Study On Dynamic Unbalance Detection Method Of Cardan Shatf In High-Speed Train Based On Variational Mode Decomposition

CRH5 is a new type bullet train designed and manufactured in china.the composition of its transmission system is the traction motor, gear box, universal shaft, etc., among them, universal shaft should not only relatively poor in working environment, but also the slender in structure, therefore universal shaft susceptible to bias in the process of train running at high speed. In addition, the train in the long run, wearing clearance of universal joint shaft and the loosening of the universal transmission shaft balance slider etc can also lead to the eccentricity of the universal transmission shaft. Eccentric make universal transmission shaft dynamic imbalance, intensifies the vibration of transmission system, which easy to rapidly destroy power transmission components such as bearings, universal joint of transmission system, serious caused power outages or accidents. In order to ensure the safety of the power transmission, to carry out the universal shaft dynamic imbalance detection technology research is very necessary. In this paper, the variational mode decomposition algorithm is applied to the universal shaft dynamic unbalance. We put forward the modification variational mode decomposition model, and overcome the shortage of the influence parameters difficult to choose, to achieve the best diagnosis effect. And introduce the idea of pattern recognition, fuzzy clustering to make the model more intelligent, which get rid of the dependence on the Fourier spectrum. Simulation and experimental results verify the effectiveness of the proposed method.The main research contents and conclusions were as follows:(1) The drive system structure of type CRH5 emu was to deeply understand, then we had setted up the corresponding dynamic unbalance experiment platform, and then collected the data of the dynamic imbalance under various operating conditions to provide data support for the later research. We had introduced foreign advanced algorithm, namely the variational mode decomposition algorithm, detailed introduced the principle and characteristics of the algorithm, and compared with empirical mode decomposition algorithm. The simulation results show that the variational mode decomposition algorithm is obvious advantages in terms of fault feature extraction.(2) We had proposed modification variational mode decomposition model, which based on the algorithm is similar to a set of adaptive wiener filter, formula for calculating the number of components was proposed, and combined with Fourier spectrum entropy difference to determine the penalty parameters, so as to solve the shortage of the influence parameters difficult to choose, to achieve the best separation effect, through the spectrum analysis to achieve the effect of dynamic unbalance diagnosis. And the definition of Fourier spectrum entropy difference was gave and capture feature of change of information contained in the mode by the Fourier spectrum entropy difference was analysis. At the same time this article had build the Fourier spectrum entropy difference a one-to-one relationship with the penalty parameter, using spectral entropy difference beating point determined penalty parameters. The simulation and experimental results show that the Fourier spectrum entropy difference can effectively ensure punish parameters and the effectiveness of the proposed modification variational mode decomposition model. With comparison to VMD, the problem of Parameter selection has been solved. With comparison to the classic ensemble empirical model decomposition, the fault detection ability has been significantly improved.(3) We had introduced the idea of pattern recognition, clustering to realize fault diagnosis, which get rid of the dependence on Fourier spectrum.The VMD- SVD- BP neural network model, the variational mode decomposition and fuzzy c-means clustering model were put forward. the core of these models are separated signal in VMD,and extract the fault feature vector by SVD, and then use neural network as a classifier for fault diagnosis or obtain the corresponding clustering center by the fuzzy c-means clustering, calculate the hamming approach degrees of test data for fault diagnosis. Experimental results show that the highest accuracy of EMD-SVD-BP neural network model is 98.33%, the mean square error is 0.0166, while the VMD-SVD-BP neural network model accuracy can reach 100%, mean square error dropped to 9.9132 e-12, diagnostic performance has improved.Empirical mode decomposition with the fuzzy c-means clustering model can diagnose serious imbalance, but the discriminant effect for the new shaft and early dynamic imbalance is very poor, and the variation mode decomposition and fuzzy c-means clustering model can well distinguish three states, accuracy of 100%.