Fan fault diagnosis based on wavelet packet and sample entropy

To accurately diagnose the mechanical failure of the fan, two diagnostic methods based on the wavelet packet energy feature and sample entropy feature are proposed. Vibration signals acquisition of 13 kinds of running states are achieved on the 4-73 No. 8D centrifugal fan test bench. The wavelet packet energy feature vector of each vibration signal is rapidly extracted through the wavelet packet denoising, decomposition and reconstruction. The vibration signal wavelet packet energy feature vector of the five measuring points in the same instantaneous running state are fused into the fan fault feature vector. Finally, the fault diagnosis of the fan is achieved by using improved SVM (Support Vector Machine) classifier, and the accuracy rate is 94.6%. A new fan fault feature vector is put forward, which is the integration of the vibration signal sample entropy of the five measuring points in the same instantaneous running state, and then the fault diagnosis of the fan is achieved by using improved BP (Back Propagation) neural network, and the accuracy rate is 99.23%. The diagnostic results show that these two methods are able to effectively diagnose the category, severity and site of the fan mechanical failures, and suitable for online diagnosis. © 2013 Universitas Ahmad Dahlan.