The Determination of the Components of the Strain Tensor in Holographic Interferometry

by C.A. Sciammarella and R. Narayanan

ABSTRACT--The general problem of obtaining the derivatives of the displacements in holographic interferometry is analyzed. Expressions for the most general case are derived and a particular solution is suggested. A data-reduction method is proposed, Application examples are given.

Introduction
The basic problem of quantitative determination of strains in holographic interferometry has been addressed in several publications. In Ref. 1, displacements of a disk under diametrical compression are obtained by the multiple-hologram technique; the strains along the diameter are determined by graphical interpolation and differentiation. Comparison of theoretical and experimental results shows a fair agreement. In Ref. 2, the same problem is solved by means of holographic moire; the strains are obtained by graphic interpolation and differentiation. Excellent agreement is obtained between theoretical and experimental results, showing the benefits obtained by increasing the density of displacement information. In Ref. 3, a purely optical technique for the determination of strains is introduced. Since this technique implies a f'mite-difference approach, a certain amount of error will be always present in the results. An expression of the components of the gradient of the fringe-function argument is derived in Ref. 4. The general derivation of the expression for the strain components as a function of the fringe-function argument is given in matrix form in Ref. 5. Although the derivation is for a homogeneous field, the extension to a general case is immediate. Further developments along the same line are given in Ref. 6. In Ref. 7, a method is proposed to utilize the fringe-vector theory of Refs. 5 and 6 for strain analysis. A different approach to obtaining strains from holographic interferometry is presented in Ref. 8, where additional references on the same subject can be found.

The relationship between fringe localization and the fringe-function gradient are utilized to obtain the components of the strain tensor. Another approach utilizing the relationships between fringe-function derivatives and localization is given in Refs. 9-11. A solution based on the calculation of local displacement derivatives is presented in Refs. 12 and 13. The method uses heterodyne holographic interferometry for the quantitative determination of the surface-displacement derivatives.

Derivatives in Holographic Interferometry

A surface has been recorded in a hologram and a reconstruction is made. A lens system (Fig. 1) produces an image. For reasons of simplicity, magnification one is assumed; x : - x~ is the image plane, x3' being the axis of the lens. Since the image plane is the image of a plane of the object space (x~", xJ9 due to the unit magnification, one can consider both planes interchangeable. That is, (x~', x29 = (x,", x~9, except for the inversion introduced by the lens. The points of the object are referred to a localcoordinate system (x,, x,, x3). The local-coordinate system is chosen in such a way that the plane x, x2 is the tangent plane to the surface at the point under consideration, e " 3 - - n , normal to the surface and ~,, is taken parallel to t~'e x,, x3, plane. In the local system the surface is described by the equation
x, = d ( x , , x , ) (D

The local and the reference systems are related to each other by the coordinate transformation

xi

L e x] + ti

(2)

where L e is an orthogonal transformation and t,. is a translation. From the surface equation
C.A. Sciammarella (SEAl Member) is Professor, Illinois Institute of Technology, Mechanical Engineering Department, Chicago, IL 60616. R. Narayanan is Deputy Director, Structural Engineering Research Institute, CISR Campus, AD YAR, Madras, 600 020, India. Original manuscript submitted: May 19, 1983. Final manuscript received: July 19, 1984.

x; = f(x/,x;)

(3)

one can obtain L e and t,.. The fundamental equation o f holographic interferometry provides the projection of the displacement of a point P.

E x p e r i m e n t a l M e c h a n i c s * 257

= (kl-g,)-

d = g. d

(4)

__.

U r

U t

(10)

The strain tensor in the local-coordinate system is where ~ is a component of the displacement, k, and k2 the illumination and viewing vectors, . respectively, d :x, . . . . . ,,,, is the displacement of pomt P and g is the sensltwlty vector. Assuming three different sensitivity vectors of
e u = L i k L j m eke'

(11)

unit vaiue,

t~

d = r~,'~, %

(5)

Calling et, e2, e3 the base vectors in the coordinate system x/, X 2 '~ X a ,' d = u'~;

(6)

The vector ~, in the base ~i"

