Abstract: To accurately diagnose the mechanical faults of fan, a new method was proposed based on wavelet packet energy analysis and improved support vector machine. Vibration signals of the fan were acquired on a 4-73 No.8D test bench under 13 different operating conditions, which were subsequently denoised, decomposed and reconstructed by wavelet packet to extract and obtain multipoint information fusion wavelet packet energy eigenvectors under various operating conditions. The sample set of above eigenvectors was trained and tested by improved support vector machine so as to diagnose and classify the mechanical faults of the fan. Results show that this method is able to effectively diagnose the category, severity and site of the fan mechanical faults with high diagnostic accuracy rate, short testing time and good online diagnosis performance.