A battery chemistry-adaptive fuel gauge using probabilistic data association

This paper considers the problem of state of charge (SOC) tracking in Li-ion batteries when the battery chemistry is unknown. It is desirable for a battery fuel gauge (BFG) to be able to perform without any offline characterization or calibration on sample batteries. All the existing approaches for battery fuel gauging require at least one set of parameters, a set of open circuit voltage (OCV) parameters, that need to be estimated offline. Further, a BFG with parameters from offline characterization will be accurate only for a "known" battery chemistry. A more desirable BFG is one that is accurate for "any" battery chemistry. In this paper, we show that by storing finite sets of OCV parameters of possible batteries, we can derive a generalized BFG using the probabilistic data association (PDA) algorithm. The PDA algorithm starts by assigning prior model probabilities (typically equal) for all the possible models in the library and recursively updates those probabilities based on the voltage and current measurements. In the event of an unknown battery to be gauged, the PDA algorithm selects the most similar OCV model to the battery from the library. We also demonstrate a strategy to select the minimum sets of OCV parameters representing a large number of Li-ion batteries. The proposed approaches are demonstrated using data from portable Li-ion batteries. (C) 2014 Elsevier B.V. All rights reserved.