Fault Diagnosis of High-speed Train Bogie Based on Spectrogram and Multi-channel Voting

Fault diagnosis of high-speed train bogie is of great importance in ensuring the safety of train operation. The multichannel vibration signals measured at different positions on the bogies characterize the dynamics of the vehicle and contain key information describing the performance of the bogie components. However, due to the complexity and uncertainty of the signals, it is hard to extract stable features that represent the characteristics of the signals. Besides, manual selection of reliable channels is indispensable in existing works. This paper presents an ensemble of methods for fault type recognition of high-speed train bogie based on spectrogram images and voting method. First, vibration signals of bogies are transformed to spectrogram images that are then taken as the input of Random Forests (RFs). In the next, four voting methods including Plurality Voting (PV), Classification Entropy (CE), Winner Takes All (WTA), as well as a novel method we proposed using neural network (NN) is applied for combining all the channels’ classification results to give a final decision on fault type. The proposed method not only avoid complicated feature extraction procedures by using a simple transform, but also make the best of multiple channels by automatic combination. Experiments conducted on the dataset based on SIMPACK simulations have verified the efficacy of the presented method in classifying key component(s) failures, with accuracy near 100%. Further, a more complex fault state in which the components of bogies only lose their effectiveness partially, instead of fully, has been tested and analyzed, where near 90% of accuracy is achieved. These results demonstrate the high robustness of the new method.