As the low carbon revolution gathers pace, the demand for new and improved battery technologies has become one of the fastest moving areas of technology development, and artificial intelligence (AI) is playing an increasingly important role in delivering the advances required.
Over the following pages we hear from some of those at the cutting edge of developments in this area: from the discovery and characterisation of new materials and chemistries, to the optimisation of manufacturing processes.
We learn how the UK’s Centre for Process Innovation is using AI tools to accelerate the scale up of new battery technologies; how Imperial college Spin-out Polaron has developed AI technology that can fine tune gigafactories; how AI software specialist Monolith is helping engineers analyse test data in ways that can cut weeks off development timelines, and how battery power systems expert Fortescue Zero is using the technology to give unprecedented insights into battery performance and longevity.
Meet the experts (clockwise from top left)
Dr Richard Ahlfeld - Founder and CEO of Monolith AI
Dr Katharina Roettger (she/her) - Principal Scientist, Digital Technologies, Centre for Process Innovation (CPI)
Ellie Coates - CEO Fortescue Zero
Dr Sam Cooper - Chief Scientist. Polaron, Associate Professor in Artificial Intelligence for Materials Design at Imperial College London.
Explain your involvement in the use of AI to drive developments in battery technology
Dr Katharina Roettger: At CPI, we use AI tools to accelerate development and scale-up of battery technologies to avoid experimental trials by re-using data where possible. We use AI for both R&D and manufacturing challenges.
For batteries and slurry development, we use a commercially available software for AI-driven adaptive Design-of-Experiments (DOE), Alchemite, to reduce the number of experiments by up to 80 per cent. This complements our automated formulation system, which allows for a wide formulation space to be investigated.
To optimise manufacturing processes, we developed soft sensors and predictive models for hard-to-measure properties, such as viscosity – a key property of battery slurries – based on simple sensor readings such as pressure. We also use physics-based modelling to generate simulated data to train AI models with as little experimental data as possible.
Additionally, we use AI to identify faults in batteries in real-time. Through techniques like anomaly detection, pattern recognition in combination with self-healing materials, AI contributes to safer and reliable battery systems.
Dr Sam Cooper: I lead a research group at Imperial College London who focus on the characterisation and design of advanced materials like battery electrodes. The performance of batteries (e.g. energy density and charging speed) is strongly influenced by the arrangement of microscopic particles in their electrodes. Microscope images of electrodes are hard to analyse, but are incredibly rich with information about manufacturing processes and performance.
AI will have a massive impact on almost every aspect of battery technology. However, in my opinion manufacturing is where the biggest impact will be seen because of how difficult it is to simulate with conventional physics-based models
Our first big success with AI was an algorithm that can learn to generate this 3D microstructure from a single 2D cross-sectional image of the electrode. This was published on the front cover of Nature Machine Intelligence and has found applications across materials science. Most recently we developed a model that can map from the manufacturing parameters used to make an electrode directly to the microstructure and the cell performance, which lead to us spinning-out a company, Polaron.
Dr Richard Ahlfeld: At Monolith, we’re tackling a real challenge facing the EV industry today. There’s mounting pressure to speed up product development, but this directly conflicts with how long it takes to test and validate batteries properly. It’s becoming even tougher as regulations get stricter – manufacturers need more thorough testing, yet they’re being pushed to deliver products faster than ever.
We’re creating tools that help engineers test more efficiently. Our AI technology analyses test data in ways that can cut weeks off development timelines, helping engineers quickly understand their results and move on to the next round of testing. We’re seeing engineering teams dramatically speed up their testing cycles while maintaining the rigorous standards needed for battery validation. In an industry where every week saved can make a crucial difference, this is transforming how our customers develop their products.
We’ve fundamentally changed how engineers learn from their test data, and AI has been key to making this happen. We’re tackling problems that have frustrated engineers for decades - like dealing with massive datasets - but we’re doing it in a completely new way with advanced machine learning.
