Multivariable Intelligent Control for M.A.G. Welding Process

Abstract

A neural control technique, applied to the MAG (Metal-Active Gas) welding process, is presented in the paper. The static nonlinear model of welding process is based on experimental determinations. The geometric parameters of the welding beam are considered as output parameters of the MAG process (Bs, a, p), and they are measured for different step-variations of the input parameters (Ve, Vs, Ua). The analysis of the output dynamics was further used to model the MAG welding process using a 3- layer neural network with 6 hidden-layer neurons. In order to reject perturbations and cancel the stationary error, an error compensator was used, which consists of the reverse dynamic model connected to a proportional integrator controller. Simulation results for the multivariable neural controller are presented.