~i %
=

where L ik is given by eq (2). Since the local-coordinate system is contained in the tangent plane, the resulting tensor is a two-dimensional tensor. Since eq (10) corresponds to the linear-strain tensor, it does not remove the effect of rigid-body rotations. This is not a peculiarity of holographic interferometry but is common to all displacement-measuring techniques. In view of the high sensitivity of holography, this effect will be noticeable even for rotations that can be considered negligible. Elimination of the rotation effect requires the use of a nonlinear tensor. Further details concerning rotation elimination can be found in Ref. 4.

aij ~'

(7)

Substituting eq (7) in eq (5) and equating to eq (6),

u"
=

Double.beam Illumination System

(8) To evaluate eq (9), a long procedure is necessary; the whole process is subject to many sources of error. The double-illumination technique offers a simple and accurate way to obtain the displacement derivatives in holographic interferometry. Having derived the basic equations with the help of the indicial notation, we can go now to t h e more traditional notation, Cartesian coordinate system x , y, z with unit vectors i , j , k . Let us assume that x = x l , 2 = Y2, z = z3', is~th~" co~ordinate system of the image plane. Let us assume (Fig. 2) that the illumination vectors are contained in the x - z plane and that the observation vector is located in the same plane. Let us assume further that both the observation and the illumination points are at infinity. The illumination vectors are

a u ~j

Equation (8) gives the displacements of the surface points in the fixed-coordinate system. From eq (8)
U t i,k = aijUj,k + ~ljaij, k

(9)

The displacement gradients in the fixed-coordinate system are not only functions of the displacement gradients in the sensitivity-coordinate-system base, but also of the displacements and gradients of the sensitivity-vector components. From eq (9) it is possible to compute the strain tensor in the fixed-coordinate system,

X,
(

-n_

LENS OBJECT

IMAGE PLANE Fig. 1--Coordinate system in the object and the image planes

258 9 December 1984

k, = (sin 0 i k;= (-sin 0/-

cos Ok) cos 0 k )

(12) (13)

Computation of the Strain Tensor

From eqs (211, (22) and (24), we obtain Ou Ox' Ou #y '

The observation vector is

Ow . From similar equations we can get O_____Lv, #v Ox " Oy #x ' (14) Ow Oy " Knowing the surface equation or using a surface Oz Oz contour we can compute Ox ' Oy and obtain Ou Oz Ou Ox ax az (25) and

ko = ko~i + ko=k

The fundamental equation of holographic interferometry applied to each of the illumination beams yields

and

80].

o5)

Similar expressions can be applied to obtain ~

Xn2 = [ k I ko].d (16)

If we differentiate eqs (15) and (16) and recall that k, and k,' are constant, we obtain an, ako ad 'x, 'X, x ax ax " d + [k, - ko] " ax (17)

a w . All the elements necessary to apply eq (lO)~'~are at az our disposal. Finally, eq (11) will yield the locai-strain tensors.

Method of Analysis of the Fringe Patterns

The holographic patterns as other fringe patterns ''''u can be analyzed as frequency-modulated signals of the form I ( x ) = 10 + 1, cos [21r ~k(x) + or] (26)

k an2
ax

Oko ax " d + t k ,

ad ' - ko] . ax ~

(18)

If we subtract eq (18) from eq (17) we obtain Od X [.an, On2] = [ k , - k ~ ] . ~ Ox ax ax If we assume d of the form d = ui.,+ [ w + f ( x , 3 ) ] k

(19)

(20)

where I ( x ) is the intensity of the fringe pattern at the point of coordinate x , Io is the background intensity, /1 the fringe-intensity amplitude, ~b(x) is a generaiization of the fringe order considered as a continuous quantity, and a is a certain initial phase which depends on the particular problem; only the fundamental harmonic has been considered.

where u, w are the displacements parallel to the x and z axes respectively, and f ( x , 3 ) is a fictitious displacement in the z direction introduced by a rotation of the reference beam, Replacing eq (20) in eq (19) a.___~u= X [an, Ox 2 sin 0 Ox
an2] (21) Y

Ox

The above equation was introduced in Refs. 3 and 15, and it is derived here under more general conditions. If we differentiate with respect to y, a.____uu_ ~ [.On, Oy 2 sin 0 Oy On2 ] Oy (22)

If we assume that ko is perpendicular to the plane y z, eq (141 becomes ko = k

(23)

-',
0 L .,,~..

