Evaluation of liquid metal embrittlement crack in resistance spot welds under intensive welding condition using industrial X-ray computed tomography and machine learning

In this work, the industrial X-ray computed tomography was exploited for measuring maximum crack depths across the weld interfaces within resistance spot welding (RSW) specimens welded under intensive welding condition. Individual crack profiles were measured for each sample and presented in terms of the total crack length, the maximum crack depth and the median crack width. A linear trend between maximum crack depth and median crack width was observed, where the associated coefficient of determination was sufficiently high ( R 2 = 0.79). Hence, the crack width is proposed to predict the crack depth. Based on the experimental findings, the effectiveness of the proposed method is demonstrated through machine learning studies, and a simple relation to predict the internal crack depth from the surface crack width was further developed. Semantic segmentation results proposed an algorithm succeeded in separating enough crack areas to calculate the crack width. Since Zn-assisted liquid metal embrittlement (LME) is now accepted as an inevitable consequence with respect to RSW process, the methodology demonstrated in this work can potentially offer the insights into an effective and automatic method of quantifying LME cracks.