Monotonicity-Induced Health Indicator for Axle-Box Bearings of Urban Rail Vehicles

Health indicators of axle-box bearings play an important role in failure prediction for urban rail vehicles. Vibration-based signal processing technology is a promising tool to reveal the degradation process with fruitful time-frequency information. However, the processes usually have non-stationary features caused by complex working conditions, which may be confused with the unknown temporal variations under different frequency bands. To address this issue, this paper proposes a new index, named monotonicity-induced health indicator, which is adaptive to weight the optimal frequency bands. First, frequency domain analysis is used to extract features from vibration signals. The monotonicity of these features is then estimated and their weights are optimized by means of maximizing the monotonicity. Eventually, all the features and weights are fused together to obtain a health indicator. The experimental result proves that the proposed health indicator shows its superiority in terms of monotonicity while accurately characterizing the degradation process of the bearings.