Fatigue life prediction based on a deep learning method for Ti-6Al-4V fabricated by laser powder bed fusion up to very-high-cycle fatigue regime

Microstructural defects and inhomogeneity of titanium alloys fabricated by laser powder bed fusion (LPBF) make their fatigue behaviors much more complicated than the conventionally made ones, especially in very-high-cycle fatigue (VHCF) regime. Most of traditional models/formulae and currently-used machine learning algorithms mainly concern fatigue behavior of LPBF-fabricated titanium alloys in high-cycle fatigue (HCF) regime, but rarely in VHCF regime. In this paper, a deep belief neural network-back propagation (DBN-BP) model was proposed to predict the fatigue life of LPBF-fabricated Ti-6Al-4V up to VHCF regime. Results obtained in this study indicate that the DBN-BP model exhibits high precision and strong stability in predicting the fatigue life of LPBFfabricated Ti-6Al-4V in both HCF and VHCF regimes. The primary hyperparameters of the DBN-BP model were optimized to further improve the prediction precision of this innovative model. Finally, the optimal DBN-BP model was applied to predict the relation between mean stress and stress amplitude, and the effect of energy density on the fatigue behavior of LPBF-fabricated Ti-6Al-4V up to VHCF regime.