If we add eqs (17) and (18) and introduce eqs (12) and (13), we get Ow Of(x,3) O--'x--+ 0--'---7= X ran, On2 ] 2(1 + cos 0) ~ ax + Ox" (24) As the term O f (~ , 3 ) is a constant given by the initial 0 x Ow pattern, ~ can be obtained by addition of the derivatives. If the illumination vectors are in the y - z plane, similar expressions can be obtained for 0 v 0v aw a y ' a x ' 0y "
~ Z

Fig. 2--Illumination and observation vectors in doublebeam-iltumination holography

Experimental Mechanics 9 259

From eq(26),

and (30).
Io

I(x) -1 ~(x) =-T~-~ (arc c o s - L

c0

(27) and

OI ( x__~= - l---L- 11 ~ ___) . Ox 2 ~r

sin 27r r
O.X

(34)

Equation (27) cannot be directly applied because of the presence of amplitude modulations caused by factors other than displacements. The problem can be solved by introducing signals in quadrature. '7 If the coordinates of eq (26) are shifted by - / o , we get

OIq(x) 1 I, Off(x) Ox = 27r ~ cos 27r r

From eqs (34) and (35), by squaring and adding

(35)

l ( x ) = 11 cos [ 2 r if(x) + c~]

and the real signal is replaced by the complex signal

(28) and (29)

Or

27rT,1 [(~)OICx) 2 + "-'--ffx--'(Olq(x)]2]"2"

(36)

I~(X) = I(x) + ilq(X)

where

I, = [I(x) 2 + Iq'(x)l ''2

(37)

Iq(x) = I1 sin [27r r

From eqs (28) and (30),
r = ~1

+ ct]

(30)

[arc t~/__~(x)._ a] o I(x)

(31)

Equation (31) has a simple graphical interpretation (Fig. 3). The intensity profile of the fringes can be assumed to be generated by a rotating vector L that moves with variable angular speed. There are two separate pieces of information in the signal: the signal amplitude and the phase. The phase and amplitude information are separated by introducing the in-phase in in-quadrature signals. From Fig. 3,

Equation (31) provides the fringe order as a continuous quantity rather than as a discrete quantity defined for integer of half-integer values. Once the fringe order is known, its derivatives can be obtained by applying either eq (33) or (36). Replacing the fringe-order derivatives in eq (21), we can obtain the strains. If the computations are repeated for the other two directions, rosette equations can be used to obtain the principal strains. The great advantage of the proposed method is that it lends itself to automation.

Data Processing
Equation (26) provides an ideal signal. The actual signal is of the form

I(x) O(x) = 2~r~b(x)- c~

(32) The angular speed of the vector gives the derivatives of the generalized fringe order
o,k(x) =
1

Io(x) + I,(x) cos [2r~b(x) - c~] +/,o,,,(x)

(38)

,OO(x)

(33)

Ox

21r

Ox

The derivative can be also obtained directly from eqs (28)

The background intensity lo is not a constant, but changes from region to region. The amplitude I, changes with the point position since the contrast of the fringes changes from region to region. Furthermore, there are many sources of noise that corrupt the signal. In the process of recovering the phase information or the

I(*)
e(x)

Fig. 3 - - V e c t o r i a l representation of a frequency-modulated signal

~--- X

e(x) = x (,U(x)
mo L_

260 9 D e c e m b e r 1984

derivative of the phase, it is necessary to remove as much of the spurious information as possible. This can be achieved by numerical filtering. The numerical filters operate on the signal and produce a filtered version of the signal or a filtered version of the derivatives. The signal is band limited; this means that the spectrum will cover a limited range of frequencies. The changes of background intensity have a much larger period than the fringes, and the noise in most cases has a frequency composition that is also removed from the frequency composition of the signal. Consequently, a band-pass filter can be utilized to filter out the unwanted portions of the signal. The process of filtering can be combined with the generation of in-quadrature signals, thus making it possible to obtain either the phase or the derivative of the phase independently from the amplitude changes. The proposed method is a method of signal detection and should not be confused with techniques of fringe interpolation such as the spline method or any other type of polynomial fitting. As a matter of fact, these methods can be employed in conjunction with the proposed method once the phase information has been retrieved and derivatives are sought--or if the derivatives have been obtained to get higher order derivatives in place of applying higher order derivative filters. The proposed scheme does not only retrieve the information, but also reduces measurement errors. Since a process of filtering has an effect on the signal component that provides the desired information, it presents the need of a careful analysis to strike a good compromise between the removal of unwanted noise and distortion of the data.

where the prima indicates the derivative (b~) and (b~)p can be obtained from bp and (bq)p by the Fourier shift method. The values of I~ and (I,;)k can be substituted in eq (36) to obtain the instantaneous frequencies.

