Rotor Faults Diagnosis Based On Vibration Signal Analysis

Rotating machinery is one of the most common mechanical equipments, such as steam turbine,aircraft engine, generator, etc. Rotor usually works with high speed and complex working conditions,so faults are easy to happen. Once some faults happen, there will be a major security hidden dangerand economic losses. Therefore, researching on rotor fault diagnosis is very important. The obviouscharacteristic of rotor fault is abnormal vibration of the machine. Vibration signal can reflect the faultinformation of the machine, from the time domain or frequency domain. The rotor fault diagnosisbased on vibration signal analysis is studied in this dissertation, the main research work includes:(1) Mechanisms of common rotor faults are studied, and the paper summarizes the vibrationcharacteristics and frequency characteristics of common rotor faults. The rotor faults of imbalance,misalignment, looseness and rub-impact are simulated by ZT-3rotor experimental device.(2) The continuous wavelet transform, multi-resolution analysis and wavelet packet theory arestudied. The experimental signals are analyzed in the frequency domain and time-frequency domainby fourier transform and wavelet scale method. The signal de-noising method based on waveletanalysis is studied, soft-threshold and Donoho threshold estimation is used to the experimental signalsde-noising.(3) The dissertation studies the methods of faults feature extraction. The method and steps of thefaults feature extraction using wavelet packet energy analysis are summarized. Against thedisadvantage of wavelet packet has no adaptability, a method of adjusting sampling frequency basedon the rotor speed is proposed, which can realize the automatic feature extraction, and unify faultsfeature. It introduces the sample entropy, a feature extraction method based on wavelet packet andsample entropy is studied. Empirical mode decomposition which can overcome the disadvantages ofwavelet analysis is used to the experimental signals processing.(4) The pattern recognition methods are studied, the paper introduces the basic theories of neuralnetwork and support vector machine. Both of the methods are used to train and classify theexperimental signal samples. Though the comparison of two methods, get support vector machine is abetter way for small sample.(5) The dissertation studies the concrete way of using the MATLAB GUI. A fault diagnosissystem is designed which has the function of reading signals, time-frequency analysis, featureextraction and fault diagnosis etc.