Bogie fault diagnosis under variable operating conditions based on fast kurtogram and deep residual learning towards imbalanced data

The condition monitoring of bogies plays an important role in the safe operation of rail transit systems. In practice, it is difficult to diagnose bogies under variable operating conditions for traditional methods. In this paper, deep residual learning is used to deal with this problem by utilizing its powerful feature self-learning characteristics and fast kurtogram is used to further enhance the robustness of the method to variable operating conditions and noises. Besides, considering the performance of data-driven deep learning method is easily to be affected by data imbalance which often occurs in practical data collection, a novel loss function called imbalance-weighted cross-entropy (IWCE) is proposed in this paper for networks training to solve the data imbalance problem. Finally, the robustness to variable operating conditions, noises and data imbalance of the proposed method is verified by two case studies.