Fault diagnosis of hydro-turbine runner based on improved masking signal method incorporate RLMD

Abstract

Sediment erosion and foreign object impacts can cause irreversible damage to hydro-turbine runner. It is proposed to use the Improved Masked Signal method combined with the Robust Local Mean Decomposition method (IMS-RLMD) to denoise the acoustic vibrational signals collected from the hydro-turbine runner under normal operating condition, sediment-landed water flow condition, and foreign object impact condition. The IMS method reduces modal aliasing, which is easily occurred in the RLMD decomposition method, and achieves excellent signal denoised by efficiently screening the components. To validate the effectiveness of IMS-RLMD method, we use the Support Vector Machine optimized by the Grey Wolf algorithm (GWO-SVM) to identify the acoustic vibrational signals denoised by different methods. The accuracy of signal set processed by the IMS-RLMD method is 96.67 %, and the accuracy of original signal set and the signal set processed by the EMD, LMD, RLMD and VMD methods are 66.67 %, 75 %, 78.3 %, 86.67 % and 90 %. This result shows that the IMS-RLMD method is not only more suitable for denoising hydro-turbine acoustic vibrational signals, but also retains more effective features in the signals. Meanwhile, the LSTM and CNN models are set as the control group, and the identification results show that the identification method, GWO-SVM, and the denoising method, IMS-RLMD, are the best adapted solutions. Furthermore, it is found that the frequency spectrum of the acoustic vibrational signals changes significantly when foreign objects are washed into the runner or when sediment-landed water flow through the runner. This study provides an effective reference for the daily condition monitoring method of hydropower units and a valuable addition to the condition monitoring and fault diagnosis system of hydro-turbine.