Journal of Sound and <ibration (2001) 247(2), 227}242

doi:10.1006/jsvi.2001.3707, available online at http://www.idealibrary.com on

ANALYSIS OF ELASTIC FLEXURAL WAVES IN

NON-UNIFORM BEAMS BASED ON MEASUREMENT
OF STRAINS AND ACCELERATIONS
L. HILLSTROG M AND B. LUNDBERG
¹he As ngstroK m ¸aboratory, ;ppsala ;niversity, Box 534, SE-751 21 ;ppsala, Sweden.
E-mail: bengt.lundberg@angstrom.uu.se

(Received 17 December 1999, and in ,nal form 15 February 2001)

Elastic #exural waves in an unloaded and unsupported segment of a non-uniform beam

were considered. A method based on Timoshenko's model was established for evaluation of
shear force, transverse velocity, bending moment and angular velocity at an arbitrary section
from four independent measurements of such quantities at one to four sections. From these
quantities, shear stress, normal stress, power transmission, etc., can be obtained.
Experimental tests were carried out with an aluminium beam which had an abrupt change in
height from 15 to 20 mm and was equipped with strain gauges and accelerometers at four
uniformly distributed measurement sections and at three evaluation sections. The distance
between the two outermost measurement sections was 600 mm, corresponding to 1)12
wavelengths at the upper end of the frequency interval 2}500 Hz considered. Bending
moments and transverse velocities evaluated agreed well with those measured at evaluation
sections located centrally among the measurement sections and at a distance of 100 mm, or
0)16 wavelengths, outside. At a distance of 500 mm, or 0)82 wavelengths, outside the
measurement sections, there was a relatively large disagreement related to a high condition
number of a matrix.  2001 Academic Press

1. INTRODUCTION
Generation of elastic #exural waves in beams occurs in di!erent technological processes,
often as an unwanted side e!ect. In percussive drilling of rock, e.g., us e is made of elastic
extensional waves, but due to eccentric impacts, imperfect drill rods [1], unsymmetrical
loading of the drill bit [2], etc., #exural waves are also generated. This gives rise to leakage
of energy from the extensional to the #exural waves, increased stress levels, and increased
generation of noise. If the predominant wavelengths are at least of the order of the
transverse dimensions of the beam, the motion of #exural waves can be examined by using
the Timoshenko beam model. If they are much longer, the wave motion can also be studied
by using the Euler}Bernoulli beam model [3].
In various applications, it is interesting to know the histories of shear force, transverse
velocity, bending moment and angular velocity associated with #exural waves at one or
several sections of a beam. From them, histories of other important quantities such as shear
stress, de#ection, normal stress, rotation of a cross-section and power transmission can also
be determined.
For elastic extensional waves, Lundberg and Henchoz [4] showed that histories of
normal force and particle velocity at an arbitrary section of a uniform bar can be evaluated
from measured strains at two di!erent sections and by solving time-domain di!erence
equations which are exact in relation to the one-dimensional theory used. A similar method

0022-460X/01/420227#16 $35.00/0  2001 Academic Press

228 L. HILLSTROG M AND B. LUNDBERG

was used by Yanagihara [5] to determine impact force. Lagerkvist and Lundberg [6],
Lagerkvist and Sundin [7] and Sundin [8] used the method to determine mechanical point
impedance. The method was also used by Karlsson et al. [9] in a study of the interaction of
rock and bit in percussive drilling. It was extended to non-uniform bars by Lundberg et al.
[10], and this version of the method was used for determination of force}displacement
relationships for di!erent combinations of drill bits and rocks by Carlsson et al. [11] and for
high-temperature fracture mechanics testing by Bacon et al. [12, 13]. The use of the method
was extended to visco-elastic extensional waves by Bacon [14, 15], and to elastic #exural
waves by Sundin and A> hrstroK m [16] who assessed frictional properties and lubricant
performance at an obliquely impacted end of a long uniform beam from acceleration
measurements at two sections.
The aim of the present paper is to develop a general method for evaluation of the histories
of shear force, transverse velocity, bending moment and angular velocity at a section E of
a non-uniform beam from measurements of such quantities at di!erent sections A, B, C and
D. It will be shown that, for an unloaded segment of the beam, this can be achieved through
measurement of four such quantities which di!er from each other in terms of either section
(A, B, C and D) or type of quantity (shear force, transverse velocity, bending moment and
angular velocity), or both.
First, the method will be developed on the basis of Timoshenko's beam model.
Then, experimental impact tests with a non-uniform beam made of aluminium and
equipped with strain gauges and accelerometers will be presented, and comparisons will
be made between (i) bending moments and particle velocities evaluated at section E on
the basis of measurements at sections A}D and (ii) the same quantities measured at
section E.

