Predictive maintenance of cutting tools using artificial neural networks

Abstract

In the manufacturing industry, preventative maintenance of cutting tools plays a critical role in ensuring operational efficiency and minimizing downtime. This paper addresses the problem of accurately predicting the wear level of a cutting tool by applying artificial neural networks. The study uses an extensive dataset derived from real-life manufacturing scenarios and uses synthetic and experimental data for illustrative purposes. The use of depth of cut, continuous vibration monitoring using an accelerometer, spindle speed, feed rate and cutting speed contribute to a holistic approach to predicting tool wear in milling processes. This comprehensive set of features is designed to capture the nuanced interactions between machining conditions and tool degradation, thereby improving the model’s predictive accuracy. The architecture, features and algorithm for training the network are described. A neural network has been created and configured to determine tool wear during the milling process. The process of training and debugging a neural network is clearly displayed on the graphs. The performance of the network was tested using test data, which allowed us to obtain good results.