Exploring shear nonlinearity of plain-woven composites at various temperatures based on machine learning

Abstract

Plain-woven composites are extensively utilized across various fields; however, it exhibits significant shear nonlinearity, especially at high temperatures. This study aims to propose a machine learning (ML) based constitutive model using Gaussian Process Regression (GPR), which is able to effectively characterize the shear nonlinearity of plain-woven composites at different temperatures. The shear nonlinearity of T800 carbon fiber-reinforced epoxy-based plain-woven composites are investigated by carrying out in-plane shear experiments, and the data sets considering temperature effects are established accordingly. Compared with traditional constitutive models, the proposed ML-based model excels in predicting shear nonlinearity, even at temperatures not included in the training set. The integration of this ML-based constitutive model into the finite element (FE) simulation framework achieves high consistency between simulation and experimental results, thereby validating the significant application of ML in complex material behavior modeling and FE analysis.