best 3.577E−02 2.351E−01
Median 1.061E−01 3.729E−01
Worst 2.830E−01 6.283E−01
Mean 1.095E−01 3.825E−01
SD 4.393E−02 9.138E−02
f avg 3.624E+04 2.015E+05
tavg 1.767E+00 8.617E+00
A Direction-Based Exponential Crossover Operator …
321
322 A. K. Das and D. K. Pratihar
4 Conclusion
References
13. Ono I, Kobayashi S (2003) A real-coded genetic algorithm for function optimization using
unimodal normal distribution crossover. In: Ghosh A, Tsutsui S (eds) Advances in evolutionary
computing. Natural Computing series, Berlin, Heidelberg
14. Tsutsui S, Yamamura M, Higuchi T (1999) Multi-parent recombination with simplex crossover
in real-coded genetic algorithms. In: Banzhaf W, Daida J, Eiben A, Garzon M, Honavar V,
Jakiela M, Smith R (eds) Proceedings of the Genetic and Evolutionary Computation Conference
(GECCO-1 1999), pp 657–664
15. Deb K, Anand A, Joshi D (2002) A computationally efficient evolutionary algorithm for real-
parameter evolution. Evol Comput J 10(4):371–395
16. Deep K, Thakur M (2007) A new crossover operator for real-coded genetic algorithms. Appl
Math Comput 188:895–911
17. Kuo H-C, Lin C-H (2013) A directed genetic algorithm for global optimization. Appl Math
Comput 219:7348–7364
18. Chuang Y-C, Chen C-T, Hwang C (2015) A real-coded genetic algorithm with a direction-based
crossover operator. Inf Sci 305:320–348
19. Deb K, Goyal M (1996) A combined genetic adaptive search (GeneAS) for engineering design.
Comput Sci Inform 26(4):30–45
Axial Deformation Characteristics
of Graphene-Sonicated Vinyl Ester
Nanocomposites Subjected to High Rate
of Loading
B. Pramanik (B)
Department of Mechanical Engineering, Montana Tech, Butte, MT 59701, USA
e-mail: bpramanik@mtech.edu
P. R. Mantena · A. M. Rajendran
Department of Mechanical Engineering, University of Mississippi, Oxford, MS 38677, USA
This is a U.S. government work and not under copyright protection in the 325
U.S.; foreign copyright protection may apply 2020
B. N. Singh et al. (eds.), Recent Advances in Theoretical, Applied, Computational
and Experimental Mechanics, Lecture Notes in Mechanical Engineering,
https://doi.org/10.1007/978-981-15-1189-9_26
326 B. Pramanik et al.
1 Introduction
2 Experimental Setup
The tensile SHPB system as shown in Fig. 1 consists of a steel barrel having a
steel hollow cylindrical striker bar. A compressed air vessel was connected with a
solenoid switch valve to one end of the barrel. The striker was mounted on two Teflon
rings for achieving frictionless sliding motion inside the barrel. The striker bar was
propelled by the rapid release of air pressure from the compressed air vessel. The
hollow cylindrical striker slides over the aluminum incident bar, inside the barrel and
impacts the aluminum collar, mounted at the end of the incident bar, causing a tensile
stress wave within the incident bar. A thin ring-shaped polyurea pulse shaper was
used for producing a trapezoidal incident pulse. The SHPB bar system was retarded
by a momentum bar, which was trapped against a block of putty-clay. The specimen
was sandwiched between an incident bar and a transmission bar and was loaded with
Fig. 1 SHPB set-up for direct tensile tests including LOEG device
328 B. Pramanik et al.
tensile stress. A part of the incident wave was reflected back to the incident bar, and
the other part was transmitted into the aluminum transmission bar.
Two batches of strain gauges were installed at a distance away from the specimen
on the incidence and the transmission bars as the strain gauges can capture the
commencement of the incident, reflected and transmitted pulses distinguishably.
Each batch consisted of two gauges positioned on the transversely reverse sides of the
bar and coupled in series network (Fig. 1). This arrangement eliminates the flexural
influence of the bars and ensures uniaxial strain measurement. A digital oscilloscope
was linked to each batch of the strain gauges, via a Wheatstone bridge circuit and a
signal conditioner. The oscilloscope displayed the signal potential history over the
entire test duration and stored in the database. Strain pulse response in SHPB bars
at the strain gauge locations was applied in the 1-D wave equation and the tensile
stress experienced by the test specimen was obtained.
A dog bone-shaped specimen and dovetail gripping mechanism [24] shown in Fig. 2a
was adopted to reduce the wave dispersion, possibly caused due to complex specimen
attachment configurations of previously attempted pinned [17], threaded [20], or
cemented [23] fixtures. A plane-stress FEA was carried out to parametrically optimize
the specimen geometry that prevented pre-mature failure of the specimen or the
fixture due to undesirable stress concentrations, and also to ensure uniform stress
distribution within the loading zone in the specimen.
The pulse width over time dictates the required least length of the loading zone
within the specimen. The pulse needs to propagate back and forth within the specimen
more than three times before yield or failure initiates. It ensures the internal stress
equilibrium within the specimen. The wave speed in the specimen was estimated
over quasi-static properties previously obtained [9] as per ASTM 638-08 [27].
The fundamental assumption for this FEA was the homogeneity of the material
integrity in case of the load transfer. Only geometric influence of the components has
been considered for this analysis. The FE model was constructed into two parts. One
part was of the dog bone-shaped specimen, and the other part was of the dovetail
end fixture. Figure 2b illustrates the load and the boundary condition considered
in those models. The geometry was modeled using ANSYS Mechanical APDL®
graphics tool and merged as an assembly (Fig. 2c). The models meshed with 10-
noded elements (SOLID187 in ANSYS® ) having quadratic displacement behavior
is appropriate for modeling irregular meshes. Quarter symmetry of the geometry
was considered to decrease the total number of nodes of the model. It allowed denser
mesh and improved the accuracy of results.
The contact elements were modeled along the interface between the fixture and
the specimen. The contact along the angular surfaces was defined with normal and
tangential components. The contact elements follow a linear constitutive behavior
Axial Deformation Characteristics of Graphene-Sonicated … 329
Fig. 2 a Development of direct tensile specimen geometry, b boundary condition applied in stress
analysis of specimen and fixture assembly, c ANSYS model, d von-Mises stress distribution in
ANSYS simulation, e specimen dimensions, and f fixture dimensions (all dimensions in inches)
normal to the surface. The aluminum fixture surface was chosen as the master sur-
face. A stiffness-penalty method was considered for explaining the friction between
the fixture and the specimen. Figure 2d shows a uniformly distributed tensile load
applied at the end of the specimen. The end of the fixture was constrained from axial
translation. The color-map of von Mises stress distribution for an imposed distributed
load is shown in Fig. 2d. The von Mises stresses were distributed uniformly along
the specimen gauge section and validated the optimization of the specimen (Fig. 2e)
and fixture (Fig. 2f) geometry. A CNC machine, equipped with a carbide tipped tool,
machined the optimized specimens for performing the test series.
fixture ends in case of direct dynamic loading, or the deforming specimen geometry
in case of indirect dynamic loading [26]. This idea was adopted from the work of
Ramesh and Narsimhan [28] on measuring radial strains in compression Kolsky bar
experiment and the later work of Chen et al. [29] on dynamic tensile response of low
strength brittle solid.
The LOEG consisted of a 635 nm wavelength laser system (©Coherent, Inc.). It
included a single line projection head with 30° fan angle. It had two anti-reflector
(MgF2 )-coated plano-convex cylindrical N-BK7 glass lenses of 25 mm diameter with
50 mm focal length (©Edmund Optics). It was equipped with an amplified, switchable
gain, silicon PIN photodetector (©Thorlabs) having 1.5 MHz bandwidth at 4.25 MS/s
sampling rate, and 0.2 mV noise, and 100 MHz. A 4-channel oscilloscope with 1.25
GS/s sampling rate (©Tektronix, Inc.) was included in this setup. It is to be noted
that the same oscilloscope is used in SHPB system. The incident pulse (sensed
at the incident bar strain-gauges) triggered the photodetector. The diode-laser unit
generated 635 nm laser beam collimated (converted from a divergent beam to a
parallel beam over an extended range of projection distance) with an elliptical cross
section of 3.8 mm × 0.9 mm. It consisted of a single line projection head with 30°
fan angle (the angular spread of the laser beam). The projection head consisted of
Powell® glass lens which spreads the collimated laser beam and converts the elliptical
projection into a 0.9 mm thick non-Gaussian line projection with almost uniform
relative intensity along the fan angle. Line thickness was reduced to about 55 µm
by the pre-focusing collimated laser beam at the target position on the specimen.
The maximum power output was 5 mW. Black anodized aluminum housing held
the entire laser unit. The plano-convex cylindrical lens used in this setup was made
from a coated N-BK7 glass substrate. It developed a parallel laser sheet of 25 mm
(~1 in.) width. The coating was of MgF2 anti-reflectant (<1.75% per surface area in
the wavelength within 400–700 nm). The light was detected through another similar
plano-convex cylindrical lens arranged in a symmetrically opposite orientation. It
converged the parallel sheet of light into a 9.8 mm diameter aperture of a photodiode
light detector placed near the focal point of the laser sheet. The photodiode detector
output was pre-amplified with a bandwidth of 1.5 MHz, and the output-voltage was
proportional to the total amount of laser light entered into the photodetector including
less than 1 mV noise level.
The incident pulse propagated along the incident bar, pulled the sample, reflected
back a part of the incident pulse into the incident bar and the rest was transmitted
to the transmission bar. In this test, the steel hollow cylindrical striker bar impacted
the aluminum collar mounted at the end of the aluminum incident bar. Due to the
striker collar impedance mismatch, the incident pulse exhibited extended restitution
with gradually reducing amplitudes similar as described in [30]. Hence, the initial
maximum amplitude portion of the incident pulse was used for SHPB analysis. The
Axial Deformation Characteristics of Graphene-Sonicated … 331
2 As ρs Cs Ai
σdt = . . E i εi (2)
Ai ρi Ci + As ρs Cs As
where E is the elastic modulus; ε denotes the strain; A, is the cross-sectional area, ρ
is the density, and C is the wave velocity for the incident bar (with a suffix, i) and
the specimen (with a suffix, s).
The laser sheet, projected horizontally above the tensile test-specimen, was partially
occluded by the ends of the specimen fixtures mounted on incident and transmission
bars. In our experimental setup, the gap between the specimen holding fixture ends
was almost equivalent to the specimen gage length. Hence, the tensile strain response
was counted from the ratio (V /V ) of the incremental change of potential along with
specimen elongation to the initial potential within the grip ends holding the pristine
test coupon.
A comparative study of the LOEG response was performed. A batch of strain
gauges was installed on the dog bone-shaped coupon. The tensile strain history was
obtained using LOEG and on-specimen strain gauges. A batch of two strain gauges
was installed on opposite sides of the coupon and connected in a quarter-bridge
circuit. The dynamic strain–time history response was plotted on an oscilloscope. The
LOEG setup acquired a comparable strain data transferred to the same oscilloscope.
It validated the applicability of the LOEG setup in tensile strain measurements.
A traditional compression SHPB setup with aluminum bars was applied for obtain-
ing the high rate compressive characteristics of the participant materials. The disk
specimen geometry was considered for compression tests and loaded axially on
the opposite faces. The compressive 1-D wave propagating through the specimen
thickness was considered for computing the dynamic compressive stress and strain
histories.
332 B. Pramanik et al.
3 Experimental Results
The typical failure of an optimized dog bone specimen under a high rate of tension
in SHPB apparatus has been studied. It showed multiple fractured surfaces on the
post-test specimen. It indicates that the dynamic stress equilibrium was achieved
before the fracture occurred. The instantaneous drop of stress indicated the onset of
a crack within the specimen. The corresponding time of the visible first stress peak
(representing the dynamic tensile strength) had been recorded as the time of crack
initiation. The energy absorbability per unit specimen volume was obtained from the
area-integral under the stress-versus-strain curve [32] until specimen failed.
Figure 3 summarizes the tensile response of vinyl ester composites containing xGnP
and CTBN subjected at about 1000 s−1 strain-rate. All nano-reinforced specimens
including the pristine vinyl ester showed brittle failure under a high rate of tension
Fig. 3 Dynamic direct tensile response a typical constitutive response, b strength comparison, and
c energy absorbability comparison for the composite system; (i) with graphite platelet, and (ii) with
additional CTBN
Axial Deformation Characteristics of Graphene-Sonicated … 333
(Fig. 3a). The tensile strength of pristine vinyl ester was decreased marginally with
xGnP (Fig. 3b(i)). The inclusion of CTBN along with xGnP into the composite
systems did not show significant improvement (Fig. 3b(ii)). The energy absorbability
of pristine vinyl ester maintained almost similar response with xGnP (Fig. 3c(i));
however, this was minimally enhanced by further addition of CTBN to the xGnP in
the composite systems (Fig. 3c(ii)).
Fig. 4 Dynamic compressive response a typical constitutive response, b strength comparison, and
c energy absorbability comparison for the composite system; (i) with graphite platelet, and (ii) with
additional CTBN
334 B. Pramanik et al.
(Fig. 4c(ii)) of pristine vinyl ester, nevertheless, a minimal increment (Fig. 4c(ii))
was observed (Fig. 4c(i)) upon comparing with that of the only xGnP added (without
CTBN) composite systems.
The constitutive materials within the composite system play an important role of
transferring mechanical load. The anisotropic nature of load transfer is significant
in composites due to the inherent differences of the bonding efficiency among con-
stituents with the matrix along the loading directions. Tensile and compressive loads
act mutually in opposite directions. Hence, the composite system responds differ-
ently for these counteracting loading systems. Comparative observation of Fig. 3 with
Fig. 4 shows that the stress–strain characteristic of these candidate materials under
high-rate tension varied considerably from its high-rate compression. The compres-
sive strength (Fig. 4b) of the candidate materials was observed to be about 300% more
than the tensile strength under high-rate of loading (Fig. 3b). The energy absorbabil-
ity was also significantly higher (about 2000%) under dynamic compressive loading
(Fig. 4c) than under dynamic tensile loading (Fig. 3c). It is evident that the vinyl ester-
based nanocomposites are brittle at high strain rates. The nano-reinforcement and
additional toughening are detrimental to the properties under dynamic compression.
Still, the contribution of the inclusions showed minimal improvement in dynamic
tension.
4 Conclusion
The significant features of the proposed techniques of the modified direct tensile test
method described in this article—
• The SHPB reverse impact technique is implemented for the mechanical charac-
terization of the candidate materials under a high rate of uniaxial tension.
• Due to the very low transmitted pulse signal, an alternative approach (using inci-
dent pulse and quasi-static properties of the candidate materials) is adopted for
estimating stress–time history. The transmission bar may also be modified with
a low impedance material and/or smaller cross-sectional area for obtaining true
dynamic tensile stress.
Axial Deformation Characteristics of Graphene-Sonicated … 335
The critical observations in this investigation regarding the effect of xGnP and
CTBN on the axial response of vinyl ester composite system are—
• The pristine vinyl ester maintains almost the similar tensile strength with addition
of xGnP and even with CTBN under quasi-static to high-strain rate tension.
• The energy absorbability of pristine vinyl ester is enhanced due to the presence of
xGnP under dynamic tension.
• The high-rate of tensile response of the candidate materials varies significantly
from the high-rate of compressive response.
• Reinforcing vinyl ester with xGnP and toughening with CTBN is found to be detri-
mental to the properties of candidate nanocomposites under dynamic compres-
sion, although a minimal improvement of these properties is observed in dynamic
tension.
References
1. Shivakumar KN, Swaminathan G, Sharpe M (2006) Carbon vinyl ester composites for ena-
hanced performance in marine applications. J Reinf Plast Compos 25(10):1101–1116. https://
doi.org/10.1177/0731684406065194
2. Ashland Inc. (2008) DERAKANE 510A-40 Epoxy Vinyl Ester Resin. Technical Datasheet,
Document 1775V2 F2, Language EN V1, Approved 2008-9-8, vol 201, pp 1–4
3. Yoshida A, Hishiyama Y, Inagaki M (1991) Exfoliated graphite from various intercalation
compounds. Carbon 29(8):1227–1231. https://doi.org/10.1016/0008-6223(91)90040-P
4. Toshiaki E, Masatsugu S, Morinobu E (2003) Graphite intercalation compounds and applica-
tions. Oxford University Press Inc, New York
5. Fröhlich J, Thomann R, Mülhaupt R (2003) Toughened epoxy hybrid nanocomposites
containing both an organophilic layered silicate filler and a compatibilized liquid rubber.
Macromolecules 36(19):7205–7211. https://doi.org/10.1021/ma035004d
6. Fukushima H, Drzal LT (2004) Graphite nanoplatelets as reinforcement for polymers: structural
and electrical properties. In: Proceedings of 17th international conference on American Society
for composites
7. Drzal LT, Fukushima H (2006) Exfoliated graphite nanoplatelets (xGnP): a carbon nanotube
alternative. The nanotechnology conference, Boston, 7–11 May 2006
8. Lu J, Do I, Drzal LT, Worden RM, Lee I (2008) Nanometal-decorated exfoliated graphite
nanoplatelet based glucose biosensors with high sensitivity and fast response. ACS Nano
2(9):1825–1832. https://doi.org/10.1021/nn800244k
9. Magableh A (2010) Viscoelastic and shock response of nanoclay and graphite platelet
reinforced vinyl ester nanocomposites. Dissertation, University of Mississippi
10. Mantena PR, Cheng AHD, Al-Ostaz A, Rajendran AM (2010) Blast and impact resistant com-
posite structures for navy ships. In: Proceedings of marine composites and sandwich structures,
2010, Office of Naval Research—Solid Mechanics Program Review, Adelphi, MD, 27–29 Sept
2010
336 B. Pramanik et al.
11. Pramanik B, Mantena PR (2011) Viscoelastic response of graphite platelet and CTBN rein-
forced vinyl ester nanocomposites. Mater Sci Appl 2(11):1667–1674. https://doi.org/10.4236/
msa.2011.211222
12. Pramanik B, Mantena PR (2012) Energy absorption of nano-reinforced and sandwich compos-
ites in ballistic and low-velocity punch-shear. Open J Compos Mater 2(3):87–96. https://doi.
org/10.4236/ojcm.2012.23010
13. Pramanik B, Tadepalli T, Mantena PR (2012) Surface fractal analysis for estimating the fracture
energy absorption of nanoparticle reinforced composites. Materials 5(5):922–936. https://doi.
