Lithium ion batteries for electric vehicles: modeling, control, and design optimization for first and second life applications

Electric vehicles are nowadays an essential technology in fighting climate change and reducing the greenhouse gas emissions produced by road transportation. The growing awareness of the climate crisis and the consequent effort by international institutions have led, in recent years, to significant advancements in electric vehicle technology. As a result, electric vehicles can nowadays deliver performance comparable to that of fossil fuel vehicles, thus enabling a larger adoption of electric vehicles worldwide. The current development stage of electric vehicles mainly relies on the development of lithium-ion batteries (LIBs) and their use as energy storage systems for mobility applications. In fact, such technology allows for higher energy and power densities, lower self-discharge, and longer cycle life than other electrochemical energy storage technologies. Thanks to these advantages, electric vehicles have been able to leave behind the ineffective storage systems of the late nineteenth century, step out of the garages of the high bourgeoisie, and return to modern roads as a mass‑market phenomenon. Nonetheless, a wider adoption of electric vehicles is still currently hampered by three main challenges: the reduced driving range, the longer charging time, and the higher cost compared to fossil fuel vehicles. All these issues are caused by LIBs inherent limitations. In fact, the driving range is bounded by LIBs energy density, while longer charging times are required as LIBs cannot tolerate high charging currents without being damaged, and thus becoming subject to premature aging. Moreover, the higher cost of raw materials used in LIB manufacturing and the ethical and geopolitical issues related to their supply contribute to increasing electric vehicles final price. In addition, battery aging and the uncertainty in assessing the battery health grade further impair the residual economic value of the electric vehicle. As a result, potential buyers may feel discouraged from purchasing used electric vehicles, whose battery performance they may not be entirely confident about, as well as new ones, whose future resale value may decrease unpredictably due to battery degradation. While these issues could be resolved or mitigated by future improvements in LIBs manufacturing, recycling processes, and by the use of alternative raw materials, it is also necessary to develop techniques that enable more effective use of existing and future technology. In particular, it is essential to develop battery models that allow the accurate estimation of the battery internal state and, thus, the accurate evaluation of the residual driving range for a certain usage profile and the detrimental effects caused by the same profile on battery health. Currently, lithium-ion battery models are commonly classified in three types: the Equivalent Circuit Models (ECMs), Physics Based Models (PBMs), and Data-Driven ones. Data driven models achieve good accuracy, but the procedures used to train the model require a large amount of data. Unfortunately, such large datasets are not available at the moment due to the wide variety of available battery types and the secrecy surrounding industrial policies and manufacturing. ECMs describe the external behavior of the battery by means of an electric circuit. These models allow good accuracy and low computational burden that make them largely used in the Battery Management Systems (BMSs) of real world applications. Nevertheless, ECMs do not provide any direct insight on the electrochemical reactions that occur inside the battery and are therefore unable to provide an accurate estimation of battery state and of residual performance. Conversely, PBMs provide a detailed description of electrochemical processes that govern lithium-ion batteries. These models allow an accurate state estimation, but this comes at the cost of high computational burden and complex parameterization procedures that prevent the use of these models in real world applications. This Ph.D. thesis investigates the improvement of the accuracy of model estimates through the use of parameterization techniques that do not resort to complex laboratory procedures, thus enabling their effective use in real-world applications. In particular, the thesis describes two of the activities carried out during the doctoral program: the development of a tuning procedure for an ECM parameter estimation algorithm and the investigation of the dependence between ECM and PBM parameter variation. The first activity consists of the development of a tuning procedure for a Moving Window Least Squares algorithm for ECM parameter identification. The proposed procedure uses actual electric vehicle usage data as collected by its BMS and does not resort to laboratory tests or equipment. The procedure was validated on real data collected from a test vehicle over more than eight months of use, achieving accuracies comparable to those obtained by other literature methods that use, instead, laboratory tests carried out under controlled conditions. In addition, the accuracy obtained allows for the identification of early signs of battery aging. The proposed procedure appears therefore able to improve battery state estimation using only vehicle usage data and without imposing on the user the disadvantages of expensive calibration procedures. The second activity uses a neural network to estimate the variations in some parameters of a PBM starting from the variations in the parameters of an ECM. In particular, the method targets a subset of PBM parameters closely linked to cell aging mechanisms. Given the lack of experimental datasets required to train the proposed network, this activity used a simulation approach to generate a synthetic dataset from a cell PBM. The trained network proved capable of estimating with high accuracy the variations in lithium-ion diffusivity parameters, and with good accuracy the parameters related to the ion intercalation in the cell electrodes. The proposed approach leverages parameter estimation capabilities of existing BMSs to enable the use of electrochemical models in real world applications, thus significantly improving battery state of health assessment and enabling advanced prognostics. After investigating the improvement of battery models estimation capabilities, the thesis discusses the use of these models to optimize electric vehicle battery utilization and management. In particular, the third activity carried out during the Ph.D. consists in the creation of a simulation platform for assessing the impact of different usage and charging profiles on battery health. The platform is based on a simple electric vehicle traction model and a detailed ECM model of its battery. In particular, the battery model is equipped with electrical, thermal, and aging models of each individual cell in the electric vehicle battery pack. The platform is thus able to describe the electrical and thermal behavior of each individual cell and assess the degradation due to the specific usage profile. The platform is then used to compare different charging profiles and assess their long-term consequences on battery health. Finally, the last contribution of the thesis discusses the repurposing of exhausted electric vehicle battery packs in second life applications. In fact, automotive applications require significantly high performance levels from battery packs, with the result that packs considered no longer suitable for automotive are sent to landfills even though they still have usable capacity. The idea of second life is to use battery packs discarded from their first automotive life in less demanding applications, such as stationary storage. This application significantly increases the residual economic value of the battery but requires more complex battery control. In fact, cells of second-life battery packs show a large performance variability that significantly compromises the performance of the entire pack, unless proper control mechanisms such as the dynamic balancing are used. This section proposes a methodology for studying and comparing the different balancing architectures focusing on their use in a dynamic balancing system for second life applications. The performed analysis has made it possible to highlight the potential and critical issues of individual topologies, qualifying the proposed methodology as a valuable support in the design of second-life battery packs. In conclusion, this work thoroughly explores the main technical limitations that currently hinder the wider adoption of electric vehicles and proposes strategies to mitigate them, thus encouraging the adoption of more sustainable mobility and energy models.