Machine learning-based prognosis of fatigue response in AA2014

Abstract Precise prediction of fatigue life is vital to ensure structural integrity in aerospace components like fuselage frames, bulkheads, and wing fittings. These parts are generally manufactured from AA2014 due to its high strength, fatigue resistance, and superior strength-to-weight ratio. The present work proposes a machine learning (ML)-driven approach for fatigue life prognosis in AA2014 to overcome the limitations of traditional fatigue life prediction models. Experimental fatigue tests were conducted at different maximum stress levels (320-350 MPa) and loading frequencies (9 Hz and 12 Hz) to generate a comprehensive dataset. This dataset was used to develop the ML models and performance metrics like R2, RMSE, and MAPE were recorded for comparative analysis of the three MLmodels: Random Forest Regression (RFR), Support Vector Regression (SVR), and Gaussian Process Regression (GPR). GPR was found to be the most reliable model for fatigue prognosis in AA2014 as it yielded the highest predictive accuracy with an R2 of 0.9874, RMSE of 618.2, and a MAPE of 1.89%. The work proves that ML can greatly enhance fatigue life predictive accuracy and drastically reduce the dependence on experimental runs. Future studies can investigate hybrid AI models by integrating deep learning and physics-informed ML approaches to enhance prediction reliability and expand applicability to other aerospace-grade alloys.