Explainable AI and data-driven modeling of tool wear in Dry–MQL-assisted turning of D2 tool steel using AIMPOD

Abstract Tool wear prediction is critical for maintaining dimensional accuracy, process stability, and tooling economy in hard turning, yet many existing approaches remain too complex or insufficiently interpretable for routine production-floor use. This study develops an explainable, data-driven framework for modeling tool flank wear in dry- and minimum-quantity-lubrication (MQL)-assisted turning of hardened D2 steel using the Artificial Intelligence-based Manufacturing Process Optimization and Design (AIMPOD) workflow. The input variables were depth of cut, feed rate, cutting speed, and lubrication mode, while average flank wear ( $$VB_c$$ ) was used as the target response. The measured wear ranged from 0.041 to 0.290 mm, providing a useful response spread for supervised learning within the investigated DOE; however, the dataset remains structured and limited to the tested turning window. Among the evaluated models, Lasso, Ridge, and linear regression showed the best predictive performance, with: Test $$R^2$$ : $$\approx 0.73$$ RMSE: $$\approx 0.028$$ mm MAE: $$\approx 0.020$$ mm. The optimized ensemble assigned 98.3% of the total weight to Lasso and improved performance by less than 1%, indicating a predominantly additive and low-dimensional wear structure. Explainability analysis identified lubrication mode as the most influential variable. Mean wear under MQL was 0.059 mm, compared with 0.145 mm under dry turning, corresponding to nearly 59% lower wear with MQL.