org/10.3390/ma5050922
14. Hopkinson B (1914) A method of measuring the pressure produced in the detonation of high
explosives or by the impact of bullets. Phil Trans R Soc Lond A 213:437–456. https://doi.org/
10.1098/rsta.1914.0010
15. Davies RM (1948) A critical study of the Hopkinson pressure bar. Phil Trans R Soc Lond A
240(821):375–457. https://doi.org/10.1098/rsta.1948.0001
16. Kolsky H (1949) An investigation of the mechanical properties of materials at very high rates
of strain. Proc Phys Soc B 62:676–700. https://doi.org/10.1088/0370-1301/62/11/302
17. Harding J, Wood EO, Campbell JD (1960) Tensile testing of materials at impact rates of strain.
J Mech Eng Sci 2(2):88–96. https://doi.org/10.1243/JMES_JOUR_1960_002_016_02
18. Lindholm US, Yeakley LM (1968) High strain-rate testing: tension and compression. Exp Mech
8(1):1–9. https://doi.org/10.1007/BF02326244
19. Albertini C, Montagnani M (1974) Mechanical properties at high rates of strain. Institute of
Physics, London
20. Nicholas T (1980) Tensile testing of materials at high rates of strain. Exp Mech 21(5):177–185.
https://doi.org/10.1007/BF02326644
21. Rajendran AM, Bless SJ (1986) Determination of tensile flow stress beyond necking at very
high strain rate. Exp Mech 26(4):319–323. https://doi.org/10.1007/BF02320146
22. Chen W, Lu F, Cheng M (2002) Tension and compression tests of two polymers under
quasi-static and dynamic loading. Polym Test 21:113–121. https://doi.org/10.1016/S0142-
9418(01)00055-1
23. Gilat A, Goldberg RK, Roberts GD (2005) Strain rate sensitivity of epoxy resin in tensile and
shear loading. NASA/TM—2005-213595, 1–33
24. Owens AT, Tippur HV (2009) A tensile split Hopkinson bar for testing particulate polymer
composites under elevated rates of loading. Exp Mech 47:799–811. https://doi.org/10.1007/
s11340-008-9192-7
25. Chen R, Dai F, Lu L, Lu F, Xia K (2010) Determination of dynamic tensile properties for low
strength brittle solids. Exp Appl Mech 6:321–326. https://doi.org/10.1007/978-1-4419-9792-
0_53
26. Pramanik B, Mantena PR, Tadepalli T, Rajendran AM (2014) Indirect tensile characterization
of graphite platelet reinforced vinyl ester nanocomposites at high-strain rate. Open J Compos
Mater. https://doi.org/10.4236/ojcm.2014.44022
27. ASTM Standard D 638-08 (2008) Standard test method for tensile properties of plastics. ASTM
International
28. Ramesh KT, Narsimhan S (1996) Finite deformations and the dynamic measurement of radial
strains in compression kolsky bar experiments. Int J Solids Struct 33(25):3723–3738. https://
doi.org/10.1016/0020-7683(95)00206-5
29. Chen R, Dai F, Qin J, Lu F (2013) Flattened brazilian disc method for determining the dynamic
tensile stress-strain curve of low strength brittle solids. Exp Mech. https://doi.org/10.1007/
s11340-013-9733-6
30. Song B, Connelly K, Korellis J, Lu WY, Antoun BR (2009) Improvements in kolsky-bar design
for mechanical characterization of materials at high strain rates. Measr Sc Tech 20(11):115–701.
https://doi.org/10.1088/0957-0233/20/11/115701
Axial Deformation Characteristics of Graphene-Sonicated … 337
31. Chen W, Lu F, Zhou B (2000) A quartz-crystal embedded split hopkinson pressure bar for soft
materials. Exp Mech 40(1):1–6. https://doi.org/10.1007/BF02327540
32. Yi F, Zhu Z, Zu F, Hu S, Yi P (2001) Strain rate effects on the compressive property and the
energy-absorbing capacity of aluminum alloy foams. Mater Charact 47:417–422. https://doi.
org/10.1016/s1044-5803(02)00194-8
State Estimation Using Filtering Methods
Applied for Aircraft Landing Maneuver
Abstract State estimation methods are the popular means of validating aerodynamic
characteristics on maneuvering Aircraft. This work deals with adaptation of familiar
filtering methods for Aircraft landing maneuvers, to estimate the aircraft touchdown
states. The mathematical model for two-point landings (main wheel in contact with
the ground and nose wheel airborne) consists of nonlinear flight mechanics equations
representing Aircraft longitudinal dynamics. A nonlinear 6 DOF pilot in loop sim-
ulation model is used for the measurement of data generation that was mixed with
process and measurement noises. These values are used for posterior state correction
in the implementation of Kalman filter. With the state values just before the initia-
tion of flare as initial conditions, filters such as Upper Diagonal factorized form of
Adaptive Extended Kalman Filter (UDAEKF) and Unscented Kalman Filter (UKF)
is implemented in Matlab environment. The estimated states and measured data are
compared using performance metrics for vertical acceleration (Nz) which brings out
the possibility of over quantification (3.5%) and under quantification (11.3%) at onset
of touchdown having an impact on landing loads. As observed, the performance of
UKF is two and half times faster than UDAEKF through superior state propagation.
1 Introduction
The vital phase of flight happens during landing, wherein “Abnormal Runway Con-
tact” is cited as the primary reason for hard landing event [1]. During initial phase of
design, specifications such as Military Specification (MIL), Joint Service Specifica-
tion Guide (JSSG) are used for prediction of the landing loads on high performance
aircraft [2, 3]. The landing gear and near attachment structures are sized according
to the predicted loads for several ground maneuvering and landing cases.
The symmetric two-point (Main Landing Gears (MLG) touchdown and Nose
Landing Gear (NLG) airborne) and three-point (Two MLG’s and NLG touchdown)
landings consists of states such as mass (m), vertical descent speed (h), pitch angle (θ ),
pitch rate (q), pitch acceleration (q̇), true vertical (az ) and longitudinal acceleration
(ax ) that governs the impact loads on Aircraft [4]. The inertial measurement sensors
measure the accelerations (ax & az ) at the touchdown state, being prone usually with
measurement issues such as noise, sensor bias and drift, location of sensor with
respect to Aircraft Center of Gravity (CG), missing data, and so on. The rotary
acceleration (q̇) is a derived response from the rotational velocity which is prone to
divergence due to integration time steps.
Aerodynamic parameter identification using state and parameter estimation
method is the most common application followed by flight mechanics engineers,
working on system identification domain [5, 6]. Several versions of Kalman filters are
tried out against position, velocity tracking problem of size (2 × 2) observation equa-
tion [7]. This work explores the adaptation of popular nonlinear filtering algorithms
on landing maneuver problems to determine the Aircraft touchdown states.
On the basis of first-order approximation of nonlinear dynamics, a well-known
recursive state and parameter estimation called Extended Kalman Filter (EKF) is
implemented in this work [8]. For landing dynamics, the filter is adapted with Upper
diagonal (UD) form and adaptive tuning process using a fuzzy logic interface for
state error covariance and process noise covariance matrices. These implementations
will prevent state error covariance matrix to diverge and support the filter tuning
process. Yet another algorithm implemented using Matlab, is on UKF which basically
propagates finite set of points, called “sigma points”, through nonlinear dynamics
and by approximating the distribution (mean and covariance) through weighted sum
and outer product of propagated sigma points [9].
A nonlinear 6 DOF flight mechanics module is coupled with the pilot in loop
interface. This interface provides the pilot stick input during landing phase of Air-
craft. Three sets of simulated landing cases were performed for measurement data
generation with different flare techniques leading to low, nominal, and high sink rate
velocity landing cases. Initially, the aircraft states during touchdown phase such as
horizontal velocity (u), vertical velocity (w), pitch (θ ), pitch rate (q), and altitude
(h) are estimated and were compared with noisy simulated measurement data. The
validity of the estimates is assessed by plotting the state error with bounds and com-
paring standard deviations. Then, the states such as vertical acceleration (az ), pitch
acceleration (q̇), horizontal acceleration (ax ), and sink rate (ḣ) which has direct
State Estimation Using Filtering Methods Applied for Aircraft … 341
impact on landing loads, are obtained. The Aircraft state vector is augmented with
aerodynamic parameters which were also estimated. These estimates can become
an initial input for performing a multi-body dynamic simulation that can be used to
predict landing impact loads.
The paper is ordered as follows: Sect. 2 describes the configuration of Aircraft and
the equation of motion, Sect. 3 provides the implementation of nonlinear filters on
landing maneuver, Sect. 4 provides the state estimation results for simulated landings,
followed by discussion and conclusion.
A typical high performance Aircraft with four Elevons (Inboard and Outboard) and
single vertical tail with Rudder is considered for the study as shown in Fig. 1. The
Aircraft Elevons are at trailing edge for lateral (Roll) and longitudinal (Pitch) control,
with a Rudder hinged to single vertical tail for directional (Yaw) control. The aircraft
has a tricycle landing gear arrangement with Main landing gears (Starboard and Port
side) symmetrically placed and positioned close to aircraft CG. The Nose landing
gear is located forward of Main landing gear positioned at the front fuselage.
respectively.
q̄s T
u̇ = C x − qw − gsinθ + cos (1)
M M
q̄s T
ẇ = C z + qu + gcosθ + sin (2)
M M
q̄s c̄
q̇ = Cmy (3)
I yy
θ̇ = q (4)
where
um = u (6)
wm = w (7)
qm = q (8)
State Estimation Using Filtering Methods Applied for Aircraft … 343
θm = θ (9)
hm = h (10)
q̄s
axm = Cx (11)
M
q̄s
azm = Cz (12)
M
q̄s
q̇m = Cmy (13)
I yy
wher e, k = 1 . . . N
where f and h are in general nonlinear functions and is the vector of unknown
parameters; x(t) is n ∗1 state vector; u(t) is p∗1 control input; z is m ∗1 measurement
vector sampled at N discrete time steps with fixed sampling interval as t and k is
the discrete-time index; F&G are the additive state and measurement noise matrices,
which are considered to be time-invariant. The noise associated with mathematical
344 P. S. Suresh et al.
where k = 1 . . . , N .
For the augmented system, the EKF consist of prediction and an update step, with
the usage of “tilde” (∼) symbol and “hat” (∧) denote the predicted and corrected
variables, respectively.
∂ fa
Aa (k) = |x =x̂ (k−1) (22)
∂ xa a a
wher e t = tk−1 − tk
∂ga
Ha (k) = |x =x̂ (k) (24)
∂ xa a a
where φa denotes discrete-time transition matrix, Aa (k) denotes the linearized state
matrix and Ha (k) denotes the linearized measurement matrix.
The UDAEKF consists of two important steps namely (i) Time propagation: a
priori states (represented by ~ tilde symbol) and state error covariance matrix pre-
dicted using a nonlinear mathematical model (ii) Measurement update: wherein
with the available measurement data, the posteriori states (represented by ˆ hat sym-
bol), and state error covariance matrix are updated using Kalman gain. The Upper
diagonal (UD) form [12] and an adaptive tuning process using fuzzy logic interface
[13] are detailed in the author’s previous work [14] along with few flight data com-
parison. Although EKF is a widely used filtering strategy, the difficulty experienced
while tuning the fuzzy-based covariance matrix for different sizes of observation
equations and the well-known limitation of EKF that linearize all nonlinear model
to enable linear filtering, had arisen the need for an alternative.
State Estimation Using Filtering Methods Applied for Aircraft … 345
The UKF classified under “sigma point filters” preserves the standard Kalman filter
form which involves propagation of finite set of points, called sigma points, through
the nonlinear dynamics and by approximating the distribution (mean and covariance)
through a weighted sum and outer (cross) product [15].
In contrast to the first-order approximation used on EKF for covariance propa-
gation, in the UKF, nonlinear dynamics are used without approximation, leading to
better performance without the effort of computing Jacobian or Hessian matrix. It
has been shown that UKF is equivalent to second-order EKF and for linear systems
the behavior of Kalman filter, UKF, and EKF are identical [11]. Initially, (2n a + 1)
sigma points are defined, where n a , is the total number of states to be estimated,
which include the basic system state, the aerodynamic parameters. Each sigma point
consists of a vector, one of the sigma vectors is the expected value of the augmented
state vector and the
remaining
2n a points are obtained from the columns of the
matrix square root ±γ Pka for k = 1, 2, . . . n a where P is the covariance matrix
of the augmented state vector (xa ). Starting from setting of parameters as described
below under UKF Parameters for weights of expected value and rest all sigma points
are assigned using scale factors chosen appropriately pertinent to the problem. The
next step is to initialize the state and covariance matrix and iterate through loop for
apriori and posterior corrections.
UKF Parameters
sigmapoints : 2n a + 1
γ = α 2 (n a + k) − n a
√
λ= na + γ
λ
Wom =
(n a + λ)
λ
Woc = + 1 − α2 + β
(n a + λ)
1
Wim = Wic = , i = 1, 2, . . . 2n a
2(n a + λ)
weight for covariance of expected value; Wim is the weight of mean of sigma points;
W0c weight for covariance of sigma points.
Initialize state and covariance matrix as:
xoa = E xoa = E xoT TO (25)
T pxo o
poa = E xoa − x̂oa xoa − x̂oa = o (26)
o pΘ
The initiation of flare maneuver at a height of ~50 ft above the ground level is taken
as initial values for the augmented state (xa ) and considered to have a bias of 10%
from the actual values. The measurement data obtained from the simulation is mixed
with random process noise of σ = 0.001 on states to represent the uncertainty in math
model and measurement noise of S N R = 10 is added to represent the uncertainty
in measurement. These noise parameters are indicative of real measurement data
obtained from several sensors on aircraft. The regular implementation of EKF suffers
from the sensitivity to the initial values of the augmented state. The presence of
adaptive tuning process (Fuzzy logic interface) in UDAEKF eases out the need for
specifying a near about exact initial values. For the UKF, (2 × Number of states + 1)
sigma points were created and appropriate scaling and tuning parameters were set.
Figure 2 presents the estimated states such as (u, w, θ, q, h) of the Aircraft during
landing for case III of flared landing. The state estimates obtained from UDAEKF and
UKF filtering methods are compared against the noisy measurement data obtained
through 6 DOF simulation. The dot-dash line in time versus h subplot indicates
the touchdown instance of Main Landing Gear (MLG). Inset in Fig. 2 shows the
performance of the filters as state error bounds near touchdown time steps. The
necessary and sufficient condition is satisfied by having the state errors within the
Fig. 2 Estimated Aircraft states and residues (inset) for case III of landing using UDAEKF and
UKF algorithms for case III of flared landing
348 P. S. Suresh et al.
√
bounds of ±2 Pk , where Pk is the state covariance matrix. Both these filters were
able to predict the true states of the Aircraft near touchdown; with residues being
minimal for UKF method as compared with UDAEKF. This is primarily due to
the linear approximation of state equations and fuzzy adaptation of process noise
covariance matrix for every time step in UDAEKF as against fair propagation of sigma
points in UKF. Table 1 shows standard deviation for the states such as (u, w, θ, q, h)
that are consistent and values being on the lower side for UKF. The computational
time for UKF is two and half times faster as compared with UDAEKF that involves
a fuzzy logic step.
Three landing cases even though initiated at the same Point In The Sky (PITS), differ
by execution of flare. For case I of flared landing, gentle flare leads to lower vertical
acceleration az among all the three landing cases due to the incremental change of ḣ
and q̇. For Case II of flared landing, the aircraft is flared at terminal phase. This has
resulted in increased acceleration levels of az &ḣ and q̇ at the time of touchdown.
t
1
xek − x p 2
Root Mean Square Error (RMSE) =
N k=1 ny
where N is the number of time steps (t), xe &x p are the estimated aerodynamic
parameters at each time step and actual aerodynamic parameters at the landing phase
n y is the number of aerodynamic parameters.
For the Case III of flared landing, an initial overshoot occurs due to the pitch
stick input which was corrected by an abrupt flare just before touchdown, thereby
leading to higher value of az &q̇, out of all the three landing cases. Figure 3 shows
the acceleration levels at aircraft CG from the time of flare initiation to touchdown
that dictates landing impact loads for higher sink rate observed in case III of landing.
Figure 4 presents the aerodynamic parameter estimates using UDAEKF and UKF
State Estimation Using Filtering Methods Applied for Aircraft … 349
Fig. 3 Aircraft states estimates at touch down for case III of flared landing
in the form of Root Mean Square Error (RMSE). The RMSE error is consistent in
using these estimation methods.
350 P. S. Suresh et al.
Table 2 Aircraft states as PRSSE (%) at touch down time stamp for several flared landing cases
Aircraft Case I of landing Case II of landing Case III of landing
states UDAEKF: UKF: UDAEKF: UKF: UDAEKF: UKF:
PRSSE PRSSE PRSSE (%) PRSSE PRSSE (%) PRSSE
(%) (%) (%) (%)
Longitudinal (+) 1.8 (+) 0.9 (+) 7.0 (+) 3.9 (−) 8.3 (−) 6.4
acceleration
(ax )
Vertical (−) 14.1 (−) 11.3 (+) 3.3 (+) 3.2 (+) 4.3 (+) 3.5
acceleration
(az )
Pitch (−) 4.0 (−) 4.0 (+) 7.0 (+) 7.7 (+) 1.2 (+) 3.0
acceleration
(q̇)
Vertical sink (−) 2.8 (−) 3.4 (+) 4.1 (+) 3.6 (+) 5.7 (+) 6.2
rate (ḣ)
(x m −x e )∧ 2
PRSSE (%) = 2
xm
∗ 100 ; Where x m is the noisy input data and x e is the estimated
states from filtering methods. (+) indicates over quantified and (−) indicates under quantified
measurements
4.3 Discussion
Table 2 presents the Aircraft state which governs the landing impact loads such as
ax , az , q̇, ḣ for all flared landing cases obtained using filtering methods. To quantify
the error, a comparative metrics called Percentage Root Sum Square Error (PRSSE)
is used. Greater emphasis is given to the estimates using UKF method as compared
to UDAEKF since the residue and standard deviation are on the lower side. In Table 2
the (+) sign indicates that the measurement is over quantified and (−) sign indicates
under quantified data.
The emphasis for aircraft true state estimation at touchdown is noticeable from
the comparative data presented in Table 2. The results of vertical acceleration (az )
are discussed in detail. The Case II and Case III of flared landing indicate that the
measurement data of az are higher than the estimated by a maximum of 4.3% for
UDAEKF and 3.5% for UKF. For Case I of flared landing, measurement data for az
is lower than estimated by 14.1% for UDAEKF and 11.3% for UKF. The over quan-
tification of vertical acceleration measurement data, during the event such as “hard
landing”, can lead to inevitable downtime of aircraft for thorough inspection [16].
On the other hand, the under quantification of measurement data leads to ignorance,
causing higher nominal stresses that can lead to accumulation of cumulative fatigue
damage.
State Estimation Using Filtering Methods Applied for Aircraft … 351
5 Conclusion
Acknowledgements The authors from Aeronautical Development Agency acknowledge “The Pro-
gramme Director” for the facility provided to publish this work. Special thanks to Technology
Director (Airframe) and Group Director (Loads, Airframe) for their technical support.
References
1. Lacagnina M (2004) Flight safety digest: avoiding hard landings. Tech Publ 23(8):249–250.
Flight Safety Foundation
2. Airplane strength and rigidity: landing and ground loads, May 1960
3. Airplane strength and rigidity: ground loads for navy acquired airplanes, May 1987
4. Sartor P, Schmidt RK, Becker W, Worden K, Bond DA, Staszewski WJ (2010) Conceptual
design of a hard landing indication system using flight parameter sensor simulation model. In:
27th international congress of the aeronautical science
5. Crassidis JL, Junkins JL (2012) Optimal estimation of dynamic systems, 2nd edn. CRC Press
6. Julier SJ, Uhlmann JK (1995) A new extension of Kalman filter to nonlinear systems. In:
Proceedings of the American control conference, pp 1628–1632
7. Chen G, Chui CK (1991) A modified adaptive Kalman filter for real-time applications. IEEE
Trans Aerosp Electron Syst 27(1):149–154
8. Mulder JA, Chu QP, Sridhar JK, Breeman JH, Laban M (2010) Non-linear aircraft flight path
reconstruction review and new advances. Prog Aerosp Sci 35:673–726
352 P. S. Suresh et al.
Abstract The study of viscous flow in a lid-driven cavity is carried out using alter-
nating direction implicit method. The conservation form of incompressible Navier–
Stokes equation in stream function–vorticity form is solved using second-order accu-
rate central difference scheme in a uniform finite-difference grid mesh. The numerical
solution is obtained up to highest Reynolds number 32,500 from the lowest 0.00001
using the grid sizes 129 × 129, 257 × 257 and 513 × 513. Good agreement of the
result is found with Erturk et al. (Int J Numer Methods Fluids 48:747–774, 2005)
[1]. The study of flow properties in the form of velocity profiles, stream function and
vorticity contour plots and location of primary and secondary eddies are carried out.
The novelty of this study is that the magnitude of the vorticity value does not cross
the theoretical limit −1.8859 (Burggraf 24:113–151, 1966 [2]).
