Different sensors exhibit distinct characteristics and sensing methods. To improve the overall classification accuracy of cavitation noise intensity in hydraulic turbines, this paper aims to present a novel multi-sensor fusion approach that combines cavitation noise signals collected by hydrophones, acoustic emission sensors and acceleration sensors. This study proposes a method for extracting cavitation noise features from hydraulic turbines, using complementary strengths of these sensors.
This paper uses hydrophones, acoustic emission sensors and acceleration sensors to extract cavitation signals from hydraulic turbines, implementing both data-level and feature-level fusion. Wavelet energy entropy, power spectral density and the Hilbert spectrum are employed to extract the features of the cavitation signals. An adaptive weighting method is applied for data fusion, while a stacked sparse autoencoder is used for feature fusion. In the results analysis, the random forest classifier is applied to classify the cavitation features collected by individual sensors. Subsequently, the fused cavitation features from multiple sensors are classified to enhance overall performance.
The results demonstrate that the overall classification accuracy significantly improves after fusion, with the highest accuracy reaching 93.3%. This enhancement can effectively boost the overall efficiency and performance of the turbine.
The research findings serve as a valuable reference for feature extraction and classification of cavitation noise intensity in hydraulic turbines.
