Integration of Micro-XCT and machine learning for defect-driven fatigue life prediction in DED-Arc titanium alloys

Defects are a non-negligible feature of direct energy deposition by arc (DED-Arc) titanium alloy components, adversely influencing their practical performance under fatigue loading conditions. This study employed micro-X-ray computed tomography (Micro-XCT) to characterize defects in DED-Arc titanium alloys and applied a support vector regression (SVR) model to precisely determine the influence of defect location, morphology, and size on stress concentration factor (SCF, K_t ). High-cycle fatigue (HCF) tests were conducted on defective DED-Arc titanium alloy. The results indicate that 51.5% of the HCF cracks in the DED-Arc titanium alloy initiate at defects. The SVR model, based on a radial basis function kernel, achieved a fitting accuracy of R² = 0.951 for the defect SCF ( K_t ) verification set. A new defect evaluation parameter, K* , integrating (area)^1/2 , K_t , and stress amplitude, was proposed. The R2 value of the linear correlation between K* and the HCF data was 0.6, representing a 62.2% improvement compared to K . The K*-N master curve ( K *=4.41 × 10⁶٠ N ^−0.4301) accurately predicted the HCF life of DED-Arc titanium alloy specimens, yielding a standard deviation of 0.38 within a 95% confidence interval.