Feature Extraction of Engine Acoustic Signal Based on Multi-resolution Approximate Entropy

The concept, main feature and fast algorithm of approximate entropy (ApEn) were introduced firstly. Then by analyzing the principle of wavelet packet and ApEn, the feature extraction method of multi-resolution approximate entropy was put forward and the selection principle of three parameters was simultaneously discussed during the course of the ApEn calculation. Subsequently, the engine acoustic signal was analyzed and processed. Further, the fault feature frequency band was confirmed by comparing the ApEn values of three layers wavelet packet decomposition under eight working conditions of normal state and fault state, and hence the engine fault features could be effectively extracted according to the variation of ApEn in sensitive frequency band. Finally, the inspection and diagnosis of engine were realized. The experiment results also proved that ApEn had the good ability of analyzing the complex signal feature and was an effective method to identify the running condition.