Rapid mechanical prediction of woven ceramic fabrics via a neural network surrogate model based on the parameterized unit cell

Ceramic fiber fabrics are vital for high-temperature morphing skins due to their exceptional thermal stability and structural adaptability. However, their mechanical properties are strongly influenced by weave architecture, necessitating detailed and systematic characterization. Current challenges include the lack of robust predictive theoretical models and the inefficiencies of experimental methods. This study tackles these issues by developing a parametric modeling framework for 2D woven fabrics using three topological parameters, combined with an automated simulation system to evaluate tensile and shear properties through Python-driven numerical analysis. The framework demonstrates high predictive accuracy, validated by experimental data. Additionally, an artificial neural network (ANN) surrogate model employs the resulting property database to reveal correlations between weave architecture and mechanical properties. A novel integrated resistance factor is introduced to comprehensively assess mechanical performance, identifying plain weave architectures as optimal for combined tensile and shear resistance. This ANN-based surrogate model approach significantly improves efficiency in material design and performance prediction.