SVM-based information fusion for weld deviation extraction and weld groove state identification in rotating arc narrow gap MAG welding

Abstract

As a high-efficiency welding method, rotating arc narrow gap MAG welding (RANGMW) is accurate with its seam tracking to make enough groove sidewall penetration at both sides. The accuracy of traditional arc sensing-based methods is impacted by the bottom shape of the welding groove, especially when the shape is irregular. To overcome this shortage, an arc sensing-vision sensing synchronous data acquisition system was developed. More specially, the vision sensing method used passive vision and worked in infrared region to extract the welding character by an automatic image processing algorithm. Furthermore, by fusing the arc sensing and vision sensing information, a statistical learning model, support vector machine (SVM), was built to predict the groove state. The method was validated by tenfold cross-validation and also compared with back propagation (BP) neural network model. The results showed that our model met the demand of welding application and outperformed BP neural network, thus can help for the further application of welding quality control.