2. THEORETICAL BASIS

2.1. FORMULATION OF THE PROBLEM

Consider a segment of a non-uniform Timoshenko beam with the cross-sectional area
A(x), moment of inertia I(x), Young's modulus E(x), shear modulus G(x) and density
o(x), where x is a co-ordinate along the straight centreline of the beam. For a rectangular
cross-section, as in the experimental part, A"=H and I"=H/12, where = is the width
and H is the height. The shear modulus is related to Young's modulus through
G"E/2(1#l), where l is the Poisson ratio. Within the beam segment, there must
be no loads, supports, joints or spots of contact. Outside, where there are no such
restrictions, the beam is assumed to interact with supports, structures and loads. The
supports and structures may have linear or non-linear responses, and they are assumed to
have the capability to absorb energy associated with vibrations. Furthermore, the loads are
assumed to act during "nite time. Otherwise, nothing needs to be known about supports,
structures or loads outside the beam segment under consideration.
The beam is quiescent for time t(0, and for t*0 it is subjected to a transverse load with
"nite duration. As a consequence, there will be a de#ection of centreline w(x,t) and
a rotation of cross-section (x, t). In the beam segment considered, these de#ections and
rotations are governed by the equations of motion

*Q *w *M *
"oA , !Q# "oI , (1)
*x *t *x *t
 ANALYSIS OF ELASTIC FLEXURAL WAVES 229

Figure 1. Angles , c , *w/*x, shear force Q, transverse velocity wR , bending moment M and angular velocity
Q at a general section xof the beam.

respectively, where Q(x, t) is the transverse shear force and M(x, t) is the bending moment.
The de#ections and rotations are related to each other through

*w
"! #c , (2)
*x 

where c is the shear strain on the centreline as shown in Figure 1. The terms ! and c on
 
the right-hand side represent the contributions to the slope of the centreline of the beam
from bending and shear, respectively, and they are related to the bending moment and the
transverse shear force through

*
M"EI , Q"iGAc , (3)
*x 

where i is a dimensionless quantity which depends on the shear stress distribution and
therefore on the shape of the cross-section. According to beam theory, this quantity can be
determined from the relation i"[(A/I)  (S/=) dA]\, where the static moment S and
in general also the width = depend on the vertical co-ordinate z from the centreline. For
a rectangular cross-section, this formula gives i"5/6+0)83, which is the value to be used
in the experimental part. According to another de"nition, i may depend also on the
Poisson ratio.
Equations (1)}(3) provide "ve relations between the "ve unknown functions Q(x, t),
w(x, t), M(x, t), (x, t) and c (x, t). Through elimination of c (x, t), these relations can be
 
transformed into the system of four "rst order partial di!erential equations

*Q *wR *wR 1 *Q *M *Q *Q 1 *M
"oA , " !Q, "Q#oI , " (4)
*x *t *x iGA *t *x *t *x EI *t
230 L. HILLSTROG M AND B. LUNDBERG

Figure 2. Unloaded section of non-uniform beam in the general case. Sections of measurement A}D and of
evaluation E.

for the four quantities Q(x, t), wR (x, t)"*w(x, t)/*t, M(x, t) and Q (x, t)"* (x, t)/*t. These
quantities constitute the elements of a state vector s(x, t)"[Q, wR , M,  ]2, which is zero for
t(0. Furthermore, because of energy absorption outside the beam segment considered,
s(x, t)P0 as tPR.
The problem to be solved is as follows. Consider the four elements Q, wR , M and Q of the
state vector s at four di!erent sections A, B, C and D of the beam segment, i.e., altogether 16
elements. Let four out of them, constituting the elements of a vector m, be known from
measurements for t*0. Then, determine the state vector s or its Fourier transform s; (which
is assumed to exist) at any section E of the beam segment. See Figure 2, where it is indicated
that the variation of the geometrical and material properties along the beam segment may
be continuous or discontinuous.

2.2. SOLUTION IN THE GENERAL CASE

Fourier transformation of relations (4) gives

s; "Rs; , (5)
where
0 iuoA 0 0
iu/iGA 0 0 !1
R" (6)
1 0 0 iuoI
0 0 iu/EI 0

is the system matrix,

QK
w(
s; " (7)
MK
Q(

is the Fourier transform of the state vector s, i.e., s; (x, u)" s(x, t)e\ SR dt, and
\
s; "*s; /*x.
The state at any section x is related to that at the "xed section x through

s; (x, u)"P(x, x, u)s; (x, u). (8)

 ANALYSIS OF ELASTIC FLEXURAL WAVES 231
Here, by equations (5) and (8), the transition matrix P(x, x, u) satis"es

P"RP, P(x, x, u)"I, (9)

where I is the identity matrix. In particular, equation (8), with x"x# and with x"x, x ,
x! and x", gives

PAE s; E"s; A, PBE s; E"s; B, PCE s( E"s( C, PDE s( E"s( D (10)

for the segments AE, BE, CE and DE.

