MACHINE LEARNING-BASED PREDICTION OF SURFACE CHARACTERISTICS AND TOOL WEAR IN Mg–AZ91D MILLING UNDER DISTINCT CONDITIONS

As industries seek greener alternatives to mineral oil-based cutting °uids due to their negative
environmental impact, vegetable oils emerge as promising substitutes. Their biodegradable and
nontoxic properties make them particularly suitable for machining operations. This study investigates the impact of di®erent lubrication and cooling (L/C) methods on machined surfaces and
machining parameters to develop sustainable milling practices for Mg-AZ91D. E®ective L/C is
crucial in regulating heat and friction to enhance product quality. Experimental milling trials involving dry, minimum quantity lubrication (MQL), and solid lubricant (SL) with MQL were conducted on Mg-AZ91D. Results show that under SL-MQL conditions, surface roughness (Ra) was
signi¯cantly lower (by 49–70%) compared to dry conditions and marginally lower (by 8–13%) than
with MQL alone. Additionally, machine learning (ML) algorithms were employed for predictive
modeling, including linear regression (LR), support vector machine (SVM), and Random Forest
(RF). These models were compared using performance metrics such as R2, MAE, RMSE, RAE, and
RRSE. The RF algorithm demonstrated the best results in predicting surface characteristics,
exhibiting notably higher R2 scores and performance metrics. Conversely, SVM outperformed other
algorithms in predicting tool wear. The ML models revealed that cooling conditions had a greater
impact on output milling parameters compared to other input variables.