Investigation Of Fault Diagnosis System Of The Engine Based On Acoustic Signal

Fault diagnosis technique of the engine is an engineering science integrating with practice, which is the reason why we should develop it. So investigation of the simple and convenient fault diagnosis technique is of great value. More and more people atta-ch importance to the acoustic signal technique for its convenience and non-contiguity. This thesis first analyzed the source and spread of the engine's noise and consid-ered the influence of acoustic attenuation and background noise. Then adopted some mode-rn techniques that can effectively and credibly diagnose the engine, according to the non-stationary characteristic of signals. The details as follows: Based on the analysis of characteristic and difficulty of signals, design the blue print of fault diagnosis system of the engine. Pick up the noise signal from the exterior of the engine, contrast to traditional libration testing, which is a breakthrough. Then find the best test spot according to the analysis of the sound field. According to the non-stationary characteristic of acoustic signals, the analysis of these signals based on the wavelet analysis showes that a minutia of signals can be magnified.Compare this frequency with the fault signals'frequency, we can identify the difference. Improve de-noising function of the non-linear wavelet analysis theory, adopt the layered threshold de-noising method to process the fault signals, and its great effectiveness of de-noising is verifyied.The power spectrum technique based on the Wavelet Transform can pick up the distinct characteristics of signals'frequency.During picking up the characteristics of signals, the reformative Wavelet Packet Decompositi-on based on sections is used to subdivide the frequency bands identified the fault easi-er and un-subdivide others. It is testified by experiment that the parameters are suitable to fault diagnosis of the engine. The fault diagnosis method based on the reformative wavelet analysis and BP neural networks (Levenberg―Marquard arithmetic) is put forward and tested on the engine, which can meet the requestment of fault diagnosis.