Abstract: For the problem of remaining life (RL) prediction of the nonlinear degradation device, existing methods have not systematically studied the degradation modeling with measurement error and random effect, the priori parameter estimation, and the corresponding RL prediction method. A nonlinear Wiener degradation model is built considering measurement error and random effect. By using historical condition monitoring (CM) data of similar device, the expectation maximum algorithm is applied to obtain the estimates of the fixed coefficient and the priori distribution of the random coefficients in the degradation model. The state space model is used to describe the current CM state of the target device. The Kalman filter algorithm is applied to iteratively obtain the posterior distribution of the random coefficients and the current real degradation state. The full probability formula is used to deduce the probability density function of the RL considering the estimation uncertainty of the implicit state. The simulation example analysis shows that this method has advantages over the existing methods in parameter estimation error and RL prediction accuracy.

Key words: remaining life (RL) prediction, nonlinear degradation model, measurement error, random effect