Efficient online estimation and remaining useful life prediction based on the inverse Gaussian process

Fast and reliable remaining useful life (RUL) prediction plays a critical role in prognostic and health management of industrial assets. Due to advances in data-collecting techniques, RUL prediction based on the degradation data has attracted considerable attention during the past decade. In the literature, the majority of studies have focused on RUL prediction using the Wiener process as the underlying degradation model. On the other hand, when the degradation path is monotone, the inverse Gaussian (IG) process has been shown as a popular alternative to the Wiener process. Despite the importance of IG process in degradation modeling, however, there remains a paucity of studies on the RUL prediction based on the IG process. Therefore, the principal objective of this study is to provide a systematic analysis of the RUL prediction based on the IG process. We first propose a series of novel online estimation algorithms so that the model parameters can be efficiently updated whenever a new collection of degradation measurements is available. The distribution of RUL is then derived, which could also be recursively updated. In view of the possible heterogeneities among different systems, we further extend the proposed online algorithms to the IG random-effect model. Numerical studies and asymptotic analysis show that both the parameters and the RUL can be efficiently and credibly estimated by the proposed algorithms. At last, two real degradation datasets are used for illustration.