A Study Of Wavelet Analysis On Fault Diagnosis Of Hydraulic Generators

With the high-speed development of sensor, signal processing, network, fuzzy math & nerve network technology, and the improved level of research & practice on device state monitoring and faults diagnosis of every country in the world, the traditional plan periodic examine and repair mode are gradually changing into state repair mode. In order to realize state repair, should fetch in and develop faults diagnosis technology. As to hydraulic generator, faults diagnosis is the basis of examine and repair, so it is necessary to put it in practice.It is the unsteady signal component analyzing that FDI (Fault Detection and Isolation) mainly according to which is the important subject of modern Data Signal Processing. Moreover, this is the sixty-four-dollar question in Condition Monitoring field without good solution. The fault is described by unsteady signal, the signal's feature vectors change correspondingly. So if there is fault, the feature information emerges out form signals. The paper dissertate unsteady signal feature extraction and fault signal isolation in detail combining theory and application.Wavelet package analysis can pick up the useful information of the Hydraulic Generators, which is regarded as evidence to diagnosis fault. The paper studies wavelet package analysis and the NN used on fault diagnosis of Hydraulic Generators.The fault character information extracting, comes down to the question how represent maximal information with minimal datum as possible as we can. This is the sixty-four-dollar question of signal processing and pattern identification. Signal is represented sparsely in wavelet domain, namely wavelet transform has the property of dimension reduction. This paper eliminates signal redundant information, de-correlates signal correlated information in wavelet domain and extracts unsteady signal character information in wavelet space, namely "entropy compression"; this paper separates mixture observation signals, namely de-correlates signal correlated information successfully.The main content as following:1. The paper use BCB to realize the communication between the workstation and the blind SQL Server in windows 2000.2.This paper discusses hypothesizes of Fourier Transform (Short Time Fourier Transform). Basing on theory, this paper explains the shortcoming of Fourier0Transform. With the information carrier of wavelet atom for unsteady signal, the Multi-Resolution Analysis of Wavelet Transformation describes unsteady signal time-frequency information adaptively which is apt to feature extraction.3. The Wavelet theory, Wavelet Frame Theory, Continuous Wavelet Transform (CWT), Discrete Wavelet Transform, Wavelet Packet Transform and Wavelet Packet Network are discussed in detail. The paper dissertate the mathematic property of wavelet basis function and their influence to application. The paper gives some principals for selecting wavelet basis and wavelet transform scheme for different signal, different signal, different signal time-frequency structure and different analysis aim.4. when the fault signal is separated, namely, fault signal re-construction is procession of fault character information extraction from measure space. The system uncertainty decreases in the Fault Detection and Isolation processes.5.As for high oscillation vibration signal, this paper discusses in detail the property of wavelet packet energy distribution, white-noise and color-noise elimination, low-dimension feature information vectors extraction and the fault diagnosis system realization with wavelet packet analysis and signal energy frequency band analysis. An example of high frequency signal of hydro-generator is given. A perfect method is designed that improves the capability of FDI of WNN, extracts the character low dimension vectors of fault information and takes the vectors as the input data of WNN, so as to simplify the structure and improve the convergence speed of the WNN and reduce diagnosis errors.