Welding Bead Inspection Using Image and Multi-Sensor Fusion

Abstract

Welding is a crucial manufacturing technique utilized in various industrial sectors, playing a vital role in production and safety aspects, particularly in shear reinforcement of dual-anchorage (SRD) applications, which are aimed at enhancing the strength of concrete structures, ensuring that their quality is of paramount importance to prevent welding defects. However, achieving only good products at all times is not feasible, necessitating quality inspection. To address this challenge, various inspection methods were studied. Nevertheless, finding an inspection method that combines a fast speed and a high accuracy remains a challenging task. In this paper, we proposed a welding bead quality inspection method that integrates sensor-based inspection using average current, average voltage, and mixed gas sensor data with 2D image inspection. Through this integration, we can overcome the limitations of sensor-based inspection, such as difficulty in identifying welding locations, and the accuracy and speed issues of 2D image inspection. Experimental results indicated that while sensor-based and image-based inspections individually resulted in misclassifications, the integrated approach accurately classified products as ‘good’ or ‘bad’. In comparison to other algorithms, our proposed method demonstrated a superior performance and computational speed.