The problem of determining the state vector s; E from the knowledge of four out of the 16
elements of the state vectors s( A, s; B, s; C and s; D can now be solved as follows. Four scalar
equations, with linear combinations of the elements of s; E in their left-hand members and
the measured elements of s; A, s; B, s; C and s; D in their right-hand members, are singled out
from the 16 scalar equations represented by equations (10). They form a system of
four linear equations for the four elements of s( E which can, at least in principle, be
solved uniquely if the determinant of this system is di!erent from zero. It should be noted
that the order of sections A}E along the beam is arbitrary, that section E may coincide with
any one of sections A}D, and also that all sections A}D need not be involved in the
measurements.
As a "rst illustration, consider the measurement of the bending moment M at each
section A}D. In this case, the third scalar equation from each of the matrix equations (10)
give the system

P# P# P# P# QK # MK 

   
P # P # P # P # wR L # MK
    " , (11)
P!# P!# P!# P!# MK # MK !
   
P"# P"# P"# P"# Q(# MK "
   

which can also be written as

P# P# P# P# sL # sL 

     
P # P # P # P # sL # sL
     "  . (12)
P!# P!# P!# P!# sL # sL !
     
P"# P"# P"# P"# sL # sL "
     

As a second illustration, consider the measurement of the transverse velocity wR at each of

sections A}D. In this case, the second scalar equation from each of the matrix equations (10)
give the system

P# P# P# P# QK # w( 

   
P # P # P # P # wR L # w(
    " , (13)
P!# P!# P!# P!# MK # w( !
   
P"# P"# P"# P"# Q(# w( "
   
232 L. HILLSTROG M AND B. LUNDBERG

which can also be written as

P# P# P# P# sL # sL 

     
P # P # P # P # sL # sL
     "  . (14)
P!# P!# P!# P!# sL # sL !
     
P"# P"# P"# P"# sL # sL "
     
In the general case, equations (12) and (14) take the form

PA# PA# PA# PA# sL # sL A

C C C C  C
PA# PA# PA# PA# sL # sL A
C C C C  " C (15)
PA# PA# PA# PA# sL # sL A
C C C C  C
PA# PA# PA# PA# sL # sL A
C C C C  C
or
Ms( E"m( , (16)
where
M "PAH# , mL "sL AH . (17)
HI CHI H CH
Here, M is a matrix with elements M singled out from the elements of the transition
HI
matrices PAE, PBE, PCE and PDE, and m( is a vector with elements mL of measured quantities at
H
di!erent sections of the beam ( j and k"1, 2, 3 or 4). The subscript e "1, 2, 3 or 4 de"nes
L H
the type of quantity (QK , w( , MK or Q , respectively) and the superscript c "A, B, C or D, the
H
section associated with mL . These quantities de"ne the measurements.
H
Equation (16) gives the state vector

s( E"Mⴚ1m( , (18)

in terms of the matrix M and the vector m( of measured quantities. Transformation into the
time domain gives the state vector sE (t) for t*0. This solves the problem.

2.3. BEAM WITH PIECEWISE CONSTANT PROPERTIES

If the beam has piecewise constant properties as indicated in Figure 3, the transition
matrices in equations (10) can be expressed as the products PAE"PsPs!12P1,2,

Figure 3. Unloaded section of beam with piece-wise constant properties. Sections of measurement A}D and of
evaluation E.
 ANALYSIS OF ELASTIC FLEXURAL WAVES 233
P "P P \ 2P , respectively, of transition matrices P , P ,2, P for beam elements with
DE p p 1 1 1 2 s

constant properties. These matrices, in turn, can be determined by "rst solving problem (9)
for matrices R which are independent of x, and then substituting appropriate values for
x and x. This procedure can be simpli"ed by replacing the coupled problem (9) for the
elements of P with an uncoupled problem as follows.
For R independent of x, equation (9a) gives

P"RP, P"R2P, P"R3P, PIV"R4P. (19)

