Composites: Part B 27B (1996) 185-191

Copyright © 1996 Elsevier Science Limited

Printed in Great Britain. All rights reserved
ELSEVIER
1359-8368(95)00050-X 1359-8368/96/$15.00

Characterization of material parameters of

composite plate specimens using
optimization and experimental
vibrational data

A. L. Ara0jo, C. M. M o t a Soares and M. J. M o r e i r a de Freitas

IDMEC--Instituto de Engenharia Mec~nica, /ST, Av. Rovisco Pais,
1096 Lisboa Codex, Portugal
(Received 26 July 1995; accepted 16 October 1995)

A numerical/experimental method for identification of material parameters of composite materials is

proposed in this paper. This method combines experimental techniques for the determination of vibration
eigenfrequencies of plates made of composite materials along with a finite element numerical method for the
determination of the corresponding numerical eigenfrequencies. The identification process makes use of
optimization techniques, through the minimization of an error measurement that estimates the deviation
between numerical and experimental eigenvalues, for a given set of material parameters. The accuracy of the
proposed technique is discussed through test cases.

(Keywords: identification of material parameters; composite materials; eigenfrequencies; identification of modal constants;
finite elements in plates; optimization)

INTRODUCTION theory. Both use a numerical model based on the

Rayleigh-Ritz approach. Other research works that
The analysis of structures through numerical methods is also use the Rayleigh-Ritz method along with the
becoming a general practice, making use of the. fast classical plate theory were presented by Wilde and Sol 5
development in personal computers. There are two and, more recently, Sol 6 and Wilde 7, using Bayesian
important parameters in the analysis: the geometric estimation to determine material moduli instead of
properties that define the structure in space, and the optimization techniques.
mechanical properties of the materials that make up the A combined numerical/experimental method for the
structure. The determination of material parameters of determination of six elastic material parameters of thick
complex materials such as fiber reinforced composites composite plates is presented. This technique may be used
presents serious difficulties when applying traditional test in composite plates made of different materials and with
methods. Problems such as boundary effects, sample size general stacking sequences. It makes use of experimental
dependencies and difficulties in obtaining homogeneous plate response data, corresponding numerical analysis
stress strain fields often arise. results and optimization techniques. The plate response is
To overcome these difficulties alternative methods a set of natural frequencies of flexural vibration. These
have been proposed by some authors. These methods experimental eigenfrequencies are measured by the
(also known as inverse methods or identification impulse technique on a completely free plate.
methods) usually combine numerical and experimental The numerical model is based on the finite element
data with the aim of determining material parameters of method using a higher order plate theory that can be
plate specimens. found in Kant and Kommineni s. Full details regarding
This work is a generalization of the work presented the model development and implementation for
by M o t a Soares e t al. 1'2 which used the Mindlin plate dynamics can be found in Arafijo 9.
theory along with a finite element model. It is closely The model is applied to the identification of the
related to the developments presented by Pedersen 3 for elastic moduli of the plate specimen through optimiza-
thin plates using the classical plate theory and tion techniques, by minimizing the difference between
Frederiksen 4 for thick plates using a higher order experimental and numerical eigenvalues.

185
Material parameters of composite plate specimens: A. L. Aradjo et al.

Figure 2 Orthotropic laminae. Material (1,2) and global (x, y)

Figure 1 Plate dimensions and global coordinate axis (x, y, z) coordinate axis

NUMERICAL MODEL The non-dimensional parameters in equation (3) are

defined as1°:
Figure 1 shows a rectangular plate of constant
thickness h and plane dimensions a and b. A Cartesian OQ = 4 - 4 E2/ E1
coordinate system (x, y, z) is located at the middle
c~3 = 1 + (1 - 2Ulz)Ez/E 1 - 4~oG12/E 1
plane as shown.
The assumed displacement field for the numerical ~4 = 1 + (l + 6Z)Iz)E2/E 1 - 4o~oGlz/E 1 (4)
finite element model is a third order expansion in the
ol8 = 4(G13 + Gz3)O~o/EI
thickness coordinate for the in-plane displacements and
a constant transverse displacement: O~9 = 4(G13 - G23)Ozo/E 1

