Creep deformation and continuous damage mechanics parameters prediction based on a meta-learning framework

The increasing operational temperatures in aerospace applications present significant challenges in predicting the long-term creep behavior of high-temperature alloys. Accurate prediction of creep deformation is crucial for ensuring the reliability and safety of engine components. This study proposes a novel continuous damage mechanics (CDM) parameter prediction framework for accurately predicting the long-term creep deformation of superalloys. The framework combines a fully connected neural network (FCNN) and a long short-term memory network (LSTM) to predict long-term creep performance by learning from short-term creep data. The key innovations of this research include the development of a meta-learning framework based on FCNN weight parameter evolution, the proposal of a hybrid loss function that combines mean square error (MSE) and total variation (TV) regularization, and the verification of the method’s effectiveness through multiple case studies. The experimental results show that the framework can accurately predict the creep deformation of various high-temperature alloys at different temperatures and stresses based on short-term creep data. The method can extrapolate the creep deformation to 2–5 times the short-term creep data, and compared with the traditional Larson-Miller method, it can control the life prediction error within ± 10 %, which shows excellent prediction performance. At the same time, it can be fitted to obtain the creep curve under the corresponding conditions, and the fitting accuracy is within 10 % of the time error.