Remaining useful life prediction based on exponential dispersion process with random drifts

Remaining Useful Life (RUL) is one of the most important indicators to detect a component failure. RUL can be predicted by historical data by adopting a model-based method. The stochastic process models have become the most popular way to model degradation data for high-quality products, such as the Wiener process, gamma process and inverse Gaussian process. However, this leads to poor reliability assessment if the model is misspecified. Application of the Tweedie exponential dispersion (TED) process, including the above-mentioned classical stochastic processes as special cases, transforms the model selection problem into a parameter estimation problem dexterously. In this paper, we propose a TED process with random drifts for degradation data and a TED process with random drifts and covariates for accelerated degradation data. A hierarchical Bayesian method is adopted to estimate the parameters of the proposed models. We also derive the failure-time distribution and the remaining useful life distribution for the proposed models. The simulation study shows that the proposed model outperforms the wrongly specified models. Two illustrative examples demonstrate the performance of the proposed TED process with random drifts and the TED process with random drifts and covariates.