u(x,y, z, t) = uo(x,y, t) + zOx(x,y, t) + z2u;(x,y, t) where So = 1 - u22E2/EI and El, E2 are the Young's
modulis in 1,2 direction (Figure 2), G12, G23 and GI3 are
+ z30*(x,y, t) the transverse shear moduli in planes 1 2, 2 - 3 and 1 3
(3 perpendicular to 1 2 plane), respectively, and Ul2 is
v ( x , y , z , t ) = Vo(x,y,t) + zOy(x,y,t) + zZv~(x,y,t) (1)
the major Poisson's ratio.
3 * (X, y, 1)
-Jr-,70y Assuming small displacements and using Hamilton's
variational principle along with an eight-node isopara-
w(x,y,z, t) = Wo(x,y, t) metric plate element with nine degrees of freedom per
node, corresponding to the nine expansion terms in the
where u0, v0 and w0 are the in-plane displacements in
displacement expression (1), we obtain the following
the x, y and z directions and Ox and 0y are the rotations
equilibrium equation for free harmonic vibrations at the
of normals to the midplane a b o u t the y axis (anti-
element level:
clockwise) and x axis (clockwise), respectively. The
functions u0, v0, 0x and 0y are higher order terms in the K e a,e e
-- Ai M
e e
ai : 0 (5)
T a y l o r series expansion, defined also in the midplane of where M e and K e are the element stiffness and mass
the plate.
matrixes, respectively, and a7 are the eigenvectors
As for the constitutive relations, considering a plane associated to the eigenvalues A~, at the element level.
stress analysis and a linear elastic material, it can be
Second degree serendipity shape functions were used to
shown that, for a laminae of orthotropic material in the interpolate the displacement field within an element 11.
(1, 2, 3) coordinate system (Figure 2), the stress-strain
The equilibrium equation for the whole discretized
relation is: plate assumed flee in space is, therefore12:
O'(1,2,3 ) =- Q e ( l , 2 , 3 ) (2)
(K + 13M)ai - (Ai +/3)Mai = 0 (6)
with the following constitutive matrix: where K and M are the stiffness and mass matrixes of the

8 0; 4 -- OL3 0 0 0
El °~4 -- a3 8 -- 2 a 2 0 0 0
Q = ~0 0 0 c~8 - c~9 0 0 (3)
0 0 0 O~8 -}- O~9 0
0 0 0 0 1 (8 -- O~2 -- 3 a 3 -- o~4)

186
 Material parameters of composite plate specimens: A. L. AraEijo et al.

plate, and ai the eigenvector associated to the numerical /I/I/1/I//Hill,

eigenvalue Ai. In order for the stiffness matrix to be / ~ Impact \ Plate
positive definite, a shift tim is applied to it (usually/3 is of SignalAnalyser Ha~er~ t~
the same magnitude as the first non-zero eigenvalue).
u /Microfone
I I, 01-- ~'~

EXPERIMENTAL METHOD \
The experimental eigenfrequencies of a completely free
Figure 3 Schematic representation of experimental setup
plate are obtained using the experimental setup shown in
Figure 3.
In practice a truly free support can not be provided. Table 1 Curve fitting results ( F R F from Figure 4)
However, a suspension which closely approximates this
condition can be achieved by supporting the test plate on Mode Frequency (Hz) D a m p i n g (%)

light elastic bands so that the rigid body modes have very low 1 86.37 2.18
natural frequencies in relation to those of the bending modes. 2 108.54 2.37
3 224.98 2.43
These elastic bands were simply glued to the plate edge 4 239.57 1.82
(approximately at the mid points of two neighbor edges). 5 319.83 0.89
The simplest and fastest way to excite the vibration 6 414.51 1.00
7 483.07 1.05
modes of a plate is the impulse technique. The hand held 8 520.76 2.00
hammer Brfiel & Kj~er (B&K) 8202 was used to impact 9 606.59 1.22
the plate. The force transducer B&K 8200 attached to the 10 677.22 1.33
11 702.26 0.38
hammer head measures the input force. 12 745.38 0.90
A condenser microphone B&K 4165 was used to
measure the plate response. The microphone is placed in
a small table stand perpendicular to the plate at a error caused by the truncation of the sampled time signal.
distance of approximately 0.01 m. The signals are fast Fourier transformed and divided to
The analog signals from the impact hammer and obtain the frequency response function (FRF). All of these
microphone were fed into a dual signal analyzer B&K steps are handled by the signal analyzer.
2032 provided with an IEC 625-1/IEEE Standard 488 The digital process described above results in a discrete
interface card. Both signals are low-pass filtered, sampled spectrum which contains a finite number of values for the
and subjected to appropriate time weighting functions. The FRF. The F R F is generally a complex valued function and
weighting is important to reduce noise and avoid a leakage a peak usually indicates the presence of a natural frequency,

