Machine learning approach for weld configuration classification within the GTAW process

Abstract

In the present study, an attempt has been made to couple experimental data with a machine learning (ML) approach to classify several weld configurations. An ML model has been developed and fed into experimental data captured by several sensors during the gas tungsten arc welding (GTAW) process. On the one hand, welding parameters (voltage, current, wire speed, welding speed, etc.) were used to monitor the control energy transmitted during welding. On the other hand, cameras coupled to an image-processing algorithm were employed to capture the weld pool contour in situ. A database was also constructed to store, label, and order the obtained information. This database was then used for the various training, validation, and prediction steps of the ML model. The welding configurations were then classified using a KNN classification algorithm, which was then analyzed for their efficiency (accuracy, processing time, etc.). It was shown that image processing combined with ML can be trained with the features which were extracted to predict the classification of welding configurations. The ultimate perspective of the current study is to realize real-time identification and modification of welding operating conditions.