1 Introduction
The lid-driven cavity is a square cavity in which the vertical walls and lower horizontal
wall are stationary. The top wall is allowed to move horizontally, and it is called the
lid of the cavity. The non-dimensional length of the cavity is unity. Initially, the
cavity is filled with fluid which is at rest. When the lid moves towards right with
non-dimensional velocity unity, the fluid flow in the cavity sets up. The velocity of
flow in the cavity becomes higher or lower depending upon the Reynolds number of
B. Dalai (B)
Faculty, Centre for Advanced Post-Graduate Studies, Biju
Pattnaik University of Technology Odisha, Rourkela, India
e-mail: banamali.2000@gmail.com
M. K. Laha
Faculty, Aerospace Engineering, Indian Institute
of Technology Kharagpur, Kharagpur, West Bengal, India
2 Formulation
where u and v are the velocities in x- and y-directions, respectively; ω is the vorticity
and Re is the Reynolds number of the flow. The Reynolds number for the flow is
defined by Re = uνL . Here L is the length of the lid and ν is the kinematic viscosity
of the fluid. The stream function (ψ) is coupled with vorticity by the equation as
∂ 2ψ ∂ 2ψ
+ = −ω (2)
∂x2 ∂ y2
t t t t
where cx = 4x , c y = 4y , dx = 2Re(x)2 , dy =
2Re(y)2
, (vω)i+1
n
j = vi+1 j ωi+1 j
n n
and t, x and y are the time step, grid widths in x- and y-directions, respectively.
Similarly, after arrangement of the discretized equation at n + 1 time step:
n+1 n+1
n+1
c y vi+1 j − d y ω n+1
i+1 j + 1 + 2d y ωij − c y vi−1 j + d y ωi−1
n+1
j
n+ 21
n+ 1 n+ 1
n+ 1 n+ 1
n+ 1
= ωi j + dx ωi j+12 − 2ωi j 2 + ωi j−12 − cx (uω)i j+12 − (uω)i j−12 (5)
tridiagonal matrix algorithm at each time step. The convergence of the solution is
obtained when the residues (LHS–RHS) of Eqs. (1) and (2) equals to 10−10 . The
numerical solutions are obtained using the grid sizes 129 × 129, 257 × 257, 513 ×
513 and 1025 × 1025.
The results obtained in the grid size 513 × 513 are validated with Erturk et al.’s
[1] (grid size 601 × 601) result in the form of u- and v-velocity profiles along y- and
x-directions, respectively, passing through the centre of the cavity as shown in Fig. 1.
Good agreement of the result is obtained.
ω = ∇ × V (6)
Numerical Solution of Steady Incompressible Flow … 357
where V = u î +v ĵ is the velocity of flow in the cavity. Expansion of Eq. (6) produces
value at the central region of the cavity because the velocity gradients ∂v ∂ x
constant
and ∂u ∂ y are linear at the centre of the cavity. The vorticity values near the walls
show large variation in comparison to the central region. This variation is due to the
presence of more viscous flow region near the wall than that of central region. The
viscous flow near the wall signifies the presence of boundary layer near the walls.
The location and strength of primary eddies at different Reynolds numbers are
compared with Erturk et al. [1] in Table 1. It is observed that both results are found
to match very well.
Numerical Solution of Steady Incompressible Flow … 359
Table 2 represents the magnitude of stream function and vorticity values at the
centre of the primary eddy at different Reynolds numbers and various grid sizes. It is
observed that the values are accurate up to two digits after decimal place in all grid
sizes. This table shows that the values are independent of grid sizes.
Figure 4 shows the variation of vorticity value at the centre of the primary vortex at
different Reynolds numbers. The magnitude of the vorticity value is also compared
with the results published in the literature. Among the compared results, Erturk
et al.’s [1] second-order accurate in the grid size 601 × 601, Erturk’s [4] second-
order accurate in the grid size 1025 × 1025 and Erturk and Gockol’s [5] fourth-order
accurate in the grid size 601 × 601 are compared with the present results in the grid
360
Table 1 Comparison of the location and strength at the centre of the primary eddy
Re Grid Location Strength
Present Erturk et al. [1] Present Erturk et al. [1]
x y x y ψ ω ψ ω
1E−04 129 × 129 0.5000 0.7656 – – −0.1000 −3.2190 – –
1000 513 × 513 0.5313 0.5645 0.5300 0.5650 −0.1188 −2.0664 −0.1189 −2.0655
5000 513 × 513 0.5313 0.5645 0.5150 0.5350 −0.1188 −2.0664 −0.1212 −1.9266
10,000 513 × 513 0.5117 0.5293 0.5117 0.5300 −0.1212 −1.9023 −0.1204 −1.8889
15,000 513 × 513 0.5098 0.5273 0.5100 0.5283 −0.1205 −1.8845 −0.1192 −1.8636
20,000 513 × 513 0.5098 0.5273 0.5100 0.5267 −0.1198 −1.8705 −0.1180 −1.8418
25,000 513 × 513 0.5078 0.5254 – – −0.1191 −1.8586 – –
30,000 513 × 513 0.5078 0.5254 – – −0.1184 −1.8472 – –
32,500 513 × 513 0.5078 0.5254 – – −0.1181 −1.8419 – –
B. Dalai and M. K. Laha
Numerical Solution of Steady Incompressible Flow … 361
Fig. 4 Plot between the vorticity values at the centre of the primary eddy at different Reynolds
numbers
sizes 513 × 513 and 1025 × 1025. The results in the present computation are possible
up to Reynolds number 22,500 and 32,500 in the grid sizes 513 × 513 and 1025 ×
1025, respectively. The present computed results up to Reynolds number 32,500 in
the grid size 513 × 513 show better than the Erturk et al.’s [1] result in the grid size
601 × 601. The results up to Reynolds number 22,500 in the grid size 1025 × 1025
of present computation shows better than Erturk’s [4] second-order accurate result in
the grid size 1025 × 1025 because the vorticity values remain within the theoretical
limit −1.8859 which was computed by the Burggraf [2] as shown in Fig. 4.
The study has been conducted up to maximum Reynolds number 32500 in the
grid size 513 × 513 (Table 3). The flow at higher Reynolds number should achieve
three-dimensional effect. Since the study is limited within two-dimensional effect,
it is very difficult to predict three-dimensionality effect in this cavity study.
Table 3 Shows the location and strength of the secondary eddies at the corners of the cavity
362
Grid ψ ω ψ ω ψ ω ψ ω ψ ω ψ ω ψ ω
size x y x y x y x y x y x y x y
1000 2.334e-4 0.3562 1.732e-3 1.1141 -8.544e-3 -5.932e-8 -5.932e-8 -8.998e-3 8.826e-9 4.283e-3 8.646e-7 0.02999
513 0.08398 0.0781 0.8633 0.1113 3.906e-3 5.859e-3 0.9922 0.0078 0.0059 0.0039 0.9844 0.0195
5000 1.372e-3 1.4975 3.091e-3 3.091e-3 -7.671e-3 -1.004e-2 0.9922 -3.624e-2 1.049e-8 4.017e-3 1.744e-6 0.0449
513 7.226e-2 0.1367 0.8027 2.7972 -7.671e-3 -1.004e-2 -1.524e-6 1.953e-2 0.0059 0.0039 0.9805 0.0215
10000 1.607e-3 2.1436 3.224e-3 2.7972 -7.671e-3 -3.186e-2 -1.496e-4 -0.3133 5.09e-11 1.098e-3 6.509e-9 6.509e-9
513 5.859e-2 0.1621 0.7734 3.8543 -3.186e-2 -3.186e-2 -1.496e-4 -0.3133 1.953e-3 1.953e-3 0.9961 3.906e-3
15000 1.663e-3 2.5064 3.049e-3 5.1527 -2.252e-5 -0.14007 -3.624e-2 -0.4755 1.452e-9 1.743e-3 1.726e-8 5.436e-3
513 5.273e-2 0.1719 0.7422 0.0488 0.0371 0.04297 0.04297 0.0898 1.953e-3 3.906e-3 0.9941 5.859e-3
20000 1.614e-3 2.9735 2.859e-3 6.4940 -8.182e-5 -0.2505 -5.022e-4 -0.5818 4.261e-4 2.932e-3 3.964e-8 7.529e-3
513 4.68e-2 0.1836 0.7188 0.0429 5.859e-2 5.468e-2 -0.25054 -0.5818 3.906e-3 3.906e-3 0.9922 7.813e-3
25000 1.345e-3 3.4415 2.697e-3 7.7831 -6.357e-4 -0.6532 -6.357e-4 0.1191 6.826e-9 2.511e-3 1.601e-7 1.338e-2
513 0.04297 0.1933 0.69727 0.0371 6.836e-2 -0.6532 -6.357e-4 -0.6532 3.906e-3 3.906e-3 0.9883 0.0195
(continued)
B. Dalai and M. K. Laha
Table 3 (continued)
30000 1.479e-3 3.9124 2.564e-3 8.9831 -1.779e-4 -0.35764 -7.609e-4 -0.6826 8.826e-9 4.283e-3 8.646e-7 0.02999
513 0.03906 0.2031 0.06796 0.0332 0.072265 0.06055 0.92969 0.1231 0.0059 0.0039 0.9844 0.0195
32500 1.446e-3 4.1611 2.506e-3 9.5568 -1.933e-4 -0.38611 -8.200e-4 -0.6974 1.049e-8 4.017e-3 1.744e-6 0.0449 -2.1e-10 -1.2e-3
513 0.0371 0.2070 0.67188 0.0313 0.076172 0.058594 0.929688 0.1250 0.0059 0.0039 0.9805 0.02148 0.9980 0.0019
Re ψ ω ψ ω ψ ω ψ ω ψ ω ψ ω ψ ω
x y x y x y x y x y x y x y
TL1 3.452e-4 1.2955 1.448e-3 2.1010 2.619e-3 2.2828 3.27e-3 2.3843 3.716e-3 2.4612 1.338e-2 4.048e-3 4.433e-3 2.5846
0.04297 0.8887 0.0625 0.9102 7.031e-2 0.9121 7.617e-3 0.9121 8.008e-2 0.9121 4.048e-3 2.5181 0.0859 0.9141
4 Conclusion
The solution for the lid-driven cavity was obtained using alternating direction implicit
method from Reynolds number 0.00001–32,500 in the grid sizes 513 × 513. The
solution is obtained up to Reynolds number 22,500 in the grid size 1025 × 1025. The
results obtained here are found to match very well with the literature. The vorticity
value obtained in the grid size 1025 × 1025 remains within the theoretical limit of
−1.8859 [5]. A remarkable result of appearance of fifth level of eddy is observed at
the Reynolds number 32,500 in the grid size 513 × 513.
References
1. Erturk E, Corke TC, Gockol C (2005) Numerical solutions of 2-D steady incompressible driven
cavity flow at higher Reynolds numbers. Int J Numer Methods Fluids 48:747–774
2. Burggraf OR (1966) Analytical and numerical studies of the structure of steady separated flows.
J Fluid Mech 24:113–151
3. Ghia U, Ghia KN, Shin CT (1982) High re-solution for incompressible flow using the Navier-
Stokes equations and a multigrid method. J Comp Phys 48:387–411
4. Erturk E (2009) Discussions on driven cavity flow. Int J Numer Methods Fluids 60:275–294
5. Erturk E, Gokcol C (2006) Fourth-order compact formulation of Navier-Stokes equations and
driven cavity flow at high Reynolds numbers. Int J Numer Methods Fluids 50:421–436
Stagnation and Static Property
Correlations for Equilibrium Flows
1 Introduction
The stagnation and static property relations for calorically perfect gases can be
derived analytically [1]. Similarly, for thermally perfect gases where specific heat
varies with temperature, the analytical derivation of relationship between stagnation
and static properties has been worked out by Zebbiche [2]. Most of the previous stud-
ies considered a single non-dissociating gas. The aim of this work is the numerical
development of stagnation–static property correlations for gaseous phase equilibrium
flows where both specific heats and compositions vary with temperature.
Equilibrium flows are often encountered in rocket engines, where combustion
products remain in equilibrium as they undergo expansion. The practical application
of the study is hence linked to rocket engines and in particular liquid engines, which
use a liquid fuel along with a liquid oxidizer.
2 Mathematical Formulation
For an isolated chemically reacting system consisting of “NS” species, the Gibbs
energy is defined as [9]
NS
p
G= Ĝ 0j + RT ln x j + RT ln nj (1)
j=1
pref
where Ĝ 0j is the standard molar Gibbs function of species “j”; n j is the number of
moles of species “j”; R is the universal gas constant equal to 8.314 J/mol K; x j is the
mole fraction of species “j”; “p” is the pressure; and pref is the reference pressure
of 1 bar. The index “j” varies as 1, 2, …, NS. Application of mass balance for each
element in the reacting system results in the following equation:
NS
ai j n j − bi = 0 (2)
j=1
where the index “i” varies as 1, 2, …, E (E being the total number of elements
involved in the chemical reaction); ai j denotes the number of atoms of element “i” in
one molecule of the species “j”; and bi denotes the total number of moles of element
“i” in the mixture.
The equilibrium problem is basically related to the minimization of Gibbs energy
subject to the constraint of mass conservation. Using method of Lagrange multipliers,
a new function L is defined as
Stagnation and Static Property Correlations for Equilibrium Flows 367
NS
p
L(n 1 , . . . , n NS , λ1 , . . . , λ E ) = Ĝ 0j + RT ln x j + RT ln nj
j=1
pref
⎛ ⎞
E NS
− λi ⎝ ai j n j − bi ⎠ (3)
i=1 j=1
and
∂L NS
= ai j n j − bi = 0 (5)
∂λi j=1
NS
n j − Ng = 0 (6)
j=1
Because summation of mole fraction is unity, combining Eqs. (6) and (7) gives
NS E
pref −Ĝ 0j + i=1 λi ai j
e RT −1=0 (8)
j=1
p
NS
pref −Ĝ 0j + i=1
E λ a
i ij
ai j N g e RT − bi = 0 (9)
j=1
p
Both Eqs. (8) and (9) represent the final decoupled set of simultaneous nonlin-
ear equations with unknowns λ1 , λ2 , . . . , λ E and N g . The number of simultaneous
nonlinear equations to be solved is now reduced from NS + E + 1 to just E + 1. In
order to determine mole fraction (or no. of moles) of each species, Eq. (7) is back
substituted with λ1 , λ2 , . . . , λ E and N g .
In problems where equilibrium temperature is known, solution of Eqs. (8) and (9)
yields the equilibrium composition. However, in most of the practical applications,
equilibrium temperature needs to be determined and hence additional equations are
required. This is achieved by following either of the two approaches, viz., (1) enthalpy
balance or (2) entropy balance.
NS
n j H j − Hreactant = 0 (10)
j=1
where H j is the molar enthalpy of jth species at temperature “T” and Hreactant is the
enthalpy of oxidizer–fuel mixture before combustion.
NS
n j Ŝ j − Sreactant = 0 (11)
j=1
where Ŝ j is the molar entropy of jth species at temperature “T” and Sreactant is the
entropy of oxidizer–fuel mixture before combustion.
Ŝ j is evaluated as given in Ref. [5],
Stagnation and Static Property Correlations for Equilibrium Flows 369
p
Ŝ j = Ŝ 0j − Rln x j − Rln (12)
p0
3 Methodology
Fig. 1 Algorithm for determination of stagnation and static property correlations for equilibrium
flows
370 S. Maurya and A. Vaidyanathan
The input parameters such as chamber pressure, i.e., stagnation pressure ( p0 ), mixture
ratio (MR), and enthalpy of reactants (Hreact. ), are specified by the user and fed into
the next subroutine.
This subroutine solves the equilibrium problem based on the stagnation condi-
tions. Equations (8), (9), and (10) are solved to obtain the Lagrange multipliers
(λ1 , λ2 , .. . ,λ E ), equilibrium temperature (T ), and the total number of moles at equi-
librium N g . The equilibrium composition (x1 , x2 , . . . , xNS ) is determined using
Eq. (7). The entropy of the combustion products (S0 ) is evaluated using Eq. (11) and
its value is passed into next step.
Equations (8), (9), and (11) are solved for each value of pressure, and the corre-
sponding equilibrium compositions are obtained from Eq. (7). The entropy remains
fixed at S0 throughout the run. The pressure at this step can be varied in the interval
(0, p0 ) and corresponding to each value of pressure, a unique equilibrium solution
is obtained. The equilibrium results for the chosen pressure value are input to next
subroutines (see Fig. 1).
The enthalpies of the equilibrium mixture obtained in first and third steps (see Fig. 1)
are used to compute the speed of equilibrium mixture. Noting that H0 (also, H 0 =
H react. ) denotes the stagnation enthalpy at chamber pressure ( p0 ), the enthalpy (H )
of the equilibrium mixture at pressure “p” (where p < p0 ) would be lesser than the
stagnation enthalpy (H0 ) in accordance with the principle of conservation of energy.
The speed “v” of the equilibrium mixture is then expressed as
H0 H
ν= 2 × 1000 × − (13)
MW0 .N g,0 MW.N g
Stagnation and Static Property Correlations for Equilibrium Flows 371
where “MW” is the molecular weight (in g/mol) of equilibrium mixture, the subscript
“0” denotes the stagnation condition and “1000” is the multiplication factor required
for consistency of units. The speed is input to the next subroutine for computing the
Mach number.
The Mach number (M = v/a) is calculated and stored for each value of pressure.
The ratios of stagnation to static properties (i.e., p0 / p and T0 /T ) are computed for a
particular value of pressure. The pressure “p” is varied in the interval (0, p0 ) and the
steps
3–7 are repeated.
As a result, a mapping between pressure “p” and parameters
p0 /p , T0 /T, M is obtained.
The ratios of stagnation to static properties along with Mach numbers are computed
at different chamber pressures and mixture ratios for LOX/LH2 and MMH/N2 O4
fuel–oxidizer combinations. The results are presented separately for each case.
4.1 LOX/LH2
Fig. 2 Ratio of stagnation to static a pressure; b temperature versus Mach number at fixed chamber
(stagnation) pressure of 10 MPa and varying mixture ratio for LOX/LH2 fuel–oxidizer combination
It is observed that both the stagnation to static ratios for pressure and temperature
increase monotonically with Mach number—the same holdstrue for thermally and
2γ
calorically perfect gases (where T0 /T ∝M 2 and p0 / p ∝ M γ −1 ). Also, the effect
of varying mixture ratio on the stagnation to static ratios is clearly evident from the
figures. In Fig. 2b, as mixture ratio increases, the T0 /T curve shifts down until the
mixture ratio gets equal to 8, at which it approaches minimum (shown in black) and
moves up thereafter. Since the value of stoichiometric mixture ratio for LOX/LH2
fuel combination is also 8 and it is known that equilibrium temperature attains maxi-
mum at stoichiometric mixture ratio, the ratio T0 /T attains minimum because of the
denominator “T” attaining maximum.
The effect of varying chamber pressure at fixed mixture ratio is illustrated in Fig. 3.
The mixture ratio is kept at 5, and chamber pressure varies from 1 to 10 MPa. The
ratio T0 /T increases with increase in chamber (stagnation) pressure because with
increase in chamber pressure, equilibrium temperature also increases but the extent
by which T0 increases outweighs the increase in T .
4.2 MMH/N2 O4
Fig. 3 Ratio of stagnation to static temperature versus Mach number at fixed mixture ratio (MR =
5) and varying chamber (stagnation) pressure for LOX/LH2 fuel–oxidizer combination
5 Conclusions
A method for determining stagnation and static property correlations for equilibrium
flows is developed. Results are presented for combustion equilibrium of two oxi-
dizer–fuel combinations, viz., LOX/LH2 and MMH/N2 O4 . The effect of variation in
chamber pressure and mixture ratio upon stagnation to static property ratio is also
studied. Unlike single non-dissociating calorically perfect gases where T 0 /T and p0 /p
depend on Mach number and specific heats ratio, in equilibrium flows, they depend
on chamber pressure (stagnation conditions), mixture ratio, and Mach number. The
intrinsic nature of equilibrium flows involves variation of specific heats and compo-
sitions with respect to temperature which in the present formulation are accounted
by adjusting both chamber pressure and mixture ratio.
There are two limitations to the current study: first, the effect of number of species
present at equilibrium upon the ratio of stagnation to static properties has not been
studied; second, while formulating the methodology, the expansion process of equi-
librium flow is assumed to be adiabatic and isentropic leading to an extra assumption
of “isentropic flow” in the current definition of stagnation to static temperature,
whereas the usual convention of defining stagnation temperature only assumes an
adiabatic flow.
