An optimal frequency band selection for bearing fault diagnosis based on squared envelope analysis

The squared envelope spectrum (SES) is widely used to rolling element bearing (REB) fault diagnosis since it is simple and effective. Generally, the performance of SES depends on the bearing fault signal that separated from the raw signal. Bandpass filtering is a commonly used approach to separate different signal components into different frequency bands. Both the centre frequency and bandwidth affect the performance of the separation. In order to separate the bearing fault signal into a frequency band well, several bandwidths are defined in advance and the distribution of the centre frequency promises no blind span throughout the entire frequency range. The kurtosis of the squared envelope (SE) and SES are usually used to select the filtered signal that contains most of the fault information. Most of the present approaches assess the kurtosis of the entire SE or SES which are easy to be interfered with by a single or a few impulses in the SE or SES. A method based on the kurtosis of the SES is proposed, which only evaluates the kurtosis of the part of SES that is closely connected to the bearing fault. Therefore, the interference of a single or several undesired impulses in the SES can be avoided. The performance of the method is tested via simulation signals and real vibration data.