Empirical wavelet transform is a tool that can effectively distinguish complex signals containing different frequency information components. However, due to the high complexity of Fourier spectrum, using this method will generate a large number of boundaries, resulting in more invalid components, so there are certain limitations in application. In this study, concise empirical wavelet transform (CEWT) was used for signal partitioning. By calculating the power spectral density of the signal instead of Fourier spectrum, this method avoids the situation of a large number of extreme points in the original method. In order to effectively extract periodic pulse information from signal components and reduce the impact of noise, this study used a periodic pulse detection index ---- harmonic spectral kurtosis (HSK). HSK can extract harmonic information from the envelope spectrum, quantify periodic pulses in the signal, and weaken the impact of noise such as random pulses. After verification by simulation and experimental signals, the proposed method can be used for fault diagnosis of rolling bearings in rotating machinery.