Research On Condition Recognition Of High Speed Train Bogie Based On Multi-view Clustering Ensemble

With the rapid development of high speed railway,more and more people travel by high speed train(HST),which has become an essential means of transportation.Nevertheless,in the condition of long-running,some of important bogie parts of HST will be degraded and damaged.As the key part of high-speed train,the bogie plays an important role in ensuring the safety and comfort of train.In order to monitor the running status of the train and avoid some hidden dangers,it is available to collect and analyze the vibration signal on diverse running conditions by installed sensors on different parts of bogie.Therefore,it is important to effectively extract the feature from the monitoring data and then recognize the typical conditions.On the one hand,taking into account that the traditional methods of recognizing condition are mostly single-feature extraction method on the single-channel vibration signal of HST,it may easily lead to incomplete information.Hence the idea of multi-view is introduced.First of all,the characteristics of bogie vibration signal are excavated in frequency domain,time-frequency domain and time domain.The multi-views are gotten by extracting the eigenvector from the four aspects,which are respectively Fast Fourier Transformation(FFT)coefficients,wavelet energy,approximate entropy(AE)of empirical mode decomposition(EMD),and the time statistical characteristics.Secondly,the clustering result of each view is obtained by the K-means.And then the two kinds of weight of the views are calculated.Consequently,the output results of multiple clustering and the weights are combined for weighted non-negative matrix factorization(WNMF)to ensemble.By experimental comparison,it is illustrated that the method is more obvious than other single feature extraction in recognition performance,and the model based on WNMF has also a certain effect than other clustering ensemble algorithms.On the other hand,in the view of the previous methods of recognizing condition are mostly analyzed by the single-channel vibration signals of HST bogie,the information of other channels can be easily ignored.Whereas there is redundancy between full-channel vibration signals of HST bogie,a Multi-view Kernel Fuzzy C-means algorithm(MvKFCM)is proposed.Simultaneously a model based on WvFCM is constructed for condition recognition of HST bogie.Firstly,FFT of bogie vibration signals of all acceleration channels are extracted.And then the Fuzzy Classification Coefficient(FCC)of every channel is calculated after Fuzzy C-means(FCM)clustering and arranged to choose the appropriate channels.Finally,the selected channels are used to cluster by MvKFCM and the conditions of HST are recognized.The experimental results show that the selection is effective to keep rich feature information and remove redundant information.Furthermore,the condition recognition rate of MvKFCM is definitely higher than single-view and other multi-view clustering algorithms on the same premise.