Feature selection and a method to improve the performance of tool condition monitoring

Tool condition monitoring (TCM) is especially important in the modern machining process. In order to distinguish different tool wear states accurately and reduce the computation cost, it is of great significance to extract and select appropriate features that can reflect changes in tool wear states but are insensitive to cutting parameters. In this work, Fisher’s discriminant ratio (FDR) is adopted as the criterion for feature selection by evaluates every feature’s classification ability. However, it is found that the continuous hidden Markov models (CHMM) trained based on the features selected by the conventional method could recognize some tool state well, but have poor ability to classify other wear states. The reasons for this phenomenon have been analyzed, then a simple and effective method that used two feature sets for TCM has been proposed to improve the recognition performance. Four tests with different cutting parameters are carried out, and the new method has been implemented and verified its usefulness and validity.