Feature extraction of gearbox vibration signals based on EEMD and sample entropy

Gearbox vibration signals are non-stationary and nonlinear. In order to identify different working conditions of the gearbox, the eigenvalues of corresponding signals should be extracted. The paper proposed for the first time a method of combining ensemble empirical mode decomposition (EEMD) and sample entropy to extract effectively the eigenvalues of corresponding gearbox vibration signals of wind turbine in three different states, namely normal gear, worn gear and gear with broken teeth. The results showed that EEMD could overcome the problem of frequency aliasing and had the advantage of strong adaptability, while the sample entropy could solve the problem of lower relative consistency of approximate entropy in small sample data processing, and improve the computing speed as well. The method of combining EEMD and the sample entropy could show the status characteristics of gearbox vibration signals well during the process of feature extraction. The new feature extraction method is fast, effective, reliable, and it can provide a reliable input vector of characteristic value for the following fault classification.