Fault Diagnosis Methods For EMUs Brake Control System

With the popularity of the high-speed railway in China, its operation security has attracted more and more attention. The issue of how to protect the high-speed train security and reliability has become a hot and difficult research subject recently.Brake control system, as the core subsystem of brake system in electric multiple units (EMUs), plays an important role in the security and reliability of train operation. Thus the research of status monitoring and fault diagnosis for brake control system is particularly important and critical. However, the brake control system failures occurs frequently of EMUs, due to the complexity of the brake control system and the shortage of technical transfer time for digestion and absorption in China, which affect the normal running of the EMUs seriously. With a depth study on potential failures of h EMUs brake control system, the improved fault feature extraction methods and intelligent fault diagnosis methods were put forward in this dissertation to improve the reliability and stability of brake control system. In addition, fault monitoring and diagnosis system was designed and applied to the brake control system of CRH2EMUs. The main researchs and innovations were summarized as follows:A fault feature extraction method for brake control unit (BCU) analog circuit soft fault of EMUs was proposed based on combined morphological filtering and wavelet packet energy entropy, solving the problem of complex electromagnetic environment of train. Firstly, the fault signals were pretreated to eliminate the Gaussian white noise, the peak pulse and unknown noise by combined mathematical morphological filter. Then the pretreated signals were multiple-level decomposed by wavelet packet method, and then wavelet packet energy entropy was extracted as fault feature vectors. The proposed fault feature extraction method could increase the correctly and sensitivity of BCU fault feature information.A fault feature extraction method for sensor failure of EMUs brake control system was proposed based on ensemble empirical mode decomposition (EEMD), in order to solve the problem of complex electromagnetic environment of EMUs. The traditional fault feature extraction method for sensor based on empirical mode decomposition (EMD) performed undesirable, because sensor failure output was always non-linear and non-stability and EMD decomposition had shortcoming of mode mixing, which decreased the accuracy of fault feature extraction. By adding uniform white Gaussian noise to EEMD, mode mixing was greatly reduced, and the accuracy and reliability of sensor fault feature extraction was improved.A least squares support vector machine (LSSVM) fault diagnosis method based on improved particle swarm optimization (PSO) algorithm for EMUs brake control system equipment failure with feature is a small sample size and the noise was proposed. Standard PSO has disadvantage of premature, slow convergence at the later stage and low accuracy of local search. Then,the multi-swarm cooperative chaos particle swarm optimization (MCCPSO) algorithm was proposed through absorbed the advantage of multi-swam cooperative particle swarm optimization(MCPSO) algorithm with good global search performance and chaos particle swarm optimization(CPSO) algorithm with good local search performance. Standard function optimization test and classification performance test had proved the superiority of proposed method. In order to improve the diagnosis accuracy and speed, MCCPSO algorithm was used for optimizing the structure parameters of LSSVM in BCU analog circuit tolerance soft fault diagnosis. The results showed that the fault diagnosis method had better performance, higher fault classification correct rate and faster training and test speed, as well as better classification performance than other classification methods in the small sample.As the multi-classification performance of LSSVM was influenced enormously by kernel function performance in the fault diagnosis of EMUs brake control system, a mixture kernel function was put forward. The mixture kernel function was combined by RBF kernel function with better local search performance and Polynomial kernel function with better global search performance, and an impact factor was introduced to balance the optimizing performance and optimizing time, which improved the fault diagnosis speed and accuracy effectively.The improved optimal binary tree (IOBT) structure LSSVM multi-classification method was proposed for multi-fault pattern recognition problems in the EMUs brake control system, and the weight was replaced by class separatory measure. Compared with one-versus-one (O-V-O), one-versus-rest (O-V-R) and decision directed acyclic graph (DDAG) multi-classification methods, the proposed multi-classification method not only needed few classifiers but also classification faster, as well as without classification dead zone and higher classification accuracy. Both the data sample test and BCU analog circuit soft fault diagnosis test had proved the effectiveness and superiority. The intelligent automatic detection and diagnosis system for EMUs brake control unit was designed and developed based on virtual instrument technology and Lab VIEW programming language. The test results showed that the system could automate detection and diagnosis the EMUs brake control system failures. It could be used for factory testing and maintenance of brake control system, which could improve the reliability and stability of brake control system.