Incipient Fault Detection Method using Kullback-Leibler Divergence for CRH5 and Some Remarks

摘要

Since incipient faults often have low amplitudes and may be easily masked by system disturbances or sensor noises, most of existing fault detection (FD) methods are ineffective for this type of faults. In order to overcome the existing problems in data-driven FD methods, this paper proposes the Kullback-Leibler divergence (KLD)-based FD method in a rotary space, specially for the electrical drive system of CRUS-type high-speed trains. Furthermore, the statistical limiting of this method is given by mathematical deviations. Finally, some concluding remarks are drawn to evaluate the practicability of the existing KLD-based FD methods.