An Online Remaining Useful Life Prediction Method With Adaptive Degradation Model Calibration

At present, there are extensive studies on remaining useful life (RUL) prediction based on degradation modeling of sensor data. However, most existing degradation models have fixed functional forms and only update the parameters for calibration. In practice, due to the influence of individual variability and dynamic environmental conditions, simply updating the model parameters may render a mismatch between degradation models with the degradation process of in-service equipment and, thus, results in bias or even errors in the predicted RUL. In this article, we propose an online RUL prediction method with adaptive model calibration for stochastic degrading equipment. The initial degradation model constructed from the historical data has been used to predict the future degradation trend, and a threshold-based triggering mechanism is then designed to determine the calibration moment for function form. A parametric model for the degradation prediction errors is established to realize calibration of the function form of the degradation model. Furthermore, the model parameters are updated online by a Bayesian method based on the degradation data of in-service equipment for the model parameters’ calibration. As such, the proposed method allows us to achieve joint adaptive calibration of both the functional form and parameters of the degradation model. Based on the calibrated model, the probability density function (pdf) of the RUL is derived in the sense of the first hitting time (FHT) to realize RUL prediction. The effectiveness and superiority of the proposed method are validated by both numerical simulations and a case study of lithium batteries.