Research On Fault Feature Extraction And Diagnosis Method For Train Sliding Plug Doors Based On Sound Signal

Due to the advantages in airtightness,sound insulation,and reliability,train sliding plug doors have been widely used in railway systems,especially in high-speed railways.As the passage for passengers to get on and off,train sliding plug doors directly affect the operation efficiency of the train and the personal safety of passengers.The failures of train sliding plug doors may not only threaten the safety of passengers,but also cause train delays.Due to the characteristics of rail transit,it may cause a large backlog of delay.However,traditional fault-based maintenance or planned maintenance methods have a series of disadvantages such as high cost,low efficiency,poor pertinence,and improper maintenance.Therefore,for the railway industry,the fault diagnosis of train sliding plug doors is of great significance.Sound signals have the advantages of convenient collection,flexible collection methods,and non-contact(no interference to the equipment).By analyzing the action sound signals of train sliding plug doors under different working conditions,this thesis proposes a non-contact fault diagnosis method,and studies data preprocessing methods and different feature extraction methods.A signal reconstruction method based on hybrid criteria is proposed to realize the reconstruction of sound signals.By introducing the idea of fractional calculus to entropy,fractional wavelet packet energy entropy and fractional multi-scale permutation entropy are put forward,and on this basis,weighted fractional wavelet packet energy entropy and weighted fractional multi-scale permutation entropy with better diagnosis effect are put forward.The main research contents are as follows:(1)Aiming at the problem of the single signal processing method that it is difficult to fully mine the fault feature information contained in the sound signal of train sliding plug doors,resulting in insufficient and incomplete diagnosis information,given the advantages of empirical mode decomposition in processing non-stationary signals and entropy in characterizing signal complexity,a fault diagnosis method based on empirical mode decomposition and entropy is proposed.Firstly,empirical mode decomposition is used to decompose the sound signal of train sliding plug doors,and a series of intrinsic mode functions are obtained.Secondly,the correlation between each intrinsic mode function and the original sound signal is analyzed,and the intrinsic mode functions with a larger correlation coefficient are selected for further research.The wavelet packet energy entropy and multi-scale permutation entropy of the selected intrinsic mode functions are extracted and analyzed respectively,and the Fisher discrimination criterion is used to select the optimal features.Finally,the particle swarm optimization algorithm is used to optimize the hyperparameters of the support vector machine.Then the selected features are input to the optimized support vector machine to realize the fault diagnosis of train sliding plug doors.The experimental results show that the proposed method has a better diagnosis effect than the single method.On the whole,the diagnosis accuracy based on empirical mode decomposition and multi-scale permutation entropy is 93.62%,which is higher than that based on empirical mode decomposition and wavelet packet energy entropy(91.49%).(2)Aiming at the problem of large noise in the sound signals of train sliding plug doors,a method for selecting the intrinsic mode functions based on both energy criterion and kurtosis criterion is proposed,and the selected intrinsic mode functions are used for signal reconstruction.By analyzing the energy and kurtosis of the intrinsic mode functions of sound samples under different working conditions,it is found that compared with the existing signal reconstruction method based on the single kurtosis criterion,the proposed signal reconstruction method based on kurtosis criterion and energy criterion can preserve the intrinsic mode functions that effectively contribute to the fault features in the sound signal of train sliding plug doors.(3)In view of the complicated and diverse failure modes of the train sliding plug door,and the difficulty of distinguishing the failure modes using traditional entropy,the fractional calculus is introduced to entropy,and the concepts of fractional wavelet packet energy entropy and fractional multi-scale permutation entropy are proposed.The range and determination method of the fractional factor are given,and the two kinds of fractional entropy of the reconstructed signal based on the hybrid selection criteria are extracted.Finally,the support vector machine optimized by particle swarm optimization algorithm realizes the fault diagnosis of train sliding plug doors.The experimental results show that the fractional wavelet packet energy entropy and the fractional multi-scale permutation entropy can greatly improve the fault diagnosis accuracy of train sliding plug doors compared with the traditional wavelet packet energy entropy and multi-scale permutation entropy.The fault diagnosis accuracy based on signal reconstruction and fractional multi-scale permutation entropy is 97.87%,which is higher than that based on signal reconstruction and fractional wavelet packet energy entropy(96.28%).(4)To improve the differentiation of similar faults,starting from enhancing the sensitivity of the fractional wavelet packet energy entropy to the energy distribution in different frequency bands,and the sensitivity of the fractional multi-scale permutation entropy to permutation entropy at different scales,by introducing the idea of weight,the concepts of weighted fractional wavelet packet energy entropy and weighted fractional multi-scale permutation entropy are proposed.To optimize the weights of fractional entropy,a synchronous optimization strategy for the hyperparameters of support vector machine and fractional entropy weights based on particle swarm optimization algorithm is proposed to realize the synchronous optimization of fractional entropy weights and the support vector machine.By applying the proposed weighted fractional entropy to the sound signal of train sliding plug doors,it is found that the fault diagnosis accuracy of train sliding plug doors can be further improved.Besides,the fault diagnosis accuracy based on signal reconstruction and weighted fractional multi-scale permutation entropy reaches 99.47%,which is higher than that based on signal reconstruction and weighted fractional wavelet packet energy entropy(97.87%).