Fatigue crack initiation mechanism and life prediction of laser-arc hybrid welded TC4 titanium alloy joints

Abstract

Titanium alloy welded structures in aerospace applications frequently undergo high-cycle fatigue (HCF) failure under high-frequency cyclic loading. Microstructural inhomogeneity in weld zones induces divergent deformation mechanisms, while welding defects compromise the accuracy of conventional life prediction models. This study systematically investigates HCF failure mechanisms in laser-arc hybrid welded TC4 titanium alloy joints under varying maximum cyclic stresses (σ_max = 410–450 MPa). Microstructural characterization reveals: The weld zone comprises predominantly acicular α′ martensite (≈43 μm) and basket-weave structures, while the heat-affected zone (HAZ) exhibits refined α′ phase (≈22 μm). At low σ_max (410 MPa), dense basket-weave structures surrounding pores obstruct dislocation slip and induce crack deflection, forming tortuous propagation paths that extend HCF life. Conversely, high σ_max (450 MPa) triggers severe dislocation pile-up at α colony interfaces, accelerating crack propagation through colony channels and significantly reducing HCF life. To address prediction uncertainties caused by defect-induced life scatter, a physics-informed neural network (PINN) model is developed. This framework integrates defect parameters (size √area, distance d, circularity Cir) with physical laws via penalty-function-constrained loss functions. Compared to conventional BPNN, the PINN model improves test-set prediction accuracy by 16 % (R² = 0.86), mitigates over-fitting, and confines nearly all predictions within triple-error bands. This research has achieved rapid and precise assessment of the fatigue life of welded titanium alloy components, providing critical technical support for lightweight design and reliability assurance in aerospace equipment.