Research On Rotor Fault Diagnosis Of Induction Motor Based On Wavelet Analysis

The induction motors, as its simple structure, low price, high reliability and convenient maintenance, are important equipments in industry, agriculture, national defense projects and daily life. Normal operations of induction motors are very significant to ensure the safety, efficiency of the production process. A motor fault can not only result in damages to the motor itself, but also can affect the work of the whole system, and even endanger the personal safety, resulting in huge economic loss. In order to prevent the occurrences of induction motor faults, to detect and eliminate them, to ensure reliable operations of motors, to reduce excess maintenances, to lower their maintenance costs and to improve the production efficiency, research on induction motor monitoring and fault diagnosis technology is of great economic and social significances.The induction motor fault signals are non-stationary transient ones. Wavelet analysis has a good capacity of reflecting local signal characteristics in the time domain and frequency domain, suitable for processing non-stationary transient signals. Based on theory of wavelet analysis and time-frequency analysis of the stator current, this paper studies rotors feature extraction and its fault diagnosis, focusing on the following aspects:(1) After studing the theory of wavelet analysis and principles of rotor fault diagnosis, based on well instantaneous frequency extraction by wavelet ridge method and the favorable conditions that air-gap eccentricity fault frequency is not very close to power frequency under normal circumstances, proposes air-gap eccentricity fault diagnosis method based on wavelet ridge line and realizes its fault diagnosis.(2) Rotor bar breaking fault frequency is difficult to extract due to its easy aliasing with the power frequency on stable operation state. To solve this problem, this paper proposes wavelet ridge line method based on park vector modulus square signals, In case of good quality power supply, especially in the three-phase power balance, this method eliminates the impact of power supply frequency thus can effectively realize rotor bar breaking fault diagnosis.(3) Traditional wavelet packet feature extraction methods are ineffective in case of multi-wavelet packet decomposition or weak signal. To solve this problem, proposes a modified wavelet packet energy feature extraction method due to the rotor fault characteristic, studies the feature extraction effect when it applied on changing load conditions, simulation results verify its validity. (4) Wavelet ridge line method based on park vector modulus square signals is ineffective in case of asymmetric three-phase power supply. To solve this problem, uses the wavelet packet neural network method to rotor fault diagnosis, takes rotor fault feature extraction results by improved method as input to the BP neural network, and completes the training, test and diagnosis. Simulation results show that it has higher accuracy and better adaptability.