Professional Documents
Culture Documents
Received 9 November 1998; received in revised form 2 March 1999; accepted 9 April 1999
Abstract
Identi®cation of elastic properties of unidirectional glass/epoxy laminates from the measured eigenfrequencies has been per-
formed. The stiness of the laminates has been investigated by a mixed numerical/experimental method employing the vibration
test of plates. Elastic constants of laminates have been determined by using an identi®cation procedure based on experiment design,
the ®nite-element method and the response-surface approach. Elastic properties of laminates with two dierent ®bre-surface
treatments have been compared. It was found that only for the transverse elastic modulus is there a statistically signi®cant dierence
between the composites with good and poor ®bre/matrix adhesion. # 1999 Elsevier Science Ltd. All rights reserved.
Keywords: Composites; B. Surface treatments; C. Laminates; C. Elastic properties
the identi®cation of parameters which can be used for types of surface treatments. The ®rst type was treated by
the model calibration as well as for the detection of epoxy dispersion with aminosilane to promote good
damaged zones in the structure. ®bre/matrix adhesion. More details about this ®bre
Numerical±experimental identi®cation methods are treatment can be found in [28,29]. For this composite
mainly used in structural applications [8±19]. For the notation EP is used. The second type was sized with
example, in [8] elastic properties of laminated compo- polyethylene to prevent ®bre/matrix adhesion. For this
sites have been identi®ed by using experimental eigen- composite the notation PE is used. A Ciba Geigy Ltd.
frequencies. The stiness parameters were identi®ed Araldite epoxy system (LY556/917/DY070) with an
from the measured natural frequencies of the laminated ultimate tensile strain of 3.3% was used as the matrix
composite plate by direct minimisation of the identi®- material.
cation functional. A similar approach to identifying the Unidirectionally reinforced test samples were pro-
stiness properties of the laminated composites was duced on a winding machine. The content of micropores
used in [18,19,22]. In [20] an improved plate model was was below 0.75% by volume. The glass ®bre-reinforced
used for identi®cation of elastic constants of the lami- plates (see Fig. 1) were prepared with given geometric
nate. In [23] it was shown that the mixed numerical± parameters. The total number of plies was eight in order
experimental method can also be used for identi®cation to get a plate thickness of approximately 2 mm. The EP
of damping properties of polymeric composites. and PE laminates were produced by following the stan-
In the present study the numerical±experimental dard cure cycle recommended by Ciba Geigy Ltd. To
method for identi®cation of mechanical properties of improve the quality of the plates and to reduce the void
laminated polymeric composites from the experimental content, the plates were placed in a vacuum before cur-
results of the structure response has been further devel- ing. Specimens were cut out with a diamond wheel and
oped. The dierence between the conventional [8,20] kept under standard conditions (23 C and 50% of rela-
and the present approach is that, instead of direct tive humidity) until testing. All tests were performed in
minimisation of the identi®cation function, experimental the same conditions. The ®bre content of the specimens
design is used, by which response surfaces of the func- was about 501.5 vol%. The density was measured for
tion to be minimised are obtained. The response surface each specimen. The specimen geometric parameters and
approximations are obtained by using information on density for the unidirectionally (UD) reinforced PE
the behaviour of a structure in the reference points of composite plates are presented in Table 1. Similar data
the experiment design. Note that ®nite-element model- for the EP composite plates were given in [7].
ling of the structure is performed only in the reference
points. The function to be minimised describes the 2.2. Vibration tests
dierence between the measured and numerically calcu-
lated parameters of the response of structure. By mini- The eigenfrequencies of the test plates were measured
mising the function the identi®cation parameters were by a real-time television (TV) holography. The samples
obtained. The method was employed to identify the having mean dimensions of 140 mm140 mm2 mm
elastic properties of the laminates from the measured were hung upon two threads in order to simulate free±
eigenfrequencies of the plates. The main advantage of the free boundary conditions. The sample was located in
present method is a signi®cant reduction of the compu- front of the holographic testing device. A piezoelectric
tational eort. Previously this method was used for the resonator (sensor) was glued on one corner to excite the
solution of the optimum design problems [24±27]. sample plate at increasing frequencies. The sensor is of
In the present study the numerical±experimental
method has been applied for identi®cation of elastic
properties of glass/epoxy laminates with two dierent
®bre-surface treatments. Experiments were performed
on laminated plates and about 16 natural frequencies
were measured. Elastic properties obtained from vibra-
tion tests have been compared with those from the
transverse tensile test.
