Balancing surface finish and ablation depth in direct laser turning of RB-SiC based on combined IPSO-SVR models with MOAHA-TOPSIS method

While direct laser turning (DLT) is a promising method for fabricating rotary components, it remains challenges to simultaneously achieve both high surface finish and high efficient material removal for reaction-bonded silicon carbide (RB-SiC), due to issues of heat accumulation and recast layers thickening. In the present work, we investigate the feasibility of simultaneously achieving low surface roughness and high ablation depth in the DLT of RB-SiC by the coupled modulation of laser parameters and turning parameters, which is carried out based on machine learning modeling and a multi-objective optimization and decision-making method. Firstly, a five-axis DLT platform is established, with which the ablation mechanisms and accompanied chemical reactions of RB-SiC under DLT are elucidated. Secondly, improved particle swarm optimization-supported vector regression models are established to predict surface roughness and ablation depth, which are trained through orthogonal experiments with varying laser power coefficient, spot overlap and circumferential overlap levels. Furthermore, the shapley additive explanation (SHAP) method is used to quantify the contributions of individual processing parameters on both surface roughness and ablation depth. Meanwhile, the detailed sampling method is utilized to elucidate the coupled effects of individual processing parameters on both surface roughness and ablation depth. Thirdly, a multi-objective artificial hummingbird algorithm, which is integrated with the technique for order preference by similarity to ideal solution is utilized to generate Pareto-optimal solutions. Consequently, optimized processing parameters are derived for obtaining a surface roughness of 2.494 μm and an ablation depth of 13.715 μm in the single-layer removal of RB-SiC by DLT. Finally, large ablation depth (1.228 mm) with synergistic low surface roughness (2.666 μm) is realized in the DLT of a cylindrical RB-SiC workpiece by multi-layer laser scanning strategy. Current work provides a machine learning-integrated optimization strategy for the DLT of difficult-to-machine materials.