Research On Fault Diagnosis Of Subway Gearbox Based On Local Mean Decomposition And Manifold Learning

With the continuous development and upgrading of science and technology and production capacity,people's life needs are constantly increasing and the urbanization process is accelerating.under the background of accelerating urbanization,as an important part of life and work,the relationship between urban rail transit and passengers lives and property are also more closely.Which gearbox as an important part of urban rail vehicles,one of the key components,it has strong theoretical and practical significance to identify the fault feature with its signal research.The main research object of this dissertation is the gearbox of the urban rail vehicle,and the specific research and analysis is carried out using the laboratory gearbox fault test bench.This paper collects the specific vibration fault data,comprehensively applies the improved local mean LMD decomposition algorithm,the extraction of sensitive components in the component,the fuzzy entropy feature index,and the manifold learning algorithm to effectively analyze and identify the fault features in the vibration signal.The main research articles as follows:(1)The local mean decomposition method is compared with the traditional analysis method,and can effectively deal with the analysis in the face of non-stationary and nonlinear signals.In the face of LMD's own end-effects,this paper presents an adaptive waveform matching continuation algorithm.The actual processing analysis of simulation and experimental signals shows that this method can effectively improve the LMD endpoint effect.(2)In the several PF components obtained after LMD decomposing,not all the PF obtained by the decomposition contain the fault information of the original signal.This paper proposes a method based on time-domain and frequency-domain multi-index sensitive component recognition.The actual signal analysis proves that this method can effectively distinguish sensitive components.(3)For the LMD decomposed sensitive components,this paper proposes a method for quantifying the energy of each PF component using fuzzy entropy.In order to be able to better calculate the value of the fuzzy entropy,the actual data is used to analyze and determine the selection range of the calculation parameters.The PF fuzzy entropy values are integrated to obtain the feature vectors of high-dimensional with faults,and the method of dimension reduction based on ISOMAP theory in manifold learning is proposed.This method can effectively reduce the complex high-dimensional fault feature vector to the low-dimensional visualized feature vector,which is easy to distinguish effectively.(4)The paper adopts the test bench to collect the vibration data,collect the fault signal vibration data of the normal gear,gear crack and tooth surface wear,and adopt the integrated fault identification method of improved LMD,fuzzy entropy,manifold learning,and the final analysis.The result proves that this method can effectively identify gear faults.