374 S. Maurya and A. Vaidyanathan
Fig. 4 Ratio of stagnation to static pressure versus Mach number at fixed chamber (stagnation)
pressure of 1 MPa and varying mixture ratio for MMH/N2 O4 fuel–oxidizer combination
Fig. 5 Ratio of stagnation to static temperature versus Mach number at a fixed chamber (stagnation)
pressure of 1 MPa and varying mixture ratio for MMH/N2 O4 fuel–oxidizer combination
Stagnation and Static Property Correlations for Equilibrium Flows 375
Fig. 6 Ratio of stagnation to static temperature versus Mach number at fixed mixture ratio (MR =
2) and varying chamber (stagnation) pressure for MMH/N2 O4 fuel–oxidizer combination
The stagnation and static properties for LOX/LH2 equilibrium mixture consisting
of eight species, viz., H, H2 , H2 O, H2 O2 , HO2 , O, O2 , and OH are evaluated by
following the steps aforementioned in Sect. 3. The detailed step-by-step procedure
is given below:
Step-1: Input parameters
p0 = 10 bar; MR = 2; H react. = −10,647 J.
Note that number of species, NS = 8 and number of elements in the mixture,
E
= 2; Molecular weight of H 2 MWH2 = 2 g/mol; Molecular weight of O2
MWO2 = 32 g/mol
Step-2: Evaluation of stagnation properties
Equation (8) results in
0 0 0 0 0
− G H +λH − G H2 +2λH − G H2 O +2λH +λO − G H2 O2 +2λH +2λO − G HO2 +λH +2λO
e RT0
+e RT0
+e RT0
+e RT0
+e RT0
0 0
0
− G O +λO − G O2 +2λO − G OH +λH +λO
+e RT0
+e RT0
+e RT0
− p0 = 0 (14)
376 S. Maurya and A. Vaidyanathan
0
− G OH +λH +λO p0
+e RT0
−2 =0 (15)
N g,0
and
0 0 0
0 0
− G H2 O +2λH +λO − G H2 O2 +2λH +2λO − G HO2 +λH +2λO − G O +λO − G O2 +2λO
e RT0
+ 2e RT0
+ 2e RT0
+e RT0
+ 2e RT0
0
− G OH +λH +λO p0 MWH2
+e RT0
−2 MR =0 (16)
N g,0 MWO2
0
0 0
0
− G HO2 +λH +2λO
− G O +λO − G O2 +2λO
− G OH +λH +λO
+ H HO2 e RT0
+ H Oe RT0
+ H O2 e RT0
+ H OH e RT0
p0
− Hreact. = 0 (17)
N g,0
ŜH , ŜH2 , ŜH2 O , ŜH2 O2 , ŜHO2 , ŜO , ŜO2 , and ŜOH are evaluated from Eq. (12) by
extracting values of ŜH0 , ŜH0 2 , ŜH0 2 O , ŜH0 2 O2 , ŜHO
0
2
, ŜO0 , ŜO0 2 , and ŜOH
0
at temperature “T 0 ”
from thermodynamic database [10].
S 0 is then used in step-3.
Stagnation and Static Property Correlations for Equilibrium Flows 377
A particular value of pressure “p” such that 0 < p < p0 is chosen and the following
equations are solved in a similar manner as in step-2.
Equation (8) results in
0 0 0 0 0
− G H +λH − G H2 +2λH − G H2 O +2λH +λO − G H2 O2 +2λH +2λO − G HO2 +λH +2λO
e RT +e RT +e RT +e RT +e RT
0 0
0
− G O +λO − G O2 +2λO − G OH +λH +λO
+e RT +e RT +e RT −p=0 (18)
0
0
− G HO2 +λH +2λO − G OH +λH +λO p
+e RT +e RT −2 =0 (19)
Ng
and
0 0 0
0
− G H2 O +2λH +λO − G H2 O2 +2λH +2λO − G HO2 +λH +2λO − G O +λO
e RT + 2e RT + 2e RT +e RT
0
0
− G O2 +2λO − G OH +λH +λO p MWH2
+ 2e RT +e RT −2 MR =0 (20)
Ng MWO2
− G OH +λH +λO −G OH + λH + λO p
+e RT . ŜOH −
0
− S0 = 0 (21)
T Ng
378 S. Maurya and A. Vaidyanathan
For p = 0.1 bar, the solution of Eqs. (18)–(21) yields T = 592.8 K, λ H = −4.68 ×
104 J/mol, λ O = −2.82 × 105 J/mol, and N g = 1 mol.
The enthalpy of mixture (H) is calculated as
H = n H H H + n H2 H H2 + n H2 O H H2 O + n H2 O2 H H2 O2 + n HO2 H HO2
+ n O H O + n O2 H O2 + n OH H OH
References
7. Reynolds WC (1986) The element potential method for chemical equilibrium analysis:
implementation in the interactive program of STANJAN Version 3. Stanford University
8. Wang J et al (2009) Chemical equilibrium calculation of complex system based on element
potential and its application. World congress on computer science and information engineering
9. Zucrow MJ, Hoffman JD (1977) Gas dynamics, vol 2. Wiley, New York
10. NASA Glenn Thermodynamic Database. https://www.grc.nasa.gov/WWW/CEAWeb/
ceaThermoBuild.htm
CFD Simulation of Hypersonic Shock
Tunnel Nozzle
Jigarkumar Sura
Abstract The design of a hypersonic nozzle for shock tunnel is of much importance
as it plays a vital role in generating the required flow field. The computational simula-
tion can help to understand the Mach number distribution inside the test section. This
can help to design the mounting assembly for the model and to limit the dimension
of the model. The flow field for an axisymmetric test section has been analysed using
commercial software. The results revealed the variation of Mach number inside the
test section and exit plane of the nozzle.
1 Introduction
The hypersonic vehicles are considered as future means of fast transportation. The
phenomena such as thick boundary layer, thin shock layer, viscous interaction, shock
wave–boundary layer interaction and shock–shock interaction are present in hyper-
sonic flow and cannot be ignored [1]. The hypersonic wind tunnel and hypersonic
shock tunnel are major experimental facilities to study flow field inside and around the
hypersonic vehicles. The hypersonic shock tunnel can simulate required surround-
ings for the vehicle with test duration of the order of millisecond. The high-speed
flow is generated through reflection of the incident shock wave in the driven section
of shock tube by sudden rupture of a diaphragm separating the driver and the driven
section [2]. The axial Mach number distribution for ideal conditions can be estimated
with one-dimensional isentropic flow equations [3]. But actual velocity direction is
not parallel to axis at all radial locations due to three-dimensional effects [4]. Because
of this, it is required to identify the useful region of the test section where velocity
remains parallel to the axial direction.
J. Sura (B)
Department of Aerospace Engineering, Amity School of Engineering & Technology, Amity
University Mumbai, Panvel, Maharashtra, India
e-mail: jigar.sura@gmail.com
Here, in this paper, an attempt has been made to identify the useful core region of
a hypersonic shock tunnel test section using commercial simulation software Ansys.
The inviscid simulation has been carried out. The diameter and length of the inlet
(before throat) are 50 mm and 50 mm, respectively. The throat diameter is 22 mm,
while exit diameter is 300 mm. The nozzle semi-cone angle is 10° and is designed
for Mach 8 (Fig. 1). The stagnation pressure is 561 kPa, and stagnation temperature
is 813 K [5].
2 Simulation Methodology
The convergence criteria for various parameters were set to 1E−6 and the solution
converged to that limit. The implicit Advection Upstream Splitting Method (AUSM)
was used for simulations, which was good for capturing shocks in fluid flow. After
the simulation was completed, the mass flux imbalance was checked. It was found
that there was no significant mass accumulation inside the computational domain.
Hence, the converged solution could be used for further analysis.
384 J. Sura
The simulation results provided insight into the flow field of hypersonic shock tun-
nel nozzle. The inviscid flow field has been analysed using commercially available
simulation software Ansys. The grid independency tests have been carried out.
The contour plot provides information about the variation of properties such as
pressure, temperature and Mach number. Figure 3 shows the contour plot for Mach
number. It can be seen that the flow field inside the test section is not uniform.
There is a variation of Mach number in both axial and radial directions. The oblique
shock wave beginning from exit of the nozzle extending into test section is also
visible. This can be attributed to the flow turning into itself because of beginning
of the constant diameter test section. The strength of this shock wave depends on
the nozzle divergence angle. This shock wave defines the dimensions of the models
that can be put inside the test section. If the shock wave hits any of the surfaces of
the model, then the flow field downstream from the impingement location will be
affected. Also, the shock boundary layer interaction will alter the flow field and the
experimental data will be erroneous.
The results show the notable variation of Mach number in radial direction at the
exit of the nozzle. For comparison, the plots for radial distribution of Mach number
in test section at five equally spaced locations (0 mm, 112.5 mm, 225 mm, 337.5 mm
and 450 mm from test section inlet) including test section entry and exit have been
given in Fig. 4. The x-axis shows the Mach number, while y-axis shows the distance
from centreline. It can be seen that at the entrance of the test section (first from the
left in Fig. 4), the flow Mach number is 8 which is the design Mach number.
The Mach number reduces when one goes away from the axis of the nozzle. This
is because of three-dimensional effects due to nozzle divergence angle. It can also
be seen that the change in Mach number is gradual, and available core region is only
up to 0.1 m from centreline above which the Mach number variation is significant.
The second plot from left in Fig. 4 also shows that there is sudden change in Mach
number at 0.13 m radial location. And that change is visible in remaining figures
(third, fourth and fifth from left) also but distance from axis is reducing. This is
attributed to shock wave generated at the beginning of the test section. It can also be
seen in Fig. 4 that the axial Mach number increases in flow direction inside the test
section.
The test section of the hypersonic shock tunnel is kept at almost zero pressure. This
helps to start the nozzle flow without any back pressure effect. Also, the volume of the
dump tank is kept very large so that throughout the tunnel operation, the back pressure
effect does not affect the nozzle performance. This leads to the underexpanded nozzle
as the expansion is not complete till nozzle exit plane. Therefore, the flow further
expands into the test section. This is the reason why the axial Mach number increases
inside the test section. For a given area ratio, the Mach number at the exit of the nozzle
is fixed. So, usually the experiments are carried out at design Mach number only. But
looking at the underexpansion of the flow, it is thought that the change in mounting
CFD Simulation of Hypersonic Shock Tunnel Nozzle 385
location for the model can help to carry out experiments at different higher Mach
numbers.
The oblique shock wave from the test section entrance plane limits the useful core
available for experiments. Not only the model cross section should be sufficiently
smaller than the core area (160 mm diameter in current case), but also the length of the
model should not be in the downstream of the oblique shock. This limits the length
of the model for experiments. Based on Mach number distribution, the available core
region can be identified. This core region will have most uniform free streamflow.
4 Conclusion
The flow-through hypersonic nozzle of the simple shock tunnel has been analysed.
The results from the simulation provided the information about the core area to mount
model without any external disturbances. Also, the fixed Mach number nozzle can
be used for higher Mach number with appropriate changes in model dimension.
386 J. Sura
References
1. Anderson JD (1989) Hypersonic and high temperature gas dynamics. McGraw-Hill, New York
2. Martin WA (1958) A review of shock tubes and shock tunnels. CONVAIR, San Diego
3. Rathakrishnan E (2012) Gas dynamics. PHI Limited
4. Crown JC, Heybey WH (1950) Supersonic nozzle design. Project NOL 159, April
5. Trivedi SD (2013) Development and validation of a force balance for a lifting model in an ultra
short duration hypersonic test facility. Ph.D. thesis
A DNS Study of Bulk Flow
Characteristics of a Transient Diabatic
Plume that Simulates Cloud Flow
1 Introduction
Clouds are complex systems, and interaction among a wide range of scales makes the
cloud parameterization problem “deadlocked” [1]. The Intergovernmental Panel on
Climate Change (IPCC) has identified clouds as one of the most urgent and difficult
problems needing attention, as the absence of a sufficiently successful model to
capture the actual dynamics of the cloud leads to significant biases (depending on the
model chosen) in climate predictions [2]. For the Indian monsoons, on which about a
half of the country’s agriculture is dependent, a significantly better parameterization
of clouds could help in improving rainfall predictions [3].
The major problem in modeling a cloud lies in the lack of our understanding of
how a cumulus cloud interacts with the surroundings. Earlier cloud fluid-dynamical
models based on steady-state plumes, thermals, and bubbles [4, 5] turned out to be
unsuccessful as the experiments did not incorporate an adequate model for describ-
ing the effects of latent heat release on condensation of water vapor into liquid
water. Also, the transient nature of cumulus clouds has to be taken into account.
Narasimha [6] and Diwan et al. [7] incorporated these two parameters and demon-
strated that a transient diabatic plume (TDP) provides an appropriate fluid-dynamical
model for cumulus cloud flow. Meanwhile, large-eddy simulations (LES) of cumulus
S. Rao (B)
Department of Mechanical Engineering, IIT Jammu, Jammu, Jammu and Kashmir, India
e-mail: samrat.rao@gmail.com
G. R. Vybhav · S. M. Deshpande · R. Narasimha
Engineering Mechanics Unit, Jawaharlal Nehru Center for Advanced Scientific Research,
Bangalore 560064, India
P. Prasanth
Michigan Technological University, Michigan, USA
convection [8] are gaining more attention in the recent past for their use in super-
parameterization schemes. In contrast to LES, our approach to model the cloud
flow as a TDP and solve it by employing a high-resolution direct numerical simula-
tion (DNS) technique has an emphasis on improving the fundamental understanding
of entrainment/detrainment and mixing mechanisms in free turbulent shear flows
including clouds.
The numerical technique involves the simulation of the TDP by solving the 3D
Boussinesq approximations of the Navier–Stokes equations of mass, momentum,
and energy balance [9]. The equations solved are
Continuity : ∇ · u = 0 (1)
∂u −1
Momentum : + (u · ∇)u = ∇ P + υ∇ 2 u + βT g (2)
∂t ρ0
∂(T ) J (t)
Energy : + (u · ∇)T = κ∇ 2 (T ) + H (x, t) (3)
∂t ρ0 C P
∂ω
Vorticity : + (u · ∇ω) − (ω · ∇)u − υ∇ 2 ω = βg × ∇T (4)
∂t
w0 d0
Reynolds number, Re =
υ
υ
Prandtl number, Pr =
κ
A DNS Study of Bulk Flow Characteristics of a Transient … 389
J d0
Heat release number, G =
ρC P w0 T0
The third parameter G plays a key role in the flow development as discussed by
Bhat and Narasimha [10]. The numerical technique employed is discussed by Pras-
anth [11]. A finite-volume technique with staggered grid arrangement has been used.
In the first sub-step, the momentum equation is solved for the velocity components
without taking the pressure term into account. This velocity field is non-divergent.
In the subsequent sub-step, the pressure Poisson equation is obtained by taking the
divergence of what ought to have been a non-divergent velocity field. The pressure
so obtained (using the HYPRE library, [12]) is used to get the final non-divergent
velocity field. This corrected velocity field is then used as the initial condition for the
next time step. This two-step procedure is continued till the simulation is completed.
The 360 TF supercomputer (Ananta) at CSIR-4PI, Bangalore, has been used for
the simulations. The total number of grids in the computational domain is ≈1.6
billion. The number of processors used is 3375, which required a total wall time of
≈16 days. The grid shown in Fig. 1b is a nonuniform Cartesian grid with higher
spatial resolution in the region where the TDP evolves. The time span of the flow is
90 flow units where each flow unit (FU) is defined as d0 /w0 .
390 S. Rao et al.
The flow schematic along with the coordinate system used in the simulation is shown
in Fig. 1, and Table 1 gives the details of the simulation.
In a TDP, the off-source heat addition which mimics the latent heat release due to
condensation of water vapor to liquid water in a cloud is simulated by dynamically
matching the nondimensional heat release number G (defined in Sect. 2). For the
present simulation, off-source heat is added over 10 ≤ z ≤ 15 and we call this the
heat injection zone (HIZ hereafter). The HIZ is equally divided into five subzones,
and varying amounts of heat are injected into each subzone (see Fig. 2). The amount
of heat injected into the HIZ may be specified by the heat release number G (x, t),
where G can be a function of space or time or both space and time (Fig. 2). In shallow
cumulus clouds, G is ~0.1–0.5 [13], where in the present simulation G varies in the
range of 0.04 to a maximum value of 0.35. The heating profile chosen to get a cloud
with a tall tower and cauliflower head is shown in Fig. 2.
As this is for the first time that the results on a fully resolved DNS study on a TDP are
being presented, it is not possible to compare it with any other study. Our objective
here is to demonstrate the effect of off-source heat addition on bulk parameters like
axial velocity (w), temperature (T ), equivalent diameter (d||eq), and mass flux
Table 1 Simulation details, where Pr is Prandtl number and Re is Reynolds number (see Sect. 2)
Simulation Domain size (x, y, z) Grid size Time step Pr Re
TDP-1 (50, 50, 40) (1007, 1007, 1577) 0.003 t ≤ 63 FU 1 2000
0.001 t > 63 FU
Fig. 2 History of the heating profile used in the present simulation. Time is nondimensionalized
using w0 and d0 . Heating starts at 43 FU and ends at 63 FU
A DNS Study of Bulk Flow Characteristics of a Transient … 391
(Q) (the latter two to be defined below). Ensemble averaging is required to obtain
the average quantities in a transient flow. This demands that the simulations in the
present study be repeated several times with different initial perturbations. However,
results are available only from a single realization, but azimuthal averaging helps in
obtaining approximate mean values of flow parameters.
The characteristic feature of a turbulent shear flow is the stochastic nature of the
vorticity field. Hence, the total vorticity modulus |ω|, defined in Fig. 3, is an appro-
priate variable for visualizing the evolution of TDP, and for determining boundaries
between turbulent flow and nonturbulent flow. The evolution of |ω| in the TDP
through t = 63, 68, and 72 FU is shown in Fig. 3a–c. From the figures, it can be seen
that the flow has an advancing cap or flattened thermal (cauliflower head) followed
by a trailing plume (stem). The figures also show the flow width ceases to grow in
any significant way with height, except in the plume head which is contrasting with a
starting plume where the flow width increases with height [14]. Comparing Fig. 3a,
c shows that even though maximum heat is injected between t = 60–63 FU, it takes
at least a few FUs for the |ω| to feel the effect of off-source heat addition [7]. For
analyzing the bulk parameters mentioned in Sect. 3.2, TDP flow at t = 72, which
(d)
1
Fig. 3 a–c Evolution of |ω| (≡ ω2x + ω2y + ωz2 , the total vorticity modulus) in the TDP at
2
different instants of time in the axial section (x-z plane, y = 0). The red strip in 10 ≤ z ≤ 15
represents the heat injection zone. |ω| is nondimensionalized by w0 and d0 . Figure 3d shows the
variation of |ω| at t = 72FU with radial distance for selected heights z
392 S. Rao et al.
has developed to a tall tower (stem) with a cauliflower head (cap), is chosen as an
appropriate candidate for further analysis.
Figure 3d is plotted to demonstrate the effect of heating on vorticity distribution
for four selected values for z. It can be seen by comparing z = 9.018 and z = 21.026
that |ω| goes up by an order of magnitude in the latter case. The reason for such a
dramatic increase in |ω| may be attributed to the baroclinic torque βg × ∇T [9],
which is a source for vorticity generation.
Instantaneous axial velocity distribution across the diametral sections (x-y plane) of
the TDP at t = 72 FU is shown in Fig. 4 on the same scale as that shown in Fig. 4d.
Addition of off-source heating increases the axial velocity by ~2–4 times: this can be
seen by comparing the plume before (Fig. 4a) and plume after (Fig. 4c, d) heating.
Another striking feature of the velocity field in Figs. 4c, d (more pronounced in the
latter) is the presence of sinking velocities within and beyond the edges of plume
head.
Radial distributions of azimuthally averaged axial velocity (w) are plotted in
Fig. 5a. It is seen that w increases with height indicating that flow is accelerating. It
shows that centerline velocity increases continually, in contrast to the decay observed
in the classical plume (wc ∼ z −1/3 ).
