Multi-cavitation states identification of a sewage pump using CEEMDAN and BOA-SVM

This study introduces a novel approach to identify multiple cavitation states in sewage pumps, crucial for urban sewage treatment. Cavitation, caused by the formation and collapse of bubbles of water vapor, hampers pump efficiency and induces vibration. The proposed method combines Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Bayesian Optimized Support Vector Machine (BOA-SVM). Vibration signals and high-speed photography from cavitation experiments on a closed-loop test rig are utilized. Initially, CEEMDAN decomposes raw vibration signals into multiple Intrinsic Mode Function (IMF) components. Features are then constructed from these components, and BOA-SVM is employed for accurate identification and classification of cavitation states. The analysis reveals the highest recognition accuracy of 98.7 % at the pump casing's x-axis monitoring point, surpassing other points. Under designed operating conditions, synchronization of vibration signals from five channels results in a 30-length feature vector input into BOA-SVM, achieving an average recognition accuracy of 99.7 %, notably surpassing the 98.7 % from single-channel sensors. For operating conditions Q = 35 m3/h and Q = 45 m3/h, recognition accuracies are 100 % and 99.3 %, respectively, with an average recognition rate of 99.7 %. This study not only provides an accurate tool for identifying multi-cavitation states in sewage pumps, but also supports the development of efficient maintenance and optimization strategies.