Stator current-based soft sensing for blade damage diagnosis in centrifugal pumps using ensemble learning

Soft sensing offers an indirect approach for fault diagnosis in centrifugal pumps, effectively ensuring their safe and stable operation. To accurately identify blade damage faults in centrifugal pumps, this paper processes the stator current signal from pumps with seven types of blade damage under various impeller and operating conditions, and prediction models for blade damage are developed based on 4-class and 63-class classification. The results show that as the damage degree of the blade increases, the RMS value of the current signal decreases. The total energy of the IMF1 decreases with increasing damage in the marginal spectrum, while the peak frequency and average frequency of the IMF2 increase as the damage degree increases. By extracting fault feature sets, the 4-class damage prediction model, developed using the Random Forest algorithm, achieves an accuracy of 91.17 % on the test set, effectively identifying the damage degree of the blade in centrifugal pumps. The 63-class damage prediction model achieves an accuracy of 80.25 % on the test set, accurately identifying varying degrees of blade damage under different operating conditions. The soft sensing method proposed in this study provides an effective solution for centrifugal pump impeller damage fault diagnosis, demonstrating significant application value for chemical process safety monitoring.