Identification of in-situ elastic modulus fields in three-dimensional angle-interlock woven composites using an iterative data-driven framework

Due to the stochastic geometries spanning from microscale to mesoscale, fiber reinforced ceramic matrix woven composites exhibit significant spatial variations in the in-situ mechanical properties. These spatial variations strongly influence in governing strain localization, damage evolution and the ultimate fracture behavior of woven composites. Accurately characterizing the full-field spatially varying in-situ properties, especially the elastic modulus fields, remains challenging for traditional methods because of the complexity of the material heterogeneity. To address this challenge, an iterative data-driven inversion framework that integrates physics-informed neural networks (PINN) with the finite element method (FEM) is proposed. By employing an invariant-based theory, which allows for the constitutive equation to be normalized by Tsai's modulus, the ill-posed nature of the inverse problem is effectively mitigated. The PINN-based results demonstrate high predictive accuracy, validated through both virtual experiments and actual tensile tests of three-dimensional angle-interlock woven silica fiber reinforced silica ceramic matrix composites. The proposed framework provides a robust methodology for the inverse identification of spatially varying in-situ elastic modulus fields in the investigated composites. The improved accuracy in local mechanical property characterization enables a more reliable understanding of material behavior under experimental or service-relevant conditions and demonstrates potential for full-field nondestructive damage evaluation of composite structural components.