A fault diagnosis method for shaft system of hydropower units based on improved symbolic dynamic entropy and stochastic configuration network

The existing research on shafting fault diagnosis of hydropower units is mainly based on the vibration sig-nal data of a single sensor. There are some problems such as lack of fault information and difficult in selecting sen-sor measurement points. Therefore, a shafting fault diagnosis method for hydropower units based on the combination of refined composite multivariate multiscale symbolic dynamic entropy (RCMMSDE) and stochastic configuration network (SCN) is proposed in this paper. First, the refined composite technique is introduced into RCMMSDE model to improve the problem of insufficient coarse-graining of traditional multivariate multiscale entro-py. Then, the RCMMSDE values of vibration signals from different sensors are extracted as fault features. Finally, the fault features are input into SCN network to realize the accurate shafting fault identification of hydropower units. Simulation results show that the RCMMSDE-SCN model achieves the highest diagnostic rates of 97.58% and 99.17% on two different data sets respectively, which verifies the good diagnostic performance of the proposed mod-el. At the same time, the diagnosis performance of different diagnosis models under different scenarios of multiple sensor signals and single sensor signals is compared, which indicates that the fusion of multiple vibration signals can effectively improve the identification performance of hydropower unit shafting fault diagnosis model. This study provides a new method for multi-sensor vibration signals of hydropower units, and has good reference value. © 2022 China Water Power Press. All rights reserved.