Abstract: The degradation or failure of spring performance due to stress relaxation can significantly impact the normal operation of the entire system. Physical models are often used to fit the degradation law of spring stress relaxation. However, the error in the prediction process has not yet been emphasized. Therefore, an error compensation model based on PSO-TCN-Attention is proposed in this study. PSO is applied to optimize the TCN-Attention hyperparameters for data at different temperatures. The prediction model is optimized to learn the difference between the empirical equations and the true value, and the prediction error is then compensated for in the empirical equations to achieve a more accurate prediction of spring degradation. The results indicate that the PSO-TCN-Attention model predicts optimally at four temperatures compared to the TCN, RNN, and LSTM models as well as the empirical equation and delay function models. Based on this, the spring’s life at a failure threshold of 0.95 is extrapolated, and the life distribution is obtained. Additionally, the relaxation mechanism of the spring is analyzed using EBSD and TEM at different temperatures before and after pressurization. The results indicate that the relaxation mechanism of the spring is mainly due to dislocation and slip. The proposed algorithmic model effectively describes the spring degradation trend, suggesting its potential for predicting spring performance degradation.

The full text can be downloaded at https://doi.org/10.1007/s40436-024-00538-5

Key words: Stress relaxation, Error compensation model, Remaining useful life (RUL), Microscopic characterization