Advanced Closed Loop Algorithms for State of Charge and State of Health Estimation in LI-ION Batteries at Wide Operating Conditions

Abstract:

The development of energy storage technologies in the last years has increased the use of batteries in a high number of applications that require working autonomously or where the use of energy can be optimized. In this framework, the use of Li-ion batteries is growing due to their advantages in energy and power density. The battery state determination is one of the most important aspects in this kind of systems because this information can be used to perform a proper management using a specific strategy. The benefits are the optimization of the charge and discharge processes, an optimal sizing in terms of cost and durability, a functionality maximization using the available energy and a lifetime extension thanks to the operation in the most appropriate working range.

The State of Charge (SOC) and State of Health (SOH) are the essential parameters in the state determination, because they indicate the available energy and the capability of supplying energy and power, but they cannot be directly measured. The difficulty of achieving an accurate estimation in any working condition has made this issue an extended research topic which is still under study. The main objective of this thesis is the definition of a state estimation technique capable of overcoming the most important limitations found in the state of the art algorithms. Among these methods, the ones based on closed loop estimations present certain advantages over the most traditional techniques as the ability of correcting the estimation results and adapting their performance to changing situations.

This thesis presents an enhanced closed loop SOC and SOH estimator for lithium-ion batteries based on Extended Kalman Filtering capable of working in wide operating conditions. The proposed parameterization technique solves the main problems identified in the well known methods concerning the identification of the model parameters. In addition, the model can be updated to precisely represent the cell response to current excitation profiles of varied dynamics and considering the hysteresis effect in the whole SOC range and different temperatures.

Then, the model has been implemented in an Extended Kalman Filter to perform the SOC estimation. The main advantage of the proposed algorithm is that the calculations from the accurate voltage prediction and the coulomb counting are concurrently used in every moment as part of the filter, obtaining an accurate and stable estimation in any circumstances. Moreover, the proposed integration procedure is essential because it enables the model implementation without any simplification.

Furthermore, the state estimation algorithm has been complemented to enable the SOH determination, based on the model parameters updating to track the changes in the cell behaviour during its lifetime. The joint SOC and SOH estimation has been achieved using a Multi Scale Extended Kalman Filter including some improvements over the existing approaches. The most important development is that each slow varying parameter is corrected according to a unique error indicator. Because of this, a new structure is defined where each parameter is estimated with separate filters which can employ different time scales. The calculation of these error measurements has also been enhanced, extending the concept of the state projection to obtain an output conformation, using the information of all the available measurements between each correction step of the filters in the slowest time scales. All these modifications contribute to reduce the influence of punctual errors in the correction actions.

Finally, the model and the state estimator have been experimentally validated with 40 Ah NMC Kokam Li-ion cells. Several experiments over the whole SOC range at various temperatures with different current profiles of varied dynamics have been carried out, including current pulses, FUDS driving cycles or a profile for an application of residential lift with energy storage. The tests have also been performed at different ageing conditions, demonstrating that the SOC estimation accuracy is maintained and the parameters which determine the SOH as the capacity or the internal resistance are concurrently updated. The SOC estimation error has been found lower than 2-3% in all cases while the error in the capacity determination is less than 1%, and 10% for the internal resistance.