Online visual monitoring triggered by arc voltage during one-side GTAW

The random variations in the fit-up gap and misalignment of groove joints result in poor process adaptability of traditional robot teaching programming, which tends to induce welding defects such as burn-through. This paper proposes a visual online monitoring method based on molten pool triggered by arc voltage for burn-through defects. Welding process stability can be monitored through the fluctuations of arc voltage. Specifically, during the occurrence of burn-through defects, significant electrical signal instability (ranging from 14 to 20 V) is observed. The unstable phenomena can be detected accurately through LSTM network. Based on the corresponding images of molten pool, it can be discerned that when burn-through defects arise, the molten pool will exhibit distinct “separation” characteristics. By analyzing the “separation” features in molten pool images with aid of CNN network, the monitoring and classification of burn-through defects can be accomplished, and its accuracy rate is as high as 94.6%.