Study On Fault Feature Extraction Of Rotating Machine Method Based On Empirical Mode Decomposition

With the development of modern industrial production, machinery fault diagnosis hasrapidly developed into an emerging discipline.The key to mechanical fault diagnosis is toexact fault features from the vibration signal. The Signal processing and analysis are theusual method for feature extraction. In recent years, there are more and more attentions tosignal processing in machinery fault diagnosis. The paper focuses on the empirical modedecomposition (EMD) and the application in mechanical fault feature extraction.Aiming at the end effect of EMD processing, the neural network ensemble for datasequence extension technologies has been put forward. Aiming at the mode fission causedby impulse noise and random noise, the de-noising algorithm based on fast independentcomponent analysis (Fast ICA) has been introduced. These approaches are applied to themechanical fault feature extraction.Firstly, the definitions of instantaneous frequency and intrinsic mode functions(IMF)are discussed, the basic principles of the EMD are analyzed and the EMD algorithm isstudied. After that, physics meanings of Hilbert/Huang Transform spectrum have beenanalyzed. Based on these two factors, end effect and mode fission in noise signal analysisby EMD, which has influenced on the accuracy and speed of EMD-phenomenon havebeen discussed.Secondly, algorithms of neuron network based data extension and mirror basedextension technologies have been discussed, meanwhile their limitations have beenpointed out, too. For this, the neural network ensemble algorithm has been presented. Inorder to eliminate phenomenon of mode fission in random noise analysis by EMD, theblind source separation de-nosing has been presented. Theoretical analysis and the realsignal analysis show that they are efficient and excellent methods to improve theeffectiveness of time-frequency analysis.Lastly, the feature extraction of rolling mill main drive fault signal is researchedbased on empirical mode decomposition and AR model. The analysis of signal shows thatthe method is efficient to extract the fault feature.