Multiscale Computational Method Of Woven Composites Based On Clustering Analysis

Woven composites are widely used in engineering because of their excellent mechanical properties,such as high specific stiffness and specific strength.Th e mechanical behavior of woven composites and their structures is directly determined by the deformation mechanism of microstructure.Using multiscale numerical method to predict the mechanical response of woven composites and its structure by using microstructure,the computational cost is huge and the efficiency is low,sometimes it can not even complete the calculation task.In order to improve the effici ency of multiscale calculation,the self-consistent clustering analysis(SCA)model is used in this thesis,and the multiscale computational methods of material level and structure level for woven composites are proposed respectively based on computational homogenization.The prediction of mechanical response of woven composites and their structures by microstructure is realized.Firstly,the parameters of theoretical derivation and numerical algorithm of SCA are studied.The numerical iterative scheme for solving the discrete Lippmann Schwinger integral equation is derived,and then the clustering algorithm is introduced to reduce the degree of freedom of the model and improves the computatinal efficiency.Finally,the whole solution process is divided into two stages:online database generation and offline iterative solution.At the same time,taking the two-dimensional circular inclusion problem as the solution object,the solution result of the finite element is taken as the reference.The influence of clustering index,clustering distance and clustering number on the calculation results is studied.Compared with the position-based clustering,the error of SCA is reduced from 4%-5% to 2% under tension,and from 4%-6% to 2% under shear using strain concentration tensor to cluster,which improves the accuracy.For three-dimensional problems,the results of FEM and SCA are coincident,which verifies the SCA.Secondly,the application of SCA in fiber-reinforced composites unit cell is studied and verified.The reinforcement phase is linear elastic and the matrix is nonlinear plastic.The FEM is used as reference.It is found that for the unidirectional fiber-reinforced composites,the solution error of SCA is less than 2% under tension and less than 4% under shear.For the woven composites,the error is less than 3%under tension and less than 5% under shear.The SCA is verified to solve the unit cell problem of fiber reinforced composites.After getting the database,by comparing the running time of the SCA and the FEM,it is found that the efficiency of the SCA is increased by 600-1000 times for different problems and algorithm parameters in this thesis.Then,by using the SCA,the equivalent yield surface of the woven composites is predicted,and the database of the relationship between elastic properties of the components and the equivalent elastic properties is generated.The high-accuracy neural network surrogate model for predicting the equivalent elastic properties is obtained.Then,the mesoscale and microscale unit cells of woven composites are solved by using the SCA,and the relationship between the two scales is realized by using computational homogenization.A multiscale method SCA×SCA for predicting the macroscale response of woven composites from microscale is established.Taking the carbon/carbon woven composite as an example,the uniaxial tensile test is carri ed out and the results of multiscale calculation and experiment are in good agreement.By using the efficiency of SCA×SCA,the database of the relationship between the material parameters of micro components and the macro equivalent performance is obtained,and a more efficient neural network surrogate model is trained.Combined with Sobol and FAST parameter sensitivity analysis method,the influence of different micro material parameters on the macro equivalent performance is obtained.Finally,a multiscale method FEM×SCA for predicting the structural response of woven composites is established by using the FEM in macroscale and the SCA in mesoscale.The multiscale damage simulation of the open-hole plate of carbon/carbon woven composites is carried out.The calculation results are in good agreement with the experimental results,and the multiscale method FEM×SCA and mesoscopic damage model are validated.Through the multiscale numerical model,the mesoscopic damage mode evolution of different positions of the plate at different loading points is given.It is found that due to the low strength of the matrix,the damage always occurs first,then in the transverse direction of the fiber bundle,and finally when the macro cracks arrive,the fiber bundle along the load direction breaks.Based on this multiscale method,the typical engineering examples such as Iosipescu shear test,biaxial tensile test and T-shaped multi-layer plate bending are studied numerically,and the deformation mechanism of mesoscopic microstructure at different positions in the structure under different loading conditions is revealed.