Integrated multi-response optimisation and predictive modelling of EDM parameters for EN8 steel using Taguchi–GRA–RSM and ANN approaches

This study investigates the multi-response optimisation and predictive modelling of electrical discharge machining (EDM) of EN8 medium carbon steel, targeting simultaneous improvement of arithmetic surface roughness ( R a ), root mean square roughness ( R q ), material removal rate (MRR), and tool wear ratio (TWR). Three process parameters — pulse-on time ( T o n : 3, 5, 7 µs), pulse-off time ( T o f f : 2, 4, 6 µs), and peak current ( I p : 10, 20, 30 A) — were investigated using a Taguchi L 27 orthogonal array with a commercially pure copper electrode and kerosene dielectric. An integrated Taguchi–grey relational analysis (GRA)–response surface methodology (RSM)–artificial neural network (ANN) framework was developed for optimisation and prediction. ANOVA identified peak current as the dominant factor for R a (58.6%), R q (49.8%), and TWR (71.8%), whereas pulse-on time governed MRR (45.3%). Multi-response optimisation via GRA (GRG = 0.8023) and RSM composite desirability ( D = 0 . 698 ) converged on the consensus optimal condition µ T o n = 7 µs , µ T o f f = 2 µs , I p = 10 A , yielding reductions of 13.3% in R a , 12.3% in R q , and 11.8% in TWR, alongside a 23.7% increase in MRR relative to the experimental mean. Second-order RSM models achieved excellent accuracy for MRR ( R 2 = 0 . 9987 ) and TWR ( R 2 = 0 . 9991 ), with moderate adequacy for R a ( R 2 = 0 . 8487 ) and R q ( R 2 = 0 . 7961 ). A feed-forward multilayer perceptron (3–10–10–4) trained on CTGAN-augmented data and benchmarked across three training algorithms showed that resilient backpropagation ( trainrp ) achieved the best overall generalisation with mean R 2 = 0 . 9991 , mean MSE = 0.0105, and mean MAPE = 3.534%, substantially outperforming RSM for surface roughness prediction ( R a : 99.95% vs. 84.87%; R q : 99.85% vs. 79.61%). The proposed integrated framework provides an effective data-driven methodology for multi-response EDM parameter optimisation of EN8 steel components used in die and mould manufacturing.