Application of Deep Learning to Root Gap Identification in GMA Welding

抄録

In pulse GMA(Gas Metal Arc) welding, it is important that welding conditions are changed due to root gap of the groove during welding to ensure quality. The visual sensor and deep learning are useful to estimate the gap. In this study, fundamental experiments were carried out and images of the molten pool are taken under various root gaps. The gap identification is carried out using Resnet50. The gap was identified under the groove welding with 8mm gap. The accuracy of the identification was 100%. Moreover, the gap identification was carried out under the groove welding with the variation gap from 4mm to 8mm. The good performance of the gap identification can be obtained. The validity of the deep learning in GMA welding was verified.