Prediction of Low‐Cycle Fatigue Life for Q345R Steel Using Machine Learning

To overcome the limitations of traditional fatigue life prediction models, this paper proposes a machine‐learning‐based framework for low‐cycle fatigue life prediction. With ultrasonic shot‐peening‐treated Q345R steel as the subject of study, three key process parameters—the shot peening time, shot diameter, and shot peening height—were selected. Response surface methodology (RSM) was used to design experiments and construct a fatigue life dataset. The genetic algorithm (GA), particle swarm optimization (PSO), and simulated annealing (SA) were used to optimize five machine learning models—support vector regression (SVR), Gaussian process regression (GPR), gradient boosting decision tree (GBDT), polynomial regression (PR), and gradient boosted tree (GBT)—to predict the fatigue life with different parameter combinations. The results indicated that the PSO‐GBDT‐RSM and SA‐GBDT‐RSM models performed best on the test set, both achieving fatigue lives of 105,754 cycles. This demonstrated that both models exhibited strong adaptability and stability in terms of parameter optimization and response prediction.