A New Rolling Bearing Work Condition Monitoring Method Based on Back Propagation Network

Rolling bearings are among the most used elements in rotating machines, thus effective bearing work condition monitoring is necessary to avoid sudden machine failures. Vibration signal analysis is a common way. In order to automatically provide accurate diagnosis results, a new method based on Back Propagation (BP) network is proposed. Wavelet ridge transform has been applied to filter part environmental noise, and then a new demodulation method combing Hilbert transform method and autocorrelation curve has been introduced to overcome the remaining noise. The autocorrelation curve of envelop signal always shows its common and relatively apparent periodicity rule when defect occurs. At last, a Back Propagation (BP) network is used and trained to identify the periodicity and give the label whether the bearing is normal (label = 0) or defective (label = 1). Thus, bearing condition can be effectively monitored. At the same time, the characteristic frequency of the defect can be directly obtained according to the autocorrelation curve. Compared to existing methods based on complicated networks, this method operates simpler. Simulated signals and experimental defect signals illustrated the effectiveness and superiority of this method.