Research On Purification,Feature Extraction And Automatic Identification Of Rotor Axis Orbit

Rotor axis orbit plays an important role in the fault diagnosis of steam turbine,and it contains all kinds of fault symptoms.However,how to get a clear axis orbit and classification and recognition is always a hot topic in the research of fault diagnosis.Under this background,the paper has made the following research on the purification,feature extraction and automatic recognition of the axis orbit:First of all,the vibration mechanism of the rotor axis orbit is introduced,and the vibration classification and formation mechanism are introduced to further understand the vibration mechanism.By introducing the rotor dynamics model,a simplified mechanical model is introduced to solve the differential equations to obtain the calculation formula of the axis orbit,and then analyzes the characteristics of the axial orbit of different rotor fault states.Secondly,in the study of shaft orbit purification,this paper introduces three methods of wavelet transform,harmonic wavelet,EEMD principle to purify the axis orbit.The simulation result shows that the three methods are able to purify the axis orbit,but by wavelet transform in wavelet decomposition level and threshold,different parameters will lead to the purification effect is not the same,and the harmonic wavelet algorithm is relatively simple,programming easily,and the purification effect is clear,however,it has a great limitation in engineering applications because the harmonic wavelet can not be in accordance with the requirements of the project to any refinement of the frequency band,so it has a great limitation in engineering applications.The EEMD algorithm can completely restore the original axis orbits,and for the future to pave the way for feature extraction,but EEMD for the original signal frequency distribution in high frequency signals,and noise mixed together,the purification effect is not ideal.Finally,based on the study of axis orbit feature extraction and automatic recognition,support vector machine is more suitable for small samples based on the basis of nonlinear identification,and this paper introduces the directed acyclic graph SVM multi classification model,and through the research from two aspects of signal characteristics of orbit and image features.In the image feature,this paper proposes an improved moment invariant method,and the original invariant moments are improved,overcome the effect of the scale factor,which satisfies the invariant requirements,and simulation results show that the proposed feature extraction method is better than the original feature invariant moments;In the signal characteristics,this paper uses the method of EEMD with the principle of combining principle of cloud model,first the original vibration data is obtained by simulating the common rotor faults in Bentley experimental station,and then the original signal is denoised by EEMD,then according to the principle of cloud model,introducing the reverse cloud generator for signal processing of digital feature computation,and as a result of the feature vector is input to the support vector machine to identify,compare the results with the traditional EEMD method by using this method,the validation of the method to identify more accurate.