## Real Time Impedance identification of Li-Polymer Battery with Kalman Filter

##### Master thesis

##### Permanent lenke

http://hdl.handle.net/11250/2488399##### Utgivelsesdato

2017##### Metadata

Vis full innførsel##### Samlinger

##### Sammendrag

In a study of a new and an aged lithium-polymer battery cell, a joint Unscented Kalman Filter coupled with a non-linear equivalent circuit battery model formed the basis for on-line identification of the battery impedance characteristics. In order to evaluate estimated impedance parameters produced by the Kalman filter under real racing conditions, a case study using a dynamic drive-cycle for the Revolve NTNU electric race car were developed. Specifically, the cell model developed describes the non-linear relationship between the open-circuit voltage and the battery State-of-Charge (SoC), the ohmic resistance, the non-linear dependency of the charge-transfer resistance on the faracaic current, and a simplified RC-network to describe the diffusion phenomenon. To verify the consistency of the estimated parameters, an impedance characteristic map was constructed at various levels of SoC and isothermal conditions for the new and aged cells using an industrial grade electrochemical impedance spectroscopy device. Moreover, DC-resistance characterization was conducted to support the analysis. In addition, to avoid introducing errors to the system, the nominal capacity of the respective cells were characterized.
Given imposed sampling conditions and load-dynamics, the respective impact of the ohmic resistance and the non-linear charge-transfer dynamics on the overvoltages, was difficult to separate due to the fast electrode kinetics of the cells at temperatures above $25 ^{\circ}C$. As a result, the estimated impedance parameters describing the rapid cell dynamics were poorly conditioned on the voltage measurements. As a proof of concept, a simplification was introduced in order to verify the impact of a model-reduction. The results indicated a similar estimation performance with respect to the expected mean of the sum-of-resistances, but with a higher confidence in its inferred value. Thus, for the considered lithium-polymer battery cells operated at cell temperatures above $25 ^{\circ}C$, the concatenation of ohmic resistance and charge-transfer dynamics into a single series resistance is likely to be the best candidate to describe the battery State-of-Health (SoH) in terms of their impedance characteristics. The outcome of the simplification is reduced computational effort due to model reduction, and increased confidence in the inferred sum-of-resistance by allowing the series resistance to represent both passive and active electrochemical effects. However, the estimation error for the aged cell with respect to impedance characteristics at low SoC, still persists. This is probably related to limited diffusion, which is not captured with the given Kalman filter decision variables, in addition to assumptions of how impedance parameters evolve in time.
Moreover, secondary results showed that SoC can be estimated very accurately for the new cell, with a maximum estimation error of about 1\%. However, due to non-modelled phenomena such as electrochemical hysteresis and thermal dependency on the open-circuit voltage, the SoC estimates for the aged cell was biased in the range of 2-4\%. The ultimate purpose of identifying the impedance is to monitor on-line the SoH of the lithium-polymer batteries in a consistent manner over the expected battery lifetime. Emphasizing minimum computational effort and satisfactory reliability, method verification suggest that interrelated phenomena with significant influence on the equilibrium conditions and impedance characteristics at low SoC need to be further explored.