L. Dong - Research Center of Fluid Machinery Engineering and Technology, Jiangsu University, Zhenjiang ,Jiangsu Province, ۲۱۲۰۱۳, China
J. C. Zhu - Research Center of Fluid Machinery Engineering and Technology, Jiangsu University, Zhenjiang ,Jiangsu Province, ۲۱۲۰۱۳, China
K. Wu - Research Center of Fluid Machinery Engineering and Technology, Jiangsu University, Zhenjiang ,Jiangsu Province, ۲۱۲۰۱۳, China
C. Dai - School of Energy and Power Engineering, Jiangsu University, Zhenjiang, Jiangsu Province, ۲۱۲۰۱۳, China
H. L. Liu - Research Center of Fluid Machinery Engineering and Technology, Jiangsu University, Zhenjiang ,Jiangsu Province, ۲۱۲۰۱۳, China
L. X. Zhang - Research Center of Fluid Machinery Engineering and Technology, Jiangsu University, Zhenjiang ,Jiangsu Province, ۲۱۲۰۱۳, China
J. N. Guo - Research Center of Fluid Machinery Engineering and Technology, Jiangsu University, Zhenjiang ,Jiangsu Province, ۲۱۲۰۱۳, China
H. B. Lin - Sichuan Provincial Key Lab of Process Equipment and Control, Sichuan University of Science & Engineering, Zigong, Sichuan Province, ۶۴۳۰۰۰,China

Cavitation monitoring is particularly important for pump efficiency and stability. It is easy to misjudge cavitation by using a given threshold of a single eigenvalue. In this work, based on the vibration signal, a method for multi-resolution cavitation status recognition of centrifugal pump is proposed to improve the accuracy and universality of cavitation status recognition., wavelet packet decomposition (WPD) is used to extract the statistical eigenvalues of multi-scale time-varying moment of cavitation signal after reducing the clutter, such as root mean square value, energy entropy value and so on. The characteristic matrix is constructed. Principal component analysis method (PCA) is employed to reduce the dimension of the characteristic matrix and remove the redundancy, which constructs the radial basis function (RBF) neural network as the input. The results show that the overall recognition rate of non-cavitation, inception cavitation and serious cavitation by using the vibration signal of one measuring point is more than ۹۷.۷%. The recognition rate of inception cavitation is more than ۸۰%. Based on the vibration signal information fusion method of two measuring points, the recognition rate of centrifugal pump inception cavitation status reaches more than ۹۹%, and the recognition rate of vibration signal information fusion method of three measuring points reaches ۱۰۰% for all three cavitation statuses. Due to the influence of factors such as change of external excitation and abrupt change of working conditions, sensor data acquisition is often subjected to unpredictable disturbance. To study the ability of single-point cavitation status recognition method to resist unknown disturbances, by constantly adjusting the value of the interference coefficient of the interference term. It is found that the recognition rate of cavitation status using single measuring point decreases almost linearly with the increase of the interference coefficient. When five measuring points are used for information fusion cavitation status recognition, the cavitation status recognition rate still reaches over ۹۰% even if the interference factor of one measuring point reaches ۵۰%.

Cavitation recognition, Centrifugal pump, Vibration, Wavelet packet decomposition, Principal component analysis, RBF neural network