The eigenvalues c of the matrix R are given by the four roots of "R!cI""0, i.e., with the
use of de"nition (6),
c#2ac!b"0, (20)
where



ou E oAu oIu
a" 1# , b" 1! . (21)
2E iG EI iGA

Thus, there are the two pairs of eigenvalues

c"$a, c"$ik, (22)

where
a"[(b#a)!a], k"[(b#a)#a] (23)

are both real and positive provided that


iGA 
"u"( , (24)
oI

which is presumed. According to the Cayley}Hamilton theorem, equation (20) for the
eigenvalues of the matrix R is also satis"ed by R, i.e.,

R4#2aR2!bI"0. (25)

Multiplication by P from the right and use of relations (19b, d) gives the fourth order
di!erential equation
PIV#2aP!bP"0. (26a)

Furthermore, relation (9b) and (19a}c) give the conditions

P(x, x, u)"I, P(x, x, u)"R, P(x, x, u)"R2, P(x, x, u)"R3
(26b}e)

for x"x. Thus, the coupled problem (9) for the elements of P has been replaced by the
uncoupled problem (26), which has the solution

P(x, x, u)"A cos[k(x!x)]#B sin[k(x!x)]

#C cosh[a(x!x)]#D sinh[a(x!x)] (27)

234 L. HILLSTROG M AND B. LUNDBERG

Figure 4. Experimental set-up. Dimensions in mm.

with
1 1
A" (aI!R2 ), B" (aR!R3),
a#k k(a#k)

1 1
C" (kI#R2 ), D" (kR#R3 ). (28)
a#k a(a#k)

3. EXPERIMENTAL TESTS

The experimental set-up is illustrated in Figure 4. A beam made of aluminium (AA

6061-T6; E"70 GPa, l"0)3 and o"2700 kg/m) with rectangular cross-section and
length 2250 mm was used. The width = of the beam was 20 mm, while its height H was
15 mm in the central third and 20 mm in the two outer thirds. The beam was held in
position by three supports with positions as shown. Each of these supports was realized
with a pair of 15 mm wide clamps clad with 1)5 mm thick rubber plates. The beam was
impacted laterally by a spherical steel ball which was guided by a tube and dropped from
a height of 600 mm. The diameter of the ball was 50 mm, and its mass was 535 g. A 3 mm
thick rubber plate at the impacted spot of the beam served to reduce the excitation at high
frequencies. The maximum stresses determined from strain measurements on the beam were
approximately 30 MPa which is well below the yield stress.
The beam was instrumented with strain gauges and accelerometers at sections A}D and
E }E as shown in Figure 4. The positions of the strain gauges coincided with these
 
sections, while those of the accelerometers were displaced by 20 mm to the right, i.e., away
from the spot of impact. The strain gauges (TML FLA-5-23-1L) were glued (Tokyo Sokki
Kenkyujo Co, Ltd, Adhesive CN) to the beam in pairs with one on the top and the other on
the bottom. The gauges of each pair were connected to a bridge ampli"er (Measurement
Group 2210) in opposite branches, so that the output of the ampli"er was proportional to
the di!erence between the two strains, and therefore to the bending moment M at the
section. Shunt calibration was used. Accelerometers (BruK el & Kjvr, three Type 4374 and
two Type 4393) were attached with thin layers of wax. The accelerometers were connected
to charge ampli"ers (three Kistler Type 5011 and two BruK el & Kjvr Type 2635).
The ampli"ed strain and accelerometer signals were fed to analogue aliasing "lters (DIFA
Measuring Systems, PDF) with cut-o! frequency 17)5 kHz. The "ltered signals were
recorded in a time interval [0, t ], with t +0)25 s, by two synchronized four-channel
PC PC
digital oscilloscopes (Nicolet Pro 20 and Pro 40) which used a sampling interval of 20 ls. At
the end of this time interval, the amplitudes of the recorded signals were reduced to about
 ANALYSIS OF ELASTIC FLEXURAL WAVES 235
TABLE 1
Cases of experimental tests

Output
Input High-pass
Case A}D E E E
  
"lter

1 M M M M No
2 M wR wR wR No
3 wR 2 wR 2 No
4 wR 2 M 2 Yes

a tenth due to the damping action of the supports. The recorded signals were transferred to
a computer for evaluation of the state vector sE (t). First, measured accelerations, if any, were
integrated to velocities. After use of the FFT algorithm, s( E (u) was determined according to
equation (18). Finally, s( E (u) was transformed into the time domain by use of the inverse
FFT algorithm. Results were produced for the same time interval [0, t ] although
PC
sometimes they may be valid only in a narrower interval [0,t ]. When, e.g., section E is
CT
located outside AD, as in some of the experimental tests, there is a certain di!erence
between t and t which is related to the travel times for #exural waves from section E to
PC CT
sections A}D.
Four test cases, labelled 1}4, are de"ned in Table 1. In Case 1, bending moments M at
sections A}D and E }E were determined from measurements of strains at the same
 
