Utilizing an artificial neural network to feedback-control gas metal arc welding process parameters

Abstract

The level of automation in welding has increased in recent decades. The motivation for the current research comes from industry, where the repair, rework and unnecessary postprocessing of welds due to varying welding conditions decreases the viability of the automated welding production. Thus, an Artificial Intelligence (AI) approach to feedback-control the Gas Metal Arc Welding (GMAW) process parameters was chosen as a research focus. An Intelligent Welding System (IWS) was developed using an Artificial Neural Network (ANN) with a Levenberg-Marquardt Algorithm (LMA) for Backpropagation (BP) as a decision-making system. The research aimed to solve the problem by addressing two research questions. The first question assessed the suitability and accuracy of the ANN with LMA for BP in welding applications. The second research question set out to evaluate the reliability of the developed IWS. First, the method to utilize Laser Triangulation Measurement (LTM) to achieve a reliable measurement of the major groove dimensions affecting weld quality. Second, the suitability of measuring the weld conditions, quality output, and data management within the IWS was addressed. Next, the effect of the learning process on the accuracy and suitability of the ANN with LMA for BP was evaluated. Finally, the suitability and reliability of the IWS were evaluated using data acquired in practical welding experiments under varying welding conditions.

The results of the research support the idea that an ANN with LMA for BP is suitable for feedback control in the GMAW process. From a statistical point of view, the reliability of the developed IWS is sufficient to reach the desired quality output according to the set measures. Further, the practical suitability and performance of the IWS are confirmed by the experimental investigation.