Intelligent Fault Diagnosis Of Rotating Machinery Based On Vibration Signal And Machine Learning

Electric power industry controls the lifeline of national economy,which is the most important key of basic energy industry and economic development strategy Coal-fired power generation is one of the most important power generation methods in China.There are many rotating machinery which is very important in thermal power plant,such as steam turbine,generator,coal mill and various fans and pumps.They are complex and usually work in high temperature,high pressure and high speed rotation.So it’s hard to diagnose for them.This research focues on fault diagnosis of the rotating machinery equipment in thermal power plant,using vibration signal to analyze and model.The specific research includes:(1)Considering the feature redundancy in the vibration signal analysis,an empricial mode decomposition(EMD)and key feature selection based fault diagnosis method is proposed.First of all,EMD is performed to decompose the vibration signal into serval sub signals.And then,statictical features which has different physical meanings in it are extracted from sub signals.Because of the redundant feature and irrelevant feature,feature selection must be performed to get key features.In the end,the selected key features are used to modeling.(2)Considering the mode mixing in the signal preprocess,a method which combines the wavelet transform and EMD is proposed.Vibration signal can be fully decomposed by this method,improving the performance of the signal preprocess.Considering that the non-stationary information in the vibration signal has not been fully mined in the previous work,the stationarity of the preprocessed sub signals is distinguished before feature selection.Fine analysis is carried out for stationary part and non-stationary part respectively,and non-stationary fault information are fully mined,improving the performance of fault diagnosis.(3)Considering the lack of generalization of artificial statistical features in traditional fault diagnosis methods of rotating machinery equipment,an EMD and convolution neural network(CNN)based fault diagnosis method is proposed.Firstly,vibration signal is decomposed into serval sub signals by EMD.Then all sub signals are input into CNN to train.The whole modeling process which extracts the fault related information adaptively does not need extract statistical features.The automatic modeling of fault diagnosis model is realized,which solves the problem of feature generalization.