sections. In Case 2, bending moments M at sections A}D and transverse velocities wR at
E }E were determined from measurements of strains and accelerations, respectively, at the
 
same sections. In Case 3, transverse velocities wR at sections A}D and E were determined

from measurements of accelerations at the same sections. In Case 4, "nally, transverse
velocities wR at sections A}D and bending moment M at E were determined from measured

accelerations and strains, respectively, at the same sections. In this case, the signals
representing accelerations and strains were passed through eight-pole Butterworth
high-pass "lters with cut-o! frequency 10 Hz. In each of the four cases, bending moments
M or transverse velocities wR at sections E }E , corresponding to the measurements made at
 
these sections, were also determined from those measured at A}D according to equation
(18). All tests were carried out at room temperature.

4. RESULTS

Figure 5 shows for Case 1 the condition number cond(M) of the matrix M as a function of
the position x# of section E and of the frequency f"u/2n. It can be seen that the condition
number is low, which implies low sensitivity to errors, if section E is inside or not too far
outside the segment AD, and if the frequency is neither too low nor too high. Also, it can be
noted that the condition number has a peak (limited in height by the step in frequency) near
850 Hz. Similar results were obtained for other cases.
Figure 6 shows results in the time and frequency domains at section E for Case 1. In

Figure 6(a), the time-domain results are based on frequencies up to 500 Hz, which means
that frequencies around 850 Hz are excluded, while in Figure 6(b) they are based on
frequencies up to 2000 Hz. As there is a signi"cantly better agreement between evaluated
and measured bending moments in the former case, the results for Cases 1}4 shown in
Figures 7}10 are based on frequencies up to 500 Hz.
236 L. HILLSTROG M AND B. LUNDBERG

Figure 5. Condition number cond (M) of matrix M versus position x# and frequency f in Case 1. (a) 1 Hz
( f(2000 Hz and (b) 0)05 Hz ( f(1 Hz.

5. DISCUSSION

It has been shown how the elements Q, wR , M and Q of the state vector s at any section E of
an unloaded segment of a non-uniform beam can be determined from independent
measurements of four such elements at up to four di!erent sections A, B, C and D of the
same unloaded segment of the beam. This has also been demonstrated experimentally. Once
the state vector has been determined, several quantities of importance can be obtained from
its elements. Thus, e.g., the shear stress q can be obtained from Q, the normal stress p from
M, the de#ection w from wR , and the rotation of the cross-section from Q . Also, the power
transmission can be obtained from the relation P"!(QwR #M Q ). Nothing needs to be
known about supports, structures and loads outside the beam segment under consideration.
The combination of sections, among A, B, C and D, and types of quantities to be
measured, among Q, wR , M and Q , can be chosen in many ways, some of which appear to be
more convenient than others. Thus, it is straightforward to determine bending moment
 ANALYSIS OF ELASTIC FLEXURAL WAVES 237

Figure 6. Bending moment M# versus time t and "MK #" versus frequency f. Comparison between bending
moment at E evaluated from bending moments measured at A}D (solid curves) and bending moment measured at

E (dotted curves) in Case 1. Results for frequencies up to (a) 500 Hz and (b) 2000 Hz.


M from measured strains and transverse velocity wR from measured accelerations, as was
done in the experimental part, while it is less convenient to determine Q or Q from
measurements. Also, it seems preferable to place at most two types of transducers at any
section A}D of the beam, and to make the same kind of measurement or measurements at
each instrumented section. Therefore, two interesting possibilities might be measurement of
(i) M at each section A}D (Cases 1 and 2) and (ii) wR at each section A}D (Cases 3 and 4) as in
the experimental part. A third interesting possibility might be the measurement of (iii) both
M and wR at each of two sections, e.g., A and B. It should also be noted that a free end A, with
Q,0 and M,0, can be used to replace two measurements.
The functions a(u) and k(u) introduced in equations (22) and (23) can be interpreted as
follows. Let the state vector have the form s( "s( * exp (bx). Then, substitution into equation
(5) gives the eigenvalue problem Rs( *"bs( *. The eigenvalues b are given by the four roots of
equation (20) with b"c. Thus, according to equation (22), the eigenvalues are b"$a and
b"$ik. Therefore, provided that condition (24) is satis"ed as presumed, a determines the
decay of non-propagating (evanescent) modes and k is the wave number of propagating
harmonic waves (u'0).
For angular frequencies "u"u "c /R, where c "(E/o) is the speed of elastic
  
