Battery Intelligence: Diagnosis & Prognosis

Battery Intelligence: Diagnosis & Prognosis hneieditor September 19, 2022

Project Contact: Matthieu Dubarry

Sponsors: Office of Naval Research (APRISES); SAFT (France); Element Energy; ACCURE (Germany)

Current collaborators: University of Warwick, University of Oviedo, Naval Research Laboratory, Dalhousie University, Oxford University, London Imperial College, Free University of Brussels, University of Bayreuth

In the last two decades, batteries have become an important part of our lifestyle, from portable electronics and electrified vehicles to renewable energy storage onto the electric grid. The art of battery engineering follows a conventional “design, build, and test” paradigm to deliver products that meet the design goals and specifications for a specific use. To warrant performance and to ensure safety and reliability, BMS (Battery Management Systems) were developed and implemented. Unfortunately, engineers who design BMS usually have little understanding of the electrochemical processes involved in Li-ion batteries that operate with much more complicated chemistries than Ni-MH or lead-acid batteries and thus require much more sophisticated controls. As a result, black-box battery models (derived from empirical correlations) tend to be implemented in BMS for both state of charge (SOC) and state of health (SOH) tracking. Although empirical models work well within the initial use of a battery, they tend to fail when they are needed the most, when cells aged. To mitigate this problem, new strategies needed to be developed. HNEI strategy includes a new measurement technology to visualize and model the internal state of a battery in order to enable monitoring for the entire battery pack.

Battery diagnosis and prognosis is a difficult task. First, lithium-ion batteries are chemically and electrochemically much more complex than traditional batteries. They rely on intercalation processes, and as such, the active materials are often non-conductive oxides or phosphates that swell and contract with lithium intercalation and deintercalation. This leads to much more complex electrode architectures with a need for two percolation frameworks to each grain of active material: a conductive one to bring electrons, and an ionic one to bring lithium ions. Some polymeric binder is also necessary to maintain the mechanical integrity of the electrodes with cycling. Moreover, their high voltage necessitates the use of organic electrolytes that are not stable. As a result, there is a multitude of degradation mechanisms operating in parallel that can all, to some degree and in different combinations, affect the longevity of a battery. Second, battery degradation is path dependent. Different usages for a battery (current, temperature, SOC range, SOC window…, etc.) can exacerbate or inhibit some of the degradation mechanisms. As a result, every battery degradation is unique since no two batteries are used in exactly the same way. This implies that it is nearly impossible to predetermine how a deployed battery will age, and thus that any prognosis must be performed onboard to be effective. Lithium-ion batteries are known to degrade slowly at first before a seemingly random and undetectable change with a rapid acceleration of the degradation. This event is dependent on how the battery was used and its prediction is essential for any useful estimation of the state of a battery. Third, large battery packs are comprised of thousands of cells and BMS have limited computing power. This precludes the practical use of complex models and the use of a multitude of sensors for each cell.

Traditionally, battery diagnosis is handled via two opposite approaches. The academic route aims for maximum accuracy and achieves it by inputting a lot of resources, with post-mortem characterization and extensive modeling. As a result, the analysis of a single battery is long, costly, often destructive and thus inadequate for deployment. The second route — the one usually used on deployed systems — is opposite. It uses as little resources as possible and must not be destructive. As a result, it is often restricted to an extrapolation of the evolution of capacity and resistance and is thus ineffective in predicting the sudden acceleration of capacity fading and thus true SOH. This assessment of state of the art led HNEI to define a third industry-compatible intermediate route to reach an accurate diagnosis with a cost-effective and non-destructive method, using only sensors already available in battery packs while requiring limiting computing power. Research conducted for this project is completed in the PakaLi Battery Laboratory.

Battery Diagnosis and Prognosis Project

Illustration of HNEI approach towards battery diagnosis.

HNEI alawa toolbox is available for licensing. Please visit https://www.soest.hawaii.edu/HNEI/alawa/ for more details.

Publications

Battery Testing:

Battery Modeling: