Artificial neural networks modelling for surface roughness in wire electrical discharge machining of Incoloy 800H

Abstract

The use of artificial neural networks (ANNs) in optimizing process parameters and model development has gained attention due to their ability to predict timely and accurate outputs. The current study uses ANNs to predict surface roughness during wire electrical discharge machining (WEDM) of Incoloy 800H. The prediction model utilizes a multilayer perception based on back-propagation neural network (BPNN). In neural network modelling, network weights and bias values were updated, and the process output was estimated using the Levenberg-Marquardt backpropagation training function (trainlm) and a hyperbolic tangent sigmoid (tansig) transfer function, respectively. The ANN model comprises one output neuron (surface roughness) and five input neurons (pulse on, pulse off, wire tension, servo voltage, and wire feed rate). WEDM experimental data is divided into training, testing, and validation in the ratio of 70%:15%:15%. The results show that the optimal ANN model with 5-11-1 topology provides an accurate estimate of surface roughness, with a correlation coefficient of 0.997. The proposed ANN model can predict WEDM process parameters quickly and effectively. © 2024 IEEE.