Monitoring and Prediction of Tool Condition Using Machine Learning

Abstract

The condition of cutting tools plays a crucial role in machining operations, impacting productivity, quality, and cost-effectiveness. This study aimed to develop an intelligent tool condition monitoring system using Support Vector Machine to estimate the tool wear of cutting tools. The motivation behind this research was to address the challenges faced by industries in effectively monitoring and predicting tool condition, enabling proactive maintenance strategies and optimizing machining processes. Vibration signals were acquired using accelerometers connected to an OneproD MVP-200 analyser during the machining process. Feature extraction involved identifying the relevant feature from the vibration signals, specifically the Root Mean Square (RMS) and Standard Deviation (Stdev), to capture patterns and characteristics indicative of tool wear. The SVM function in MATLAB was utilised to train a model using the extracted feature as input and surface finish as the label. The trained model was then used to estimate the tool wear of the cutting tool. The findings show that the RMS feature exhibited better accuracy compared to the stdev feature. Notably, the Gaussian SVM kernel achieved the highest accuracy of 83.07% for the RMS, surpassing the Linear (77.78%) and Polynomial (81.82%) SVM kernels.