Investigating key operational factors for robotic grinding quality using ensemble machine learning

Robotic grinding quality remains difficult to guarantee due to the inherent flexibility and high vibration of industrial robots, which leads to compromised surface finishes. Current industrial practices rely heavily on subjective manual expertise and inefficient trial-and-error procedures. This study was motivated by the critical lack of predictive classification frameworks specifically tailored for automated grinding parameters in aluminum machining. A comprehensive dataset consisting of 350 experimental points from aluminum workpieces was analyzed across seven primary settings: height, force, time, diameter, grinding, press, and speed. The methodology employed Multiple Linear Regression (MLR) as a baseline, followed by tree-based ensemble methods, specifically Random Forest (RF) and Extreme Gradient Boosting (XGBoost), to categorize surface quality and capture complex nonlinear interactions. Statistical analysis confirmed that Height, Force, and Diameter were the most significant predictors for quality classification. While the linear model exhibited low explanatory power with an R 2 ≈ 0.105, both RF and XGBoost models achieved classification accuracies of approximately 75%. This study contributes a robust predictive foundation for autonomous robotic deployment by quantifying the interplay of operational variables. The identification of Path Geometry (height) and Contact Mechanics (force) as dominant factors offers an essential framework for replacing time-consuming manual corrections with proactive, data-driven quality control.