60

44

28

12-

-4

-20 I I I
200 400 600 800
Frequency (Hz)
Figure 4 Example of an experimentally measured F R F

187
Material parameters of composite plate specimens: A. L. AraOjo et al.

however, the frequency of maximum response is not an where °E I is the initial guess for the Young's moduli in
accurate estimate of the natural frequency. First of all, the direction 1, for material i.
spectrum is discrete with a certain resolution and the peak The objective function is a weighted least squares
value may not rely entirely on a single point. The second error estimator, expressing the difference between the
most serious reason is that neighboring modes contribute a measured eigenvalues of the plate specimen and the
certain amount to the total response at the resonance of the corresponding eigenvalues of the numerical model, i.e.:
mode being analyzed. Finally, the third reason is that the
values of the F R F include damping, which affects slightly I (~i ~ )~i~2
~ = ~-'~wi (9)
the resonant frequencies. To deal with these problems,
refined modal analysis methods have been developed. In
this work the rational fraction polynomial (RFP) method as w h e r e /~i = 03~ are the experimentally obtained eigen-
described in Richardson and Formenti 13 was used for the values and "~i = ~M2are the eigenvalues of the numerical
curve fitting process on a personal computer suited with an free simulated composite plate model, with I the total
IEEE Standard 488 interface card. number of eigenvalues considered in the analysis and wi
Figure 4 is an example of a measured FRF. Table 1 the weight factors, used to express the confidence in a
contains the corresponding frequency and viscous particular experimental eigenvalue, assuming values
damping values which were extracted by the R F P curve between 0 and 1. The degree of confidence in a particular
fitting method. experimental eigenfrequency is determined by the modal
analysis described in the previous section. In this work
the weight factors wi were all set equal to unity because
all the natural frequencies were very well determined by
THE OPTIMIZATION PROBLEM
the modal analysis technique.
The use of experimental vibrational data in order to For the material's constitutive matrix (3) to be
obtain the material parameters is carried out through positive definite some constraints g(p) have to be
the minimization o L a n error function • (~>0) over imposed on the material parameters for each material i:
f lT MT'~I
p=~c~ ,...,& ~ , where M is the total number = - < .<o
of Materials present in the specimen and 6 d =
{ ~ ] , . . . , c~} T is the vector of material parameters for gi=
material i. These parameters, for a given material i, are c~ - 4
related with the usual mechanical composite material
moduli through the following relations: g~= 8-~i-3a~-~<0

El= °el< -i -i i ~
i Ol4 -- OL3 _
Q/E = (4 - 64)/4 g4 = 8 -- 2~I- ~<0 (10)

G]2/E] = (8 - c~ - 38~ - c~)/16c~ -i

• = --0% ~0
(8~ - c ~ ) / ( 8 - 28~) (7) -8~
• -i -i
G]3/E i = ( ~ ; -~- O%)/80~ 0 i +
g6 -- ~<0
(~,~ __ - i -i
6 3/EI = O%)/80~0
The constrained minimization is carried out using the
non-linear mathematical programming algorithms of
program ADS 14. A two step strategy was adopted: in the
-i 1 (0~ -- ~ ) 2
O~0 first step the modified feasible directions (MFD) method
16(4 - ~ )
was used on the first three iterations and the second step
where the non-dimensional parameters present in (7) are makes use of the B r o y d e n - F l e t c h e r - G o l d f a r b - S h a n n o
related to the non-dimensional parameters in (4) by: (BFGS) method. A detailed description of these methods
i
can be found in Vanderplaats 15.
The optimal material parameters are achieved when
the difference in the objective function between two
consecutive iterations is less than a certain quantity s.
The sensitivities of eigenvalues with respect to
= (8) constitutive material parameters of the laminate plate
layers makes use of objective function derivatives in
< = <
order to determine variables p:

