Non-stationary Feature Extraction Method For Small Amplitude Hunting In High-speed Trains

Along with the rapid development of China's high-speed railway in recent years,the running speed of trains is getting faster and faster,and the stability of train operation is also appearing.Therefore,real-time safety monitoring of train operation status and timely prediction of high-speed train operation status are the key factors to ensure the safe operation of high-speed trains.During the running of the train,the pair of wheels of the machine will make a side-shifting movement of the head around the center line of the track,which is a hunting movement.The existing method of monitoring hunting movements is to monitor whether the vehicle has a hunting instability by monitoring the lateral acceleration of a position sensor,and can only detect the sharp and large hunting anomaly of the train.Moreover,the author team analyzed and processed the lateral acceleration data of the train steering frame,and found that the train was prone to small amplitude hunting movements at high speed.The small amplitude hunting can be divided into two kinds of running states: small amplitude hunting convergence and small amplitude hunting divergence.The small amplitude hunting convergence state will gradually converge to normal operation,while the small amplitude hunting divergence state will continue to diverge to a large amplitude hunting instability,and thus harm the safety of the train.The existing research rarely involves two evolution states of small amplitude hunting.Therefore,it is urgent to monitor the two states of small amplitude hunting evolution to ensure the safe operation of trains.In this paper,the small amplitude hunting evolution state monitoring problem caused by trains in high-speed operation is taken as the research object of high-speed train frame lateral acceleration signals at multiple locations.Various nonlinear signal processing methods are adopted to extract the lateral acceleration of trains in high-speed operation.The signal is extracted to monitor the running state of the high-speed train in time to ensure the safe operation of the train.The main work of this paper is:(1)A framework for extracting small-amplitude hunting features of high-speed trains based on Ensemble Empirical Mode Decomposition(EEMD),manifold learning and Singular Vector Decomposition(SVD)is proposed.Considering the complex background of highspeed train operation,the collected lateral acceleration signal is thus seriously polluted,the source signal shall be pre-processed.The EEMD algorithm is used to extract the processed data,and considering the resulting modal aliasing and white noise residual,the obtained IMF component is denoised using SVD.Then the energy matrix is obtained.The state clustering and further feature extraction of high dimensional nonlinear data are carried out by manifold learning method.Finally,the least squares support vector machine(LSSVM)is used to determine the fault.The experiment proves the superiority of the feature frame and can detect the running state of the high-speed train in time.(2)For the problem that the existing diagnosis and identification method for the highspeed train cannot identify the running state of the train quickly and accurately,this paper uses the ensemble empirical mode decomposition method to decompose the pre-processed signals,convert the results into energy matrices.The joint approximation diagonalization process under non-stationary condition is performed to process energy matrices of multiple sensors into fusion feature.By incorporating the fusion feature into the least squares support vector machine for training and recognition,it is verified that the method can quickly and accurately operate the high-speed train normal signal,small amplitude hunting convergence signal,small amplitude hunting divergence signal and large amplitude hunting signal,so as to ensure train operation safety.