An integrated multi-stage pipeline for precision tool manufacturing process modelling and optimization under multi-objective constraints

Form-tool manufacture from high-speed steel remains difficult because productivity, surface quality, and dimensional accuracy must be achieved simultaneously in a hard-to-machine geometry. This study addressed that problem through wire electrical discharge machining (WEDM) of a tapered high-speed-steel (HSS) forming-tool profile using a full-factorial design and an Artificial Intelligence for Manufacturing Process Optimization and Design (AIMPOD) workflow that integrates surrogate modelling, explainability, and multi-stage optimization. Material removal rate (MRR), surface roughness, and four angular-error responses were modelled using response-specific surrogates. The results showed that material thickness was the dominant productivity factor, with a correlation coefficient of $$r = 0.97$$ with MRR and a 102.0% increase in mean MRR from 7.15 to 14.45 mm ³ min ⁻¹ as thickness increased from 12.7 to 25.4 mm. In contrast, quality responses exhibited localized and response-specific behaviour; for example, the angular error e₂ increased by 46.1% with thickness, whereas the angular error e₄ decreased by 29.5% as open voltage increased from 85 to 105 V. Among the machine-learning models, SVR provided the best MRR prediction with test R² = 0.96 , RMSE $$= 0.63$$ mm ³ min ⁻¹ , and MAE $$= 0.50$$ mm ³ min ⁻¹ ; the optimized ensemble achieved RMSE $$= 0.6313$$ mm 3 min -1 and R² = 0.9592 . The framework identified candidate compromise solutions for simultaneous optimization of productivity, finish, and angular accuracy in toolmaking-relevant WEDM.