Correlation-guided temporal deep learning model for the prediction of fatigue life of printed AlSi10Mg alloy

Abstract

Accurate prediction of fatigue life in additively manufactured components remains challenging due to the complex interplay of mechanical properties, surface states, and microstructural factors. Correlation analysis enables the identification of key influencing parameters, while deep learning models offer the capability to capture complex temporal patterns associated with fatigue behaviour. This study presents a correlation guided temporal deep learning approach for predicting the fatigue life of AlSi10Mg alloy fabricated using laser powder bed fusion. The data are collected from independent literature from the analysis of V-notched specimens. The mechanical, microstructural, and surface characteristics including yield strength, elongation, hardness, residual stress, and surface roughness data are collected and corelated. The prediction shows elongation as the most influential parameter, whereas traditional strength indicators and surface hardness are statistically insignificant under cyclic loading. Guided by these findings, deep learning models are developed, among which the temporal convolutional network (TCN) achieved the highest predictive accuracy with an R² of 96.6%. Its ability to model long term dependencies without recurrence proved highly effective for capturing fatigue behaviour. The proposed framework demonstrates the potential of combining statistical insights with temporal deep learning to improve fatigue life prediction in additively manufactured components.