Intelligent milling tool wear estimation based on machine learning algorithms

This study introduces an innovative approach to estimate tool wear in milling operations across diverse operational settings, employing a multi-sensor signal feature analysis. The method’s novelty lies in its selection of varied milling operational conditions for tool wear estimation and the utilization of distinct data sources to extract features. Tool wear estimation was conducted by analyzing multi-sensor signal data collected and processed under different working conditions. Features from both the time and frequency domains of sensor signals were extracted, and tool wear estimation was conducted comparatively using contemporary machine learning algorithms. The features obtained from different sources were incorporated into the dataset in single, paired, and triple combinations, with subsequent evaluation of the results. The proposed approach underwent validation using the NASA Ames milling dataset and the 2010 PHM Data Challenge dataset. The results showcase the remarkable success of this method in regression, especially across diverse operational conditions. The highest regression success rates were achieved using the VAM-ANN model (R = 0.981290, MSE = 0.0044047) for case 1, and the VFA-ANN model (R = 0.985628, MSE = 0.002943) for case 2. It was observed that the combination of different signal sources significantly enhances the model’s overall performance.