Date of Award
Open Access Thesis
College of Engineering and Computing
Prognostics and Health Management (PHM) is the discipline involving diagnostics and prognostics of components or systems, with the primary objective of increasing the overall reliability and safety of these components or systems. PHM systems convert raw sensor data into features, and utilize state observers to estimate the current damage state online. Popular state observers are the traditional Kalman filter, along with its non-linear extensions, and the particle filter. Each technique has differing advantages. This thesis investigates the fusion of results from different techniques in order to achieve a more trustworthy probability of detection (PoD) during diagnosis and a more reliable remaining useful life (RUL) prediction in prognosis. Models for extended Kalman filter (EKF) and particle filter (PF) are developed from the feature data. The results from EKF and PF are then fused using an application of Dempster-Shafer theory (DST). Different models are utilized for EKF and PF in order introduce multi-model PHM, and to optimize the performance of each technique for both aging detection and RUL prediction. Prognostics is triggered when one-step-ahead predictions compared against the healthy battery demonstrate aging. DST is then applied to the prognostic results from EKF and PF. The result of DST is a density function whose performance can be compared with that of EKF and PF. DST allows for the fusion of multiple sensors and state estimates.
Weddington, J.(2017). An Application of Dempster-Shafer Fusion Theory to Lithium-ion Battery Prognostics and Health Management. (Master's thesis). Retrieved from http://scholarcommons.sc.edu/etd/4182