2. Experimental
2.1. Material
circular shape with a diameter 25 mm located at the 3. Parameters of identi®cation and criterion
coordinates x a ÿ 12:5 mm and y b ÿ 12:5 mm (see
Fig. 1). The mass of the sensor is ms 3:5 g. The numerical±experimental method proposed for the
The plate is illuminated by laser light and imaged by identi®cation of the elastic constants from the vibration
CCD (charged couple device) array, resulting in a tests consists of the following stages. In the ®rst stage
speckled image on the PC monitor. When the plate is the physical experiments are performed. Also the para-
deformed (excited), this interference pattern is slightly meters to be identi®ed, the domain of search and criter-
modi®ed. Digital substraction of two consecutive inter- ion containing experimental data are selected. In the
ference patterns yields a fringe pattern depicting the second stage the ®nite-element method is used in order
surface displacements of the plate. The nodal lines of to model the response of the structure and calculations
the vibration modes can be easily identi®ed on the are performed in reference points of the variables to be
monitor in the form of white lines on the speckled identi®ed. The reference points are determined by using
image. The digital substraction of two consecutive pic- a method of experiment design. In the third stage the
tures helps to minimise noises, such as rigid body numerical data obtained by the ®nite-element solution
motion of the hung plate. This measurement technique in the reference points are used in order to determine
was described in more detail in [4,30]. simple functions (response surfaces) for a calculation of
Experiments were performed for all plates considered the structure response. In the fourth stage, on the basis
(see Table 1) and about 16 ¯exural eigenfrequencies of the simple models and experimentally measured
were measured. In Table 2 the most frequently found values of the structure response, the identi®cation of the
experimental plate ¯exural frequencies for the PE com- elastic properties is performed. For this a corresponding
posite are presented. The mode shapes were also recog- functional is minimised by using the method of non-
nized in the experiment. Experimentally measured linear programming.
frequencies for the EP composite were presented in [7].
3.1. Parameters of identi®cation
Table 1
Geometric parameters and density of the PE composite plates Identi®cation of parameters is performed from the
Sample a (m) b (m) 3
h 10 (m) (kg/m ) 3 experimentally measured eigenfrequencies. It is assumed
that the plate dimensions (see Fig. 1), plate mass and the
PE01 0.1399 0.1401 2.095 1884 layer stacking sequence are known. The parameters to
PE02 0.1401 0.1401 2.072 1889 be identi®ed are ®ve elastic constants of a transversely
PE03 0.1400 0.1399 2.060 1890
PE04 0.1400 0.1401 2.070 1900
isotropic plate [8]:
PE05 0.1400 0.1404 2.010 1890
PE06 0.1401 0.1403 2.060 1900 . two Young's moduli, E1 , E2 E3
Mean 0.1400 0.1401 2.061 1892 . shear moduli, G12 G13
Table 2
Experimental plate ¯exural frequencies fi (Hz) for the PE composite
. Poisson's ratio, 12 13 where !~ 1 is the ®rst numerical eigenfrequency calculated
E2 with the prior selected value of a longitudinal Young's
. shear modulus, G23
2
1 23 modulus E01 (initial guess value) of the layer [8].
The plate is composed of unidirectionally reinforced 3.2. Identi®cation functional and minimization problem
layers with material axis 1±2±3, where 1 is the ®bre
direction and 2, 3 are directions transverse to the ®bres. The functional to be minimised describes deviation
The unidirectional layer is assumed as homogeneous between the measured !i and numerically calculated
and transversely isotropic with respect to the ®bre !~ i
x frequencies [8]
direction. The parameters to be identi®ed can be h i2
expressed in terms of dimensionless constants [8] X
I !2i ÿ C!~ i
x2
x
6
E2 i2
!4i
2 4 ÿ 4 ;
E1
It is seen that criterion (6) is a nonlinear function of
E2 G12
3 1
1 ÿ 212 ÿ 4 0 ; the identi®cation parameters x. The identi®cation of the
E1 E1 elastic constants x is performed on the basis of infor-
E2 G12 mation obtained from the measurements of the I lowest
4 1
1 612 ÿ 4 0 ; frequencies. The identi®cation problem is formulated as
E1 E1
follows
0
5 4
G23 G12
1
E1 min
x
7
x
where
Subject to constraints
E2
0 1 ÿ 212
2
E1 E2
g 1
x 4 ÿ 2 > 0 or >0
8
E1
The inverse relations of Eqs. (1) are as follows
8 ÿ 2 ÿ 33 ÿ 4
g 2
x " # > 0 or
E2 4 ÿ 2 G12 8 ÿ 2 ÿ 33 ÿ 4 4 ÿ 3 2 4 ÿ 2
; ; 16 1 ÿ
E1 4 E1 160 8 ÿ 22 4
9
4 ÿ 3 G23 25 ÿ 0:5
8 ÿ 2 ÿ 33 ÿ 4 G12
12 ;
3 >0
8 ÿ 22 E1 80 E1
The optimum values for the dimensionless material used by Mota Soares et al. [8] and also in [19]. On each
parameters i (i 2; 3; 4; 5) that satisfy Eq. (7) and iterative stage the eigenvalue problem for the linear
constraints (8)±(12) were obtained by using a random system was solved by the ®nite element method or by
search method [24]. The value of Young's modulus of the Ritz method and the nonlinear programming algo-
the layer in the ®bre direction E1 can be easily eval- rithm applied in order to minimise directly the identi®-
uated, since C and 0 are known. The steps of this eva- cation functional (6). Such a procedure requires very
luation are shown below. large computational eorts.
