Abstract:Aiming at the problem that traditional cyclostationary signal processing method based on second order statistics can not effectively deal with impulsive noise interference, a new fault detection method based on cyclic multiple kernel correntropy is proposed. Firstly, the definition of the multiple kernel correntropy is given, the calculation formulas of the cyclic multiple kernel correntropy function and the cyclic multiple kernel correntropy spectral density function are deduced. The denoising mechanism of the cyclic multiple kernel correntropy is analyzed. Secondly, the denoising performance of the cyclic multiple kernel correntropy under lower signal to noise ratio is verified with simulative signal, the result shows that the cyclic multiple kernel correntropy can not only effectively suppress the Gaussian noise, but also effectively suppress the nonGaussian noise. The cyclic multiple kernel correntropy provides a robust processing method for Gaussian and nonGaussian noise. Finally, the cyclic multiple kernel correntropy method was applied to the fault diagnosis of gearbox tooth wear. The experiment results show that the cyclic multiple kernel correntropy has demodulation function, can accurately characterize the spectral characteristics of gear tooth wear fault, and effectively extract the weak signal submerged in strong noise environment. The cyclic multiple kernel correntropy method is proved to be an effective method for gear fault diagnosis. .txt