3.5 Temperature (T )
Radial distributions of azimuthally averaged temperature are plotted in Fig. 5b. It can
be seen from Fig. 5b that due to off-source heat addition, T decay in TDP above z > 10
is much less than in the classical plume (T c ∼ z −5/3 ). Buoyancy flux is a conserved
variable in classical plume. However, in a TDP buoyancy flux increases (not shown
here) due to the increase in ∇T and w. The dramatic increase in the vorticity (Fig. 4d)
is the result of the increase in ∇T and the accompanying baroclinic torque (Eq. 4),
as found in the temporal simulations of a cloud flow in a periodic box by Basu and
Narasimha [9].
It is the average diameter of the plume at a given height. It is defined as the diameter
of the circular area of same magnitude as covered by the |ω|thr edge or boundary of
the TDP at that height,
A DNS Study of Bulk Flow Characteristics of a Transient … 393
Fig. 4 Contour plot showing diametral section (x-y plane) of instantaneous vertical velocity, w at
different
√ heights at t = 72 FU. w is nondimensionalized using exit velocity at hot patch w0 =
gβd0 T0 . Superimposed pink (|ω| = 0.25) and black (|ω| = 0.5) curves represent the magnitude
of total vorticity |ω| on the boundary. The inner boundary (black curve) separates turbulent flow
from nonturbulent flow (T\NT interface), based on the data shown in Fig. 3d. Outer boundary
(pink curve) separates rotational and irrotational flow. |ω| is nondimensionalized using w0 /d0 . a the
w distribution below HIZ, b middle of HIZ, c above the HIZ in the cylindrical stem, and d above
the HIZ in the plume head part
|ω| ≥|ω|
i, j thr
2
deq =√ xi, j yi, j , (3.2)
π i, j=1,1
Fig. 5 a Azimuthally averaged distributions of vertical velocity w (a) and azimuthal averaged
√ Curves are plotted for different heights at t = 72 FU. Velocity is nondimension-
temperature T (b).
alized by w0 = gβd0 T0 , temperature by T0 and r by d0 . Black, yellow, blue, and orange lines
represent the w distribution, respectively, below HIZ, inside HIZ, inside the cylindrical stem above
HIZ, and plume head part above HIZ
(a) (b)
Regime -IV Regime -IV
Regime -III
Regime -III
Fig. 6 Axial variation of a turbulent mass flux (Q) and b equivalent (turbulent) diameter (deq )
both at t = 72 FU. Q and deq are nondimensionalized by w0 and d0 , respectively. |ω| = 0.5 is used
as the threshold to define the T\NT interface
rate of change of mass flux with height. The axial variation of the mass flux is often
obtained from an arbitrary choice of radial location on the mean velocity distribution
but is here defined as
|ω|i, j ≥|ω|thr
Q= wi, j xi, j yi, j , where |ω|thr = 0.5.
i, j=1,1
It must be noted that this mass flux is obtained by integrating the axial velocity
in the horizontal plane (x-y) from plume axis (r = 0) to r = |ω|thr at the relevant
azimuthal angle (Fig. 4). A more detailed justification of such thresholds based on
|ω| will be found in Prasanth et al. [15]. The Q value so reported here adds up
to the true turbulent mass flux. While computing Q no assumption is made on the
axial velocity profile or on azimuthal symmetry. Figure 6a shows the variation of
Q with axial distance and is broadly consistent with the results of Diwan et al. [7].
In regime-I, the flow is still a classical plume and Q increases linearly with height.
Regime-II is in the HIZ where the increase in velocity over that in regime-II is only
slightly higher than in regime-I (yellow curve in Fig. 5a), however, the flow width
is nearly constant, and hence the Q hardly increases. Q goes up by a decade in
regime-III and rapidly falls to zero in regime-IV. The reason for the former is that the
flow begins to experience the effect of heat, as a result w goes up, resulting in more
entrainment. In the later stages, even though w goes up there are sinking velocities
near the edges which contribute to negative mass flux (Figs. 4d, 5a), resulting in
appreciable detrainment. This behavior is in striking contrast to that in a classical
self-preserving plume, where the mass flux increases with height at a constant rate.
4 Conclusions
This paper describes a fully resolved study of a transient diabatic plume (TDP).
We have simulated a first-order minimalistic model of a cumulus cloud flow by
incorporating a dynamically matched off-source heat addition. The results show that
the addition of off-source heating makes the evolution of a transient diabatic plume
strikingly different from that of a classical plume. It causes the flow to accelerate
and enhances fluctuating vorticity above HIZ by an order of magnitude through the
action of the baroclinic torque. We have presented here the true turbulent mass flux
in TDP at one time instant. The results on the bulk parameters discussed above are
broadly consistent with the previous studies on diabatic jets and plumes [7, 11].
A more detailed analysis of DNS results for the TDP and comparison with the
results of experiments on TDP and LES results on clouds might help in understanding
some of the more important factors that govern entrainment/detrainment and mixing
in clouds.
396 S. Rao et al.
References
Subhankar Sen
1 Introduction
S. Sen (B)
Department of Mechanical Engineering, IIT (ISM) Dhanbad, Dhanbad, India
e-mail: subhankars@gmail.com
an empirical formula yielding the lowest value of Re or reduced speed for galloping
to occur. Sen and Mittal [4] also listed the total number of kinks (in characteristic
curves) or transition regimes and provided possible explanations for the absence of
asymmetric wake modes. Jaiman et al. [5] showed that galloping does not occur in
a freely vibrating square cylinder once the sharp corners are smoothed out. He et al.
[6] numerically explored the single and two-degrees-of-freedom VIV of a square
cylinder and also studied the effects of mass ratio in brief.
In the earlier studies by us [2–5], the Reynolds number was coupled to the reduced
speed, U* (defined in Sect. 2) by an empirical formula. This imposed dependency
resulted in reduction in the number of controlling parameters by one. In the current
set of computations, the reduced speed is varied independent of Re. A fundamental
study concerning free vibrations of a rigid square cylinder at low Reynolds number
is still unavailable in the literature. This forms the motivation of the current work.
Direct numerical simulations of the governing differential equations are performed
at Re = 250 assuming the flow to remain two-dimensional since vibrations delay all
the transitions [7].
2 Methodology
The governing differential equations for the coupled fluid-rigid body system include
the continuity equation as well as those of motion of fluid and cylinder. The equations
of motion are derived from Newton’s second law; these are the Navier–Stokes equa-
tions for fluid medium and a second-order ODE for the solid medium, respectively. In
strong form, the governing momentum and continuity equations for incompressible
fluid flow are expressed as
∂u
ρ( + u.∇u − f ) = ∇.σ (1)
∂t
∇.u = 0. (2)
In vector Eqs. 1 and 2, t, ρ, u (= u, v), f, and σ , respectively, stand for the time,
density of the fluid, velocity vector for flow, body force vector per unit volume, and
stress tensor. The contribution from body force is not considered in the present work.
In context of the present problem, the flow domain in which the oscillator resides is
of rectangular shape. Free-stream inlet (u = U, v = 0), stress-free exit and slip lateral
boundaries define the boundary conditions for the rectangular domain. At the fluid–
solid interface, no-slip on velocity is considered. A solenoidal or divergence-free
velocity field is used as initial condition for the above system of PDEs.
Transverse-Only Vibrations of a Rigid Square Cylinder 399
For undamped translations of the square oscillator across the flow, Newton’s
second law of motion reduces to
d2 Y
+ (2π FN )2 Y = Cl /2m ∗ (3)
dt 2
where Y is the displacement of the oscillator measured from its initial (0, 0) location.
Here, F N = (f n D)/U is the reduced or normalized natural frequency of the oscillator,
C l is the instantaneous lift force, and m* denotes the mass ratio of the oscillator. Mass
ratio or relative density of the oscillator signifies the ratio of the mass of the oscillator
and mass of the displaced fluid per unit length. The dimensional natural frequency
is denoted by the symbol f n . The reciprocal of reduced natural frequency is known
as the reduced speed. For the ODE corresponding to Eq. 3, the initial condition is
Y = 0 and dY dt
= 0 at t = 0. However, this initial condition is used only for the lowest
value of reduced speed, i.e., U* = 3.5. For other U*, the solutions for Y and its first
derivative at the nearest U* are considered as the initial conditions.
A stabilized space-time finite-element formulation [8, 9] accommodating equal
order bilinear interpolation for velocity and pressure is used for discretization of the
conservation equations of mass and linear momentum in two dimensions. A rigid
square cylinder of edge length D resides in a rectangular computational domain. In the
absence of viscous damping, the cylinder executes free transverse-only vibrations.
The mass ratio of the cylinder is 10. A blockage of 5% is used for all the calculations.
For a Reynolds number of 250 based on the edge length of the cylinder, the reduced
speed is varied from 3.5 to 9.5. Motion of the cylinder is depicted relative to a fixed
or inertial frame of reference, the origin (0, 0) of which coincides with the center of
the cylinder when it is stationary.
A multi-block, non-uniform, and structured finite-element mesh containing
24,149 nodes and 23,780 bilinear quadrilateral elements has been used for the com-
putations. The mesh is composed of five component blocks—a central block and four
rectangular blocks surrounding the central block. The central block accommodates
the cylinder. The mesh is reconstructed at each time step. During the reconstruction,
the central block remains undeformed (but it moves), while the other blocks deform
due to movement of the central block along with the cylinder. A detailed discussion
of the finite-element mesh is available in Sen and Mittal [2].
Table 1 establishes mesh insensitivity of the computed results. The free transverse
vibrations of a square cylinder of m* = 10 at zero incidence are computed on two
Table 1 Flow past a freely vibrating (transverse-only) rigid square cylinder m* = 10 at Re = 250
and U* = 15: details of meshes M1 and M2
Mesh Nodes Elements Y max /D Mean C d C lrms
M1 24,149 23,780 0.6772 2.0686 0.8627
M2 47,800 47,280 0.6697 2.0703 0.8645
Also listed are the values of maximum response and fluid forces obtained from these meshes
400 S. Sen
Table 2 Flow past a freely vibrating rigid square cylinder of m* = 3 at Re = 100 and U* = 5:
comparison of the predicted maximum transverse response with those reported by Zhao et al. [1]
Studies Y max /D
Zhao et al. [1] 0.3500
Present 0.3231
For both cases, the blockage equals 0.025 and the cylinder executes simultaneous in-line and
transverse translations
meshes M1 and M2 at Re = 250 and a much higher U* of 15. At such high U*,
the cylinder executes galloping motion (not discussed in this paper). As apparent
from this table, the resolution of mesh M2 is about two times the resolution of
M1. Columns four through six of Table 1 ascertain that the characteristic vibration
(maximum cross-stream displacement, Y max ) and flow quantities (mean drag and
r.m.s. lift) do not exhibit any noteworthy deviation when obtained from meshes M1
and M2. Mesh M1, therefore, is considered for all computations in this work.
For two-degrees-of-freedom translations of an m* = 3 square cylinder at Re =
100 and U* = 5, Table 2 compares the predicted maximum transverse response
with the one obtained by Zhao et al. [1] via finite-element computations. For both
cases, a blockage of 0.025 is considered. The closeness of the peak response values
establishes the accuracy of the predicted results.
The variation of Y max , pressure recovery at cylinder base, and drag on the cylinder
are presented as a function of reduced speed. Some flow features are also discussed
for a representative U* of 4.6 belonging to the initial branch of response. For a pair
of fixed square cylinders at incidence and in tandem arrangement, [10] studied the
flow topology.
marking the transition from initial to lower branch of response. The response attains
the maximum value of 0.12D at U* = 5 in the initial branch.
The relationship between pressure recovery at cylinder base and reduced speed is
shown in Fig. 2. Pressure recovery is the difference between the time-mean forward
stagnation pressure, C p0 , and time-mean base pressure, C pb . A lower value of (C p0 -
C pb ) indicates higher recovery of pressure and vice versa. A higher recovery implies
delayed separation of boundary layer, narrow wake, and low drag. The minimum
pressure recovery occurs at U* = 5.1 marking the onset of lock-in.
Figure 4 plots the drag-lift Lissajous diagram at U* = 4.6. The presence of multiple
curves of similar profiles ensures that the flow is quasi-periodic at this reduced speed.
The phase plot is asymmetric about the zero-lift line. This implies that mean lift is
Transverse-Only Vibrations of a Rigid Square Cylinder 403
Fig. 6 Instantaneous vorticity field at U* = 4.6 showing the one-sided S 1 and S 2 vortices. The
cross symbol (x) marks the origin of the inertial (fixed) frame of reference
4 Conclusions
The free transverse vibration of a rigid square cylinder of mass ratio 10 is investigated
numerically at Re = 250. The motion is purely VIV and devoid of galloping for the
U* range of 3.5–9.5. For the first time, existence of asymmetric wake modes is noted
for a square cylinder executing 1:1 VIV. A new wake mode corresponding to one-
sided shedding is identified. This newly identified mode is denoted by the symbol
S 1 S 2 where two vortices of same sign are shed from each oscillation cycle.
References
8. Tezduyar TE, Behr M, Liou J (1992) A new strategy for finite element computations involving
moving boundaries and interfaces-the DSD/ST procedure: I. The concept and the preliminary
numerical tests. Comput Methods Appl Mech Eng 94:339–351
9. Tezduyar TE, Behr M, Mittal S, Liou J (1992) A new strategy for finite element computations
involving moving boundaries and interfaces-the DSD/ST procedure: II. Computation of free-
surface flows, two-liquid flows, and flows with drifting cylinders. Comput Methods Appl Mech
Eng 94:353–371
10. Dash SM, Lee T-S (2012) Impulsively started flow topology around tandem arrangement of
two square cylinder at incidence. Int J Mod Phys Conf Ser 19:100–108
Steady Flow Past Two Square Cylinders
in Tandem
1 Introduction
Despite being simple in geometry, due to its vast applicability and abundance of
flow physics, flow over a circular cylinder has been the most explored problem
in the field of bluff body flow. A detailed review for the same can be found in
Zadravkovich [1], Sumner [2], and many more. The knowledge of aerodynamic
forces on the structure and subsequent changes in flow field is very important before
designing any structure experiencing fluid–structure interaction. These structures
may include from tube bundles of heat exchangers, wires in the transmission line,
floating platforms in oversea, pipelines near seabed, chimney stacks skyscrapers, and
many more. Presence of another bluff structure in the vicinity makes the analysis even
more challenging. Two or more than two cylinders can be arranged in three possible
arrangements, namely, tandem, side-by-side, and staggered [1]. Depending upon the
closeness of two circular cylinders and interaction of vortices of one cylinder with
the other, Zadravkovich [3] defined three different regimes of flow. (i) The proximity
interference regime (cylinders are adjacent to each other), (ii) wake interference
regime (rear cylinder is in the wake of the front cylinder), and (iii) no-interference
regime (cylinders are far from each other and are not affected by the presence of
the other). Patil et al. [4] conducted numerical experiments on two circular cylinders
placed in tandem arrangement. They studied the effect of varying shear parameter,
spacing ratio, and Reynolds number on flow separation and aerodynamic forces.
Singha and Sinhamahapatra [5] reported steady and unsteady flow past two circular
cylinders for varying Re and S/D. They found the flow to be fully steady irrespective
of the spacing ratio for Re = 40. However, as the Re is increased to a value of 70, the
flow progressively switches from steady to unsteady as the spacing ratio advances
from 0.2 to 4.
As compared to circular cylinder, the square counterpart is relatively less explored.
Flow past a pair of identical square cylinders at low Re has been studied numerically
by Sohankar [6]. He investigated the effect of spacing ratio for selected values of Re
and found a critical spacing ratio beyond which the fluid forces increased consider-
ably. He also presented the effect of Reynolds number for a fixed spacing of S/D = 4.
Based on the study, Sohankar [6] found three major regimes, namely, single-slender
body regime, reattach regime, and co-shedding regime. Later, Shyam and Chhabra
[7] conducted numerical experiments to depict the effect of Prandtl number on in-
line cylinders of square cross section and immersed in power-law fluid for Re range
1–40. S/D is varied from 2 to 6. Ehsan et al. [8] investigated steady and unsteady
flow over a pair of square cylinders in in-line arrangements for Re range 1–200 and
for shear parameter varying from 0.5 to 1.8. They found that the onset of separation
from the leading edge occurs early for shear-thinning fluid and is delayed for shear
thickening fluid. Most of the numerical and experimental studies on flow around
pair of square cylinders are confined to high Re flow. The flow regimes in all the
reported literature are defined for unsteady flow based on the patterns of shedding
of vortices from the upstream and downstream cylinders. Therefore, the obvious
questions which come to the mind are: Can we identify different regimes of flow
in the absence of vortex shedding from the cylinders, i.e., when the flow is steady?
What will be the criterion for defining the regimes? The lack of literature for low Re
flow past two square cylinders in tandem arrangement in steady flow regime and the
above-unanswered questions motivated us to explore the problem further. Therefore,
in the present investigation, our objective is to identify the flow structure formed by
varying S/D of two identical square cylinders at a Re = 40. An attempt to understand
the effect of spacing on the streamlines variation, in-line velocity distribution on the
wake centerline in the gap region and the corresponding effect on wake structure
is also made. Structure of the remaining paper is as follows: Governing equations
are discussed in Sect. 2. Section 3 describes the problem setup and mesh informa-
tion. Validation study and grid independence test are discussed in Sect. 4. Results
are presented in Sect. 5. The first Sect. 5.1 in Sect. 5 describe overall flow. Critical
Steady Flow Past Two Square Cylinders in Tandem 409
2 Governing Equations
The motion of incompressible fluid in steady flow is governed by Eqs. 1 and 2 listed
below:
ρ(u.∇u − f ) − ∇σ = 0 (1)
∇u = 0 (2)
Here, u vector is the velocity of the fluid, f is body force, and σ represents the
Cauchy stress tensor. Since the flow is advection dominated, therefore the contribu-
tion of body force is not considered in the computations. The stress tensor consists
of isotropic and deviatoric parts:
σ = − p I + T, T = 2με(u), ε(u) = 1/2 (∇u) + (∇u)T (3)
Here, p is the pressure, I is the identity matrix, μ is the dynamic viscosity of the
medium, and ε is the strain rate.
Figure 1 shows the schematic of the problem. The rectangular box represents the flow
domain. Two identical square cylinders whose centers are separated by a spacing S are
fixed in the domain. The inlet and exit of the domain are provided with free stream and
stress-free boundary conditions, respectively. Cylinder surfaces are no-slip surfaces.
The sidewalls are free-slip walls. A collocated, multi-block, structured, nonuniform
finite-element mesh consisting of 94,391 nodes and 93,440 elements for S/D = 2 is
shown in Fig. 2a. The mesh consists of total ten blocks. This arrangement allows to
control the resolution of the mesh locally. The cylinders are placed in two central
mesh blocks. The enlarged view of central block consisting of upstream cylinder is
shown in Fig. 2b. The other blocks adjacent to the central blocks are stitched with the
central blocks. The origin of Cartesian coordinate system is located at the geometric
center of upstream cylinder. The distance of inlet and exit walls from the origin of
the coordinate system is 30D and 95D, respectively. Both the upper and lower walls
are placed at a distance of 25D from the origin, thus offering a blockage (B, ratio of
cross-stream dimension of the cylinder to the total cross-stream width of the domain)
of 2%.
410 D. Kumar et al.
Fig. 1 Sketch of the problem statement and associated boundary conditions for steady flow past a
pair of identical square cylinder in tandem arrangement
Fig. 2 a The finite-element mesh consisting of two identical square cylinders for S/D = 2.
b Enlarged view of the central mesh block consisting the upstream cylinder. The two orthogonal
arrows represent the coordinate system
The results reported by Sharman et al. [9] for a pair of circular cylinders and Bao
et al. [10] for two square cylinders in tandem are compared with the results of present
computation to establish the accuracy of the numerical model used (see Table 1).
