Research on tool wear prediction based on temperature signals and deep learning

Tool condition monitoring is an important part of tool prediagnosis and health management systems. Accurate prediction of tool wear is greatly important for making full use of tool life, improving the production efficiency and product quality, and reducing tool costs. In this paper, an intelligent cutting tool embedded with a thin-film thermocouple collected the temperature signals during the tool life cycle. As a deep learning method, stacked sparse autoencoders model with a backpropagation neural network for regression was proposed to predict tool wear based on raw temperature signals. An improved loss function with sparse and weight penalty terms was used to enhance the robustness and generalizability of the stacked sparse autoencoders model. To confirm the superiority of the proposed model, the predictive performance was compared with that of traditional machine learning methods, such as backpropagation neural network and support vector regression with manual features extraction. The root mean square error and coefficient of determination were calculated to evaluate the prediction model. The experimental results showed that the proposed model outperformed the traditional methods with a higher prediction accuracy and better prediction stability. The feasibility and accuracy of temperature signals for tool wear prediction were also confirmed.