Online remaining useful life prognostics using an integrated particle filter

The lifetime evolution of mechanical equipment with complicated structure and the harsh operating environment cannot be accurately expressed due to the dynamics of the failure mechanism. However, the performance monitoring of equipment, with the information characterizing the failure process from the sensed data, can be used to assess the failure time and then the online remaining useful life. Because of the existence of nonlinearity and non-Gaussian for most real systems, for online assessment, unscented Kalman filter combined with particle filter is studied, instead of the standard particle filter with importance sampling, which is modified to update the states iteratively. Meanwhile, Markov chain Monte Carlo is performed after resampling to improve the prediction accuracy. In the modeling, state–space model is developed to quantify the relationship between the information from online observation and underlying degradation, and the unscented particle filter is investigated to realize the assessment of remaining useful life. In particular, the sufficient statistic method is presented to obtain a joint recursive estimation on both the system state and model parameters for those state–space model with unknown time-invariant ones. At the end of this article, the acoustic emission signals of a milling cutter are illustrated as a case study for cutter online remaining useful life estimate. The milling cutter example demonstrates the effectiveness of the proposed method for online estimate and provides useful insights regarding the necessity of online updating and the assessment.