Estimation of battery aging using an interacting multiple model Kalman Filter

Abstract

Battery management systems are responsible for ensuring that lithium-ions cells are operated within safe limits that will preserve the charge capacity and power delivery of a cell, while still meeting the demands of the system it is connected to. Not only must a battery management system seek to prevent premature aging, but as aging processes unavoidably occur, they impact the operational boundaries the battery management system seeks to enforce. While there are methods for identifying how aging has reduced the capacity or increased the resistance of a cell, there is no well-defined method for non-destructively identifying which chemical processes contributed to cell aging. This thesis presents a method to estimate the state-of-health of a lithium-ion battery cell, as well as estimate the chemical mechanisms that contributed to cell aging. A method to estimate changes to critical cell parameters due to two significant cell aging processes is established, as well as a method to estimate parameter changes due to a blend of both aging processes. The aged cell parameters are used in a process that produces a reduced-order physics-based model in a state-space form. A selection of possible aging configurations are modeled, representing the state-of-health of a cell from its beginning-of-life parameters to a defined end-of-life condition, as caused by “pure” aging mechanisms and blended aging mechanisms. These are used as the system model within a set of nonlinear Kalman filters to produce an estimate of cell voltage and state-of-charge. The interacting multiple model Kalman filter method is utilized to blend the results of the Kalman filters, and produces a probability mass function that identifies the aging model that best fits the system measurements. The ability of the interacting multiple model Kalman filter to estimate aging and aging mechanisms is shown for a variety of input datasets and tuning configurations. A method for reducing the computational requirements of the interacting multiple model method is also presented.