Resistance spot welding defect detection based on vectorized dynamic resistance signal and LightGBM classifier

Abstract The problem of real-time detection of welding defects is a difficult problem in resistance spot welding. It is found that the dynamic resistance has a strong connection with the growth of the nugget. The dynamic resistance signals with welding defects are significantly different from those of normal welding, and the dynamic resistance signals between different welding defects show different characteristics, and the dynamic resistance signals of the same kind of welding defects may also differ from each other. The most common practice today to realize the detection of resistive defects is by extracting the time-domain features of the signal waveforms. However, this approach is highly subjective, so this article proposes a double-size mesh division method to process the dynamic resistance signal. Experiments prove that the method can retain the characteristics of the signal curve well, and it is also improves the training speed and accuracy compared with the mesh division method. Finally, the processed signals are classified using the light gradient boosting machine classifier with an accuracy of 98.55%.