A new method to select frequency band for vibration signal demodulation and condition estimation of rolling bearings

The narrowband amplitude demodulation of vibration signals is widely used to extract components carrying information about rolling bearing faults for condition estimation. However, precisely selecting the frequency band determines the quality of demodulation. Although transient detection based methods have achieved satisfactory performance in some scenarios, they may be severely disturbed by strong random impulses or harmonics. Additionally, frequency division strategy is always considered as rough. Aiming at these problems, a new method is proposed to select the optimal frequency band by combining the frequency division strategy, genetic algorithm (GA) optimization, and an automatic fault characteristic order (FCO) search algorithm. Aimed at a signal, its sub-signals are extracted according to the frequency division strategy and the order tracing technology is implemented to obtain the order spectrum of each sub-signal at first. Then, the FCO and its multiples (FCOs) are automatically searched from each order spectrum, and their amplitude summation is calculated. Afterwards, the ratio of FCOs summation to noise amplitude (RFN) is measured as the basis to construct an RFNgram for selecting the primary optimal frequency band Finally, GA is implemented to select the optimized optimal frequency band where the signal is demodulated for estimating the bearing condition. RFN measures both the impulsiveness and periodicity of signals, and GA optimizes the result from a RFNgram, essentially making up for the shortcomings of previous methods under the condition of multiple interferences. RFN is measured through the order spectrum, which also allows it to be applicable to varying speeds, thus achieving a more comprehensive range of industrial applications. Three case studies are implemented to present the superiority of the proposed method in condition estimation for rolling bearings by comparison with four classical or advanced methods.