Abstract:In view of the important position of CNC lathe cutter in machining system, on-line detection and recognition of tool wear fault of CNC lathe are of great significance. Taking the Central China CNC lathe as the research object,the methods of means of average empirical modal decomposition (EEMD), chaotic particle swarm optimization (CPSO) and nuclear limit learning machine (ELM) were proposed to diagnose the lathe tool wear fault.The basic principles and processes of EEMD, CPSO and ELM were introduced. The acquired tool wear signals were pre-processed, and the IMF component was obtained after EEMD decomposition.Taking the Kurtosis, peak value and root-mean-square value as a selection criterion,several IMF contained more fault information were selected for signal recombination and calculation.After composing the feature vectors,the calculation results were input into CPSO-ELM, SVM and BP neural network and other classifiers for fault identification and comparison. The experimental results show that compared with the traditional BP neural network and SVM classifier, the CPSO-ELM classifier has the characteristics of fast, accurate and effective identification, and can effectively detect and identify tool wear faults.