Computer Program
A computer program has been developed for band-pass smoothing and first-derivative filtering. Martin-Graham filter weights have been utilized (see Appendix). In addition to the smoothing and first-derivative filtering, a modification incorporating phase determination and differentiation through the use of spline interpolation has been added. The various options are automatically introduced by data indicators. The filter properties are determined by the parameters: central frequency coo, cut-off frequency wc and termination frequency cot (Fig. 4). The difference coo- coc gives the band pass of the filter. In order to determine the values of these parameters the following procedure ",t8 is applied. The instantaneous angular frequency for a general angular-frequencymodulated signal is given by
i=oo

co(x) = coo +

]~ Acoi COS coralx

i=1

(43)

where At0,. are the angular-frequency changes corresponding to each modulating harmonics. If we assume a single dominating harmonic (Fig. 3) for a region of the pattern, calling X. . . . X,,i, the maximum and minimum wavelengths in the region and X,,,odthe wavelength of the dominating harmonic, we have
COrn = ~mod

Numerical Filters for Band-pass Smoothing and Band.pass-smoothing First-derivative Filters

The subject of numerical band-pass filters has been dealt with in Refs. 16-18. Some basic notions will be reviewed here. We work with a sampled version of I(x) produced by a scanner which we call I t , k being an integer k = 1 . 2 . . . m. Since I(x) is a band-limited function, the sampling frequency f , is selected in such a way that fs>2fe where fe is the highest frequency contained in l(x). The expression for digital tering of the sampled version of a signal in phase and quadrature is

2x

(44)

The angular-frequency excursion is given by Ace = ~r( 1

k,,,i.

1 .) Xma~

(45)

The angular central frequency is

(]_L_ + 1___)
(DO ~--- ~ Xmax Xmin

(46)

The cut-off frequency is given by

coc = ~o co,.oa

(47)

L=
(,~)k =

p= -M

p=M ~ b.!t,~,
r, (bq). *t+.

(39)

p=M p= - M

(4O)

B(w)
1

bp is the band-pass-smoothing filter weight in phase. (bq)p is the corresponding weight for the in-quadrature signal. M is the number of filter weights in one side of the central frequency. The resulting phase can be obtained by applying eq (31). In the case of the smoothing and first-derivative filters, only the filter weights have to be changed.
p=M

I~

~ bp If,+p p= - M p=M

(4])

( i,~)k =

~ (bq)oI~+o p= - M

(42)

I>0kl UJo

Wc WT

Fig. 4--Martin-Graham band-pass s m o o t h i n g filter

Experimental Mechanics 9 261

and ~ called modulation index is = ~--~(~

)~min

1 ) ~f~.rNax

(48)

Knowing o~mand 6, it is possible to have an idea of the spectral distribution of a frequency-modulated signal, and consequently, of the required band-pass width. For the type of fringe patterns under consideration, in order to filter out the noise without distorting the signal, it is necessary to change O~o and A~0 following the pattern changes. This process is called filter tuning. Filter tuning is accomplished automatically by the filter. The program scans the wavelengths, determines kin=x, kmi,, ),~d and divides the pattern into regions where the changes in modulation index are less or equal to five percent. Once OJoand Ao are known, o~c is determined and the termination frequency is calculated by the following expression:
~T = o~ + A~o, (49)

where A ~ , is the roll-off transition required to take care of the Gibb's phenomenon. The quantities OJo, ~0c, ~0~and ~, sampling frequency are related to the optimum number of filter weights. The optimum number of filter weights is the number of filter weights that passes a given frequency with minimum distortion. A parametric study was carried out to establish the relationship among the above variables and M. The result of this study has been incorporated in the program so that the optimum values are selected in each case.

