Development of artificial intelligence technology for robotic GMAW quality improvement

Arc commencement and termination conditions are critical to producing high quality welds. For gas metal arc welding (GMAW), incorrect arc initiation can lead to defects, such as macroporosity, spatter and lack of fusion. Inappropriate arc termination can generate the same, in addition to solidification cracking. This in turn will diminish mechanical properties of the welded components. There are a multitude of factors which can be applied within a robotic or mechanized process, such as ignition current, ignition time, current used at the end of the process and duration of the currentless wire withdrawal after the welding current has been switched off. Often, statistical methods such as the Taguchi Method are employed over a narrow range of viable process parameters, but these methods fail if welding conditions necessitate large changes in processing parameters. To address this, a machine learning algorithm has been developed in this study to build the relationship between welding parameters and welding start and end quality for a robotic GMAW system for steel. A statistical matrix has been used to quantify the quality of welding start and end based on digitization of a series of bead geometries. Quality factors, including the symmetry and uniformity of the bead and the amount of spatter identified have been incorporated. After extracting features using this quality matrix, a surrogate model was produced based on the developed machine learning algorithm for weld quality optimization. The surrogate model was able to significantly reduce the number of trials required to obtain a target quality metric. A workflow of extracting experimental results and adopting this implementation of artificial intelligence to obtain predictions of welding quality is illustrated.