A machine-learning fatigue life prediction approach of additively manufactured metals

•Support vector machine (SVM) model was used to characterize the defect population.•Defect location, size and morphology collaboratively determine the high cycle fatigue life.•Synchrotron X-ray tomography can well acquire the geometric features of the defects. The defects retained during laser powder bed fusion determine the poor fatigue performance and pronounced lifetime scatter of the fabricated metallic components. In this work, a machine learning method was adopted to explore the influence of defect location, size, and morphology on the fatigue life of a selective laser melted Ti-6Al-4 V alloy. Both the high cycle fatigue post-mortem examination and synchrotron X-ray tomography were combined to acquire the geometric features of the critical defects, which were trained using a support vector machine (SVM). To accelerate the optimization process, the grid search approach with cross validation was selected for fitting the model parameters. It is found that the coefficient of determination between the predicted and experimental fatigue lives can reach up to 0.99, indicating that the SVM model shows strong training ability.