Abstract: Welding process is modelled and controlled by artificial neural networks. Taking GTAW as the object, it is realized in the process. The relationship between face bead width, back bead width, which is the pxpression of the quality, and welding parameters is established. It is static model. The feasibility is tested from both the quality and quatity aspects. The modelling and controlling method for the GTAW dynamic process is put forward by two networks. Experiments show the validity of the method. This method is totally different from the traditional one which is basesd on fixed structure mathematical model in which it does not need make any assumption. The established model fits well with the reality. Experiments show that the method has the advantages of good fault tolerance, generality and disturbance resistance etc. It finds a new way for the modelling and intelligent controlling of the welding process.

Key words: Artificial neural network, GTAW, Intelligent controlling