Real-time seam defect identification for Al alloys in robotic arc welding using optical spectroscopy and integrating learning

Abstract

Accurate on-line weld defect detection in robotic arc welding manufacturing is still challenging, due to the complexity and diversity of weld defects. In this study, a new real-time defect identification method is proposed for Al alloys in robotic arc welding, using arc optical spectroscopy and an integrated learning method. Spectrum feature was extracted, based on the absolute coefficients of the principal components. Feature importance was quantitatively evaluated using the mean decrease accuracy of Principal Component Analysis-Random Forest (PCA-RF). A new indicator, e.g., Importance Factor, was proposed, based on the variance of the out-of-bag test error of RF to select the optimal feature subset. The proposed PCA-RF proved to effectively identify five classes of weld defects with better performance than support vector machine and back propagation neural network. Finally, the selection pattern of spectrum feature subset was investigated, before revealing the correlation mechanism of the selected lines spectrum and weld process.