The performance and lifetime of battery cells are determined not only by material properties but also to a large extent by how they are used. Even though all batteries inevitably age, their lifetime, safety, and efficiency can be improved by using a sophisticated Battery Management System (BMS). Daniel Jakobson, PhD student at Chalmers University of Technology within the SEC project “Machine Learning Assisted Ageing Prediction and Adaptive Modelling for BMS”, is investigating how physics-based models and machine learning can optimise battery usage and prolong battery life-time.
The models used in today’s BMSs are relatively simple and must therefore act through more or less fixed constraints when it comes to current, voltage, and temperature. The margins for these constraints can be too conservative at times while not being constraining enough at other times. This normally results in an underutilization of the cells together with potentially prematurely ageing batteries.
“Simply put, you could say that all batteries are kept within the same boundaries like a one-sized box. However, most batteries have variations and would operate better with a different or varying set of boundaries. By using physics-based models called pseudo two dimensional (P2D), instead of the electric circuit models we have now, the hope is to make better matching ‘boxes’ leading to a more efficient use of batteries”, says Daniel Jakobsson.
Capturing the internal electrochemical processes
Today’s BMSs are built around electric circuit models, which are easy to implement, but these models do not capture the internal electrochemical processes within the battery. They are also less accurate under extreme conditions. Physics-based models on the other hand simulate the actual physical and chemical processes inside a battery like lithium-ion transport, electrochemical reactions, heat generation, and degradation phenomena.
“A battery can deteriorate in many different ways, and with a better understanding of what is happening in the battery you can set-up more efficient usage boundaries. The challenge with physics-based models is keeping them up-to-date with the changed parameters due to the ageing process. If you don’t update the model, and use it as if the battery was new, the error margins will not be accurate. You could compare it with updating the city roadmap of an expanding city. If your map is outdated you can still drive, but the journey will be less efficient and perhaps also less safe”, Daniel explains.
Machine learning helps solving complex ageing behaviours
“The problem we are trying to solve is highly complex and difficult to address with traditional methods. Machine learning makes it possible to estimate nonlinear behaviours very fast with few datapoints. This gives us an opportunity to handle the complexity of ageing behaviors effects like SEI growth, lithium-plating, cracking and loss of active material. If we know that an SEI-layer has formed on the anode you will not be able to transport ions as efficiently in the electrolyte. This limitation is caused by a more complex pathway for ions to travel and limit the output current. To use another traffic analogy: if the road has developed some bumps, you need to slow down the traffic to avoid congestion or crashes”, says Daniel.
Promising results
“Our initial results are very promising and show that this is a plausible method. We are still working on various things, like optimising input, reducing parameter space, post processing and investigating accuracy in critical model variables we want to control. We plan to publish the first results during the end of 2025 or the beginning of 2026.”
About Daniel Jakobsson
“I’m born and raised in Gothenburg and moved to Luleå to study the electro and physics engineeering program. Initially, I had no plan of any further studies, at least not engineering, since I took the economy program in upper secondary school. But then I felt like moving from home and someone told me that engineering physics was too advanced for, me so I just had to try it. It went well and after graduation I spent a couple of years in the industry, and then I decided to take on doctoral studies. What I like about this project is that it is a good combination of advanced physics and industrial application. In my spare time I enjoy doing sport activities like sea kayaking and snowboarding.”