Abstract This study employed machine learning (ML) and optimization approaches, with support vector regression (SVR), artificial neural networks (ANNs) simulations and response surface methodology to study surface roughness of wire electrical discharge machined/machining of (WEDM) aluminum alloy C355 hybrid composite samples. The samples were strengthened by silicon nitride (Si₃N₄) and graphene nanoparticles (GNPs). The composites surface roughness was investigated using real-time WEDM experiments conducted with varied control settings, including servo-voltage, maximum current, wire feed rate and on/off pulses. The grid-based search approach was used to modify the support vector machine variables, and the layers (input-hidden-output) of the ANN architectural design were achieved. The correlation coefficient and mean absolute percentage error (MAPE) were used to assess the generated models’ prediction ability. SVR outperformed ANN ( R = 0.991350) and RSM ( R = 0.985320) in terms of accuracy, with an R-value of 0.997603 and the lowest MAPE of 0.0748. According to ANOVA results, peak current was the most significant WEDM parameter, accounting for 60.21% of the variation in surface roughness. The suggested method, combining support vector machine and ANN algorithm, can efficiently and accurately analyze and predict WEDM surface roughness on aluminum alloy C355 with Si₃N₄ and GNPs hybrid composites. Hence, this innovative study leveraged application of simulation, statistical and ML techniques to advance substrative manufacturing/WEDM process for the benefits of machining industries.
Vellingiri et al. (Fri,) studied this question.