THEORETICAL & APPLIED MECHANICS LETTERS 4, 061001 (2014) Numerical-experimental method for elastic parameters identification of a composite panel Dong Jiang, 1, 2, a) Rui Ma, 1, 2 Shaoqing Wu, 1, 2 1, 2, Qingguo Fei b) 1) Jiangsu Key Laboratory of Engineering Mechanics, Southeast University, Nanjing 210096, China 2) Department of Engineering Mechanics, Southeast University, Nanjing 210096, China (Received 1 December 2013; revised 15 April 2014; accepted 5 June 2014) Abstract A hybrid numerical-experimental approach to identify elastic modulus of a textile composite panel using vibration test data is proposed and investigated. Homogenization method is adopted to predict the initial values of elastic parameters of the composite, and parameter identification is transformed to an optimization problem in which the objective function is the minimization of the discrepancies between the experimental and numerical modal data. Case study is conducted employing a woven fabric reinforced composite panel. Three parameters (E 11, E 22, G 12 ) with higher sensitivities are selected to be identified. It is shown that the elastic parameters can be accurately identified from experimental modal data. c 2014 The Chinese Society of Theoretical and Applied Mechanics. [doi:10.1063/2.1406101] Keywords composite panel, elastic modulus, parameter identification, modal data, numerical-experimental method Carbon fiber reinforced ceramic substrate composites have been broadly used in aerospace engineering because of their excellent properties. It is an active research area to predict composites elastic properties, 1,2 since the properties are very important for mechanical analysis of composite structures. Investigations for estimating the equivalent elastic modulus of fiber reinforced composites can be classified into three main categories, including experimental methods, analytical methods and numerical methods. 3 5 Generally, several elastic constants of composite may be accurately tested by experimental method, but experiments are usually expensive and time consuming. Furthermore, it is even difficult to obtain the whole orthotropic stiffness matrix including nine independent coefficients. Analytical approaches and numerical simulation methods are based on the homogenized unit cell model of composites, 6 which is an alternative solution for evaluating the equivalent elastic modulus. Numerous investigations have been performed on these methods. 1,2,7 9 Bystrom et al. 10 studied two simple and convenient analytical models for calculating woven fabric composites elastic properties and the results show that the unit cell model based on iso-strain/iso-stress assumptions for in-plane/out-of-plane components gives solutions in good agreement with numerical predictions. However, idealization assumptions have to be made in analytical and numerical methods before estimating equivalent elastic modulus from a a) Email: jiangdonal@gmail.com. b) Corresponding author. Email: qgfei@seu.edu.cn.
061001-2 D. Jiang, et al. Theor. Appl. Mech. Lett. 4, 061001 (2014) homogenized representative volume model, which will inevitably introduce errors. Hallal et al. 11 indicated that accurate predictions could not be given by a model based on only an iso-strain assumption. Pochiraju and Chou 12 revealed that the errors of predictions of elastic properties are within 10% compared to the experimental values. Modal frequencies, mode-shapes, or frequency response functions are different from static test data can be used to identify parameters. Model updating was described in detail by Mottershead et al., 13,14 and in recent years has been developed promptly and applied successfully in engineering. On the premise of initial finite element model with good understanding of the structural physical significance, 15 model updating can be effectively used for identifying parameters and detecting mechanical behavior changes of structures. 