Forward and reverse mappings of the tungsten inert gas welding process using radial basis function neural networks

Abstract

In the present study, forward and reverse mapping problems of the tungsten inert gas (TIG) welding process have been solved using radial basis function neural networks (RBFNNs), which is required to automate the welding process. The performance of an RBFNN depends on its structure and parameters. Here, a few approaches are proposed to optimize its structure and parameters simultaneously. A binary-coded genetic algorithm (GA) has been used for the said purpose. The GA strings carrying information of the networks might have varied lengths, and consequently, it becomes difficult to implement the conventional crossover operator. To overcome this difficulty, a new scheme has been adopted here. The performances of the developed approaches are tested to conduct both forward and reverse mappings of a TIG welding process. Cluster-based approaches are found to perform better than the non-cluster-based ones.