Particle Filtering for Prognostics of a Newly Designed Product With a New Parameters Initialization Strategy Based on Reliability Test Data

In particle filtering-based prognostic methods, state and observation equations are used in which one or more parameters are uncertain. These parameters are estimated with collected monitoring data. The choices of the initial value ranges and distributions of the unknown parameters in the state and observation equations influence the performance of the particle filtering approaches, in terms of convergence, speed, and stability of prognostic results. For new products with little or even no degradation process data, uniform distributions over experience-based value ranges are the most common choice for parameters initialization. In this paper, the failure times’ data collected during reliability tests executed before volume production are used for defining the initial value ranges and distributions of uncertain parameters. This is expected to increase the convergence speed of the parameters estimation with monitored data and to reduce the uncertainty of the predicted remaining useful life. Numerical experiments on synthetic degradation processes of PEM fuel cells and lithium-ion batteries are considered. The convergence speed of the parameters estimation and the sensitivity of the proposed method to the duration and number of product samples in the reliability tests are analyzed. Comparisons with particle filtering methods with standard initialization are also carried out to verify the effectiveness of the proposed new strategy for parameters initialization.