16 In this study, a methodology is proposed to identify the elastic modulus of a fiber reinforced composite using modal data. Homogenization method is adopted to predict the initial values of the equivalent elastic parameters. The so called layer-to-layer angle-interlock woven composite is shown in Fig. 1. We interlace the warp yarns and weft yarns orthogonal in the x y plane, which leads to binding of warp yarns by interlocking weft yarns. Warp yarns is bound to different depth where various layer of weft yarns are placed, and warp weavers travel one layer to the neighboring layer, thus a set of warp weavers hold all the layers of the fabric. In order to obtain the elastic properties of a fiber reinforced composite, a homogenization problem is firstly formulated for a unit cell or the representative volume element (RVE). Predictions of the mechanical properties are significantly relying on the two geometric parameters, cross-sectional shape of yarns and curved line of the warp weaving path. The corresponding geometric hypothesis are (1) supposing the cross-sectional shapes of the weft weavers and warp yarns to be respectively ellipsoidal and rectangular and (2) simulating the warp weaving path by a curve and a tangent straight line. Secondly, the numerical model of the unit cell or the RVE can be constructed. Consequently, the macroscopic stiffness matrix and the macroscopic compliance matrix is calculated by using iso-strain or iso-stress method. 1 The essence of parameter identification is to minimize the residuals between the predicted and measured modal data. We define the objective function and the constraint as MinJ(p)=ε T W ε = W 1/2 (z m z a (p)) 2 2, p 1 p p 2. (1) Here p R N is a vector of N parameters to be identified, ε is the numerical modal data s error vector, W, representing the relative weight of each error, is a diagonal weighting matrix, the superscript T denotes the matrix transpose, and z m and z a (p) R n are the vectors measured and analytical modal parameters with n dimensions. The numerical and measured data have to be matched using the modal assurance criterion (MAC). 13,14 We employ gradient-based method to solve Eq. (1). The problem at the j-th iteration step is described as W 1/2 (z m z a j )=S j (p j+1 p j ) with the weighted sensitivity matrix of modal data with respect to structural parameters having the form of S j = W 1/2 z a j / p j. When the measured and numerical modal data (z m and z a j ) are eigenvalues, we can determine the term z a j / p j in the weighted sensitivity matrix may by the expression z a j / p j = Φ T j ( K(p j )/ p j z j M(p j )/ p j )Φ j, where K, M R s s denote
061001-3 Numerical-experimental method for elastic parameters identification stiffness matrices and the finite element model mass, respectively, and Φ j is the j-th mode shape; s is the number of degrees of freedom (DOFs) of the model. The implementation procedure of the parameter identification is defined in Fig. 2. Initial values of elastic parameters Calculate modal data from finite element model Vibration test modal data Take the value of elastic parameters of the j-th step as the initial values of the ( j+1)-th step Pair mode shapes by MAC-value Calculate sensitivity matrix Warp yarn Weft yarn Unit cell Calculate change of the elastic parameters of the j-th step, p j+1 p j Fig. 1. Microstructure of a woven composite. No Accuracy satisfied? Yes Precise elastic parameters Fig. 2. Procedure for parameter identification. A case study of a fiber reinforced composite panel is conducted. The size of the panel is 300 mm 300 mm 3 mm. By using the homogenization method, the equivalent elastic modulus of the composite are predicted and shown in Table 1. An initial finite element model of the composite plate is built using the geometrical and material parameters. In experimental modal test, we simulate the free-free boundary condition by hanging the plate with soft ropes, and the first eight modal frequencies and shapes are obtained. Table 1. Equivalent elastic modulus of the composite. Elastic parameters E 11 /GPa E 22 /GPa E 33 /GPa G 12 /GPa G 23 /GPa G 31 /GPa μ 12 μ 23 μ 13 Predicted values 117.09 126.18 99.38 37.67 35.66 34.81 0.2 0.25 0.25 Parameter selection is crucial in model updating. 17,18 For selecting proper parameters, there is always the need of considerable mechanical insight of the structure to ensure not only correlations between measured and numerical modal data, but also physical significance of parameters to be identified. It is common to use relative sensitivity analysis for selecting parameters due to its advantage of avoiding the influence of the quantity or unit employed for parameters. Comparison of sensitivities for different parameter types becomes feasible and effective because of this feature. We can define
