Gas metal arc welding of butt joint with varying gap width based on neural networks

Abstract

Arc welding is still the dominant welding process in industry and a major challenge in this context is robotisation and automation, which require sensor based adaptive control. Therefore, the present paper describes the application of neural network technology for gas metal arc welding control. A system based on a multilayer feed forward network has been developed for modelling and online adjustment of welding parameters appropriate to guarantee a certain degree of quality in the field of butt joint welding with full penetration when the gap varies during the welding process. It has been shown that operating in an open loop, the developed system can tackle gap variations relevant to many welding situations. However, to improve robustness to uncertainties, disturbances, etc. a closed loop control system based on a ‘single neuron self-learning proportional, sum, and differential’ control algorithm, which compensates for non-monitorial changes in welding conditions by feeding back information on the realised front bead geometry, has also been developed and tested. The Levenberg–Marquardt algorithm for non-linear least squares has been used with a back propagation algorithm for training the neural networks, and a Bayesian regularisation technique has been successfully applied for minimising the risk of inexpedient overtraining.