Tell me about your thesis … on the aging of electric vehicle batteries – Release

Our “Tell me about your thesis” series returns with a future-oriented study topic: how to make better use of the soon-to-be-installed lithium batteries (in any case, is the goal) most of our cars, buses and motorcycles? Inès Jorge has been working on the subject since 2019 at the National Institute of Applied Sciences (INSA) in Strasbourg, in a laboratory team. iCube. His thesis is part of the project VEHICLEfunded by the European Union as part of the Interreg Upper Rhine program and co-funded through the Offensive Science program of the Upper Rhine Metropolitan Region.

“The aim of my thesis is to study the aging of lithium batteries in electric vehicles. It is a fairly easy topic to popularize because it has specific applications. The new electric vehicles all use a lithium battery, because it is currently the most efficient technology, with the highest energy density (a large amount of energy can be stored in a small volume) and the most cost-effective. What I do is observe the operation of the batteries, to predict when the batteries will die and adapt the behavior of the driver to extend its life.

learn to program

“It doesn’t seem like it, but it’s really a lot of programming. I do data processing, algorithm development … It’s a computer science thesis, in fact. Initially, however, I was not a developer: I studied electrical engineering. I have technical knowledge about components, development of electronic circuits … Then, during my studies, I did a project where we had to develop a hybrid power source for a small car to transport equipment, with lithium batteries and supercapacitors. From there, I became interested in drums. Then I did my internship at a company that develops high-end speakers, and my mission was to see if we could develop a new product that would run on lithium batteries, instead of being connected to the mains. , keeping the same. power to the speakers. After these internships, I had a good training in batteries: I knew how they work, how they are loaded and unloaded, the advantages and disadvantages …

“At first I didn’t feel like doing a thesis, but the professors-researchers I worked with during my studies proposed a thesis on lithium batteries in electric vehicles with computer tools. I had programming knowledge, it was a daily tool for me. And it was good for me to dig a little deeper into the direction of the programming. It is quite easy to develop skills and autonomy in this area, especially with the Python programming language, which I use, because there are many online resources and a fairly open community. When you get stuck, you find help in the forums. I thought it might be a good challenge to start my career this way, because knowing how to program is a skill that is in high demand right now. I learned at work. I studied on my own, went to the bathroom and progressed every day.

Simulates battery life

“Specifically, I use artificial intelligence tools. This is very practical when you want to study a fairly complex system such as lithium batteries, because it is difficult to model its operation on a computer. Batteries are electrochemical systems, which store electrical energy in chemical form. We can try to make equations to try to understand the physical rules that govern them, but there are many factors that come into play: the outside temperature, the environment in which the batteries are used, the chemical reactions … In conventional battery simulations . , researchers are forced to simplify the problem by leaving aside certain factors in order to have reasonably sized models that can run on computers we use every day.

“Artificial intelligence, and more specifically machine learning, allows us to avoid this problem. It is no longer necessary to understand in depth all the mechanisms of a battery and put them into an equation. I have my computers with their computing power, and I show them a large volume of data recorded in the lab about the behavior of different batteries, with different types of use in different situations. The algorithm loops through this input data, over and over again, to learn information. Once the model is trained, it is able to predict the behavior of the battery from a sample of data that it had never seen before.

Correct driver behavior

“With my trained model, I look at a part of the aging curve of a battery, for example, the first charge and discharge cycles, and I try to predict the rest of the curve. I want to predict when there will be a failure or when Battery health will deteriorate.The ultimate goal is to try to correct the driver’s behavior in advance.These are common sense questions: when driving with sudden acceleration or braking, or when using the battery in an overheated environment or too cold, when you discharge it too much … This affects its operation … If I detect, for example, that this or that use deteriorates the battery more quickly and that it will age prematurely in 100 or 200 charge cycles, we can try give feedback to the driver to rectify their behavior and postpone the end of battery life.

Explain the reasoning of AI

“The important thing for me is to try to quantify how the parameters influence the battery life. Is it better to discharge it at a lower temperature, or to make fewer trips? It’s not easy to quantify, because the models of intel The artificial intelligence we use is black boxes, we pass data through the algorithm and tell us if the battery will age more or less quickly, but we are not sure how this result was determined. what exact parameters influenced the neural network in their conclusions.

“It simply came to our notice then. While I use these black boxes to say “with this usage pattern, the battery will reach the end of its life in 200 cycles”, other people on my research team are adding a layer of explanation. Uncheck the result so that it can be interpreted by a user.

Search for public data

“The hardest part of my job is collecting data on the operation of batteries to power artificial intelligence, because I’m doing a public domain thesis and not in the private lab of a car manufacturer. Manufacturers have a lot of data on the operation of their batteries in real conditions, recorded on the road by the on-board computer of electric cars. But they are confidential, because there are big economic bets linked to improving the performance of their batteries. This data is not publicly available to the research community. What we can use is data from major research labs that test battery aging.

“I work a lot, for example, with a dataset published by the Massachusetts Institute of Technology (MIT) in late 2018, just before I start my dissertation. Today it is the largest public data set in the field of battery aging. They aged 124 battery cells, these individual cells are a bit the size of a remote control battery, with charging and discharging protocols in a controlled environment at 30 ° C. They take new batteries and charge and discharge them continuously until the end of their useful life. There is also a data set published by Nasa but dating from the early 2010s. It was the first of its kind, so it is a benchmark today. And then there are other datasets published by Oxford University, for example, grouped in public download sites.

“We can’t produce this data ourselves because it’s expensive to generate in time and money. Battery cell aging can take months or even years, enough resources to support it. to this project.The result of the test benches of MIT, NASA or Oxford is a bit abstract compared to the actual conditions of use of a battery in an electric vehicle, but it allows me to get an idea.

Establish partnerships

“Since last year, we have been working with a French electric scooter manufacturer, Mob-ion, which is quite focused on developing vehicles with“ planned durability ”(as opposed to planned obsolescence). Their scooters are designed to be disassembled and repaired. The manufacturer is willing to share with us your battery performance data, so that in return we can move forward with them in improving the autonomy of their vehicles. Its aim is to reduce the cost of having scooters and give a second life to the batteries without having to disassemble them. They are in the process ofopen science to share knowledge.

“Other manufacturers, often start-ups or small businesses, contact us for help with research and development. I have the impression that there are few laboratories in France that study the aging of batteries with artificial intelligence tools. Sometimes we can establish a partnership with these companies. Other times, it’s more complicated, due to lack of human resources, because we’re a small team. Finally, we would like to be able to use real data. What I do in my thesis is to develop models that work on theoretical data, but it is a very small stone that will need to be validated and consolidated with real-life experience. In the field of public research, I believe that we will never be able to obtain satisfactory and representative results without access to real vehicles and their usage data.

“There is a lot of competition in the field of electric vehicles, which is very good today. I keep track of the means to keep up to date and keep up to date with the state of the art of work on predictive battery maintenance … And I find that quite disturbing. Every day I read new articles on the same research topic, with the same tools and approaches. It’s great that it arouses so much enthusiasm, but it would be nice if we could communicate better with each other and give each other advice on the data instead of working in parallel. We need to move forward because we are all moving in the same direction. In the long run, the goal remains that we manage to save on the resources extracted to make the batteries, and that we develop cleaner mobility.

Previous episodes

Leave a Comment