Identification and predictive control of laser beam welding using neural networks

Abstract

Welding with laser beams is an innovative technique, which leads to higher penetration depth and a narrower seam compared to conventional welding techniques. One significant criterion of the quality of a junction is the penetration depth. Within this article a predictive control scheme is presented that optimises the process' input laser power by taking the future welding speed into account. For modelling the non-linear process an Artificial Neural Network (ANN) is applied. The GPC-algorithm with a linear model obtained by instantaneous linearization of the network is used. For this reason, an extended training of the ANN is introduced. First results of the application on a real laser welding system are described.