Abstract:Aiming at the limitation and insufficiency of traditional machine tool wear state detection method, a new method based on the threephase input average effective current signal of machine tool inverter is proposed to detect tool wear state by using an indirect quantity. Firstly, the working principle of the new method of tool wear for current detection of lathe inverter was introduced, and the calculation method of threephase average effective current signal was better than that of traditional Root Mean Square(RMS) calculation method by comparison and verification, and secondly, the calculated effective current signal was obtained by using the threephase average effective current calculation method and the traditional RMS calculation method. Combining wavelet noise reduction and Ensemble Empirical Mode Decomposition (EEMD) method to reduce noise, decompose and reconstruct the signal, the energy and mean variance of each frequency band of the effective current value signal calculated by the two methods in the reconstructed timefrequency domain were analyzed and the eigenvectors were composed. Finally, the eigenvectors were input Chaos Particle Swarm OptimizationExtreme Learning Machine(CPSOELM) and Productbased Neural Network (PNN) and other classifiers for fault identification and comparison. The experimental results show that CPSOELM has a fast and accurate recognition effect compared with PNN neural network classifier, which verifies the validity and feasibility of a new method of detecting lathe tool wear by using inverter current.