Instead of direct minimisation of the criterion (6), a
method based on experiment design can be used. The
4. Finite-element solution selection of points in the domain of interest where the
response must be evaluated is commonly called design
The eigenvalue problem for harmonic vibrations of of experiments. The choice of experiment design can
the plate can be represented by have a large in¯uence on the accuracy of the approx-
imation and the cost of constructing the response sur-
Ku !2 Mu
13 face [33]. A commonly used experiment design is the D-
optimality criterion [34]. Other experiment design
Here K is the stiness matrix of the plate, M is the methods require an a priori knowledge of the functional
mass matrix and u is the displacement vector. The form of the response. This knowledge might not be
eigenvalue relation (13) for the mode u1 which corre- available. The method of experiment design in which
sponds to the ®rst experimental eigenfrequency !1 can the a priori knowledge of the functional form of the
be written in an equivalent form placing E1 in evidence: response is not necessary was proposed in [24]. This
method previously was used in the solution of optimi-
E1 K u1 !21 Mu1
14 sation problems for sandwich and laminated composite
structures [25±27]. The identi®cation and optimum
Here E1 K K is the stiness matrix. Taking into design problems are very similar. In both cases the
account relation (5) this equation can be written as solution can be obtained by minimisation of the function
considered. The method of experiment design for the
CE01 K u1 C!~ 21 Mu1
15 solution of the identi®cation problems was outlined in [7].
Further, as an example the method is shown for
hence identi®cation of elastic constants for PE composite by
the use of the mean values of plate ¯exural frequencies
E1 CE01
16 (see last column in Table 2). The mean values of the
geometric parameters and density of the plates are taken
where E01 is the initial guess value given to the Young's from Table 1.
modulus in the ®bre direction of the layer and E1 is the Firstly the experiment design with four variables
corresponding identi®ed mechanical property. After (n=4) and 35 reference points is selected. The second
evaluation of the optimum value of x the remaining step is selection of the domain of interest. The upper
mechanical properties are calculated by inverse relations and lower limits of the identi®cation parameters are
(3). taken as follows
The eigenvalue problem (13) was solved by the sub-
space iteration method [31] and using a triangular ®nite 5 GPa4E2 440 GPa; 5 GPa4G12 430 GPa;
element of laminated thick plate with a shear correction
[32]. In order to avoid `shear locking' a selective inte- 5 GPa4G23 430 GPa; 0:2412 40:4
17
gration technique was applied. A 2222 regular mesh
(968 ®nite elements) was considered in order to achieve The initial guess value of Young's modulus is taken
the necessary accuracy for at least 16 ®rst eigenvalues of E01 =40 GPa, which is related to E1 by expression (16).
the laminated plate with FFFF (all edges free) bound- After selection of the domain of interest the experiment
ary conditions. design is employed. The matrix ij of the experiment
design for n 4 and k 35 is used and the parameters
of identi®cation in the reference points of the domain of
5. Method of experiment design for identi®cation interest (17) are calculated by the expression
problems
1 max ÿ
x
j i xmin
j xj ÿ xmin
j Bij ÿ 1
18
For direct identi®cation of the elastic properties of a kÿ1
material it is necessary to perform a multiple iterated
®nite element procedure. Such a direct procedure was Here i 1; 2; :::; k and j 1; 2; :::; n.
