Surface roughness prediction of wire electric discharge machining (WEDM)-machined AZ91D magnesium alloy using multilayer perceptron, ensemble neural network, and evolving product-unit neural network

Abstract Magnesium (Mg) alloy parts have become very interesting in industries due to their lightness and high specific strengths. The production of Mg alloys by conventional manufacturing methods is difficult due to their high affinity for oxygen, low melting points, and flammable properties. These problems can be solved using nontraditional methods such as wire electric discharge machining (WEDM). The parts with a quality surface have better properties such as fatigue, wear, and corrosion resistance. Determining the surface roughness (SR) by analytical and experimental methods is very difficult, time-consuming, and costly. These disadvantages can be eliminated by predicting the SR with artificial intelligence methods. In this study, AZ91D was cut with WEDM in different voltage (V), pulse-on-time (µs), pulse-off-time (µs), and wire speed (mm s −1 ) parameters. The SR was measured using a profilometer, and a total of 81 data were obtained. Multilayer perceptron, ensemble neural network and optimization-based evolving product-unit neural network (EPUNN) were used to predict the SR. It was observed that the EPUNN method performed better than the other two methods. The use of this model in industries producing Mg alloys with WEDM expected to provide advantages such as time, material, and cost.