We’ve built our platform based on years of working directly with engineers in test labs, understanding their real challenges and needs. This has led to us developing a powerful suite of machine learning tools that actually solve practical engineering problems. Right now, some of the world’s top automotive companies are using these tools daily to speed up their development while keeping their high standards. This goes beyond efficiency gains – we’re giving engineers capabilities they’ve never had before, and that’s game-changing.
Ellie Coates: At Fortescue Zero, we’ve developed a groundbreaking approach that combines traditional battery science with advanced artificial intelligence, giving us unprecedented insights into battery performance and longevity.
Typical AI applications within battery monitoring often rely on purely data-driven techniques. Many advanced predictive models operate as ‘black boxes,’ relying on complex algorithms where the factors driving predictions often don’t have clear connections to the real-world. A battery lifetime prediction is useful but if we can’t understand what is driving degradation, our insights become very limited. The inherent flexibility of data driven methods also makes them susceptible to overfitting, and when they are faced with the often messy reality of real world data they can quickly fall over. Fortescue Zero’s battery intelligence technology, Elysia, combines a deep understanding of battery electrochemistry with probabilistic AI to push the limits of battery technology. We call this approach “bottom-up” relying on comprehensive first-principles knowledge of the cells and the system first, before leaning on AI to describe relationships.
What is the potential impact of your work on battery performance?
KR: Using our AI tools for R&D and manufacturing, we can minimise the number of trials and reduce waste to optimise battery formulations and processes and predict scale-up behaviour to achieve the best possible battery performance, even for completely new materials. This can have significant impact on cost of R&D for new materials and slurry optimisation.
The AI-enhanced signal processing and sensor data integration we’re currently developing as part of a Horizon Europe project will enable monitoring of degradation, fault prediction, and self-healing strategies, enhancing battery life and safety by preventing issues like thermal runaway.
AI will impact every aspect of batteries manufacturing from the selection and development of raw materials during R&D, scale-up, real-time control of the manufacturing process to real-time information about battery performance throughout a battery’s life
Our new Horizon Europa project, FULLMAP, integrates AI, Big Data, Autonomous Synthesis, and High-Throughput Testing to accelerate next-gen materials and batteries. By implementing advanced and privacy-preserving techniques across distributed datasets, parameters like State of Charge (SoC) and State of Health (SoH) will be optimised. This digital transformation in battery manufacturing will lead to safer, more reliable, and high-performance batteries.
SC: One of the major challenges in the battery industry is fine tuning all of the thousands of settings in your gigafactory in order to make high performance cells consistently. To make things worse, physics-based simulations of the various manufacturing processes really struggle to capture what is going on with sufficient accuracy to guide decision making, so trial-and-error is relied on heavily, which is very expensive and very slow!
Our data-driven AI model learns these relationships directly from data, which allows engineers to rapidly identify optimal manufacturing parameters. This process can be used to directly improve performance metrics like energy density, charging speed, and battery longevity, while significantly cutting development times. For example, the designs our model can explore in under a day would take 50 years to simulate with state-of the-art physics-based models.
RA: When engineers can better understand their test data, they can truly optimise performance for specific applications. Our platform enables this deeper understanding by helping teams analyse their results more effectively and build more accurate models of component behaviour.
We’ve seen this impact firsthand. One of our automotive clients used our platform to develop more precise models of their component temperatures. This improved accuracy allowed them to reduce their design safety margins and uncertainties in their power delivery calculations. With greater confidence in their thermal predictions, they could push their systems closer to their true performance limits while maintaining safety standards. This kind of insight is transformative – it’s not just about testing faster, it’s about unlocking better performance through deeper understanding of test data.
EC: Our combined approach gives us three major advantages over other methods: First, it’s transparent. Our technology can explain exactly why a battery might fail or degrade, not just predict that it will. Second, it’s adaptable - it works reliably across different use cases and real-world conditions, not just in controlled laboratory settings. Finally, it’s efficient, requiring less data to make accurate predictions than traditional AI methods.
How far away is your technology from having a commercial impact?