2020 A.K. Bledzki et al. / Composites Science and Technology 59 (1999) 2015±2024
geometric parameters and density of the samples (see points and the function (response surface) obtained by
Table 1) are almost the same for all specimens and the the approximation have been selected. Such selection is
scatter of the measured frequencies is rather small (see necessary in order to minimise the approximation
Table 2). However, since frequencies are very sensitive errors. The vibration modes selected for identi®cation
to thickness h and density of the plate, in the case of are given in Table 4. In the last row (Mean) the vibra-
larger scatter of the measurements the ®rst approach tion modes selected for identi®cation from the mean
must be used, i.e. the identi®cation of the elastic con- values (the second approach) are presented.
stants should be performed for each specimen. Veri®cation of the results was performed by the ®nite-
For the plate specimen with a sensor there are some element method (FEM) and through the independent
dierences in calculation of the mass matrix M in Eq. experiments. For the ®nite-element analysis the elastic
(13) in comparison with the plate without sensor. In constants obtained by the identi®cation procedure were
order to represent more accurately the inertia forces of used [see Table 3, the mean values
2 ]. Results are
the plate, the mass of sensor ms should be taken into shown in Table 5. Residuals were calculated by the
account. In the ®nite-element modelling it is assumed expression
that the ®nite elements where the sensor is located have
the same thickness h as the plate, but for these ®nite f FEM
i
x ÿ f exp
i
i 100
21
elements an equivalent density eqv is calculated: f exp
i
ms
eqv
Fs h
Here Fs is the area of the sensor. 7.1. Comparison with independent tests
Results of identi®cation are presented in Table 3.
Utilising the ®rst way the elastic constants were identi- It is of interest to compare the elastic constants
®ed for each specimen using the data from Tables 1 and obtained from the vibration test through the identi®ca-
2, and then the mean values
1 and standard devia- tion (see Table 3) with the values obtained by the inde-
tions s
1 were calculated. In the second approach at the pendent tension test. The same PE composite was tested
beginning the mean values of frequencies (see Table 2) in static tension transverse to the ®bres [4]. The compo-
and geometric parameters (see Table 1) were calculated. site also was tested in the ®bre direction.
These mean values were used as input data and then the Results obtained by independent tests are compared
identi®cation procedure was applied. The results in Table 6. For the tension test in the ®bre direction,
obtained by identi®cation utilising the second way are six specimens were tested. In the tension test transverse
also given in Table 3 and denoted by
2 . to the ®bre direction, eight specimens were tested. The
In Table 3 it is seen that results for the in-plane elastic data reduction was performed according to DIN 53457
constants E1 ; E2 ; G12 and 12 obtained by both approa- and DIN 29971, i.e. the modulus E1 was calculated as a
ches are in good agreement. The transverse shear mod- slope between the points for the strain level 0.05 and
ulus G23 can not be reliable determined from the 0.25%, and the modulus E2 was calculated as a slope
measured frequencies since the plates were too thin between the points for the stress at 10 and at 50% of the
(h=a=1/70) for identi®cation of this property. In this maximum stress [4]. The sample mean and standard
case thick plates should be used. deviations are presented.
It should be noted that the number of frequencies In order to evaluate the dierences between the values
which were selected for the identi®cation was dierent obtained by independent tests the methods of statistics
for dierent specimens. For the identi®cation only the should be applied. The t-test [38] for comparing mean
modes with a high values of the coecient of correlation values of two normal distributions (variances being
(correlation higher than 98%) between the data equal) was employed. Assuming the two mean values of
obtained by the ®nite-element solution in the reference the test groups 1 (vibration test) and 2 (tension test)
Table 3
Elastic constants of the PE composite obtained by identi®cation
Property PE01 PE02 PE03 PE04 PE05 PE06 1 s 1 2
Table 4
Vibration modes used for identi®cation
Plate Mode i
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
PE01 + + + ÿ + + + ÿ ÿ ÿ ÿ + + + ÿ ÿ
PE02 + + + ÿ + + + ÿ ÿ ÿ ÿ + + + ÿ ÿ
PE03 + + + ÿ + + ÿ ÿ ÿ ÿ ÿ + ÿ + ÿ ÿ
PE04 + + + ÿ + + ÿ ÿ ÿ ÿ ÿ + + + ÿ ÿ
PE05 + ÿ + + + ÿ ÿ ÿ ÿ ÿ ÿ + + ÿ ÿ ÿ
PE06 + + ÿ + + + ÿ ÿ ÿ ÿ ÿ + ÿ ÿ ÿ ÿ
Mean + + + + + + + ÿ ÿ ÿ ÿ + + + + ÿ
Table 7
Comparison of dierent properties
9. Conclusions
Acknowledgements
8. Comparison of elastic properties for PE and EP
composites The investigations concerning the development of the
identi®cation method for elastic properties of laminated
In order to compare the laminates with good ®bre/ composites were sponsored through Grant 96.0504. The
matrix adhesion (EP composite) and poor ®bre/matrix authors are pleased to acknowledge the ®nancial sup-
adhesion (PE composite), dierent material properties port by the Latvian Council of Science. Thanks are also
can be used. For example, in [4] the transverse strength due to German Federal Ministry of Science and Tech-
and in [37] the interlaminar fracture toughness properties nology (Grant WTZ LET010.97) and Latvian Ministry
have been compared for the same composites con- of Education and Science (Grant 6292/98) for their
sidered here. In the present paper elastic properties are generous support of this paper.
