Chatter detection for milling by using wavelet transforms and convolutional neural networks

Milling is one of the most widely used machining methods in today's industry. Different types of chatter vibrations directly affects the efficiency of milling operations by causing the cutting tools to wear faster and especially the surface quality of the workpiece to decrease. Chatter can be defined as abnormal vibrations that occur due to the forces between the workpiece and the tool during cutting. Chatter occurs depending on the dynamic structure of the cutting machine tool, the characteristics of the workpiece and the tool, and the cutting process parameters. Within the scope of this study, a novel chatter detection method has been developed that enables the detection of chatter vibrations in milling processes using the CNN model. In addition, self-renewing chatter vibrations were successfully labeled and classified with the CNN. During milling, vibration data are collected via a noninvasive data acquisition strategy. In the processing of the data collected in the experiments, wavelet transforms, one of the most advanced signal processing tools, were used considering the nonlinear and complex nature of chatter vibrations. Images that are obtained by wavelet transforms are given as inputs to the CNN model. When the results were examined, an average of 99.88% chatter classification accuracy was achieved for 3 different scenarios.