Data Driven Prognostics of Milling Tool Wear :A Machine Learning Approach

Abstract

Tool wear in a milling process affects the finished product's overall quality, which results in rejection. With an increase in tool wear, cutting power decreases that affects the load on the machine. This results in damage of the equipment. Conventional manufacturing system lacks the way of forecasting the tool wear and its effects. Machine Learning (ML) model-based techniques with data-driven prognostics convert conventional manufacturing systems into smart manufacturing systems. This research paper focuses on the comparison of data-driven predictive models that predict tool wear based on the analysis of various sensor signals. In this study, eight algorithms such as Linear Regression (LR), Support Vector Regression (SVR), Naïve Bayesian (NB), Gradient Boost (GB), XG Boost (XGB), CatBoost (CB), Random Forest Regression (RFR), and Artificial Neural Network (ANN) are applied and compared their performance evaluation. The comparative study of regression algorithms provides an overview of tool wear prediction. Evaluation metrics chosen show conclusive evidence that the ANN model performs better than other models. The obtained predictive performance of the ANN model outperforms the existing models reported in the literature. The proposed ANN model for tool wear prediction uses the sensor information and exposes hidden patterns that completely fit the dataset.