Real-Time Remaining Useful Life Prediction for a Nonlinear Degrading System in Service: Application to Bearing Data

Remaining useful life prediction for degrading systems plays a key role in prevalent prognostics and health management discipline. Compared with the offline remaining useful life prediction methods, online methods are usually more attractive in practice since they can be implemented dynamically adaptive to the real-time health conditions of the interested system in service. However, most of the existing studies for online remaining useful life prediction were limited to the linear degradation patterns. Aiming at this problem, this paper proposes a novel online remaining useful life prediction method under the framework of a generalized nonlinear degradation model with deterministic and stochastic parameters. Based on the historical degradation data of other similar systems from the same batch, the deterministic parameters and the hyperparameters in the prior distribution of the stochastic parameter are estimated through the maximum-likelihood estimation method, while the stochastic parameter in the degradation model can be dynamically updated by the Bayesian paradigm each time a new piece of degradation measurement of the interested system in service is observed. This makes the predicted remaining useful life dependent on the real-time health conditions of the interested system in service. The proposed online remaining useful life prediction method is then applied to a simulated example, and a practical case of bearings to demonstrate its effectiveness and superiority. Experimental results reveal that, comparing with two commonly used methods in the literature, our proposed method can improve the remaining useful life prediction accuracy dramatically.