A corrected circular graph-driven fault diagnosis method for tidal stream turbine blades

Abstract

Tidal stream turbine (TST) blades are susceptible to failure due to biofouling, accurate blade faults diagnosis is of great significance. However, the insignificance of multi-blade fault features reduce diagnosis accuracy, and the random swell effects further increase the fluctuations of multi-blade fault features. Therefore, a corrected circular graph-driven fault diagnosis (CCGFD) method is proposed to improve the multi-blade faults diagnosis accuracy under the swell effects. In this approach, a corrected circular graph construction (CCG) method is proposed to capture the multi-blade faults features, and the radius circular graph interactive network (RGN) is proposed for multi-blade faults classification. Specifically, the CCG method converts one-dimensional stator current signals into two-dimensional corrected circular graphs, and the correction of vector trajectory improves the discriminability of the blade fault features under the swell effects. Furthermore, the RGN method enriches the multi-blade fault features by fusing one-dimensional circular radius and two-dimensional corrected circular graph, which contains fine-grained fault features in both the time and angular domains. Finally, several experiments on a 230 W TST prototype verified the effectiveness and the robustness of the proposed method.