Tool Wear Monitoring Using Machine Learning

Abstract

Monitoring tool wear conditions and estimating tool lives are crucial for automatic metal cutting manufacturing processes. Accurate estimation of tool wear status can optimize tool usage and tool replacement, which in turn leads to improvement of product quality and reduction of downtime and costs. Various methods have been proposed to evaluate tool wear conditions using multiple sensory signals during the cutting process. However, because the signal-to-noise ratio is extremely low in the machining process, the accuracy of tool wear evaluation still needs to be improved. In this paper, machine learning methods were explored to estimate the tool wear conditions based on the experimental data provided by the 2010 PHM society conference data challenge. A self-organizing map was designed to identify the tool wear conditions into 16 levels. According to the correlation coefficients between sensory datasets and the tool wear levels, 14 features were selected for tool wear estimation. A feedforward backprop neural network and a support vector machine with the fine Gaussian kernel were constructed for tool wear estimation. The testing results showed that the root mean square for neural network model was 0.9866 and for the support vector model was 1.4985.