Cavitation identification based on feature extraction

Cavitation is a common phenomenon in centrifugal pumps and can result in severe damage to impellers and mechanical components, ultimately leading to significant economic losses. Therefore, the timely and accurate identification of cavitation states is critical to ensuring the reliable operation of such systems. This study presents a novel cavitation detection approach based on time-frequency feature analysis. A dedicated cavitation test bench was constructed to acquire vibration acceleration signals under various cavitation numbers. Singular Value Decomposition (SVD) was applied to reduce signal noise, and salient time-frequency features were extracted from the processed data. These features were subsequently used as input to a Support Vector Machine (SVM) classifier for the identification of cavitation states. Experimental results demonstrate that the proposed method accurately captures the cavitation characteristics embedded in vibration signals and achieves superior real-time performance compared to conventional techniques, with a classification accuracy exceeding 95.65%.