Investigation of surface morphology and surface roughness of AISI 304 stainless steel in WEDM: Experimental optimization and prediction using machine learning

WEDM is a non-conventional method for making complex shapes of hard metals whose surface finish is significantly dependent on process parameters. The influence of WEDM parameters on surface roughness (R a ) at AISI 304 steel was studied in this work by using an innovative approach for predicting the quality of machined surfaces with four inherent prediction models, including ANN-LOOCV (Artificial Neural Networks and Leave-One-Out Cross Validation), PCR (Principal Component Regression), RSM (Response Surface Methodology) and MLR (Multiple Linear Regression). For machining the samples, three input factors are identified: peak current, pulse-on time and the pulse-off time of the wire. Nine experiments are performed based on an L9 orthogonal array. The SEM test contrasts materials with low and high surface roughness in order to analyse surface morphology.From SEM analysis, Sample 9 reveals a recast microstructure with a thin, uniform layer and evenly distributed microcracks, which resulted from thermal loading within a controlled range. ANOVA shows that the most significant factor affecting surface roughness during EDM is peak current with a percentage contribution of 40.5%, followed by the pulse-on time (17.6%) and the pulse-off time (38.2%). The prediction accuracy for the four models is highest (78%) with the ANN model at a RMSE of 0.325 μm and an MAE of 0.253 μm. These results highlight the superior predictive capability of ANN models for surface roughness, even with limited data. The findings provide valuable insights for optimising WEDM parameters, contributing to improved machining outcomes.