Van-Hai Nguyen

This study develops an integrated monitoring system with a human–machine interface (HMI) for groove grinding on the outer surface of the inner ring of a bearing type 6205 in an industrial environment. The proposed system is specifically designed to ensure stable real-time monitoring under progressive grinding wheel wear evolution, rather than maximizing prediction accuracy under fixed and stationary operating conditions. The system combines a grinding wheel wear measurement setup using pneumatic sensors, machine learning algorithms, and multi-objective optimization methods. The input data are collected in real time from the pneumatic measurement system (PMS) to monitor wheel wear ( G w ), and from inductive proximity sensors to count the number of processed parts ( N p ) in each grinding cycle. Subsequently, machine learning models (ANN, XGBoost, CatBoost, LightGBM) tuned with Bayesian optimization (BO) and combined with the NSGA-II algorithm are used to determine the optimal solutions for grinding conditions that consist of wheel feed rate ( f w ), workpiece speed ( v w ), depth of cut ( a e ), and N p , based on the objectives of maximizing material removal rate ( Q w ), and minimizing surface roughness ( Ra ) and ovality ( O p ). The machine learning hyperparameters are optimized using the Bayesian algorithm. The selected solutions are adapted to user preferences and are used as input cutting parameters for the monitoring system. The SHapley Additive exPlanations (SHAP) technique is used to interpret the influence of grinding parameters on the output. The entire system is integrated into the HMI to display predictions in real time and to issue warnings when Ra exceeds 0.3 µm, O p exceeds 3 µm, or G w exceeds 12 µm, at which point a dressing alert is triggered. In factory trials, the HMI maintained stable latency and prediction consistency, with MAPE values within 5% across the full wear range. These results reflect the actual behavior observed on the line and confirm that the system can operate reliably under industrial conditions.

This study employs an integrated, data-driven approach to optimize and rank optimal solutions for unbiased cutting parameter selection in turning AA6063 alloy. The objectives are to achieve an optimal trade-off between surface roughness (Ra) and material removal rate (MRR) using entropy and Gini weighting methods. Accordingly, dry turning experiments were conducted, and data were collected. Four machine learning (ML) models, including Kolmogorov–Arnold network (KAN), Artificial Neural Network (ANN), XGBoost, and CatBoost, are used to evaluate the predictive performance. Through the model performance metrics RMSE, MAE, and R ² , the best models are embedded in the NSGA-III algorithm to find Pareto solutions. Then, two multi-criteria decision-making (MCDM) methods (MABAC and EDAS) are combined with different weighting schemes to rank the optimal solutions. The results show that KAN outperforms the other models. Thirteen Pareto solutions are generated and ranked roughly according to the prior of MRR. Overall, the proposed method demonstrates the benefits of integrating interpretable ML with evolutionary optimization and rational decision-making, providing a robust solution for machining process planning under conflicting objectives.