Unsteady computations are carried out at Re = 100 and for a spacing ratio of S/D =
4 and 5 two identical circular and square cylinders, respectively. Table 2 lists the grid
independence test for the present computations. Two grids M1 and M2 have been
tested and based on the results obtained, mesh M1 is chosen for all the computations
in the present work.
Steady Flow Past Two Square Cylinders in Tandem 411
Table 1 Unsteady flow past a pair of identical circular and square cylinder at Re = 100. The
comparison of aerodynamic coefficients as reported by Sharman et al. [9] and Bao et al. [10] with
the present computations
Geometry Study S/D Blockage Upstream cylinder Downstream
cylinder
C D AVG C L RMS C D AVG C L RMS
Circle Sharman et al. 4 0.02 1.2756 0.3028 0.7033 0.9869
[9]
Circle Present 4 0.02 1.2602 0.3019 0.7007 0.9700
Square Bao et al. [10] 5 0.02 1.4260 0.2890 1.0990 1.2110
Square Present 5 0.02 1.3910 0.2901 1.0956 1.2060
Table 2 Steady flow past a pair of identical square cylinder at Re = 40 and for B = 0.02. Grid
independence test for S/D = 6
Mesh Nodes Elements Upstream cylinder Downstream cylinder
C Dp C Dv C Dp C Dv
M1 98,577 97,600 1.32156 0.27975 0.22733 0.17848
M2 200,063 198,620 1.33518 0.26598 0.23106 0.17575
5 Results
The results of steady flow around two bluff obstacles of square cross section are pre-
sented for a fixed Re value of 40 and for S/D ranging between 2 and 10. The equations
governing the flow are discretized using stabilized finite-element formation.
Flow patterns in terms of streamline contours for spacing ratio range considered are
shown in Fig. 3. As apparent from Fig. 2a, b, for low S/D (= 2 and 4), wake of
the upstream cylinder reattaches on the downstream cylinder. Initially, the classical
Fig. 3 Flow past a pair of square cylinders in tandem at Re = 40: streamline plots for S/D = a 2,
b 4, c 6, and d 8
412 D. Kumar et al.
To understand the changes in the flow structure near the critical regime (i.e., when a
closed wake is observed for upstream cylinder in the gap region), computations are
carried out at intermediate spacing ratio values between S/D = 4 and S/D = 6. The
variation is shown with respect to streamline contours in Fig. 4. As apparent from
Fig. 4a, at S/D = 5.5, the wake of upstream cylinder shows a tendency to detach from
downstream cylinder, thereby making a very narrow structure near the frontal edge
of downstream cylinder. As the spacing ration approaches six (i.e., the critical value),
a completely closed wake structure of the upstream cylinder is observed in the gap
region for the first time. Interestingly, another weak yet observable counterrotating
recirculation zone above and below the wake centerline also appears adjacent to the
wake of upstream cylinder and frontal edge of downstream cylinder. This is caused
by the flow reversal in this zone. To understand the flow direction in the gap region,
the variation of in-line component of the velocity is plotted with in-line spacing for
S/D = 6. Figure 5 clearly depicts the reversal of flow as the in-line component of
flow velocity changes its sign from positive to negative near the frontal edge. The
first negative u-velocity region represents the wake of upstream cylinder.
The effect of spacing ratio on the flow structure is presented via variation of wake
length in Fig. 6a. While the upstream cylinder shows a monotonic decrease in the
wake length after its evolution at S/D = 6, the wake length of the downstream cylinder
initially decreases up to S/D = 6 and then increases gradually. Figure 6b presents
the distribution of surface vorticity on the downstream cylinder. The intersection of
vorticity curve with zero line represents either the separation or reattachment points.
Fig. 4 Streamline plots for S/D = a 5.5, b 6, and c enlarged view of downstream cylinder. A weak
recirculation zone near can be clearly observed near the frontal edge of downstream cylinder
Steady Flow Past Two Square Cylinders in Tandem 413
Fig. 5 Steady flow past two identical square cylinders at Re = 40 and S/D = 6. Variation of in-line
velocity component (u) in the gap region. The inset shows enlarged view of the variation of u-x
near the frontal edge of downstream cylinder
Fig. 6 The variation of a length of the wake of both the cylinders and b surface vorticity distribution
around the downstream cylinder as a function of S/D. The angular position 0° and 180° represents
the forward stagnation point and base points (moving in counterclockwise direction) of the square
cylinder, respectively
It is worth noting that for S/D = 2 through 6, when the wake of the upstream cylinder
has interaction with its downstream counterpart, the reattachment point is not fixed
and it gradually move toward forward stagnation point (on the leading edge).
414 D. Kumar et al.
6 Conclusions
References
1. Zdravkovich MM (1977) Review of flow interference between two circular cylinders in various
arrangements. ASME J Fluids Eng 99:618–633
2. Sumner D (2010) Two circular cylinders in cross-flow: a review. J Fluids Struct 26(6):849–899
3. Zdravkovich MM (1985) Flow induced oscillations of two interfering circular cylinders. J
Sound Vib 101(4):511–521
4. Patil RC, Bharti RP, Chhabra RP (2008) Steady flow of power law fluids over a pair of cylinders
in tandem arrangement. Ind Eng Chem Res 47(5):1660–1683
5. Singha S, Sinhamahapatra KP (2010) High-resolution numerical simulation of low Reynolds
number incompressible flow about two cylinders in tandem. J Fluids Eng 132(1):011101
6. Sohankar A (2012) A numerical investigation of the flow over a pair of identical square cylinders
in a tandem arrangement. Int J Numer Methods Fluids 70(10):1244–1257
7. Shyam R, Chhabra RP (2013) Effect of Prandtl number on heat transfer from tandem square
cylinders immersed in power-law fluids in the low Reynolds number regime. Int J Heat Mass
Transf 57(2):742–755
8. Ehsan I, Mohammad S, Reza NM, Ali J, Tashnizi ES (2013) Power-law fluid flow passing two
square cylinders in tandem arrangement. J Fluids Eng 135(6):061101
9. Sharman B, Lien FS, Davidson L, Norberg C (2005) Numerical predictions of low Reynolds
number flows over two tandem circular cylinders. Int J Numer Methods Fluids 47:423–447.
10. Bao Y, Wu Q, Zhou D (2012) Numerical investigation of flow around an inline square cylinder
array with different spacing ratios. Comput Fluids 55:118–131
A Robust and Accurate
Convective-Pressure-Split Approximate
Riemann Solver for Computation of
Compressible High Speed Flows
1 Introduction
schemes. While the FVS schemes work by decomposing the total Euler flux vector at
an interface into an upwind and downwind contribution based on local wavespeeds,
the FDS schemes seek the solution of a Riemann problem to estimate the numerical
flux at an interface. The most cost-effective implementation of the FDS schemes is the
approximate Riemann solvers and the most popular of them include the Roe scheme
[1], the HLL scheme [2], the HLLC scheme [3] etc. Although the superior accuracy
has rendered these schemes more popular than the FVS schemes over the years, they
still lack the inherent robustness of the FVS schemes against the phenomenon of
numerical shock instability [4].
A particular scheme that represents a combination of the philosophy of the FVS
and the FDS schemes, yet remains distinct from each of them, is the AUSM scheme
of Liou et al. [5]. The AUSM scheme is notable for pioneering the philosophy of
Convective-Pressure Splitting (CPS) by arguing that since the Euler equations com-
prise of separable convective and pressure systems with distinct physics of prop-
agation, it is meaningful to decompose it accordingly and discretize these compo-
nents independently. The AUSM scheme rivaled most of the approximate Riemann
solvers in its accuracy while closely retaining the robustness of the FVS schemes in
shock-capturing. The philosophy of CPS that was put forward by this scheme had
subsequently inspired several other schemes [6–8]. Most of these schemes used a
one-sided differencing for the convective component and a Mach number or velocity-
based polynomial approximations for the pressure component.
Recently, Mandal et al. [9] proposed an interesting CPS based approximate
Riemann solver called the HLL-CPS scheme. In this scheme, the Euler flux is first
decomposed into its respective convective and pressure systems based on either
AUSM-type or Zha-Bilgen-type [6] splitting. The resulting convective system was
governed by a single linear wave and the pressure system was governed by two non-
linear waves and a stationary contact wave. While the convective system is upwinded
based on locally averaged fluid velocity, the pressure system is subjected to HLL-type
discretization inspired by the AUFS scheme of Sun et al. [10]. The contact ability
was introduced explicitly by replacing the density jumps appearing in the numerical
dissipation of the pressure system to pressure jumps using an isentropic condition.
The individually discretized convective and pressure systems are then coupled to
each other using the slowest and fastest wavespeeds that occur as the solution of
the local Riemann problem. The HLL-CPS scheme was shown to be as accurate
as the HLLC scheme at least on several inviscid problems. It also inherited several
appealing features like positivity, entropy satisfaction without explicit entropy fixes
and flux differentiability from the HLL scheme. Further, all versions of the scheme
were shown to avoid numerical shock instability on selected problems. The HLL-
CPS formulation has been shown to work effectively even on the Toro-Vazquez-type
flux splitting [11] and has been easily extended to its genuinely multidimensional
version [12], which clearly demonstrates the versatility of the method. However,
recent studies reveal that the HLL-CPS scheme has two major drawbacks. Firstly,
the numerical dissipation tuning employed to resolve stationary contacts adversely
affected the capability of the scheme to resolve shear waves [13]. Secondly, our
A Robust and Accurate Convective-Pressure-Split Approximate Riemann Solver … 417
experience shows that the HLL-CPS may succumb to numerical shock instability on
certain problems.
To alleviate these shortcomings, in this work, we propose a new low diffusion
approximate Riemann solver based on the CPS philosophy. Although we reuse the
framework made available by the HLL-CPS scheme, the proposed scheme differs
from it specifically in the discretization of the pressure system. In particular, we
show that full accuracy on the linearly degenerate wavefields can be recovered by
simply introducing an HLLC-type anti-diffusive term into the formulation. Further,
we demonstrate that the proposed scheme could be made robust against numerical
shock instability by simply controlling this anti-diffusive term in the vicinity of a
numerical shock front.
2 Formulation
∂U ∂F(U)
+ =0 (1)
∂t ∂x
where U and F(U) are the vector of local conserved variables and grid normal
fluxes at any interfaces given respectively as U = [ρ, ρu, ρv, ρE]T and F(U) =
T
ρu, ρu2 + p, ρuv, (ρE + p)u . In these expressions, ρ, u, v, p and E stands,
respectively, for density, normal velocity, tangential velocity, pressure, and specific
total energy. The system of equations are closed through the equation of state. In the
above expression the grid tangent fluxes are neglected in a finite volume discretiza-
tion because they do not contribute to the flux across an interface. A semi discretized
form of Eq. (1) that uses a conventional Finite Volume based two-state approximate
Riemann solver on a two-dimensional quadrilateral mesh with a cell element i of
area |Ωi | can be written as,
1
4
d Ui
=− [FRiemann (UL , UR )]sk (2)
dt |Ω|i
k=1
where UL , UR indicates the initial conditions of a local Riemann problem across kth
interface of cell i and FRiemann is the Riemann flux operator at any interface with face
length sk . In this work we use the Zha-Bilgen type flux splitting strategy [6] to split
the total Euler flux vector into its convective (FC ) and pressure (FP ) systems as,
F = FC + FP
418 S. Simon and J. C. Mandal
where, ⎛ ⎞ ⎛⎞
ρ 0
⎜ρu⎟ ⎜p⎟
FC = u ⎜ ⎟
⎝ρv ⎠ FP = ⎜
⎝0⎠
⎟ (3)
E pu
For the proposed scheme, the convective part is discretized using a simple local
average velocity based upwinding similar to [9].
⎞ ⎛
ρ
⎜ρu⎟
FC = Mk ⎜ ⎟
⎝ρv ⎠ ak (4)
E k
L if ū ≥ 0
k= (5)
R if ū < 0
uL + uR
where the local average velocity ū =
2
⎧
⎪ ū
⎨ if ū ≥ 0
Mk = ū − SL (6)
⎪
⎩
ū
if ū < 0
ū − SR
and
uL − SL if ū ≥ 0
ak = (7)
uR − SR if ū < 0
Now, instead of tuning the term (UR − UL ) to recover accuracy on linearly degenerate
waves like contact waves and shear waves as advocated in the HLL-CPS scheme, we
simply add an HLLC-type anti-diffusive term to the pressure discretization. Hence,
we write,
SR FLP − SL FR
P
+ SR SL (UR − UL )
FP = + ωAHLLC (9)
SR − SL
A Robust and Accurate Convective-Pressure-Split Approximate Riemann Solver … 419
Fig. 1 Stencils for evaluation of the anti-diffusion control parameter ω. a On horizontal interfaces.
b On vertical interfaces
where, SM is the estimated speed of the contact and shear wave and can be obtained
as suggested in [14]. Since this anti-diffusive term is responsible for triggering shock
instability [15], a pressure-ratio based control parameter ω is introduced in order to
control this term in the vicinity of strong normal shocks. The control parameter ω is
given as,
The term fk denotes a pressure-ratio based function that uses a predefined stencil
around a particular interface as shown in Fig. 1a. At any interface k, fk is defined as,
5
pR pL
fk = min , (12)
pL pR k
A M = 0.1 laminar flow over a flat plate is used as a test case to demonstrate the
capability of the proposed HLLC-CPS-Z scheme to satisfactorily resolve a shear
dominated viscous flow. The domain is divided into 31 × 33 Cartesian cells with
fifteen cells retained in the boundary layer. Viscous fluxes are discretized using simple
averaging. The CFL number is taken to be 0.7. In Fig. 2, we plot the√normalized
longitudinal velocity profiles ( uu∞ ) against the Blasius parameter η = y u∞ /μL. It
can be clearly seen that HLLC-CPS-Z scheme has a marked improvement over the
HLL-CPS-Z scheme in capturing the boundary layer.
This problem investigates the inviscid contact capturing ability of a given scheme by
simulating a fluid system consisting of two different density fluids sliding over each
other at different speeds [16]. The top fluid is chosen to have conditions (ρ, p, M )top
= (1, 1, 2) while the bottom fluid is chosen to have conditions (ρ, p, M )bottom = (10,
1, 1.1). The domain of 1.0 × 1.0 is coarsely discretized using 10 × 10 cells. The
simulation is run for 1000 iterations using a CFL number of 1.0. All simulations
are plain first-order accurate. Figure 3 shows a comparison of the density variation
along y-location at the center of the domain. The plot clearly shows that while the
HLL-CPS-Z scheme diffuses the interface to an unacceptable level, the proposed
HLLC-CPS-Z scheme is able to retain it exactly.
Fig. 2 Comparison of
velocity profiles computed
by the HLL-CPS-Z scheme
and the proposed
HLLC-CPS-Z scheme for a
M = 0.1 laminar flow over a
flat plate
A Robust and Accurate Convective-Pressure-Split Approximate Riemann Solver … 421
1
HLLC-CPS-Z
0.8 HLL-CPS-Z
0.6
y
0.4
0.2
0
0 1 2 3 4 5 6 7 8 9 10
ρ
Fig. 3 Variation of density along the y-direction at the center of the domain in the two-dimensional
supersonic shear flow
This simple test case evaluates the ability of a numerical scheme to compute an
isolated normal shock front without allowing random perturbations to grow and dis-
tort the initial shock structure or produce contaminated post-shock values [17]. A
normal shock of strength M = 7 is located in the middle of an unit-dimensional
domain divided into 26 × 26 Cartesian cells. The CFL number is 0.5. First-order
solution is sought. Figure 4 shows thirty density contour lines equally spanning val-
ues from 1.0 to 7.4 for this experiment at t = 50 units. From the figure, it is evi-
dent that HLL-CPS-Z scheme produces this variant of shock instability while the
HLLC-CPS-Z scheme is completely free of this.
422 S. Simon and J. C. Mandal
1 1
0.5 0.5
0 0
Y
-0.5 -0.5
-1 -1
-1.5 -1.5
-2 -2
-1.5 -1 -0.5 -1.5 -1 -0.5
X X
4 Conclusions
References
1. Roe P (1981) Approximate Riemann solvers, parameter vectors, and difference schemes. J
Comput Phys 43:357–372
2. Harten A, Lax PD, van Leer B (1983) On upstream differencing and Godunov-type schemes
for hyperbolic conservation laws. SIAM Rev 25:35–61
3. Toro EF, Spruce M, Speares W (1994) Restoration of the contact surface in the HLL-Riemann
solver. Shock Waves 4:25–34
4. Pandolfi M, DâAmbrosio D (2001) Numerical instabilities in upwind methods: analysis and
cures for the carbuncle phenomenon. J Comput Phys 166:271–301
5. Liou MS, Steffen CJ (1993) A new flux splitting scheme. J Comput Phys 107:23–39
6. Zha G-C, Bilgen E (1993) Numerical solutions of Euler equations by using a new flux vector
splitting scheme. Int J Numer Methods Fluids 144:115–144
7. Liou MS (2010) The evolution of AUSM schemes. Def Sci J 60(6):606–613
8. Shima E, Kitamura K (2011) Parameter-free simple low dissipation AUSM-family scheme for
all speeds. AIAA J 49(8):1690–1709
9. Mandal J, Panwar V (2012) Robust HLL-type riemann solver capable of resolving contact
discontinuity. Comput Fluids 63:148–164
10. Sun M, Takayama K (2003) An artificially upstream flux vector splitting scheme for the euler
equations. J Comput Phys 189(1):305–329
11. Xie WJ, Hua L, Sha P, Tian ZY (2015) On the accuracy and robustness of a new flux splitting
method. Acta Phys Sin 64
12. Mandal J, Sharma V (2015) A genuinely multidimensional convective pressure flux split Rie-
mann solver for Euler equations. J Comput Phys 297:669–688
13. Kitamura K A further survey of shock capturing methods on hypersonic heating issues. 21st
AIAA computational fluid dynamics conference, San Diego, CA, pp 1–21
14. Batten P, Clarke N, Lambert C, Causon DM (1997) On the choice of wavespeeds for the HLLC
riemann solver. SIAM J Sci Comput 18:1553–1570 Nov.
15. Simon S, Mandal J (2018) A cure for numerical shock instability in HLLC Riemann solver
using antidiffusion control. J Comput Phys 174:144–166
16. Wada Y, Liou M-S (1997) An accurate and robust flux splitting scheme for shock and contact
discontinuities. SIAM J Sci Comput 18:633–657
17. Dumbser M, Moschetta JM, Gressier J (2004) A matrix stability analysis of the Carbuncle
phenomenon. J Comput Phys 197:647–670
Numerical Investigation of Flow Through
a Rotating, Annular, Variable-Area Duct
1 Introduction
With the increasing use of renewable sources of energy and their time-varying energy
output, grid demand for steam-power plants is becoming increasingly more variable
[1]. This poses a severe constraint on conventional power plants to operate frequently
and over prolonged duration at small fractions of the design output power (the so-
called low-load operation).
Stator and rotor blades of a steam turbine are designed to work under favorable
pressure gradient fields. Low-load operation of turbine involves a low throughput
mass flow rate. However, angular velocity of the rotor has to be maintained constant
to match the grid frequency. Under such conditions, the flow field inside a turbine
rotor is found to be associated with a complex vortical structure around the rotor
P. Saini (B)
Department of Aerospace Engineering, PEC University of Technology, Chandigarh 160012, India
e-mail: palaksaini.beaero14@pec.edu.in
S. Saroha · S. Shukla · S. S. Sinha
Department of Applied Mechanics, Indian Institute of Technology Delhi, New Delhi 110016, India
© Springer Nature Singapore Pte Ltd. 2020 425
B. N. Singh et al. (eds.), Recent Advances in Theoretical, Applied, Computational
and Experimental Mechanics, Lecture Notes in Mechanical Engineering,
https://doi.org/10.1007/978-981-15-1189-9_35
426 P. Saini et al.
blades. Vigorous action of viscous heating within these vortical zones leads to high
temperature which can be detrimental for blade material and can cause severe dam-
age to the turbine. Thus, for safe and economically viable operations of modern
steam turbines at low-load conditions, it is imperative to investigate and understand
the origin and the behavior of the vortical structures that develop under low-load
conditions inside a turbine rotor.