0"_ 2Zwi 1- (11)

= Op i=1 \ ~i Op

188
 Material parameters of composite plate specimens. A. L. Ara6jo et al.

Table 2 Material moduli results (Example 1) Table 6 Experimentally measured frequencies and residuals (Example 2)

Present Frederiksen 4 i ~bi (Hz) r i (%)

Identified Identified* Strain gauge 1 96.31 0.32
2 163.18 0.28
E 1 (GPa) 138.2 137.5 133.3 5:4.9
3 202.66 0.29
E2 (GPa) 123.4 122.0
Gl2 (GPa) 49.2 47.5 4 252.17 -0.63
5 285.23 0.10
GI3 (GPa) 22.9
G23 (GPa) 21.4 -- -- 6 467.86 0.08
7 474.70 -0.13
1212 0.262 0.257 0.244 + 0.016
8 540.02 -0.15
* Classical Plate Theory (CPT) 9 549.50 -0.33
10 632.63 0.15

Table 3 Experimentally measured frequencies and residuals (Example 1)

Table 7 Identified material moduli (Example 3)
i ~3i ( H z ) r i (%)

1 564.51 -0.03 Present Frederiksen 4

2 837.01 -0.02
E l (GPa) 42.8 42.7
3 1026.2 0.02
E2 (GPa) 12.2 12.4
4 1427.7 0.17
G12 (GPa) 4.8 4.8
5 1491.0 -0.12
G13 (GPa) 4.2 5.1
6 2440.1 -0.04
G23 (GPa) 4.9 4.0
7 2638.4 -0.03
//12 0.301 0.277
8 2704.6 0.09
9 2882.0 0.00
10 3242.0 0.09
Each identification set of results is compared with other
authors and with strain gauge measurements, when
Table 4 Identified material moduli (Example 2) available.
Present Frederiksen .4 The value o f e used in the stop criterion defined in the
previous section is (for all the examples) 10 -9 for the
E 1 (GPa) 118.8 118.5
E 2 (GPa) 9.1 9.1
B F G S method. Weight factors wi = 1 were used in all
G12 (GPa) 5.6 5.4 examples because full confidence was achieved for all
GI3 (GPa) 10.7 -- natural frequencies.
G23 (GPa) 10.7 --
u12 0.318 0.304

* Classical Plate Theory (CPT) Example 1

Plate made of SiC fiber/calcium-alumino-silicate

Table 5 Strain gauge results for specimens made of the material of (CAS) glass-ceramic matrix, with all fibers oriented at
Example 2. Linear regression from 0% strain to varying end ~train
level 4 90 ° was studied. The plate dimensions and density are:
End strain level (%) 0.05 0.1 0.2 0.5 a = 0.1594m, b = 0.1554m,
E I (GPa) 120.4 122.7 125.8 129.5
h = 0.00317 m, p = 2696.0kg/m 3
The initial guess for the elastic constants corresponds to
a typical steel:
where the components of O A i / O p a r e :
°E 1 =°E 2 = 200 GPa; °G12 = ° G 2 3 = ° G 1 3 = 80 GPa;
0
OAi_ ,~-E,a~T0Ke ae (12) /J12 ~---0.3
Results are presented in Tables 2 and 3 and are in good
where N E is the total number of elements and OKe/Opj is agreement with those proposed by Frederiksen 4.
obtained analytically.
Example 2

Carbon fiber reinforced epoxy plate made of unidirec-

APPLICATIONS
tional fiber plies with stacking sequence [0°, 4- 602°]s was
studied. The plate dimensions and density are:
The following examples illustrate the application of the
method described above. F o r each example the experi- a = 0.2227 m, b = 0.1851 m,
mental eigenfrequencies are presented, as well as the h = 0.00126m, p = 1540.3 kg/m 3
corresponding residuals at the optimum:
The initial guess for the elastic constants corresponds to
COi -- 03i a typical prepreg layer of Structil CTE 2 45R367
ri -- - - (13)
c0i unidirectional carbon with 50% fiber volume:

