Many electric technology applications rely heavily on the efficiency of their battery systems. However, in cases like electric cars, security stands above performance improvements which can be limiting. Luckily, machine learning and AI can help to create better batteries while also enhancing safety.But how?
Read in this week’s Helmholtz AI project showcase how researchers at the Jülich Institute of Energy and Climate Research (IEK-9) at Forschungszentrum Jülich (FZJ), the Steinbuch Centre for Computing (SCC)at Karlsruhe Institute of Technology (KIT) and the Theory Department at the Fritz Haber Institute of the Max Planck Society in Berlin are working jointly on this challenge.
Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?
My name is Josef Granwehr, I work at the Jülich Institute of Energy and Climate Research (IEK-9) at FZJ. I currently coordinate the i2Batman (Intelligent, individual battery management using spectroscopy and machine learning) project, which aims to integrate advanced AI techniques in an intelligent battery management system (BMS).
For this project, my team from IEK-9 is collaborating with an interdisciplinary team of natural scientists, engineers and mathematicians from the Steinbuch Centre for Computing (SCC) at KIT and the Theory Department at the Fritz Haber Institute of the Max Planck Society in Berlin. Three PhD candidates, two of them with an engineering background and a mathematician, and two physics Master students, are working closely with chemists with specialization on theoretical, analytical and electro-chemistry.
The project follows a data-centered approach, which permits such a combination of expertise at three locations distributed across Germany.
In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?
Many key technologies developed to achieve the sustainable development goals are electric, and thus, usually rely on its battery’s efficiency. Fast charging capability, lifetime extension and safety are crucial elements for a broader application of innovations like electric vehicles, which could transform the mobility and transport ecosystem towards carbon neutrality.
In these battery systems, a crucial element is the battery management system (BMS) that controls the charging and discharging of a battery. The BMS typically consists of a large number of individual pieces or cells that are constantly checking on the battery’s performance and balancing its activity. Since the development time of a battery is shorter than its expected lifetime in a car, these BMS are tuned with the main objective of ensuring safe operation. Any improvement of performance that compromises safety is not suitable for its use.
At the i2Batman project, we aim to create an AI supported BMS that allows the integration of new enhancements on performance and longevity while ensuring an improvement in safety. Our focus is set at the individual cells that form the BMS. If we could predict the decrease in performance or failure of each cell, and combine this information with new operational data of the battery, we could create a novel safety checkpoint for the BMS. This improved BMS could open the door to further enhancements that otherwise would have been too risky to implement. .
Our AI supported BMS combines continuous intelligent data collection in the physical battery with a numerical "digital twin" of the battery, like a continuously updated simulation of its performance. This way we achieve a better and more accurate charging protocol on the level of individual battery cells. The model is parametrized robustly, and parameters are updated continuously in an automatized fashion during the use of the battery. To quantify the uncertainty of the digital model (how well it is simulating the real life scenario) we use Gaussian process regression techniques.
Concept of the AI supported BMS. A physical battery is combined with a predictive digital twin model, which are linked by an AI. Data is continuously collected over the battery lifetime, which facilitates a dynamic adaptation of AI and battery model based on the actual state of the battery.
Another issue that can arise in some energy applications is that data acquisition, AI and modeling are considered jointly, as data and information flow in both directions between all three elements. There are many powerful individual tools available for these tasks, but there is no infrastructure in place to control all parameters together (an automatized, data-driven processing, instrument control and digital twin parametrization). Furthermore, most of these tools or involved steps require some form of user intervention.
A successful implementation of the proposed work could demonstrate a new powerful approach towards battery management able to overcome these limitations. However, there are, in addition to the challenges provided by the AI development itself, several non-trivial sub-steps necessary to be solved for a successful demonstration of the concept, like data acquisition, preprocessing, etc.
How important has the Helmholtz AI funding and platform been to carry out this project?
Helmholtz AI funding was central for two reasons. First, as a seed for this project, to bring the research teams from the different centers together. And secondly, as a unique program that supports high-risk basic research projects, without prior proof-of-concept which would have been unrealistic to develop unfunded in projects like ours.
Here, Helmholtz AI funding has been central as a door opener – not only for the project, but also to gain high-level support in the different research centers.