Experimental investigation of different machine learning approaches for tool wear classification based on vision system of milled surface

Tool condition monitoring (TCM) is of great concern for the quick and accurate detection of tool wear in the machining industry to avoid tool change errors and perform tool wear compensation, directly affecting product quality, production time, and costs. Thus, this article aims to develop a machine vision-based TCM system based on the machined surface images to detect and classify the tool wear status during milling process. The acquired images of the machined surface texture have been analyzed using multi-step image processing algorithms. Then, 12 texture features have been extracted using grey-level co-occurrence matrices (GLCM) to characterize the machined surface texture. Based on the statistical measures of the extracted features, the tool wear state can be classified into three levels: sharp, semi-sharp, and dull. The collected datasets have been used to train models via three different machine learning (ML) algorithms, including support vector machine (SVM), multilayer perceptron (MLP), and generalized feedforward neural network (GFFNN), to predict the tool wear classification. The performance accuracy for the three ML classification models has been compared to determine the most effective tool wear classification approach. The results showed that SVM model exhibited the highest accuracy of about 98%, followed by the GFFNN model with an accuracy of 96% of the total average actual images learned to the model have been correctly classified. MLP model had the lowest percentage of 93% and failed to classify most of the tool wear status, which means that the MLP algorithm had poor classification performance.