Electrochemical machining (ECM) prediction and control via multi-physics-constrained bi-layer machine learning with alloy electrochemistry knowledge

Optimizing surface roughness (Ra) and material removal rate (MRR) remains a critical yet challenging objective in electrochemical machining (ECM), owing to its inherently complex multi-physics interactions. This study integrates our original ECM experimental data (based on ECM of additive manufacturing-fabricated IN718 alloy in NaNO 3 electrolytes) with literatures to construct a reliable ECM database. Building on this, we propose a bi-layer machine learning (ML) framework to predict ECM performance with high fidelity. Our approach comprehensively incorporates key ECM control parameters, including electric field, flow field, machining conditions, and corrosion electrochemistry, through physics-informed feature engineering as ML inputs. Uniquely, the ML framework embeds alloy corrosion electrochemistry as its first hierarchical layer, enforcing essential electrochemical constraints to enhance ECM predictive accuracy. A detailed analysis of these ECM control parameters evaluates their relevance to ECM performance, enabling dimensionality reduction of ML inputs without compromising model integrity. Moreover, SHAP analysis was employed to investigate the underlying mechanistic dependence on these input features. Through this methodology, our framework delivers robust predictions and actionable insights, identifying priority features in ECM control for optimizing surface quality and material removal efficiency. Ultimately, the bi-layer ML model offers a novel perspective on the governing mechanisms of non-traditional manufacturing, where fundamental physics (particularly alloy electrochemistry) plays a pivotal role amid highly intricate machining environments.