061001-4 D. Jiang, et al. Theor. Appl. Mech. Lett. 4, 061001 (2014) the relative sensitivity as S r,ij = f i p j p j, (2) in which f i is the i-th order modal frequency, and p j is the j-th element of p. The orthotropic material property of the composite has nine independent elements, expressed as p =(E 11,E 22,E 33,G 12,G 23,G 31, μ 12, μ 23, μ 31 ) T. S r consists of the first nine modal frequencies with respect to elastic parameters. Table 2 shows the results of nine elastic parameters calculated by using Eq. (2), and three elastic parameters such as the in-plane elastic modulus E 11, E 22, and the shear module G 12 with higher sensitivity are selected to be identified. R MAC is used for testing the correlation between experimental and numerical mode orders as R MAC ij = Φ mt i Φ m j 2 /[(Φ mt i Φ m i )(Φ at j Φ a j)]. Here the subscripts i, j indicate the mode orders. Results from Table 3 show good agreements between the results of mode shapes from the experiment and the numerical analysis. The first four experimental modal frequencies and mode shapes are adopted in identification procedure. Selecting the elastic modulus E 11, E 22, and the shear module G 12 in the initial finite element model of the composite plate to be identified. Convergence of the selected parameters is shown in Fig. 3. The values of parameters after identification are shown in Table 4. It is seen that the selected parameters are converged after 25 iterations, after convergence, the highest ratio of change of the three parameters is no more than 4%, which makes the first four computational modal frequencies highly accurate with bounded error of 0.5%. The 5-th to 8-th modal frequencies are used for validating the identification results. Comparison of modal frequencies between experimental and computational results is shown in Table 5. The accuracy of modal frequencies from the 5-th to 8-th order are elevated, even though they are not used in parameter identification; the highest error is decreased to 2.32%. After parameter identification, a highly accurate finite element model of the plate is obtained, which can reflect the dynamic characteristics (the macro-mechanical performance of stiffness) of the plate more precisely. A method is proposed to predict the elastic modulus of a fiber reinforced composite using Table 2. Relative sensitivities of the first eight modal frequencies with respect to elastic parameters. Elastic Mode order parameters 1 2 3 4 5 6 7 8 9 E 11 /GPa 2.82 11.10 6.31 9.97 43.20 35.26 14.06 75.07 78.50 μ 12 0.075 2.19 4.67 0.50 0.50 4.62 4.62 14.2 31.61 μ 13 1.08 1.37 5.63 1.51 8.40 4.06 0.96 19.61 22.04 E 22 /GPa 3.81 21.08 20.34 43.21 6.97 62.04 35.50 75.09 78.17 μ 23 0.13 0.52 4.35 6.06 1.12 0.99 24.22 15.00 13.67 E 33 /GPa 0.20 1.15 6.12 4.88 5.67 25.73 28.55 21.64 22.11 G 12 /GPa 10.62 3.39 12.37 46.15 46.16 12.88 22.89 25.73 96.01 G 23 /GPa 0.062 0.035 0.024 0.454 0.335 1.712 0.184 3.439 1.601 G 31 /GPa 0.058 0.033 0.023 0.316 0.428 0.173 1.616 3.241 1.507
061001-5 Numerical-experimental method for elastic parameters identification Change of elastic parameters/% 0-1 -2-3 -4 E 11 E 22 G 12-5 0 10 20 30 40 50 Iteration Fig. 3. Convergence of elastic parameters. Table 3. Correlation of the experimental and numerical mode shapes before identification. Experimental Computational mode order mode order 1 2 3 4 1 0.974 8 0.000 1 0.004 9 0.008 8 2 0.000 7 0.971 1 0.005 5 0 3 0.000 2 0.005 4 0.967 0 0.009 5 4 0 0.003 8 0 0.961 1 Table 4. Elastic parameters after identification. Elastic parameters E 11 /GPa E 22 /GPa G 12 /GPa After identification 114.16 124.01 36.24 Table 5. Comparison of modal frequencies between experimental and computational results. Mode Experimental Computational data/hz order frequencies/hz Before parameter identification Error /% Identified results Error /% 1 154.24 156.88 1.71 154.0 0.15 2 255.00 257.6 1.02 253.96 0.41 3 282.85 287.07 1.51 283.94 0.38 4 424.98 432.11 1.68 425.13 0.03 5 761.45 790.06 3.76 779.15 2.32 6 845.72 862.62 2.00 848.98 0.38 7 902.54 895.06 0.83 882.06 2.25 8 949.77 980.27 3.21 967.7 1.88 Error = ( f a f e )/ f e 100%, f a is the computational frequencies, f e are the test frequencies. vibration test data. Homogenization method is employed for predicting equivalent elastic modulus of the periodic composite materials, and the errors arising from the idealization assumption is
