Multiscale sensitivity analysis of elasticity properties for woven composites using data-driven self-consistent clustering analysis and artificial neural network

Sensitivity analysis of material microscopic properties like matrix and fiber parameters on macroscopic performance is critical for the refined design of the mechanical elastic equivalent properties. However, this work remains challenging due to the high computational cost of traditional multiscale methods. In this study, a method integrating data-driven self-consistent clustering analysis (SCA) and artificial neural network (ANN) is proposed to address this issue. Firstly, the SCA × SCA multiscale framework is employed to predict macroscale elastic properties through microscale and mesoscale representative volume element based on homogenization. A high-accuracy database linking microscale and macroscale elasticity properties is efficiently generated. Subsequently, an ANN surrogate model is established and trained using this database for rapid prediction of macroscale properties. Last, combined with the improved Fourier amplitude sensitivity test, the cross-scale parameter sensitivity correlation is obtained. The results indicate that the in-plane equivalent modulus of the material is mainly influenced by the axial modulus of the fiber monofilament, while the out-of-plane equivalent modulus and in-plane shear modulus are mainly influenced by the transverse modulus and Poisson’s ratio of the fiber monofilament. Poisson’s ratio is mainly influenced by the modulus and Poisson’s ratio of the fiber monofilament in the corresponding direction.