Machine learning framework for fatigue life prediction of laser powder-bed fusion additively manufactured FCC, BCC and HCP crystallographic structures

Accurate fatigue life prediction remains as one of the biggest hurdles in the widespread adoption of laser powder-bed fusion (LPBF) additively manufactured (AM) components. Conventional stress-based models struggle to capture the fatigue response due to the presence of internal defects and high surface roughness. Machine learning models, such as artificial neural networks (ANN), have been used in various engineering applications and offer a reliable alternative. In this work, an ANN model was used to predict the stress-based fatigue response of several polycrystalline materials with different R-ratio, orientation, post-processing conditions, and crystal structures (FCC, BCC, and HCP). Crystallographic, quasi-static and fatigue data was used as inputs to the proposed ANN model to predict the fatigue life under different conditions. The model showed high predictive accuracy (R2 = 0.932) and low mean square errors (0.672%). In addition, the model was also used to predict the Basquin equations for various alloys under different post-processing conditions. The predicted S-N curves showed strong agreement with the experimental results. This work provides a robust ANN framework for accurate predictions of fatigue response from multiple materials.