KR: We have tools across all technology readiness levels (TRLs). Adaptive DOE is routinely used for customer projects to develop new formulations and optimise processes. Materials developed using Alchemite were already successfully scaled up to production by customers. Our soft sensors have been tested in other sectors such as plastics recycling and food and will soon be transferred to the new batteries manufacturing line hosted by AMBIC.
Our hybrid modelling approach using CFD generated data has been tested in-house and is ready to be used externally to support customers in their scale-up and predictive modelling challenges.
We are currently building a federated learning testbed, which will be ready to be used by customers soon. Federated learning is a privacy-preserving technique to share knowledge without sharing data, to further enhance our AI toolset.
In the wider ecosystem, core AI-enhanced tools are already validated in pilot battery manufacturing projects across Europe and some commercial applications.
SC: Polaron is already in the hands of manufacturing engineers across the world. We would hope to see cells that have been designed using Polaron in production as early as 2026. It is an extremely exciting time in the battery industry and manufacturers are hungry for tools that can speed up their design process.
RA: Our technology is already delivering commercial impact today. We’re working with leading automotive manufacturers who are seeing significant benefits - from accelerated testing timelines to substantial cost savings in their development programmes. The impact is particularly powerful given the massive operational costs of test facilities; catching an issue weeks earlier or reducing testing time by even a few weeks can save hundreds of thousands of dollars.
Beyond automotive, we’re seeing opportunities across multiple engineering fields, from electric aviation to energy storage solutions.
EC: The Elysia Cloud product, which harnesses AI insights, increases lifetime by up to 30 per cent and is already deployed today on every single Fortescue Zero Battery, including the Liebherr T 264 truck. We also recently announced a strategic collaboration with JLR which sees Elysia Cloud deployed on all future JLR Electric Vehicles to improve battery longevity, safety and performance as well as faster charging, improved reliability and increased range.
How do you envisage AI driving the development of battery technology in the future?
KR: AI will impact every aspect of batteries manufacturing from the selection and development of raw materials during R&D, scale-up, real-time control of the manufacturing process to real-time information about battery performance throughout a battery’s life. This will make battery development and production smarter, more efficient, and adaptable to rapidly evolving market demands. In the future, it may be possible to model the entire manufacturing process using data-driven and hybrid modelling for real-time control of battery production. All steps during the manufacturing process have an effect on the final cell performance, and hence it is critical to keep control of material properties at all stages by making use of all available data.
Another area that will be increasingly important is the adoption of AI tools for digital passports for batteries parts to allow optimal use for second life applications and increased recyclability.
SC: I think AI will have a massive impact on almost every aspect of battery technology, from discovering new materials at the atomistic scale, through to optimising charging at the pack level. However, in my opinion manufacturing is where the biggest impact will be seen because of how difficult it is to simulate with conventional physics-based models. Imagine, inside the mixing, coating, and drying machines used to make electrodes, trillions of ceramic particles are interacting with solvents and polymers - cracking, binding, reacting, flowing, evaporating, shearing, dissolving… the level of complexity is simply intractable to physics-based models. But data-driven AI models can learn the relevant relationships without needing to know all of the underlying physics. This will allow engineers to rapidly optimise their processes and get their products to market faster. It’s a wonderful time to be working at the interface of these two fields!
RA: AI is simply essential to cracking the intractable physics of battery development, and delivering products at the required rate for businesses to stay competitive. Self-learning algorithms help engineers optimise the design of batteries, pinpoint the best materials, offer intricate analytics, and deliver cutting-edge simulation.
Of course, there are engineers who will be sceptical about AI and machine learning, and won’t want to trust its conclusions. But what’s crucial to remember is that the technology delivers robust data explainability, and clearly marks the inputs from which it reached its results. That means engineers can save time taken talking to colleagues about how they arrived at their conclusions – regularly by hunch.
30 years ago, early simulation tools were also subject to scepticism, but they’re now the accepted standard. To progress, we need to transform those standards – and the adoption of AI is an invaluable part of that process.
Will AI end or enhance a career in engineering?
<blockquote>‘Go and get me a number of feasible designs that will satisfy these constraints so that I can add my greater creativity.’ </blockquote>If...