2024 A.K. Bledzki et al. / Composites Science and Technology 59 (1999) 2015±2024
References [19] Frederiksen PS. Experimental procedure and results for the
identi®cation of elastic constants of thick orthotropic plates. J
[1] Mahiou H, BeÂakou A. Modelling of interfacial eects on the Compos Mater 1997;31(4):360±82.
mechanical properties of ®bre-reinforced composites. Composites [20] Frederiksen PS. Application of an improved model for the iden-
Part A 1998;29A:1035±48. ti®cation of material parameters. Mech Compos Mater and
[2] Steen M, ValleÂs JL. Interfacial bond conditions and stress dis- Struct 1997;4:297±316.
tribution in a two-dimensionally reinforced brittle-matrix com- [21] Frederiksen PS. Numerical studies for the identi®cation of elastic
posite. Composites Science and Technology 1998;58:313±30. constants of thick orthotropic laminates. Eur J Mech A/Solids
[3] Wacker G, Bledzki AK, Chate A. Eect of interphase on the 1997;16:117±40.
transverse Young's modulus of glass/epoxy composites. Compo- [22] ArauÂjo AL. MeÂtodo numeÂrico/experimental para carazterizacËaÄo
sites Part A 1998;29A:619±26. mecaÏnica de materiais compoÂsitos. MSc. thesis, Technical Uni-
[4] Kessler A, Bledzki, AK. Does the interface control the transverse versity of Lisabon, Mech. Eng. Dept., IST, 1995 (in Portuguese).
modulus of GFRP? Composite Interfaces, in press. [23] De Visscher J, Sol H, De Wilde WP, Vantomme J. Identi®cation
[5] Mai K, MaÈder E, MhuÈle M. Interphase characterization in com- of damping properties of orthotropic composite materials using a
posites with new non-destructive methods. Composites Part A mixed numerical experimental method. Appl Compos Mater
1998;29A:1111±9. 1997;4(1):3±33.
[6] Keusch S, Queck H, Gliesche K. In¯uence of glass ®bre/epoxy [24] Rikards R. Elaboration of optimal design models for objects
resin interface on static mechanical properties of unidirectional from data of experiments. In: Pedersen P, editor. Optimal design
composites and on fatigue performance of cross ply composites. with advanced materials. Proceedings of the IUTAM Sympo-
Composites Part A 1998;29A:701±5. sium, Lyngby, Denmark, 18±20 August, 1992. Amsterdam: Else-
[7] Rikards R, Chate A, Steinchen W, Kessler A, Bledzki AK. vier Science Publishers, 1992. p. 148±62.
Method for identi®cation of elastic properties of laminates based [25] Rikards R, Chate A, BaÂcklund J. Optimal design of sandwich
on experiment design. Composites Part B 1999;30B:279±89. plates based on planning of experiments. In: Olho N, Rozvany
[8] Mota Soares CM, Moreira de Freitas M, ArauÂjo AL. Identi®ca- GI, editors. Proc. 1st World Congress of Structural and Multi-
tion of material properties of composite plate specimens. Com- disciplinary Optimization, 28 May±2 June 1995, Goslar, Ger-
posite Structures 1993;25:277±85. many. Pergamon, 1995. p. 569±74.
[9] Sol H, De Visscher J, De Wilde WP. Identi®cation of the viscoe- [26] Rikards R, Chate A. Optimal design of sandwich and laminated
lastic material properties of orthotropic plates using a mixed composite plates based on planning of experiments. Structural
numerical/experimental technique. In: Brebbia CA, Carlomagno Optimization 1995;10(1):46±53.