189
Material parameters of composite plate specimens." A. L. AraEijo et al.

Table 8 Experimentally measured frequencies and residuals (Example 3) °E 1 = 45 GPa; 0E 2 • 4.5 GPa;
i o~1i (Hz) ri (%) °G12 =°G23 =°G13 = 3.7 GPa; 0u12 = 0.28
1 812.45 0.19 Results are presented in Tables 7 and 8. The low-level
2 1589.3 0.08 residuals and the excellent agreement with the results
3 1947.34 0.02
4 2245.4 0.13 reported by Frederiksen 4, also using a higher-order
5 2459.6 -0.31 theory, is noted.
6 3908.8 -0.04
7 4167.3 0.01
8 4638.8 -0.08
9 4963.6 0.09 Example 4
10 5397.8 0.30
11 6168.4 0.15 Carbon fiber reinforced epoxy plate made of unidirec-
12 6640.8 -0.26 tional fiber plies with stacking sequence I0°]8 was studied.
The plate dimensions and mass are:

Table 9 Material moduli results (Example 4)

a = 0.24275 m, b = 0.1525m,
h = 0.00339m, p = 1531.5 kg/m 3
Present Frederiksent
Identified Strain gauge* Identified:~ The initial guess for the elastic constants corresponds to
a typical prepreg layer of Structil CTE 2 45R367
E 1 (GPa) 107.8 113.2 107.1
E2 (GPa) 8.3 7.4 8.3
unidirectional carbon with 50% fiber volume:
Gl2 (GPa) 4.2 -- 4.2
G13 (GPa) 4.4 -- -- °E 1 = 117.25 GPa; °E 2 = 8.8 GPa;
G23 (GPa) 5.3 °G12 ----°G23 = ° G 1 3 = 3.1 GPa; °u12 = 0.35
ut2 0.421 0.348 0.246
Results are presented in Tables 9 and 10. Once again,
* Only one experiment (no standard deviation available)
low level residuals might indicate a good identification.
t Results kindly reported by P. S. Frederiksen (Technical University of
Denmark) Identified results are in good agreement with those
Classical Plate Theory (CPT) kindly supplied by Frederiksen. As for the strain gauge
measurements, they also agree with the identified
results, in spite of their standard deviations not being
Table 10 Experimentally measured frequencies and residuals
(Example 4) available.

i a3i (Hz) ri ( % )

l 161.23 0.04 CONCLUSIONS

2 347.14 -0.07
3 476.93 -0.14
4 495.69 0.35
A non-destructive numerical/experimental method for
5 589.09 0.22 the identification of up to six elastic constants on
6 895.68 0.08 composite plate specimens has been discussed. The
7 957.08 0.42
8 1074.0 -0.38
method is based on a numerical higher-order finite
9 1363.0 -0.39 element model as well as on experimental non-destructive
10 1444.0 -0.19 free vibration analysis.
11 1462.0 0.05
The mechanical properties evaluated through this
technique are valid on average for the entire specimen,
whereas the classical strain gauge measurements are only
°E~ = 117.25 PGa; °E2 = 8.8 GPa; valid in a specific point of the test specimen.
°G12 = ° G 2 3 = ° G I 3 = 3.1 GPa; °ul2 = 0.35 The mechanical properties El, E2, G12 a r e always
Results are presented in Tables 4-6 which are in good evaluated without major discrepancies.
agreement with those presented by Frederiksen 4. Some discrepancies are found in the values of the
transverse shear moduli G13 and G23. This is due to the
fact that transverse shear deformations are only notice-
Example 3 able for very thick plates. In the examples presented it
Glass fiber reinforced epoxy plate made of unidirec- can be seen that the best agreement is achieved for
tional fiber plies with all layers stacked at [0°1 was Example 3, which is the thickest plate analyzed. In the
studied. The plate dimensions and density are: remaining examples the identified values of the trans-
verse shear moduli can be disregarded as the sensitivity
a = 0.203 m, b = 0.136m, to these moduli is very low, due to the fact that the plates
h = 0.014m, p = 1886.9 kg/m 3 used are relatively thin.
Some discrepancies are also found in the values of the
The initial guess for the elastic constants corresponds to
major Poisson's ratio ul2. This can be explained if we
a typical prepreg layer of Structil VEE 2 45R365
consider the results of a sensitivity analysis carried out by
unidirectional glass with 60% fiber volume:
Frederiksen 4, according to which the sensitivity of the