Probably the work of Ref. [2] is the first published report in open literature on
low-load behavior of steam turbines. Subsequently, several other experimental stud-
ies have been reported on the subject. Reference [3] performed a series of experiments
on single- as well as multi-stage steam turbines and demonstrated the occurrence of
vortices under low-load conditions. Reference [4] performed detailed measurements
of the flow field inside low-pressure steam turbine. Reference [5] conducted mea-
surements on a four-stage model air turbine and confirmed the presence of vortices
and associated recirculation zones. Recently Ref. [6] has conducted measurements
of the flow field in a 7-stage model high-pressure (HP) steam turbine blading in order
to develop better understanding of part and low-load operation of HP steam turbines.
In recent years, several researchers have also attempted to numerically simulate
low-load behavior of steam turbines. References [7, 8] employed a finite element
(FE) procedure to numerically simulate the steam flow in the meridional plane of
a turbine cylinder. They further simulated several cases with mass flow rate as low
as 20% of the design value and demonstrated good agreement of their results with
the corresponding experimental measurements. Reference [9] proposed a through-
flow code based on the primary variable method and reached up to 6.1% of design
mass flow rate. Reference [10] presented a full three-dimensional computational
fluid dynamic (CFD) analysis of a two-stage low-pressure steam turbine. Reference
[11] performed simulations of a four-stage model air turbine during low-load oper-
ation. Reference [1] performed an extensive three-dimensional CFD study employ-
ing Reynolds-Averaged Navier–Stokes (RANS) equations in conjugation with sev-
eral turbulence models, and specifically studied temperature field under low-load
conditions.
Indeed, these studies have provided convincing evidence of the existence of com-
plex vortical structures and associated pressure and temperature fields around turbine
rotor blades under low-load conditions. However, the studies have been restricted to
some specific turbine designs, and not much attention has been paid to the reasons
behind the origin of these vortices. The focus of these studies has been primarily on
identifying and visualizing the vortical patterns. Since it is further desirable (i) to
have a more clear understanding of the origin of these vortices and (ii) to predict the
flow behavior in response to more generalized changes in the geometry of flow pas-
sage and the boundary conditions, further research efforts in the form of fundamental
studies can prove to be quite useful. Such is the objective of this work.
From a computational point of view, a full three-dimensional CFD simulation
which accounts for all details of the rotor blade geometry would be the ideal platform
to study and understand the vortical structures in a turbine operating at low-load
conditions. However, setting up such simulations is extremely expensive and time
consuming. To avoid huge demands in terms of time and resource, and yet make a
Numerical Investigation of Flow Through a Rotating … 427
This paper is organized into five sections. In Sect. 2 we describe the flow domain
considered in this work and justify how despite its simplicity, the domain still retains
the fundamental aspects of the actual flow through a steam turbine rotor under low-
load conditions. In Sect. 3 we present the governing equations. This is followed by
Sect. 4 which contains the results and discussions.
Section 5 presents major conclusions of the study.
2 Computational Domain
Fig. 1 Geometry of the computational domain and the related boundary conditions
Numerical Investigation of Flow Through a Rotating … 429
9.4 * 104 elements was employed. Results of simulations 1 and 5 are studied for
performing a grid convergence study. Grids 2–4 are used to perform study on the
influence of body force on nozzle flow.
3 Equations
The governing equations of the flow field under consideration are the continuity,
radial, and axial momentum equations along with the energy and state equations of
a perfect gas. These equations are written with radial (r) and axial (x) coordinates as
the independent variables:
(4)
P
ρ= (5)
RT
The symbols p, ρ, T represent pressure, density, and temperature of the fluid,
respectively. vx and vr are the axial and radial components of velocity, respectively.
Fluid is taken to be an ideal gas with a specific heat ratio of 1.4 and gas constant R
as 287 J/kg-K. The last term in the radial momentum Eq. (3) is the body force per
unit volume. The body force is taken to be proportional to the radius with C being
a constant for a simulation. This form of the radial force accurately captures the
variation of centrifugal acceleration experienced by fluid particles passing through
a turbine. Since the governing equations are written with respect to the rotor frame,
the body force appears as a pseudo force.
430 P. Saini et al.
The outward radial direction of the force represents the rotating effect of the annu-
lar duct on the flow. In continuum mechanics, the Froude number is a dimensionless
number defined as the ratio of the flow inertia to the external field. Froude number
will play a significant role whenever there is a significant body force. In this case
the Froude number appears because of the centripetal force. In accordance with the
computational setup of this work, Froude number (F) of the flow can be defined
using various geometric and boundary parameters in the following manner:
ṁ
F=
√ (6)
A ρ LC A
where ṁ is the mass flow rate through the duct, L is the inlet radius, A is the inlet
area, ρ is the density at the inlet, and C is the coefficient of body force per unit
volume. The quantity C actually represents the magnitude of angular velocity of
the rotor. With Froude number being inversely proportional to the coefficient C, an
increase in the value of C results into a decrease in Froude number and equivalently
an increase in the influence of body force on flow dynamics. The limiting case of
C being zero (no body force) implies an infinite value of Froude number. Similarly,
relevant geometrical and boundary parameters can be used to define a representative
Reynolds number as well:
ṁ L
Re = (7)
Aμ
Low-load operation of steam turbines involves mass flow rates which are small
fractions of the design mass flow rate while the rotation rate of the rotor remains the
same as that at designed condition. With the derived expression for the Froude number
(6), it is clear that the low-load flow through a steam turbine can be characterized
by low values of Froude number itself. Thus to systematically understand low-load
behavior of flow through turbines, in this work, we perform several simulations with
varying Froude number. A summary of these simulations is presented in Table 1.
The commercial software ANSYS R FLUENT is used to obtain the numerical
solution to the two-dimensional compressible Navier–Stokes equations, in general,
curvilinear coordinates (x, r). Rest of the flow modeling details have been taken from
a previous study done at IIT Delhi [26, 27]. Three convergence criteria were used to
ensure that the results were accurate. First, the convergence criteria for all residuals
were set to 10−6 . In addition to this, the net mass flux was monitored to ensure that
there was no imbalance. Finally, primary variables (velocity, density and pressure)
were monitored through the iterations at inlet and exit to ensure that they reached a
steady value.
Numerical Investigation of Flow Through a Rotating … 431
Our study is based on several simulations performed with Froude number as param-
eter. Description of these simulations is provided in Table 1. All these simulations
are performed with identical mass flow rates (25 kg/s) specified at the inlet of the
nozzle. Reynolds number in each simulation is 1.7 × 105 .
Before employing our results to understand the underlying physics, we have sub-
jected all our results to extensive grid convergence studies. Here, we present compar-
isons for zero-force case (Simulation 1 and 5) only. Case 1 and Case 5 are identical
in all aspects except the grid. Case 5 simulation is performed with a grid element
size that is smaller than that of Case 1. In Figs. 2 and 3 we present streamlines and
pressure contours from Case 1 and 5 simulations. Close agreement is seen between
results of the two simulations. Based on our convergence studies, we deem the mesh
size with 2.3 × 105 elements adequate for our study. All the remaining results in
this paper have been obtained with this finer mesh. In the following sections, we
first study the general flow characteristics with zero body force. Subsequently, the
influence of varying Froude number on the flow is studied.
In simulation case 1 which represents our baseline case without any body force,
the flow accelerates in the converging portion, remains subsonic at the throat and
subsequently decelerates in the diverging portion. Further, in the diverging section
of the nozzle, the primary flow separates from the outer nozzle wall and forms a
recirculation zone. However, the primary flow remains attached to the inner wall
throughout the flow domain and maintains a predominantly axial direction.
Significant variations in pressure are only shown by the primary flow in the
variable-area part of the duct. In the constant area duct pressure is almost the same
as that in the ambience. At the outlet boundary of the flow domain, backflow occurs.
A recirculation zone with very low velocities is formed here. The recirculating flow
enters at farther radial locations (near the exit wall) and joins the primary flow stream
at smaller radial locations. With this simulation as our base case study, in the next
section we discuss how this flow is influenced by varying Froude number.
Fig. 4 Streamlines colored by normalized velocity magnitude (a, c, e) and pressure contours (b,
d, f) for simulations 2, 3, and 4
the deflection induced by larger body force, the fluid tends to move radially outward.
The acceleration of the flow in a direction away from the axis causes the pressure
to drop in the vicinity of the annulus. The obstruction offered by the wall stagnates
the flow, thus explaining the reason for the higher pressures observed near the outer
walls.
In summary, our results clearly show that the Froude number has profound effects
on the flow pattern inside an annular nozzle. As the Froude number decreases, the flow
characteristics start following a trend. In successive simulations, as the magnitude
of body force increases (i) the primary flow shows a clear tendency to separate from
the inner annulus wall, (ii) the separation point at the inner wall moves upstream,
434 P. Saini et al.
(iii) the tendency of the primary flow to turn radially outward becomes stronger, (iv)
magnitude of velocity in the recirculation zone becomes larger and (iv) pressure near
the outer annulus wall becomes larger.
These observed changes in the flow field can be explained as follows. As the
Froude number decreases, a fluid particle is subjected to more intense radial accel-
eration, which in turn leads to the development of substantial radial velocity in the
diverging/constant area section of the nozzle. This substantial radial velocity makes
the flow detach from the inner nozzle wall creating a recirculation zone near the inner
wall wherein flow enters the nozzle from outlet ambiance. As the primary flow in
the diverging and constant area section of the nozzle is forced to become radial over
the bulk of the nozzle cross section, it can now exit the nozzle axially only through a
thin annular area adjacent to the outer wall of the nozzle (see Fig. 5). This behavior
of the primary flow is akin to what has been observed in full three-dimensional CFD
simulations and experiments performed on steam turbines (Fig. 11 of Ref. [1]). Thus
the study presented in this paper seems to provide a plausible explanation of the
steam behavior in turbines under severe low-load conditions.
5 Conclusions
to the rotor. The motivation behind choosing such a canonical domain is to isolate,
identify, and understand the influence of a radially-dependent body force on the flow
pattern through a variable-area duct. This study in turn is expected to aid in explain-
ing the flow behavior in steam turbines working under severe low-load conditions,
which is indeed characterized as a low-Froude number flow. Our simulations show
that as the magnitude of body force increases, the pattern of streamlines gets sub-
stantially modified in the diverging and the constant area parts of the nozzle. The
presence of a radially outward body force induces severe separation tendency near
the inner nozzle wall. This behavior is in stark contrast to the separation induced
by high-divergence angle, which happens at the outer wall of ducts. We find that in
our simulations streamlines turn radially outward leading to the formation of a large
recirculation zone near the exit of the constant area section. As the strength of body
force increases, the separation from the inner walls occurs further upstream and the
size of the recirculation zone increases. This behavior is akin to the flow pattern
observed in experiments on steam turbine rotors and also in full three-dimensional
CFD simulations of the flow over the exact rotor geometry. Thus, despite having
employed a relatively simple flow domain, the study seems to successfully isolate
and identify the essential body force influences which are responsible for the develop-
ment of flow separation and a massive recirculation zone near the rotor exit of steam
turbines operating under low-load conditions. The authors believe that the results of
this study make a strong case for the viability of further employing the presented
flow domain to perform detailed studies of the influence of other parameters like duct
divergence angle, viscosity, etc. on rotating annular flows. These parameters will be
the subjects of our future work.
References
Abstract This paper demonstrates the newly developed particle method in the Direct
Simulation Monte Carlo (DSMC) framework, and we refer to it as the Maxwellian–
DSMC (M–DSMC) method. The M–DSMC solver is utilized for various fluid flow
problems with different length scales varied from continuum to transitional regime.
The hypersonic flow of argon over a cylinder simulation is carried out using M–
DSMC and the same simulation results are compared with the regular DSMC method.
The present work shows that the M–DSMC results are quite a good match with the
DSMC results. The computational cost is significantly reduced by using M–DSMC
method and the same is compared with regular DSMC method.
1 Introduction
In the present work, computational models and associated numerical methods are
developed to focus on minimization of the computational cost by instigating effective
and robust algorithm. The same aspect has motivated us to develop a simplified
DSMC method, which works more efficiently than the regular DSMC method, and
is within acceptable limits of accuracy. To this end, there have been some earlier
works, which attempted to extend the DSMC method to continuum regime. In these
works, researchers had either restricted the number of collisions, or relaxed cell size,
time step restrictions, which are typically followed in the regular DSMC method.
Breuer et al. [1] utilized the CFL [2, 3] number and artificial viscosity [1] in their
work to extend the DSMC method into continuum regime. Lengrand et al. [4] used
approximate method to the regular DSMC method, which employs limited number
of collisions in a cell for expanding plume flows.
G. Malaikannan (B)
Department of Aerospace Engineering, SRM Institute of Science and Technology, Chennai, India
e-mail: malaikannan.g@ktr.srmuniv.ac.in; malaikag@srmist.edu.in
R. Kumar
Department of Aerospace Engineering, Indian Institute of Technology Kanpur, Kanpur, India
e-mail: rkm@iitk.ac.in
Similar to Lengrand et al. [4], Bartel et al. [5] also used an approximate scheme for
the DSMC method to simulate high-density flows through nozzle. Lengrand et al. [4]
and Bartel et al. [5] used the collision limiter technique to extend the DSMC method
in the continuum regime for expanding flows through nozzle. There have been some
works focusing on extending the applicability range of the DSMC method to higher
density flows. Pullin [6] developed Equilibrium Particle Simulation Method (EPSM)
for compressible inviscid ideal gas flows. Recently, Titov and Levin [7] used the
eDSMC method to simulate high-pressure flows, which is also based on the collision
limiter scheme.
In order to extend the range of applicability of DSMC approach to higher density
near-equilibrium flows, in the present work, we have developed a new approximate
method in the DSMC framework, and we refer to it as the Maxwellian–DSMC (M–
DSMC) method, which is analogous to the EPSM method [6]. M–DSMC is capable
of solving flows ranging from continuum to rarefied regime.
2 Maxwellian–DSMC Approach
The M–DSMC is similar to the regular DSMC [8, 9] method, except the collision
modeling part. In contrast to the binary collision modeling of dilute gas flows in the
conventional DSMC setup, the collisional modeling is completely done away with
in the M–DSMC framework. Instead, the simulated particles are assigned properties
sampled from the Maxwellian distribution at the local temperature. It is expected
that the simplified calculation procedure for collisions would result in huge savings
in the computational cost. The calculation procedure of the M–DSMC method is
discussed below.
Particle velocities are averaged over each cell at each time step. The instantaneous
particle velocity data in each cell is first averaged out to calculate bulk velocity using
Eq. 1.
Cx
ux =
Nc
Cy
uy =
Nc
Cz
uz = (1)
Nc
where C x , C y , and C z are the components of particle velocity in x-, y-, and z-directions
at a given time; ux , uy , and uz are the components of bulk velocity; and Nc is the total
number of particles in a particular cell at each time step.
On the other hand, the translation kinetic temperature is computed by referring to
the equipartition theorem, given as follows:
Development of M–DSMC Numerical Algorithm for Hypersonic Flows 439
3 1
k B Ttrans = m cx2 + c2y + cz2 (2)
2 2
where T trans is the translational temperature, ci is the random translational velocity
component in the ith direction, and k B is the Boltzmann constant. Equation 2 can
now be rephrased as follows:
C x2 − u 2x + C y2 − u 2y + C z2 − u 2z
Ttrans = m (3)
3k B
where, cx , cy , and cz are the thermal velocity components in the x-, y-, and z-directions,
R1 , R2 , R3 , and R4 are random numbers, cmps is the most probable speed. Later,
particle velocities are obtained by using corresponding thermal velocities and the
bulk velocity. The local cell-based most probable speed, cmps , is calculated from the
known translational temperature and is given by the following equation:
2k B Ttrans
cmps = (5)
m
C x = cx + u x
C y = cy + u y
C z = cz + u z (6)
To validate the M–DSMC numerical algorithm, we have simulated one of the classical
fluid flow problems. Hypersonic flow of argon over a cylinder is simulated at two
different freestream conditions, and the results are compared with the regular DSMC
method. The diagrammatic view of the problem is illustrated in Fig. 1. The freestream
parameters for both cases are given in Table 1. Noteworthy is the fact that the two
440 G. Malaikannan and R. Kumar
Table 1 Freestream
Freestream Freestream values Freestream values
parameters for hypersonic
parameters (Case-I) (Case-II)
flow over a cylinder
Gas Argon Argon
Temperature 200 K 200 K
Number density 4.247E+21 m−3 4.247E+20 m−3
Knudsen number 0.001 0.01
Mach number 10 10
Velocity of gas 2634.5542 m/s 2634.5542 m/s
cases differ only in terms of number density or Knudsen number (based on cylinder
diameter). The time step, cell size, and other numerical parameters are set as per
the DSMC procedures, and they remain the same for both M–DSMC and DSMC
approaches. In both cases, simulation for first 25,000 time steps was done till the
system reached steady state, beyond which sampling was done for another 40,000
time steps to give a similar level of statistical error (<3%). The detailed investigation
is carried out for both flow and surface properties of the cylinder.
Development of M–DSMC Numerical Algorithm for Hypersonic Flows 441
Flow properties are obtained for hypersonic flow over a cylinder using M–DSMC
at two different global Knudsen numbers (based on the cylinder diameter), and are
compared with the in-house DSMC solver. Contour plot of x-component of velocity
and temperature are given in Fig. 2, whereas Fig. 3 shows the same properties at a
Knudsen number of 0.01, based on cylinder diameter. M–DSMC method is found
to clearly capture the strong detached shock and the wake region. The variation of
x-component of velocity and temperature across the domain compares well between
Fig. 2 Comparison of
x-component of velocity and
temperature contour of
DSMC and M–DSMC for
flow over a cylinder at a
Knudsen number of 0.001
Fig. 3 Comparison of
x-component of velocity and
temperature contour of
DSMC and M–DSMC for
flow over a cylinder at a
Knudsen number of 0.01
DSMC and M–DSMC methods for both cases. Small deviations can be seen in the
wake region, particularly for a Knudsen number of 0.01. This is understandable as
the local degree of rarefaction in the wake region is quite large as compared to that
shown by the global Knudsen numbers for these cases. Linear plot of x-component
of velocity and temperature along the stagnation streamline are given in Fig. 4 for
the first test case with a Knudsen number of 0.001. On the other hand, Fig. 5 shows
the same properties along the stagnation streamline for the second test case with a
Knudsen number of 0.01. From Figs. 4 and 5, it is distinctly visible that M–DSMC
results are quite good match with the regular DSMC method. Small deviations can
Development of M–DSMC Numerical Algorithm for Hypersonic Flows 443
Fig. 4 Comparison of
x-component of velocity and
temperature along the
stagnation streamline for
flow over a cylinder at a
Knudsen number of 0.001
be noticed in the wake region, which is the zone of large rarefaction, however, shock
is found to be well captured.
Similar to flow properties, the surface properties such as heat flux and pressure
coefficient are obtained using the M–DSMC approach and the results are compared
with regular DSMC method for the two case studies.
444 G. Malaikannan and R. Kumar
Fig. 5 Comparison of
x-component of velocity and
temperature along the
stagnation streamline for
flow over cylinder at a
Knudsen number of 0.01
q
CH = (7)
1
ρ U3
2 ∞ ∞
Heat flux along the surface of the cylinder is normalized and is given as follows,
where q is the convective heat flux and ρ ∞ and U ∞ are the freestream density and
velocity, respectively.
Figure 6 shows the variation of surface heat flux along the length of the cylinder
for the two cases. From Fig. 6a, it is distinctly visible that normalized heat flux
Development of M–DSMC Numerical Algorithm for Hypersonic Flows 445
variation along the surface of the cylinder obtained from M–DSMC approach shows
same trend as that obtained from DSMC. There is a maximum deviation of ~15%
from the DSMC results at a Knudsen number of 0.01 (based on cylinder diameter),
whereas the deviation for a Knudsen number of 0.001 is less than 10%. Noteworthy
is the fact that actual local degree of rarefaction is much more than as shown by the
global Knudsen numbers. Thus, it can be said that the M–DSMC approach works
quite well for near-continuum to semi-rarefied flow regime.
