A Particle-Filter-Based Online Method for Degradation Analysis with Exponential-Dispersion Process

As a flexible stochastic process for degradation mod-eling and analysis, the exponential-dispersion process includes several widely used stochastic processes as its special cases, such as Wiener, gamma and inverse Gaussian processes. The Tweedie exponential-dispersion model is an important class of the exponential-dispersion model, which is flexible and can describe various types of degradation processes. Recently, several offline methods have been introduced for degradation analysis with the Tweedie exponential-dispersion model. However, when dealing with continually updating degradation observations, these offline methods may have the difficulty of storing and recalling historical observations everytime for analyzing newly observed data. Furthermore, to provide real-time results for in-situ decision making, a novel degradation analysis method equipped with a flexible degradation model and online inference method is promising. This paper proposes an online degradation analysis method based on the Tweedie exponential-dispersion model and a particle filter method. The Tweedie exponential-dispersion model is adopted to model the degradation process and the particle filter method is leveraged to develop the online inference method. The particle filter method can update the parameters' estimators in an iterative process by using each observed data only once in the process. The estimators would be updated after receiving new data without need to recall the historical data. The online method can meet the real-time requirement for in-situ degradation analysis. A simulation study is presented to show the effectiveness of the online inference. A case study with the GaAs lasers data is used to demonstrate the proposed method for degradation modeling and online inference.