Deep learning-based thermomechanical fatigue life of nickel-based superalloys

This paper introduces two deep learning-based approaches developed to predict the thermomechanical fatigue (TMF) life of nickel-based superalloys subjected to complex service conditions involving the interaction of fatigue, creep, and oxidation damage. The first method utilizes a convolutional neural network (CNN) to extract fatigue-relevant features from half-life hysteresis loops transformed into image representations. The CNN model is pre-trained on a DD6 single-crystal superalloy dataset and fine-tuned for application to DZ406 and DZ125 directionally solidified alloys. The second method integrates 1D-CNN, long short-term memory (LSTM) networks, and an attention mechanism to construct a spatiotemporal model capable of learning directly from complete loading sequences without needing image preprocessing. Comparative experimental evaluations demonstrate that both approaches achieve high prediction accuracy and strong correlation with experimental results. These findings confirm the feasibility and effectiveness of data-driven modeling for TMF life prediction and highlight its potential for broader engineering applications under multivariate loading conditions.