Similarly, differential pressure is normalized to the dynamic pressure and given as
the pressure coefficient, CP. The normalized pressure coefficient is given as follows:
446 G. Malaikannan and R. Kumar
Table 2 Comparison of
Parameter DSMC M–DSMC
computational cost of each
method for hypersonic flow CPU time (hours) 203 103
over a cylinder for Knudsen Number of cells 640,000 640,000
number of 0.001
Number of particles 2,797,968 2,798,111
Table 3 Comparison of
Parameter DSMC M–DSMC
computational cost of each
method for hypersonic flow CPU time (hours) 224 126
over a cylinder for Knudsen Number of cells 640,000 640,000
number of 0.01
Number of particles 3,042,676 3,056,466
P − P∞
CP = (7)
1
ρ U2
2 ∞ ∞
4 Conclusions
References
1. Breuer K, Piekos E, Gonzales D (1995) DSMC simulations of continuum flows. In: 30th
thermophysics conference, p 2088
2. Courant R, Friedrichs K, Lewy H (1928) On the partial differential equations of mathematical
physics. Math Ann 100(1):32–74
3. Courant R, Friedrichs KO (1948) Supersonic flow and shock waves. In: Applied mathematical
sciences, vol 12
4. Lengrand J-C, Raffin M, Allegre J (1981) Monte-Carlo simulation method applied to jet-wall
inter-actions under continuum flow conditions. Rarefied Gas Dyn 994–1006
5. Bartel TJ, Sterk TM, Payne J, Preppernau B (1994) DSMC simulation of nozzle expansion flow
fields. In: 6th AIAA and ASME, joint thermophysics and heat transfer conference. Colorado
Springs, CO
6. Pullin D (1980) Direct simulation methods for compressible inviscid ideal-gas flow. J Comput
Phys 34(2):231–244
7. Titov EV, Levin DA (2007) Extension of the DSMC method to high pressure flows. Int J Comput
Fluid Dyn 21(9–10):351–368
8. Chinnapan AK, Malaikannan G, Kumar R (2017) Insights into flow and heat transfer aspects of
hypersonic rarefied flow over a blunt body with aerospike using direct simulation Monte-Carlo
approach. Aerosp Sci Technol 66:119–128
9. Malaikannan G, Kumar R (2017) Hybrid particle-particle numerical algorithm for high speed
non-equilibrium flows. Comput Fluids 152:24–39
Fluid–Structure Interaction Dynamics
of a Flexible Foil in Low Reynolds
Number Flows
C. Bose · S. Gupta
Department of Applied Mechanics, Indian Institute of Technology Madras, Chennai 600036, India
e-mail: cb.ju.1991@gmail.com
S. Gupta
e-mail: gupta.sayan@gmail.com
S. Sarkar (B)
Department of Aerospace Engineering, Indian Institute of Technology Madras, Chennai 600036,
India
e-mail: sunetra.sarkar@gmail.com
1 Introduction
Research into design and development of bio-inspired futuristic devices like Micro
Aerial Vehicles (MAVs) or Autonomous Underwater Vehicles (AUVs) has recently
gained an impetus due to the necessity for futuristic surveillance tools. A crucial
feature in investigating the behavior of these systems is to gain clear understanding
of the underlying fluid mechanics behind flapping-wing/fin propulsion [1–3]. The
inspiration for the design of MAVs/AUVs comes from the natural bio-propulsion
systems like insect flight or swimming of fishes. Both insect wings and fish fins are
very flexible structures that undergo significant deformation due to the interactions
with the surrounding fluid flow during their locomotion. Furthermore, the fluid vis-
cosity and the flexibility of the structure play a key role in generation of propulsive
and maneuvering forces as the body inertia is comparable to the fluid inertia. Hence,
the intricacy of the fluid–body interactions of flexible foils/filaments in the surround-
ing flow has become a central problem over the past two decades. A comprehensive
survey of existing literature in this regard can be found in the review paper by Shelley
and Zhang [4].
On the other hand, passive flexibility of wings/fins is seen to play a pivotal role
in boosting their aerodynamic and hydrodynamic performance. The role of passive
flexibility in propulsion has been studied experimentally in the literature in terms
of the dynamics of a flag or a filament. Taneda [5] was the first to explore the
flapping dynamics of a flag through wind tunnel experiments. Zhang et al. [6] revisited
the problem by performing soap film experiments. A variety of flapping modes
(“stretched straight” (SS), coherent flapping) with different flapping frequencies were
reported by them. The “SS” mode was accompanied by a von Kármán wake, whereas
the coherent flapping mode showed an undulating vortex street having mostly same-
sense fine-scale vortices which can be attributed to the Kelvin–Helmholtz instability
of the shear layer vortex street. They also reported bistability and hysteresis behavior
between different flapping modes. Moreover, a transition to irregular flapping was
observed with the increase in the length scale or the flow velocity. The other important
experimental works include Shelley et al. [7] and Elloy et al. [8].
However, the experimental studies are limited in the parametric space. Therefore,
we have to rely on numerical simulations to get better insights into the essential
features of the FSI dynamics with respect to the key parameters such as mass ratio,
bending rigidity, and Reynolds number. Due to prohibitive computational cost of
full-fidelity FSI simulations, some of the numerical studies have considered the sur-
rounding fluid to be inviscid while modeling the structure as an elastic continuum [9,
10]. Alben et al. [10] modeled a fin as an elastic continuum, and the flow was modeled
by the inviscid potential theory with some empirical modification to incorporate the
effects of viscous drag. However, the applicability of the model was limited espe-
cially at low Reynolds numbers. On the other hand, some works relied on lower order
structural models such as flexibly mounted rigid components coupled with fluid [11,
12]. Notably, the FSI behavior of such systems is much different from that of a fully
elastic continuum. With the increase in the computational resources in recent days,
Fluid–Structure Interaction Dynamics of a Flexible Foil … 451
viscous simulation of a flexible filament has been carried out in a few studies [13–
16]. Some of the very recent numerical simulations of self-propelled flapping system
[17–19] showed that the flexibility can enhance the propulsive performance which
is also in accord with the experimental observations [20, 21]. However, most of the
abovementioned studies have been performed in uniform flow condition. The effect
of flow fluctuations onto the FSI dynamics of a flexible filament has not received
much attention except in a few recent studies [22–24].
Therefore, the primary focus of this study is to unravel the physics associated
with the FSI behavior of such systems subjected to a fluctuating flow field in the
low Reynolds number regime, relevant for natural locomotion. The corresponding
coupled systems are phenomenologically rich in dynamics due to complex FSI. This
paper aims to numerically investigate the FSI behavior of these systems by consid-
ering complexities that include the flexibility of the structure and the viscous flows
in the low Reynolds number regime. The flow topology over the flexible structure
is investigated along with its different vibrating modes during the passive flapping
motion. The remainder of the paper consists of the following sections: Sect. 2 dis-
cusses the computational methodology. The deflection envelope of the flexible fila-
ment, the vortex interactions as well as the generation of aerodynamic forces have
been discussed in Sect. 3. The salient outcomes of this study are summarized in
Sect. 4.
2 Computational Framework
∇ · u = 0,
∂u ∇p
+ u − um · ∇ u = − + ν∇ 2 u.
∂t ρf
Here, u is the flow velocity, u m is the grid point velocity, p is the pressure, ρ f
is the fluid density, ν is the kinematic viscosity, and ∇ is the gradient operator. The
details of the flow solver can be found in [25, 26]. The flexible filament is modeled
as a solid continuum. The equation for conservation of linear momentum is given by
∂
ρs vd V = n · σ d + ρs bd V,
∂t
452 C. Bose et al.
where V is the volume of the foil, bounded by the surface with unit normal n, v
is the velocity vector, σ is the Cauchy stress tensor, ρs is the density of the material
of the structure, and b is the body force per unit mass. Assuming large strain, the
material behavior of the filament can be modeled using Kirchhoff–St. Venant hyper-
elasticity theory. The linear momentum equation can therefore be written in terms
of Piola–Kirchhoff stress tensor S as
1
σ = F · S · FT ,
J
where F = I + ∇w is the deformation gradient, I is the second-order identity tensor,
and w is the deformation vector. A large strain elastic stress analysis solver based on
Lagrangian displacement formulation [28] is used to solve the structural governing
equations. A strong coupling methodology [28], illustrated through the flowchart
shown in Fig. 1, is adopted in this study using a quasi-Newton coupling algorithm
with an approximation for the inverse of the residual’s Jacobian matrix from a least-
square model (abbreviated as IQN-ILS). The present FSI solver is well validated
with the existing literature [29]. A rectangular computational domain (see Fig. 2a)
has been discretized using structured grids. The mesh (Fig. 2b), having 89,052 grid
points, has been chosen after a grid convergence study. Standard boundary conditions
are applied: a zero pressure gradient and a constant free stream at the inlet; a zero
velocity gradient and atmospheric pressure condition at the outlet; no-slip and zero-
normal pressure gradient condition on the horizontal walls and traction boundary
condition on the flexible flapper.
The FSI solver has been quantitatively validated in the present study with the
benchmark case of a flexible splitter plate attached with a rigid cylinder given by
Turek and Hron [30]. The parameters corresponding to “FSI2” case presented in [30]
have been considered for the benchmark validation. It can be clearly seen from Fig. 3
that vertical tip displacement time history of the flexible plate shows an excellent
match with the results presented by Turek and Hron [30] in the self-sustained periodic
oscillatory state.
The present computations are performed in the following
parametric space:
Reynolds number (R E D ) = 500, mass ratio μ = ρs /ρ f = 1, nondimensional
Young’s modulus Ē = E/ρ f u 2∞ = 5600, and nondimensional length (l¯ = l/D) =
2.5. Here, ρs and ρ f are the solid and fluid density, respectively, E is the dimensional
Young’s modulus of the structure, l and D are the length of the filament and diameter
of the cylinder, respectively, and u ∞ is the uniform velocity at the inlet.
The present work focuses on the role of chord-wise flexibility of a filament in enhanc-
ing the aerodynamic loads in comparison to a rigid foil subjected to a flow fluctuation
in terms of the wake of a rigid stationary cylinder. Figure 4 presents the vorticity
contour past the rigid foil at t = 10 s. It is observed that the interaction of shear
layers separated from the rigid cylinder is delayed due to the presence of the rigid
foil; the immediate formation of the von Kármán vortex street is inhibited which
454 C. Bose et al.
Fig. 4 Vorticity contour behind a rigid filament in the wake of a rigid cylinder
eventually occurs in longer time. Figures 5a and 5b show the time history of the lift
coefficient (CL ) and drag coefficient (CD ), respectively. Since the rigid foil is placed
symmetrically at zero angle of attack in the fluctuating flow field, CL has a zero mean
and CD has a very small mean value in the order of 10−5 .
Thereafter, the FSI dynamics of a flexible filament with same length and similar
inflow condition has been simulated. The filament is kept fixed at its leading edge. In
this low Reynolds number regime, the filament exhibits a passive pitching in terms
of its different natural bending modes. The deflection envelope and the time history
of the tip displacement have been plotted in Figs. 6a and 6b, respectively. At RE D
= 500, the shear layers past a rigid cylinder roll up and consequently vortex cores
are formed evolving with time, thus gaining more strength. Subsequently, the vortex
cores convect over the structure, acting as a periodic forcing and subsequently shed in
the downstream forming a vortex street of alternate shedding vortices. It is observed
that the first mode shape is initially triggered when the vortex cores start convecting
over it. However, it attains a mixed-mode vibration between the first and second
mode shapes once the vortices impinge on it and start shedding from the trailing
edge. Figures 6c and 6d present the CL and CD time histories for the flexible filament.
It can be evidently seen that the aerodynamic loads have been significantly increased
in this case with a high maximum CL value along with a high mean CD (negative)
Fluid–Structure Interaction Dynamics of a Flexible Foil … 455
Fig. 6 FSI dynamics of a chord-wise flexible filament: a deflection envelope, b tip deflection time
history, c lift coefficient, and b drag coefficient time histories
value. It can be noted that the fluid–body interaction results in a symmetric wake
that results in zero-mean lift. Besides, it produces significant thrust on the structure
desirable for efficient propulsion.
The impingement of the vortex core on the flexible filament causes it to bend
in different modes depending on the location of impingement and the Strouhal fre-
quency of the fluctuating flow. The vortex–filament interactions and evolution of
the trailing edge flow topology have been investigated by the vorticity contours at
different time instances as shown in Fig. 7a. To get more insight into the vortex inter-
action, the Lagrangian coherent structures (LCS) are presented in the near field along
with the vorticity contours (Fig. 7b). LCSs are obtained by plotting the backward
finite-time Lyapunov exponents (FTLE) of the velocity field which clearly show how
the impingement of the vortex cores cause the filament to bend in different modes
augmenting the propulsive efficiency.
To understand the fluid–body interactions in more detail, the flow physics is
analyzed through a sequence of vortex contour snapshots at different time instances,
see Fig. 8. At t = 2.68 s, a large clockwise vortex core “4” (generated as the primary
wake vortex of the cylinder in the previous cycle) is seen to be convecting over
the elastic structure along with a small weak counterclockwise vortex “3” at the
leading edge. Besides, two opposite sense vortices “1” and “2” are seen to be formed
through rolling up of the separated shear layers from the cylinder. Clockwise vortex
“4” gradually moves forward and impinges on the structure at the tip bending it
downward. At the same time, the counter clockwise vortex “2” impinges in the middle
456 C. Bose et al.
(a) (b)
t = 0.2 sec
t = 1.0 sec
t = 2.0 sec
t = 4.0 sec
t = 6.0 sec
Fig. 7 a The instantaneous vorticity field and b the corresponding LCS (bFTLE ridges)
Fluid–Structure Interaction Dynamics of a Flexible Foil … 457
Fig. 8 Change of mode shapes due to vortex impingement on the flexible filament
of the structure making it bend in its second mode shape; see Fig. 8c. Thereafter,
the vortex “4” sheds downstream and becomes a part of the vortex street, while the
vortex “2” convects over the structure and reaches to the tip. On the other hand,
newly formed clockwise vortex “1” deforms the weaker counterclockwise vortex
“3” and comes near the leading edge of the structure. As a result, the flapper again
bends upward in the first mode shape at t = 2:86 s (see Fig. 8d); thus, a mixed-mode
oscillation is observed. These interactions take place periodically in the subsequent
cycles.
In case of a shorter flapper (l¯ = 1), it appears to oscillate predominantly in
its first natural mode shape; see Fig. 9a. This can be attributed to the fact that the
wake vortices impinge at the tip of the structure, thus making it bend in the upward
and downward directions periodically in its first natural mode. However, it attains
a mixed-mode vibration for l¯ = 2.5 as the length of the flapper becomes greater
than the formation length of the wake vortices; see Fig. 9b. As a result, the vortices
impinge in the middle of the longer elastic structure leading to a periodic mixed-mode
oscillation.
458 C. Bose et al.
Fig. 9 a First mode oscillation (l¯ = 1); b mixed-mode oscillation (l¯ = 2.5)
4 Concluding Remarks
References
1. Shyy W, Aono H, Chimakurthi SK, Trizila, Kang CK, Cesnik CES, Liu H (2010) Recent
progress in flapping wing aerodynamics and aeroelasticity. Prog Aerosp Sci 46:284–327
2. Platzer MF, Jones KD, Young J, Lai JS (2008) Flapping wing aerodynamics: progress and
challenges. AIAA J 46:2136–2149
3. Triantafyllou MS, Techet AH, Hover FS (2004) Review of experimental work in biomimetic
foils. IEEE J Oceanic Eng 29:585–594
4. Shelley MJ, Zhang J (2011) Flapping and bending bodies interacting with fluid flows. Annu
Rev Fluid Mech 43:449–465
5. Taneda S (1968) Waving motions of flags. J Phys Soc Jpn 24:392–401
6. Zhang J, Childress S, Libchaber A, Shelley M (2000) Flexible filaments in a flowing soap film
as a model for one-dimensional flags in a two-dimensional wind. Nature 408:835–839
7. Shelley M, Vandenberghe N, Zhang J (2005) Heavy flags undergo spontaneous oscillations in
flowing water. Phys Rev Lett 94:094302
8. Eloy C, Lagrange R, Souilliez C, Schouveiler L (2008) Aeroelastic instability of cantilevered
flexible plates in uniform flow. J Fluid Mech 611:97–106
Fluid–Structure Interaction Dynamics of a Flexible Foil … 459
9. Alben S (2008) Optimal flexibility of a flapping appendage in an inviscid fluid. J Fluid Mech
614:355–380
10. Alben S, Witt C, Baker TV, Anderson E, Lauder GV (2012) Dynamics of freely swimming
flexible foils. Phys Fluids 24:051901
11. Spagnolie SE, Moret L, Shelley MJ, Zhang J (2010) Surprising behaviors in flapping locomotion
with passive pitching. Phys Fluids 22:041903
12. Zhang X, Ni S, Wang S, He G (2009) Effects of geometric shape on the hydrodynamics of a
self-propelled flapping foil. Phys Fluids 21:103302
13. Zhu L (2007) Viscous flow past a flexible fibre tethered at its centre point: vortex shedding. J
Fluid Mech 587:217–234
14. Huang W-X, Shin SJ, Sung HJ (2007) Simulation of flexible filaments in a uniform flow by
the immersed boundary method. J Comput Phys 226:2206–2228
15. Zhu L (2009) Interaction of two tandem deformable bodies in a viscous incompressible flow.
J Fluid Mech 635:455–475
16. Olivier M, Dumas G (2016) A parametric investigation of the propulsion of 2d chordwise
flexible flapping wings at low Reynolds number using numerical simulations. J Fluids Struct
63:210–237
17. Lee J, Lee S (2013) Fluid–structure interaction for the propulsive velocity of a flapping flexible
plate at low Reynolds number. Comput Fluids 71:348–374
18. Yeh PD, Alexeev A (2014) Free swimming of an elastic plate plunging at low Reynolds number.
Phys Fluids 26:053604
19. Tang C, Lu X (2015) Propulsive performance of two-and three-dimensional flapping flexible
plates. Theor Appl Mech Lett 5:9–12
20. Heathcote S, Gursul I (2007) Flexible flapping airfoil propulsion at low Reynolds numbers.
AIAA J 45:1066–1079
21. Cleaver D, Calderon D, Wang Z, Gursul I (2016) Lift enhancement through flexibility of
plunging wings at low Reynolds numbers. J Fluids Struct 64:27–45
22. Jia L-B, Yin X-Z (2009) Response modes of a flexible filament in the wake of a cylinder in a
flowing soap film. Phys Fluids 21:101704
23. Zhu L (2014) Numerical investigation of the dynamics of a flexible filament in the wake of
cylinder. Adv Appl Math Mech 6:478–493
24. Kundu A, Soti AK, Bhardwaj R, Thompson MC (2017) The response of an elastic splitter plate
attached to a cylinder to laminar pulsatile flow. J Fluids Struct 68:423–443
25. Kumar SK, Bose C, Ali SF, Sarkar S, Gupta S (2017) Investigation on a vortex induced vibration
based energy harvester. Appl Phys Lett 111(24):243903
26. Bose C, Kumar SK, Sarkar S, Gupta S (2019) Interaction of a flexible splitter plate with vortex
shedding past a rigid circular cylinder. Recent Adv Struct Eng 2:841–852
27. Ferziger JH, Peric M, Leonard A, Computational methods for fluid dynamics (1997)
28. Cardiff P, Tukovic Z, Jasak H, Ivankovic A (2016) A block-coupled finite volume methodology
for linear elasticity and unstructured meshes. Comput Struct 175:100–122
29. Tukovic Z, Cardiff P, Karac A, Jasak H, Ivankovic A (2014) OpenFOAM library for fluid
structure interaction. In: 9th OpenFOAM workshop. Zagreb, Croatia
30. Turek S, Hron J (2006) Proposal for numerical benchmarking of fluid-structure interaction
between an elastic object and laminar incompressible flow. In: Fluid-structure interaction.
Springer, Berlin, Heidelberg, pp 371–385