extensional waves and R"(I/A) is the radius of inertia of the cross-section, equations
(23) can be approximated by a+k+b and equation (21b) by b+(oA/EI)u. As the
wave number k is related to the wavelength j by k"2n/j, one obtains a+2n/j with
j+2n(c R/"u")2nR. In terms of the frequency f"u/2n and the height H"2(3R of

a beam with rectangular cross-section, the corresponding relations are " f " f "c /hH,
 
a+2n/j, and j+(c hH/" f ")hH, with h"n/(3. These low-frequency

approximations of a and j represent the limiting case of the Euler}Bernoulli beam.
238 L. HILLSTROG M AND B. LUNDBERG

Figure 7. Bending moments (a) M# versus time t and "MK #" versus frequency f, (b) M# versus time t and "MK #"
versus frequency f, and (c) M# versus time t and "MK #" versus frequency f. Comparison between bending moments
at E }E evaluated from bending moments measured at A}D (solid curves) and bending moment measured at
 
E }E (dotted curves) in Case 1. Results for frequencies up to 500 Hz.
 

For the aluminium beam used in the experimental tests, the highest frequency normally
considered, 500 Hz, corresponds to the wavelengths 0)606 and 0)525 m in the segments with
heights 20 and 15 mm respectively. Similarly, the frequency 2000 Hz, considered in
Figure 6, corresponds to the wavelengths 0)300 and 0)260 m, respectively. Thus, in the tests
carried out, the wavelengths were much larger than the heights of the beam, and " f " f .

This means that the above approximations are accurate and also that condition (24) is
satis"ed as it can be written as " f "(f [i/2(1#l)].

In the time domain, solution (18) corresponds to a deconvolution. When the condition
number of the matrix M is high (cf. Figure 5), the sensitivity of this solution to measurement
errors and numerical errors may be high. Such situations are related to the transition
matrices PAE,2, and PDE, which determine the elements of the matrix M. Therefore, it is
useful to note from equation (27) that, for a uniform beam and relatively low frequencies
with a+2n/j, these transition matrices contain terms proportional to eLV#\V H,2, and
eLV#\V"H respectively. In the discussion which follows, it is assumed that
x(x (x!(x" and that x(x#.
 ANALYSIS OF ELASTIC FLEXURAL WAVES 239

Figure 8. Transverse velocities (a) wR # versus time t and "w( #" versus frequency f, (b) wR # versus time t and "w( #"
versus frequency f, and (c) wR # versus time t and "w( #" versus frequency f. Comparison between transverse velocities
at E }E evaluated from bending moments measured at A}D (solid curves) and transverse velocities measured at
 
E }E (dotted curves) in Case 2. Results for frequencies up to 500 Hz.
 

Figure 9. Transverse velocity wR # versus time t and "w( #" versus frequency f. Comparison between transverse
velocity at E evaluated from transverse velocities measured at A}D (solid curves) and transverse velocity

measured at E (dotted curves) in Case 3. Results for frequencies up to 500 Hz.

240 L. HILLSTROG M AND B. LUNDBERG

Figure 10. Bending moment M# versus time t and "MK #" versus frequency f. Comparison between bending
moment evaluated at E from transverse velocities measured at A}D (solid curves) and bending moment measured

at E (dotted curves) in Case 4. Results for frequencies up to 500 Hz.


The matrix M may be ill-conditioned at low frequencies corresponding to long wave

lengths which make the four scalar equations, represented by equation (11), nearly the same.
It may also be ill-conditioned at high frequencies corresponding to short wavelengths which
make the magnitudes of the elements in the di!erent rows of the matrix M very di!erent.
Thus, e.g., the ratio eLV"\VH of the exponential factors of the "rst row to those of the
fourth row of the matrix M in equation (11) quickly grows with the ratio (x"!x)/j. In the
experimental tests, the distance AD (600 mm) was only slightly longer than one wavelength
(+1)12j) at the highest frequency 500 Hz, corresponding to the ratio eL+1)1;10. The
matrix M may also be ill-conditioned near discrete frequencies which make distances
between adjacent sections A}D equal to an integral multiple of a half wavelength. In Case 1,
the distances AB and BC in the thinner central section of the beam are one half wavelength
at a frequency of about 860 Hz, which may explain the high sensitivity to errors around
850 Hz illustrated in Figure 5. This problem and those at high frequencies were avoided in
the experimental tests by considering only frequencies below 500 Hz. Furthermore, the
matrix M may be ill-conditioned if section E is far outside the segment AD. This has the
e!ect that the coe$cients of the four scalar equations represented by equation (11) become
very large. The largest coe$cients, containing exponential factors eLV#\VH, occur in the
"rst of these equations. In the experimental tests, the distances AE , AE and AE were
  
