Predictions on surface finish in electrical discharge machining based upon neural network models

Abstract

Predictions on the surface finish of work-pieces in electrical discharge machining (EDM) based upon physical or empirical models have been reported in the past years. However, when the change of electrode polarity has been considered, very few models have given reliable predictions. In this study, the comparisons on predictions of surface finish for various work materials with the change of electrode polarity based upon six different neural-networks models and a neuro-fuzzy network model have been illustrated. The neural-network models are the Logistic Sigmoid Multi-layered Perceptron (LOGMLP), the Hyperbolic Tangent Sigmoid Multi-layered Perceptron (TANMLP), the Fast Error Back-propagation Hyperbolic Tangent Multi-layered Perceptron (Error TANMLP), the Radial Basis Function Networks (RBFN), the Adaptive Hyperbolic Tangent Sigmoid Multi-layered Perceptron, and the Adaptive Radial Basis Function Networks. The neuro-fuzzy network is the Adaptive Neuro-Fuzzy Inference System (ANFIS). Being trained by experimental data initially screened by the Design of Experiment (DOE) method, the parameters of the above models have been optimally determined for predictions. Based upon the conclusive results from the comparisons on checking errors among these prediction models, the TANMLP, RBFN, Adaptive RBFN, and ANFIS model have shown consistent results. Also, it is concluded that the further experimental results have agreed to the predictions based upon the above four models.