Abstract: Remaining useful life (RUL) prediction is the key technique to implement the health management of stochastic degrading equipment. As one of the typical methods for RUL prediction, statistical data-driven methods generally adopt the stochastic model to characterize the evolving progression of the equipment’s performance degradation variable and provide the probabilistic distribution of the RUL facilitating the uncertainty quantification of the RUL prediction. In existing studies with such methods for degradation modeling and RUL prediction, the fixed functional form of the degradation model is adopted and the model parameters are estimated or updated by the degradation monitoring data of the concerned equipment to perform the model calibration. However, selecting the functional form of the degradation model is itself a challenging problem. More importantly, when the selected functional form of the degradation model is inappropriate, it is difficult and ineffective to calibrate the degradation model simply by updating the model parameters, and the prediction accuracy will be thus affected. In this paper, a RUL prediction method for equipment is developed based on dynamic calibration of degradation model. First, the stochastic degradation model is constructed based on the nonlinear diffusion process and the model parameters are estimated through the degradation monitoring data of the equipment to predict the future degradation trend. Then, the parametric model for the degradation prediction errors is established to compensate for the degradation model to calibrate the functional form of the degradation model. At the meanwhile, the parameters of the calibrated model are updated based on the method to achieve the simultaneous calibration of the functional form and parameters of the degradation model. With the calibrated model, the RUL distribution of the equipment is derived for the RUL prediction. Finally, the developed method is validated by the numerical simulations and lithium battery data.

Key words: Remaining useful life, Degradation process model, Dynamic calibration, Bayesian method