An explainable AI model with adaptive feature constructions for the prediction and analysis of surface roughness in anhydrous textured-wheel polishing

Precise prediction of surface roughness is crucial for optimizing manufacturing processes and improving component performance. However, current mainstream artificial intelligence prediction models are predominantly ‘black-box’ models, incapable of revealing the intrinsic physical mechanisms and higher order interactions between complex process parameters and surface roughness. To address this issue, this paper proposed an explainable artificial intelligence (XAI) framework for predicting and analyzing surface roughness during waterless texture-consolidated grinding wheel polishing. The framework first constructs features using third-order polynomials and open feature engineering (OpenFE) and then employs a differential evolution (DE) algorithm for adaptive feature selection, ultimately driving the Stochastic Gradient Descent Regressor (SGDR) model to achieve high-precision predictions. Experimental results demonstrate that compared to multiple benchmark models, the proposed method achieves significantly improved prediction performance, reducing the Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) by 49% and 19%, respectively. More importantly, interpretability analysis based on this framework successfully reveals the influence patterns and coupling effects of key process parameters on surface roughness. This provides clear theoretical guidance and process optimization pathways for achieving efficient, high-quality batch manufacturing of optical components.