Fatigue life prediction of ceramic matrix composite materials based on physical constraint multi-scale attention network

In response to the complexity of fatigue behavior of ceramic matrix composites in high-temperature and high stress environments, the author proposes a fatigue life prediction method based on a physically constrained multi-scale attention network (PC MSANet). The study obtained fatigue data of Cf/SiC materials under multiple working conditions through experiments, and extracted features such as macroscopic stress, time-domain statistics, and frequency-domain energy. The results showed that PC MSANet was significantly better than the comparison model in terms of prediction accuracy and stability, with RMSE and MAE of 477.4 and 338.5, R² of 0.975, and CSR satisfaction rate of 99.0%, which was superior to LSTM (RMSE=787.5, R² =0.932), MLP (RMSE=849.1, R² =0.921), and SVR (RMSE=928.7, R²=0.906). Further analysis shows that introducing physical constraints can reduce RMSE from 905.6 to 630.0, MAE from 647.7 to 408.3, and R² to 0.957. This method improves the physical rationality of prediction results through soft constraints in the loss function while maintaining high accuracy, providing a new approach for evaluating the service life of composite materials.