Abstract: Since the vibration signal of gear is complicated and degradation features are hard to extract, a novel method based on local characteristic-scale decomposition (LCD) and composite spectrum (CS) entropy is proposed. The CS algorithm is modified, and Fourier transform is replaced by discrete cosine transform so as to reduce the information missing and improve the feature sensitivity. On this basis, LCD-CS algorithm is proposed for degradation feature extraction. Vibration signal is decomposed by using LCD method with high frequency harmonics. Bayesian information criterion and kurtosis time series cross correlation coefficients are used to screen the intrinsic scale components, so as to abandon the unnecessary components and extract the feature information effectively. In order to improve the degradation measurability of features, the selected ISC components are fused by using the modified CS algorithm, and the CS entropy is extracted as feature vector. The proposed method is applied to the gear run-to-failure degradation experiment, and the feature extraction and degenerative status recognition of the measured signals are carried out. The results show that the modified composite spectrum entropy has a good ability to characterize the gear degenerative state. Key

Key words: gear, performancedegradation, featureextraction, localcharacteristic-scaledecomposition, compositespectrumanalysis