Data-driven deep learning models for predicting off-axis tensile damage of 2.5D woven composites at elevated temperatures

While finite element (FE) simulation has proven effective in predicting damage in fiber-reinforced composites under mechanical loads, it still remains time-consuming and resource-intensive, making it less suitable for tasks requiring extensive parametric analyses. This study introduces deep learning models employing a predictor- decoder architecture that can directly predict the damage state of 2.5D woven composites under coupled elevated temperature and off-axis tensile loads based on temperature and off-axis angle. The results demonstrate that these deep learning models, trained on comprehensive damage datasets, not only achieve a prediction speed approximately 104 times faster than FE simulations but also reduce the average prediction error by about 30 % across ten test cases. The deep learning-assisted damage analysis reveals that off-axis angle influences constituent damage by altering the load distribution while temperature affects damage by degrading the matrix performance. This research highlights the significant potential of data-driven deep learning models in predicting damage of textile composites with complex microstructures, which is crucial for accelerating optimization design and enabling online health monitoring of the related composite structures.