Development and assessment of machine learning models for predicting fatigue response in AA2024

Accurate prediction of fatigue life is vital in the design of aerospace components subjected to varying stress levels and loading frequencies. In the current research, machine learning (ML) models were developed to predict the fatigue life of AA2024-T6, a popular aerospace grade alloy, under different stress levels and loading frequencies. The aim was to reduce the reliance on expensive and time-consuming experimental fatigue testing. Stress-controlled fatigue tests were conducted, followed by fractographic analysis using a scanning electron microscope to assess failure mechanisms. It was observed that fatigue life decreases with increasing loading frequency, with failure modes transitioning from ductile at higher stress levels to a combination of brittle and ductile at lower stress levels. Three ML models namely Elastic Net, k-NN, and Random Forest were evaluated using the experimental fatigue results as input. The Random Forest model, optimized with an 85%-15% training-testing data split and nine decision trees, outperformed other models with a Root Mean Square Error (RMSE) of 101.62, Mean Absolute Percentage Error (MAPE) of 5.23% and an R-squared value of 1.0. Confirmation experiments validated the model, showing an average deviation of 7.57% between predicted and actual fatigue lives. These results highlight the potential of ML models to accurately predict fatigue life, offering a reliable alternative to tedious and costly experimental methods.