GM, editors. Computational methods and experimental mea- [27] Rikards R. Minimum weight design of sandwich and laminated
surements VI, vol. 2: stress analysis. London/New York: Elsevier composite structures. Mech Compos Mater 1995;31(1):51±64.
Applied Science, 1993. p. 131±42. [28] Wacker G. Experimentell gestuÈtzte Identi®kation ausgewaÈahlter
[10] Sol H. Identi®cation of anisotropic plate rigidities using free Eigenschaften glasfaserverstaÈrkter Epoxidharze unter Ber-
vibration data. PhD thesis, Free University of Brussels, 1986. uÈucksichtigung der Grenzschicht. PhD Dissertation, University
[11] Bolognini L, Riccio F, Bettianli F. A modal technique for the of Kassel, Kassel, Germany, 1996 (in German).
identi®cation of stiness and mass parameters in large structures. [29] Bledzki AK, Lieser J, Wacker G, Frenzel H. Characterization of
In: Brebbia CA, Carlomagno GM, editors. Computational meth- the surfaces of treated glass ®bres with dierent methods of
ods and experimental measurements VI, vol. 2: stress analysis. investigation. Composite Interfaces 1997;5(1):41±53.
London/New York: Elsevier Applied Science, 1993. p. 337±52. [30] Yang LX, Steinchen W, Schut M, Kupfer G. Precision measure-
[12] Frederiksen PS. Identi®cation of temperature dependence for ment and nondestructive testing by means of digital phase shift-
orthotropic material moduli. Mechanics of Materials 1992;13:79±90. ing speckle pattern shearing interferometry. Measurement
[13] De Wilde WP. Identi®cation of the rigidities of composite sys- 1995;16:149±60.
tems by mixed numerical/experimental methods. In: Vautrin A, [31] Bathe K-J. Finite element procedures in engineering analysis.
Sol H, editors. Numerical identi®cation of composites. London/ Englewood Clis (NJ): Prentice Hall, 1982.
New York: Elsevier Applied Science, 1991. p. 1±15. [32] Rikards R, Chate A, Korjakin A. Damping analysis of laminated
[14] Grediac M, Vautrin A. Mechanical characterization of aniso- composite plates by ®nite element method. Engineering Compu-
tropic plates: experiments and results. Eur J Mech A/Solids tations 1995;12:61±74.
1993;12(6):819±38. [33] Roux WJ, Stander N, Haftka RT. Response surface approxima-
[15] Link M, Zou Zh. A two steps procedure to identify physical para- tions for structural optimization. Int J Numer Meth Engng
meters of composite material structures using vibration test data. 1998;42:517±34.
In: Proc. Int. Conf. on Vibration Engineering, ICVE 94, Beijing. [34] Van Campen DH, Nagtegaal R, Schoofs AJG. Approximation
Beijing: International Academic Publishers, 1994. p. 217±24. methods in structural optimization using experimental designs for
[16] Moussu F, Nivoit M. Determination of elastic constants of multiple responses. In: Eschenauer H, Koski J, Osyczka A, editors.
orthotropic plates by a modal analysis method of superposition. J Multicriteria design optimization. Berlin: Springer, 1990. p. 205±28.
Sound and Vibrations 1993;165(1):149±63. [35] Box GEP, Draper NR. Empirical model-building and response
[17] Pedersen P. Optimization method applied to identi®cation of surfaces. New York: Wiley, 1987.
material parameters. In: Eschenauer HA, Thierauf G, editors. [36] Vanderplaats GN. Numerical optimization techniques for engi-
Discretization methods and structural optimizationÐprocedures neering design with applications. New York: McGraw Hill, 1984.
and applications. Berlin: Springer Verlag, 1989. p. 277±83. [37] Rikards R, Korjakin A, Buchholz F-G, Wang H, Bledzki AK,
[18] Frederiksen PS. Identi®cation of material parameters in aniso- Wacker G. Interlaminar fracture toughness of GFRP in¯uenced
tropic platesÐa combined numerical±experimental method. PhD by ®bre surface treatment. J Compos Mater 1998;32(17):1528±59.
thesis, Department of Solid Mechanics, The Technical University [38] Lederman W, editor. Handbook of applicable mathematics,
of Denmark, 1992. vol. 6. statistics. Part A. New York: Wiley, 1984.