Intelligent Fault Diagnosis Of Rotating Machine Based On Hilbert Spectrum And Novelty Detection

In modern production, the structure of machinery and equipment is becoming more and more complex. Once a component is failure, it is easy to trigger chain reaction, resulting in great damage to the equipment. So fault diagnosis technology of mechanical equipment is more and more important. Any dynamic mechanical device will have a certain vibration. When an exception occurs, the vibration will change as amplitude increases usually, and will show a strong non-linear and non-stationary nature of this characteristic. So that diagnostic information should be obtain from the vibration signals, and intelligent fault diagnosis could realize. In this paper, the intelligent fault diagnosis of rotating machinery based on the Hilbert spectrum is studied through vibration signals of the rotor experimental instrument. the main jobs are as follows:Firstly, the paper introduces the background and the significance of rotating machinery fault diagnosis, and makes an overview of rotating machinery fault diagnosis methods. In particular, time-frequency analysis methods of rotating machinery fault diagnosis are described briefly. After contrast of several methods, the Hilbert-Huang Transform in rotating machinery fault diagnosis shows superiority.Secondly, the paper introduces the basic theory, processes and characteristics of the Hilbert-Huang transform, including the empirical mode decomposition (EMD) principles, Hilbert transform theory, Hilbert spectrum calculation of simulation signal, as well as the existing problems and improve the method of the Hilbert-Huang transform. HHT method shows the effectiveness of signal decomposition, through the analysis of the simulation signals.Thirdly, using ZT-3 multiple-function experimental instrument acquires rotor fault signal, and the Hilbert spectrum of fault signals has been obtained through Hilbert-Huang transform. Then feature of fault signal is extracted from Hilbert spectrum using the PCA method.Finally, for there is limited real fault signal of instrument, so the Hilbert spectrum features are classified using the novelty detection, and parameters of novelty detection are optimized using particle swarm optimization algorithm and the adaptive optimal parameters are obtained. The experiment using experimental data is carried out, and the results show the effectiveness of the method.