Research On The Representation And Decoupling Algorithms Of Urban Rail Train Wheelset-axle Box Fault

The wheelset-axle box system is a key organization of urban rail trains,and its healthy service status is of great significance to the safe and efficient operation of trains.After the wheelset-axle box system fails,it will increase the maintenance cost,or cause traffic safety accidents.The more complex mechanical structure makes the system fault non-linear,difficult to be characterized by vibration signals,and system fault coupling Seriously,the fault features are confused and difficult to extract.Therefore,the effective monitoring,characterization and decoupling diagnosis of the system is an important part of maintaining the safe service status of urban rail trains,and also an important guarantee for urban rail trains to avoid economic losses and safety accidents.In view of the traditional algorithms for the characterization of urban rail train wheelset-axle box failures,such as nonlinear characterization difficulties and reliance on prior knowledge,this paper discusses the statistical characteristics based on the form and signal characteristics of urban rail train wheelset-axle box faults.On this basis,a two-dimensional signal representation algorithm based on frequency domain characteristics,a compound fault decoupling algorithm and a train speed estimation algorithm based only on vibration signal decoupling are proposed.For the statistical feature-level representation method,statistical feature selection is carried out and the multi-state test bench data is used to verify the selection.First,analyze the mechanical mode and signal characteristic level of the multi-state wheelsetaxle box failure,and explain the mechanical movement modes and corresponding signal characteristics of the wheelset failure,axis failure,and composite failure.Secondly,by selecting multiple Time domain and frequency domain statistical characteristics,comprehensive analysis of the fault mode,speed influence,feature type and other factors in the fault state characterization of the sensitivity,and through the proportional test bench for multi-state fault simulation combination,the time domain,frequency After normalizing the statistical features in the same scale,we obtained the conclusion that the frequency-domain features are generally better than the time-domain statistical indicators for the discrimination of wheelset-axle box fault status,and the optimal statistical indicators are obtained.Aiming at the problem of insufficient intelligence of traditional vibration signal image representation algorithms and relying on prior knowledge,a representation algorithm based on the angle field of the Gram matrix in the frequency domain is proposed.First,analyze the shortcomings of the traditional Gram matrix characterization method for the periodic impact signal characterization of rotating machinery,and derive the specific algorithm;second,based on the aforementioned frequency domain characteristics that have strong characterization capabilities for wheelset-axle box multi-state faults,it is proposed The Gram algorithm in the frequency domain takes the frequency domain signal as input for angular fieldization,and uses the correlation extraction ability of the Gram matrix to extract the frequency domain impact correlation as a two-dimensional image feature to realize the image representation of the wheelset-axle box multi-state fault;finally,Based on the migration learning algorithm,intelligent fault diagnosis based on the angle field of the Gram matrix in the frequency domain is realized.Through the data of the proportional test rig,the validity of the proposed algorithm for the characterization and diagnosis of wheelset faults,axle box bearings and compound faults is verified.Aiming at the shortcomings of classical vibration signal characterization algorithms relying on artificial experience,a characterization algorithm based on spectral Markov transfer field is proposed.First,analyze and derive the traditional time-domain Markov transition field algorithm's ability to extract the dynamic transition probability of the system state,and explain the reason for the insufficient characterization ability of the classic algorithm for wheelset-axle box failure;secondly,propose the Markov state transition field for spectrum signals The characterization algorithm of the system realizes the statistical field analysis of the spectrum characteristics and the two-dimensional image by matrixing the state transition probability of the statistical system spectrum;finally,the migration learning network based on small sample data input realizes the Markov transition based on the spectrum Intelligent diagnosis of field images.Through the data of the proportional test rig,the validity of the proposed algorithm for the characterization and diagnosis of wheelset faults,axle box bearings and compound faults is verified.Aiming at the problem of failure to extract fault characteristics caused by fault coupling in the wheelset-axle box composite fault vibration signal,a symplectic geometric modal decomposition-mutual approximate entropy decoupling algorithm based on biaxial vibration is proposed.Firstly,the principle of a novel algorithm for deriving the symplectic geometric modal decomposition is introduced,which demonstrates the excellent decoupling ability of the novel algorithm for complex signals;secondly,it analyzes the vertical and axial biaxial signals in the wheelset-axle box state characterization A composite fault decoupling algorithm based on the symplectic geometric decomposition of the biaxial vibration signal and mutual approximate entropy reconstruction is proposed.Finally,the biaxial symplectic geometric modal decoupling algorithm is used to separate the vertical and axial signals.Fault feature extraction and comparison.Through the data of the proportional test bench,the superiority of the proposed algorithm for the decoupling and diagnosis of wheelset-axle box compound faults is verified.Aiming at the problem that the speed measurement system of urban rail train wheelset-axlebox signal processing is complicated and depends on the wheel diameter value,a speed estimation algorithm based on decoupling processing and correlation analysis that only relies on vibration signals is proposed.First,the decoupling analysis of single-channel vibration signals is carried out,the periodic track impact period of a fixed distance is extracted by autocorrelation,and real-time speed calculation is realized by combining the track length;secondly,the decoupling of the fixed-distance dual-channel vibration signal is carried out,using the maximum After the kurtosis principle realizes the component selection,the dual-channel impact time lag is obtained based on the cross-correlation extreme value of the fixed-distance dualchannel signal,and then real-time vehicle speed estimation is realized based on the known fixed distance;finally,the actual vehicle speed data collection of a domestic subway line By comparing the two proposed speed estimation algorithms with the displayed speed in the driver's cab,the reliability of the two speed estimation algorithms is verified without relying on the wheel diameter value only by the vibration signal.The research results will provide algorithmic basis for the analysis and diagnosis of real train failures of the wheelset-axle box system of urban rail trains,and have positive significance for the maintenance of urban rail transit equipment.