Abstract: The sifting stopping criterion of empirical mode decomposition is one of the key factors affecting the accuracy of fault diagnosis of rotating machinery in high-speed trains. The current method of determining thresholds in advance is generally adopted, which is not adaptive. Such methods lead to the problem of mode mixing in empirical mode decomposition, which affects the fault diagnosis results. The influence of sifting stopping criterion on the empirical mode decomposition results are fully demonstrated, and then a soft sifting stopping criterion that can adaptively monitor the sifting process is proposed. This criterion is used to suppress the mode mixing problem and improve the accuracy and efficiency of empirical mode decomposition. Aiming at the target signal, the criterion defines an objective function that describes the global energy and local impact characteristics, and combines a heuristic mechanism to realize the optimization of the sifting iterations number in each sifting process, so as to guarantee the empirical mode decomposition to obtain the optimal decomposition results. Based on the simulation data and the Case Western Reserve University Bearing Data, the improved empirical mode decomposition is compared with the two traditional methods in different decomposition and diagnostic performance dimensions. Finally, the proposed improved EMD is successfully applied to the fault diagnosis case of rotating machine simulation test rig of high-speed trains.

Key words: EMD, fault diagnosis, high-speed train, mode mixing, sifting stopping criterion