Abstract: In view of problems that fault signal of centrifugal pump is easily submerged by noise and data analysis is difficult, a wear fault analysis method of centrifugal pump impeller based on multi-source signal fusion was proposed. Three vibration signals of volute, outlet and base of centrifugal pump, and electrical signals of prime mover terminal are collected when centrifugal pump impeller is in normal condition and wear state. Wavelet packet decomposition is used to extract characteristic frequency bands of the vibration signals, it is determined that the base can be used as the best detection point by laterally comparing energy values of each frequency band, and frequency analysis range is narrowed by longitudinally comparing the energy values of each frequency band. Based on narrowed frequency analysis range, chirp Z transform is used to analyze electrical signals of prime mover terminal and separate fault characteristics frequency and the third harmonic frequency, so as to extract fault characteristic frequency accurately. Test result verifies effectiveness of the method.