190
 Material parameters of composite plate specimens. A. L. Ara~jo et al.

Poisson's ratio is very small when compared to the anisotropic plates--A combined numerical/experimental
method. Ph.D. thesis, Department of Solid Mechanics, The
sensitivities of the other in-plane mechanical properties. Technical University of Denmark, 1992
This is especially pronounced for strongly anisotropic 5 Wilde, W.P. and Sol, H. Anisotropic material identification
plates which is the case of the plates of Example 3 and, using measured resonant frequencies of rectangular composite
plates. In 'Composite Structures 4, Vol. 2' (edited by Marshall,
particularly, Example 4 where the discrepancies are I.H.). Elsevier Applied Science, London, 1987, pp. 2317-2324
relatively large. 6 Sol, H. Identification of the complex moduli of composite
materials by a mixed numerical/experimental method. In 'Com-
posite Materials Design and Analysis' (edited by Wilde, W.P.
ACKNOWLEDGEMENTS and Blain, W.R.). Springer-Verlag, Berlin, 1990, pp. 267 279
7 Wilde, W.P. Identification of the rigidities of composite systems
The authors wish to thank Prof. Pauli Pedersen and by mixed numerical/experimental methods. In 'Mechanical
Dr Per S. Frederiksen from the Department of Solid Identification of Composites' (edited by Vautrin, A. and
Sol, H.). Elsevier Applied Science, Barking, 1991, pp. 1-15
Mechanics, Technical University of Denmark, for their 8 Kant, T. and Kommineni, J.R. Geometrically non-linear analy-
valuable cooperation in this research work. sis of symmetrically laminated composite sandwich shells with a
This work has been partially sponsored by J N I C T - - higher-order theory and C ° finite elements. Compos. Struct.
1994, 27, 403 418
Junta Nacional de Investigaqfio Cientifica e Tecnoldgica 9 Arafijo, A.L. M6todo num6rico/experimental para caracteriza-
(Proj. STRIDE-STRDA/C/TPR/592/92), Commission qfio mecfinica de materiais comp6sitos. MSc. thesis, Instituto
of E.C. (Proj. HCM CHRX-CT93-0222) and F L A D - - Superior T6nico, Technical University of Lisbon, 1995 (in Por-
tuguese)
Fundaqfio Luso Americana para o Desenvolvimento. 10 Frederiksen, P.S. Natural vibrations of free thick plates and
identification of transverse shear moduli. In 'Proc. IUTAM
Symp. Optimal Design Advanced Materials' (edited by Peder-
REFERENCES sen, P.). Elsevier Science Publishers, Amsterdam, 1993, pp.
131-148
1 Mota Soares, C.M., Moreira de Freitas, M., Arafijo, A.L. and 11 Bathe, K.J. "Finite Element Procedures in Engineering Analy-
Pedersen, P. Identification of material properties of composite sis'. Prentice-Hall, Englewood Cliffs, N J, 1982
plate specimens. Compos. Struct. 1993, 25, 277 285 12 Zienkiewicz, O.C. 'The Finite Element Method in Engineering
2 Mota Soares, C.M., Moreira de Freitas, M.J. and Meira Science'. McGraw-Hill, New York, 1971
Afonso, J.M. Identificaqfio de propriedades mecS.nicas de mater- 13 Richardson, M.H. and Formenti, D.L. Parameter estimation
iais compdsitos utilizando t6cnicas de optimizaqfio. Revista from frequency response measurements using rational fraction
lnternacional de M~todos Num&icos para Calculo y Dise~o en polynomials. In 'Proc. IMAC I', Orlando, Florida, 8 10
lngenieria 1993, 9 (4), 357 373 (in Portuguese) November 1982
3 Pedersen, P. Optimization method applied to identification of 14 Vanderplaats, G.N. ' A D S - - A Fortran Program for Automated
material parameters. In 'Discretization Methods and Structural Design Synthesis. Version 2.01'. Vanderplaats Research and
Optimization Procedures and Applications' (edited by Esche- Development Inc., Colorado Springs, Colorado, USA, 1987
nauer, H.A. and Thierauf, G.). Springer-Verlag, 1989, pp. 277- 15 Vanderplaats, G.N. 'Numerical Optimization Techniques for
283 Engineering Design with Applications'. McGraw-Hill, New
4 Frederiksen, P.S. Identification of material parameters in York, 1984

191