approximately 0)57, 1)28 and 1)94 wavelengths at the highest frequency 500 Hz. Thus,
E was located at the centre of the beam segment AD, while E and E were located at
  
distances 0)16 and 0)82 wavelengths, respectively, outside this segment. These locations
correspond to the exponential factors eL+3)6;10, eL+3)1;10 and
eL+2)0;10 respectively.
The e!ects of noise may be reduced by use of Wiener "ltering techniques [17]. Also,
a conceivable way of avoiding integral multiples of half wavelengths between adjacent
sections A}D, which might allow frequencies considerably higher than 500 Hz, would be to
make use of more than four measurements, so that system (16) for the elements of s( E would
be overdetermined, and of a non-uniform distribution of instrumented sections. This way of
eliminating critical frequencies has been found to be e!ective in an application involving
viscoelastic extensional waves [18].
For Case 1, Figure 7 shows that there is an excellent agreement in the frequency range
2}500 Hz between (i) the bending moment evaluated at section E from measurements of

bending moments at sections A}D and (ii) the bending moment measured at the same
section E . For section E the agreement is very good in the same range of frequencies. For
 
section E there is relatively large disagreement, in particular at high frequencies

 ANALYSIS OF ELASTIC FLEXURAL WAVES 241
# 
corresponding to short wavelengths and large exponential factors eLV \V H. In the time
domain there is particularly large disagreement near t+t +0)25 s.
PC
For Case 2, Figure 8 shows that there is a good agreement in the frequency range
10}500 Hz between (i) the transverse velocity evaluated at section E from measurements of

bending moments at sections A}D and (ii) the transverse velocity measured at the same
section E . For section E , the agreement is fair in the same range of frequencies. For
 
section E , there is again relatively large disagreement, in particular at high frequencies

corresponding to short wavelengths and large exponential factors eLV#\VH, and in the
time domain near t+t +0)25.
PC
Cases 1 and 2 show that the quality of the evaluated results is generally high when
section E is located within the segment AD, whereas it rapidly decays outside. This is
consistent with the rapid increase of the condition number outside the segment AD shown
in Figure 5 for Case 1. The large disagreement near t+t +0)25 s for section E in both
PC 
cases is believed to be due to the di!erence between t and t mentioned in the
PC CT
experimental part. Thus, where the large errors occur, information from section E may not

yet have reached sections A}D to the extent required. This error can be avoided by
recording signals till all waves are damped out. Then, t and t can be considered to be
PC CT
arbitrarily large.
For Case 3, Figure 9 shows that there is a fair agreement in the frequency range
20}500 Hz between (i) the transverse velocity evaluated at section E from measurements of

transverse velocities at sections A}D and (ii) the transverse velocity measured at the same
section E . For Case 4, Figure 10 shows that there is a fair agreement in the same range of

frequencies between (i) the bending moment evaluated at section E from measurements of

transverse velocities at sections A}D and (ii) the bending moment measured at the same
section E .

It should be noted that the accelerometers used are quite inaccurate below 10 Hz. This
inaccuracy is re#ected by the disagreement between evaluated and measured quantities
below 10}20 Hz in Cases 2}4. The fair agreement in Case 4 was obtained by using the
Butterworth high-pass "lters which reduced this disagreement.
In this paper, use has been made of transition matrices which relate the state vectors (with
elements Q, wR , M and Q ) at the two ends of a beam element. As an alternative, it would be
possible to make use of the corresponding dynamic sti!ness matrices which relate the
generalized forces (Q and M) to the generalized velocities (wR and Q ) at the two ends of the
same beam element. For the beam segment considered (the structure), this approach would
result in a system Zv"F, where Z is the dynamic sti!ness matrix, v is the vector of
generalized nodal velocities, and F is the vector of generalized nodal forces. This is the
approach of the so-called exact displacement method which was developed by Kolousek
[19] in the early 1940s and implemented for plane frame analysis by A> kesson [20]. If the
number of nodes of the beam segment considered is to be n, the matrix Z would be 2n;2n
and the vector v would contain 2n generalized velocities. Also, in addition to zeroes, the
vector F would contain four generalized nodal forces (two at each end node of the beam
segment considered). Thus, there would be a system of 2n relations between 2n#4
generalized velocities and forces. If four of these quantities are measured, then generally the
system Zv"F can be solved for those which remain. Although the uses of transition
matrices and dynamic sti!ness matrices are formally equivalent, an advantage of the former
approach, which was chosen here, is that the state vector s( E, with its four elements, and the
vector m, with the measured quantities as its elements, appear explicitly in the system
Ms( E"m( . Another advantage is the connection between the wave phenomena in the beam,
the properties of the matrix M, and the accuracy of the results brought about by the
functions a(u) and k(u).
242 L. HILLSTROG M AND B. LUNDBERG

ACKNOWLEDGMENTS

Funding of this research from the Carl Trygger Foundation, the Swedish Council for
Engineering Sciences and AB Sandvik Tamrock Tools is gratefully acknowledged.