power input is provided by a power supply to maintain stable intensity. The intensity is measured by a Reticonpin-type diode compatible to a photometer amplifier. The area whose intensity is measured is imaged by a microscope with 10x magnification on the diode. The film or plate holding the pattern rest on a movable carriage. The carriage motion is imparted by a finethreaded rod provided with a flexible nut to average pitch errors. The motion of the carriage defines the c o ordinate positions accurately. The output can be observed on a strip-chart recorder and can be digitized, The analog to digital conversion is achieved with the help of a digital PDP-11 minicomputer. The data are filed on a floppy disk; the floppy disk data are then fed into the mainframe computer. The specimen is a cantilever beam of cross section 11.5 x 12 mm and length 116 ram. A concentrated load of 35.6 N is applied at 63 mm from the fixed end. The beam was calibrated through the measurement of deflections under applied loads. Figure 6 shows the experimental setup. The patterns were photographed from the monitor. Figure 7 shows the left and right patterns. These two patterns were differentiated by utilizing first derivative and smoothing filters. Figure 8 shows the obtained results which are in excellent agreement with the theoretical

values.

Conclusions
The technique described in this paper provides a direct solution to the problem of obtaining strains from h o l o graphic interferometry. It eliminates the need to record several holograms as well as the tedious processing required to obtain, besides the gradient of the displacementvector, the displacement-gradients of the sensitivity vectors. Combining the double-beam illumination with the digital method of signal detection is shown to improve the resolution and accuracy that can be achieved in obtaining the displacement derivatives. To solve the problem of studying the complete strain field in a general threedimensional object, four beams and a contouring technique are necessary. In this sense, the experiment described in the paper is limited, since it does not illustrate all the

Experimental
To test the feasibility of the proposed method to obtain the displacement derivatives in holographic interferometry, the methodology previously described was implemented. First, a fringe scanner was designed and built (Fig. 5). The scanner is essentially a moving-sample, transmission-type microdensitometer. The light source consists of a subminiature lens end lamp producing a 1/16-in.-diam spot of 350-ft candle intensity. Regulated

Fig. 5--Schematic
r e p r e s e n t a t i o n of the automatic f r i n g e s c a n n e r PS

LI
,JL~ SUPPLY DETECTOR CHART CL CARRIAGE ILLUMINATION FM RECORDER

SC ..........I
PM PHOTOMETER FOCUSSING MICROSCOPE

PS

POWER PHOTO STRIP

PD SC -

262 9 December 1984

different aspects of the theory that has been derived. However, the main objective of the experiment has been accomplished since it demonstrates the feasibility o f accurately obtaining strains from holographic patterns and provides a check against known results. The same technique of pattern differentiation has been applied to circular tubes. The corresponding results are reported in another paper. Since this paper was written, the contouring technique described in Ref. 19 has been imple-

mented. This technique utilizes the same optical system described in this paper as well as the same computer program. Consequently all the necessary elements for a more general case are available; the solution of a general case will be the object of future research. The improvements described and implemented in the research work reported in this paper can be considered the first step towards the goal of achieving fully automated systems for holographic-fringe analysis.

VC L

0
O - OBJECT
L -

TVC

Fig. 6 - - E x p e r i m e n t a l setup showing the T.V. imaging system and video recording

F. g

TV
IB - I L L U M I N A T I O N RB
-

BEAM

M - MIRROR TVC
-

IMAGING LENS

REFERENC BEAM VIDEO CASSET

TV CAMERA C - CAMERA

TV - VIDEO MONITOR

VC -

RECORDER

A.

LEFT

BEAM

PATTERN

B.

RIGHT

BEAM

PATTER~

Fig. 7 - - L e f t and right fringe patterns of cantilever beam under concentrated toad

Experimental Mechanics 9 263

35

THEORY
NTAL

30
r I

0 X 25 Z
< rr" I.CD

20

15

10 .. ~5

26

2~

28

29

30

31

32

33

34

35

36

37

DISTANCE IN mm FROM SUPPORT Fig. 8--Comparison oftheoreticaland experimentalstrainsobtained bydigitaldetection

