Abstract:To resolve the deficiency of vibration fault diagnosis method, it was proposed that kernel principal component analysis (KPCA) fault diagnosis method based on sound signal. The basic theory of KPCA and its basic procedures for fault detection were introduced and sound signal preprocessing was depicted, multidomain feature vector was extracted from time, timefrequency and frequency domain, faults were diagnosed with KPCA method. The new KPCA fault diagnosis method based on sound signal processing was tested on axial piston pump, the result shows that this method is effective and it can overcome the deficiencies of fault diagnosis method based on vibration signal.