Classification of metal transfer mode using neural networks

To develop a control strategy for a metal inert gas (MIG) welding system it is necessary to classify several parameters in order to describe the process state. Neural networks have been identified as an appropriate processing technology because of the noisiness of weld data and the nonlinearity of the relationships between many of the process parameters. This paper describes the application of neural networks to the classification of metal transfer mode. We report on the analysis of the data, network selection, network development and evaluation of the final system