061001-6 D. Jiang, et al. Theor. Appl. Mech. Lett. 4, 061001 (2014) inevitable. Effectiveness of the presented approach has been verified by adopting an experimental composite panel. This work was supported by the Program for New Century Excellent Talents in University (NCET- 11-0086), the National Natural Science Foundation of China (10902024), the Doctoral Program of Higher Education of China (20130092120039), and the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD-1105007001). 1. A. Mahmood, X. Wang, C. Zhou. Modeling strategies of 3D woven composites: A review. Composite Structures 93, 1947 1963 (2011). 2. A. Hallal, R. Younes, F. Fardoun. Review and comparative study of analytical modeling for the elastic properties of textile composites. Composites Part B: Engineering 50, 22 31 (2013). 3. A. Dalmaz, D. Ducret, R. El Guerjouma, et al. Elastic moduli of a 2.5D C-f/SiC composite: Experimental and theoretical estimates. Composites Science and Technology 60, 913 925 (2000). 4. J. Zheng, W. Wen, H. Cui, et al. Geometric model of 2.5 dimensional woven structures. Acta Materiae Compositae Sinica 25, 143 148 (2008) (in Chinese). 5. C. Kong, Z. Sun, X. Gao, et al. Unit cell of 2.5 dimension C/SiC and its stiffness prediction. Journal of Aerospace Power 26, 2459 2467 (2011) (in Chinese). 6. L. C. Pardini, M. L. Gregori. Modeling elastic and thermal properties of 2.5D carbon fiber C/SiC hybrid matrix composites by homogenization method. Journal of Aerospace Technology and Management 2, 183 194 (2010). 7. P. Tan, L. Tong, G. P. Steven. Modelling for predicting the mechanical properties of textile composites A review. Composites Part a-applied Science and Manufacturing 28, 903 922 (1997). 8. A. Dixit, H. S. Mali. Modeling techniques for predicting the mechanical properties of woven-fabric textile composites: A review. Mechanics of Composite Materials 49, 1 20 (2013). 9. I. A. Jones, A. C. Long, J. J. Crookston. A summary review of mechanical properties prediction methods for textile reinforced polymer composites. Journal of Materials: Design and Applications 219, 91 109 (2005). 10. J. Bystrom, N. Jekabsons, J. Varna. An evaluation of different models for prediction of elastic properties of woven composites. Composites Part B-Engineering 31, 7 20 (2000). 11. A. Hallal, R. Younes, F. Fardoun, et al. Improved analytical model to predict the effective elastic properties of 2.5D interlock woven fabrics composite. Composite Structures 94, 3009 3028 (2012). 12. K. Pochiraju, T. W. Chou. Three-dimensionally woven and braided composites. II: An experimental characterization. Polymer Composites 20(6), 733 747 (1999). 13. J. E. Mottershead, M. I. Friswell. Model updating in structural dynamics a survey. Journal of Sound and Vibration 167, 347 375 (1993). 14. J. E. Mottershead, M. Link, M. I. Friswell. The sensitivity method in finite element model updating: A tutorial. Mechanical Systems and Signal Processing 25, 2275 2296 (2011). 15. Q. G. Fei, J. F. Ding, X. L. Han, et al. Criteria of evaluating initial model for effective dynamic model updating. Journal of Vibroengineering 14, 1362 1369 (2012). 16. Q. G. Fei, Y. L. Xu, C. L. Ng, et al. Structural health monitoring oriented finite element model of Tsing Ma bridge tower. International Journal of Structural Stability and Dynamics 7, 647 668 (2007). 17. G. Kim, Y. Park. An automated parameter selection procedure for finite-element model updating and its applications. Journal of Sound and Vibration 309, 778 793 (2008). 18. N. A. Husain, H. H. Khodaparast, H. Ouyang. Parameter selection and stochastic model updating using perturbation methods with parameter weighting matrix assignment. Mechanical Systems and Signal Processing 32, 135 152 (2012).