REFERENCES

1. R. BECCU, C. M. WU and B. LUNDBERG 1996 Journal of Sound and <ibration 191, 261}272.
Re#ection and transmission of the energy of transient elastic extensional waves in a bent bar.
2. I. CARLVIK 1981 International Journal of Rock Mechanics, Mining Science and Geomechanical
Abstracts 10, 167}172. The generation of bending vibrations in drill rods.
3. K. F. GRAFF 1975 =ave motion in elastic solids. Mineola, New York: Dover Publications,
Reprinted edition, 1991.
4. B. LUNDBERG and A. HENCHOZ 1977 Experimental Mechanics 17, 213}218. Analysis of elastic
waves from two-point strain measurement.
5. N. YANAGIHARA 1978 Bulletin of the Japan Society of Mechanical Engineers 21, 1085}1088. New
measuring method of impact force.
6. L. LAGERKVIST and B. LUNDBERG 1982 Journal of Sound and <ibration 80, 389}399. Mechanical
impedance gauge based on measurement of strains on a vibrating rod.
7. L. LAGERKVIST and K. G. SUNDIN 1982 Journal of Sound and <ibration 85, 473}481. Experimental
determination of mechanical impedance through strain measurement on a conical rod.
8. K. G. SUNDIN 1985 Journal of Sound and <ibration 102, 259}268. Performance test of
a mechanical impedance gauge based on strain measurement on a rod.
9. L. G. KARLSSON, B. LUNDBERG and K. G. SUNDIN 1989 International Journal of Rock Mechanics
Mining Sciences and Geomechanical Abstracts 26, 45}50. Experimental study of a percussive
process for rock fragmentation.
10. B. LUNDBERG, J. CARLSSON and K. G. SUNDIN 1990 Journal of Sound and <ibration 137, 483}493.
Analysis of elastic waves in non-uniform rods from two-point strain measurement.
11. J. CARLSSON, K. G. SUNDIN and B. LUNDBERG 1990 International Journal of Rock Mechanics
Mining Sciences and Geomechanical Abstracts 27, 553}558. A method for determination of in-hole
dynamic force-penetration data from two-point strain measurement on a percussive drill rod.
12. C. BACON, J. CARLSSON and J. L. LATAILLADE 1991 Journal de Physique III (Suppl., Colloque C3)
1, 395}402. Evaluation of force and particle velocity at the heated end of a rod subjected to impact
loading.
13. C. BACON, J. FAG RM and J. L. LATAILLADE 1994 Experimental Mechanics 34, 217}223. Dynamic
fracture toughness determined from load-point displacement on a three-point bend specimen
using a modi"ed Hopkinson pressure bar.
14. C. BACON 1998 Experimental Mechanics 38, 242}249. An experimental method for considering
dispersion and attenuation in a viscoelastic Hopkinson bar.
15. C. BACON 1999 International Journal of Impact Engineering 22, 55}69. Separation of waves
propagating in an elastic or viscoelastic Hopkinson pressure bar with three-dimensional e!ects.
16. K. G. SUNDIN and B. O. A> HRSTROG M 1999 Journal of Sound and <ibration 222, 669}677. Method
for investigation of frictional properties at impact loading.
17. T. SOG DERSTROG M 1994 Discrete stochastic systems. Estimation and control. Cambridge: Prentice
Hall International (UK) Limited.
18. L. HILLSTROG M, M. MOSSBERG and B. LUNDBERG 2000 Journal of Sound and <ibration 230,
689}707. Identi"cation of complex modulus from measured strains on an axially impacted bar
using least squares.
19. V. KOLOUSEK 1943 Der Stahlbau 16, 5}6 and 11}13. Berechnung der schwingenden
Stockwerkrahmen nach der Deformationsmethode.
20. B. A> KESSON 1976 International Journal for Numerical Methods in Engineering 10, 1221}1231.
PFVIBAT*A computer program for plane frame vibration analysis by an exact method.