References
1. Sciammarella, C.A. and Gilbert, J.A., "'Strain Analysis of a Disk Subjected to Diametral Compression by Means of Holographic Interferometry, "'Appl. Opt., 12, 1951-1956 (Aug. 1973). 2. Sciammarella, C.A. and Gilbert, J.A., "',4 Holographic Moir~ Technique to Obtain Separate Patterns for Components of Displacement," EXPERIMENTALMECHANICS, 16 (6), 215-220 (June 1976). 3. Sciammarella, C.A. and Chawla, S,K., "'Holographic Moir~ Technique to Obtain Displacement Components and Derivatives," EXPERIMENTALMECHANICS, 18 (10), 373-381 (Oct. 1978). 4. Bigl, D. and Jones, R., ",4 New Theory for the Practical Interpretation of Holographic Interference Patterns Resulting From Static Surface Displacements, '" Opt. Acta, 21 (2), (1974). 5. Stetson, K.A., "'Homogeneous Deformations: Determination by Fringe Vectors in Hologram lnterferometry, "Appl. Opt., 14 (1975). 6. Stetson, K.A., "'The Relationship Between Strain and Derivations of Observed Displacements in Coherent Optics Metrology," Appl. Opt., 14 (1975). 7. Pryputniewies, R.G., "Holographic Strain Analysis: An Implementation of the Fringe Vector Theory, ""Appl. Opt., 17 (1978). 8. Schumann, W. and Dubas, M., "Holographic Interferometry,'" Springer Series in Optical Sciences, Springer (1979). 9. Charmet, J.C., "lnterferometrie holographique appliqu~e b I'analyse des structures sous contrainte: Determination de l'etat de deformation, en particulier par l'etude du contraste de l'interferogramme, "" Thesis, Universit~ P. and M. Curie, Paris (1977). 10. Charmer, J.C., "Interferometrie holographique sur objets dispersant: Application de la mesure du contraste It la determination des gradients de displacement, ""Rev. Phys. Appl., 12 (1977). 11. Ebbeni, J. and Charmet, J.C., "Strain Components Obtained from Contrast Measurement of Holographic lnterferometry Patterns, '" Appl. Opt., 16 (1977). 12. Dandliker, R., Ineichen, B. and Mastner, T., "'Quantitative Strain Measurement through Holographic lnterferometry,'" 1977 SESA Spring Meeting, Dallas, TX (May 1977). 13. Dandliker, R. and Eliasson, B., "'Accuracy of Heterodyne Holographic Strain and Stress Determination, "" EXPERIMENTALMECHANICS, 19 (3), 93-101 (March 1979). 14. Durelli, A.J. and Parks, V.J., "'Moir6 Analysis of Strain," Prentice-Hall (1970). 15. Sciammarella, C.A. and Chawla, S.K., "Holographic Moir~ Technique to Obtain Displacement Components and Derivatives, "' Mech. Res. Comm., 4 (5), (1977). 16. Sciammarella, C.A. and Sturgeon, D.L., "'Digital Filtering Techniques Applied to the Interpolation o f Moire, '" EXPERIMENTAL MECHANICS, 7 (11), 468-475 (Nov. 1967). 1Z Sciammarella, C.A., "'.4 Numerical Technique of Data Retrieval from Moir~ or Photoelastic Patterns," Proc. SPIE Seminar-in-depth Pattern Recognition Studies, 18 (1969). 18. Sciammarella, C.A. and Ronlands, E., "Numerical and Analog Techniques to Retrieve and Process Fringe Information," Proc. 5th Int. Conf. in Experimental Stress Analysis, Udine, Italy (1974).

19. Sciammarella, C.A., "Holographic Moirt, an Optical Tool for the Determination o f Displacements, Strains, Contours, and Slopes of Surfaces," Opt. Eng., 21 (3), (1982).

APPENDIX: Martin-Graham Smoothing and First Derivative Filter

Transfer Function

I H( )l

~',tO

0 I'U C UJT Fig. (A1)--Martin-Graham smoothing and first-derivative filter

Filter Weights
h'=
60s k r c 0 s ( 2 7rkrn) + ~ c o s ( 2 x h c n ) - h , ( 1 - 12k~n 2) n(1 - 4 h ~ n 2) n=0 X.)]

hg=O

h . = c o s ( n l r k . ) s i n [ n lr(2 ~ + n ir(1 - 4 h i n 2) hm = 2 k ~ s i n [ r ( 2 ~ + h " ) ] 2 )~R ho = 2 h c + k ,

m =

1 2h.

n = 0

264 = December 1984