An empirical comparison of chatter classification methods in turning

Abstract The avoidance of chatter is an important issue in metal cutting. Due to variation in material quality, clamping, and wear in machine tool or tool insert the process conditions may change. A work piece that has been of sufficient quality earlier in the run can no longer be cut to fulfill the requirements for surface integrity, topography and part geometry. Many industrial operators rely on experience and trial-and-error decide suitable cutting parameters. The variation in conditions during the entire production run makes this challenging. However, it is possible to automatically detect or predict the onset of chatter, potentially shortening the process to find suitable parameters or relax requirements for the parameters. Special care must be taken to extract useful features from the cutting data. Simply measuring spindle power or torque may have issues with detecting minor or different directional vibration. When extracting the subtler vibration-related features, care must be taken to avoid interference from other phenomena, such as chip breakage, which may create similar features as chatter. In this study, several methods for detecting or predicting chatter are examined. A library of samples recorded from earlier experiments is used to test different methods. An experienced machinist has subjectively graded the samples. The methods for detecting or predicting chatter are applied to the data set and results are compared with each other and the expert data. Based on this comparison, applications and challenges for various methods are identified.