Abstract: Focusing on the non-stationary characteristics of mechanical vibration signals of high voltage distribution equipment, a novel feature extraction method of vibration signals was proposed based on a joint analysis of wavelet packet decomposition and reconstruction, Hilbert transform and normalized energy spectrum which all can be performed on the platform of LabVIEW. Simulating experiments on distribution equipment were conducted and vibration signals of normal condition, opening spring loose and friction in solenoid were collected. The vibration signals were analyzed based on the characteristics of a time-frequency and K-nearest neighbor algorithm for different conditions feature quantity pattern recognition can make a detailed analysis of the signal. Experiment results show that each element of the normalized energy spectrum vector of normal signal in the distribution equipment is evenly distributed; while the elements of fault signal normalized energy spectrum vectors are remarkably varied. The accuracy of KNN is 93.3% and its recognition speed is fast, and the results verify the feasibility of the mechanical fault diagnosis approach for high voltage distribution equipment employing K-nearest neighbor algorithm scheme.