Automatic Identification of Tool Wear Based on Convolutional Neural Network in Face Milling Process

Monitoring of tool wear in machining process has found its importance to predict tool life, reduce equipment downtime, and tool costs. Traditional visual methods require expert experience and human resources to obtain accurate tool wear information. With the development of charge-coupled device (CCD) image sensor and the deep learning algorithms, it has become possible to use the convolutional neural network (CNN) model to automatically identify the wear types of high-temperature alloy tools in the face milling process. In this paper, the CNN model is developed based on our image dataset. The convolutional automatic encoder (CAE) is used to pre-train the network model, and the model parameters are fine-tuned by back propagation (BP) algorithm combined with stochastic gradient descent (SGD) algorithm. The established ToolWearnet network model has the function of identifying the tool wear types. The experimental results show that the average recognition precision rate of the model can reach 96.20%. At the same time, the automatic detection algorithm of tool wear value is improved by combining the identified tool wear types. In order to verify the feasibility of the method, an experimental system is built on the machine tool. By matching the frame rate of the industrial camera and the machine tool spindle speed, the wear image information of all the inserts can be obtained in the machining gap. The automatic detection method of tool wear value is compared with the result of manual detection by high precision digital optical microscope, the mean absolute percentage error is 4.76%, which effectively verifies the effectiveness and practicality of the method.