Analysis of GMAW process with deep learning and machine learning techniques

Abstract

GMAW (Gas Metal Arc Welding) is widely applied industrially in ferrous and non-ferrous materials. The interrelation of its initial and final parameters of GMAW process has a non-linear behavior. This is one of the main problems that researchers have to model the GMAW process. Arc parameters capture is difficult due to image process time with the most classic image analysis techniques. Presenting the arc parameters in a model is of high importance because they reflect the disturbances that occur in the GMAW process. Current advances in image processing and computational model predictions areas, help to optimize processes and save costs. These techniques can obtain good results in welding analysis. Deep Learning techniques obtain excellent results in the classification of complex images. Moreover, a machine learning technique can contribute to the predictive model, using the arc parameters obtained in the image processing with deep learning technique. This research aims to develop a framework for weld bead geometry prediction of GMAW process, joining two machine learning techniques. The results obtained demonstrated the effectiveness of these algorithms to predict GMAW process and its potential for real-time analysis.