A recurrent neural network based health indicator for remaining useful life prediction of bearings

In data-driven prognostic methods, prediction accuracy of bearing remaining useful life (RUL) mainly depends on the performance of bearing health indicators, which are usually fused from some statistical features extracted from vibration signals. However, many existing bearing health indicators have the following two shortcomings: (1) many statistical features do not have equal contribution to construction of health indicators since the ranges of these statistical features are different; (2) it is difficult to determine a failure threshold since health indicators of different machines are generally different at a failure time. To overcome these drawbacks, a recurrent neural network based health indicator (RNN-HI) for RUL prediction of bearings is proposed in this paper. Firstly, six related-similarity features are proposed to be combined with eight classical time-frequency features so as to form an original feature set. Then, with monotonicity and correlation metrics, the most sensitive features are selected from the original feature set. Finally, these selected features are fed into a recurrent neural network to construct the RNN-HI. The performance of the RNN-HI is verified by two bearing data sets collected from experiments and an industrial field. The results show that the RNN-HI obtains fairly high monotonicity and correlation values and it is beneficial to bearing RUL prediction. In addition, it is experimentally demonstrated that the proposed RNN-HI is able to achieve better performance than a self organization map based method.