This study conducted fatigue crack growth rate (FCGR) experiments on TC4 titanium alloy specimens under different stress ratios (R=0.06, 0.1, 0.5) before and after ultrasonic impact treatment (UIT) conditions. Based on this, a Physics-Informed Neural Network (PINN) model was employed to predict the FCGR of TC4 titanium alloy. As a control group, the traditional Walker FCGR formula was fitted using experimental data. The results indicate that the stress ratio R does not significantly affect the FCGR of TC4 titanium alloy. Under the same range of stress intensity factor (ΔK), the FCGR (da/dN) decreases with increasing R. The FCGR prediction established by the PINN model can well reflect the nonlinear characteristics of the crack growth process rate. The coefficient of determination R2 for fitting the experimental data reached above 0.9900, generally higher than the 0.9545 of the Walker formula. Although the degree of dispersion increased after UIT, the PINN model demonstrated stable prediction accuracy compared to the Walker formula. In addition, when calculating the fatigue crack growth life using the cycle-by-cycle method, the PINN model also exhibits stable prediction accuracy compared to the Walker formula, and this advantage becomes more apparent with increased data dispersion.
Read full abstract