Real-time welding defect classification using peak count analysis of current signals with statistical validation

Abstract Welding is a critical process in heavy industries such as construction, automotive, and oil and gas, where weld quality directly impacts structural performance and safety. Traditional non-destructive testing (NDT) methods, although effective, are often labour-intensive, costly, and reliant on operator expertise. This study investigates an alternative approach using real-time monitoring of welding current signals to identify defects based on peak count variations. Under controlled laboratory conditions, welding current signals were captured and segmented into 1 mm intervals for detailed analysis. Statistical evaluation using Analysis of Variance (ANOVA) and Tukey’s post-hoc tests in R Studio revealed significant differences in peak distributions across various defect types. Good welds consistently exhibited 8–17 peaks per segment, while defects such as Lack of Penetration (LOP), Lack of Fusion (LOF), Burn-through, and Excess Weld displayed distinctive peak count deviations. These results confirm that peak count analysis is a statistically significant and reliable metric for real-time weld quality assessment. The findings lay the foundation for future development of intelligent welding systems capable of automated defect detection and adaptive process control.