Study On The Vibration Signal Procession And Fault Feature Extraction Based On LMD

Rotating machinery plays a very important role in modern industrial production, which should not be overlooked. It is an important guarantee for safe production. So, using new signal processing and fault diagnosis of rotating machinery to monitor and diagnosis faults, to a certain extent, can ensure the safety and efficient operation of mechanical equipment systems. Rolling bearing is an important part of rotating machinery and a typical representative. In the paper, it mainly studies the signal processing and fault diagnosis of rolling bearing fault vibration data.Actually the collected vibration signals include not only bearing vibration signals, but also contain a lot of noise signals, which would seriously affect the signal decomposition and subsequent feature extraction. Therefore, it is very necessary to pretreat the signal noise by the use of a certain method. The paper introduces some de-noising discrimination indexes, such as noise ratio, the minimum mean square error. Besides, it opts for wavelet function decomposition level, threshold selection rule and the threshold function which is the most suitable for vibration signal in this article as denoising parameters. It adopts wavelet threshold denoising to pretreat measured bearing vibration signals. Taking it into account that local mean decomposition algorithm has the advantage of adaptive signal decomposition, it can get a series of product function components after decomposing the vibration signals after noise reduction by this method. By analyzing the characteristics of fault vibration signal, the paper puts forward a method for extracting fault characteristics and diagnosing faults. This method includes some theories, such as energy entropy, singular value entropy, kurtosis and Lemple-Ziv complexity. Subsequently, it extracts fault features form the product function components from LMD. The final results suggest, that both of energy entropy and singular value entropy bearing of normal operating are greater than the corresponding value of three fault states. At the same time, Kurtosis and Lemple-Ziv complexity index in the normal state are less than that of fault condition.In summary, the method proposed in this paper, including wavelet threshold de-noising as well as energy entropy、singular value entropy, kurtosis, Lemple-Ziv complexity extracted from the product function component after LMD, can extracts fault characteristics and diagnosis faults very well.