The steady rise of electric vehicles and renewable energy systems has pushed batteries to the limit. With cars, drones, and even global grids relying ever more on rechargeable cells, battery management systems have emerged as the unsung heroes that keep them safe, efficient, and durable. But while systems today can tell you how much charge remains, they frequently don’t give you a more helpful response to a more realistic question: Can I finish my trip?
Researchers at the University of California, Riverside think they’ve developed a smarter solution. Led by engineering professors Mihri and Cengiz Ozkan, the team has developed a new diagnostic tool called the State of Mission, or SOM, that gives more than a percentage reading.
Instead of merely announcing charge levels, SOM predicts whether a battery can perform a specific task — such as powering a car up steep hills, flying a drone in a windy storm, or getting a house through a cloudy day.
“It’s a mission-aware metric that combines data and physics to predict whether the battery will have enough to complete a planned task in real-world conditions,” said Mihri Ozkan.
Graphic abstract of the state of mission (SOM), a mission-aware diagnostic metric that quantifies whether a battery can successfully complete a specific operational task. (CREDIT: iScience)
From Guesswork to Guidance
If you have an electric car, you’ve probably noticed that the “miles remaining” reading on your dashboard is not to be trusted. It will tell you that you have 40 percent left, but it won’t tell you if that will be enough to carry you over the mountain pass with the heater on. That uncertainty is what SOM is attempting to eliminate.
In contrast to traditional battery management systems, which react to what’s happening in the moment, SOM anticipates. It factors in battery data in relation to outside factors such as temperature, terrain, and even traffic. The result is an informed prediction that answers a straightforward but critical question: Can the battery safely and efficiently power this trip?
“SOM closes that gap,” Ozkan said. “It enables us to move from reactive surveillance to proactive, goal-based decision-making.”.
The innovation in SOM is how it combines two worlds that do not usually mix: physics-based modeling and artificial intelligence.
Comparison of traditional battery state functions (SOC, SOH, SOP, and SOE) with the SOM across five key performance metrics. (CREDIT: iScience)
Combining Physics and Machine Learning
Engineers have relied for decades on physics-based models like equivalent circuit models or electrochemical equations to explain battery behavior. They are science-based models that can explain why things happen, but they also require heavy computation and hard-to-measure parameters. This makes them too complex and slow for real-time use in electric vehicles or consumer electronics.
Data-driven models, on the other hand, can process huge amounts of data quickly. Machine learning systems can learn from thousands of charge and discharge cycles, spotting subtle patterns that humans might miss. But they come with their own problem — they don’t always explain why something happens. Without the laws of physics built in, they can make mistakes when conditions change or data gets noisy.
The Riverside team’s approach takes the best of both worlds. Their system uses advanced techniques called Neural Ordinary Differential Equations (Neural ODEs) and Physics-Informed Neural Networks (PINNs). These allow a computer to learn from data while still being bound by the underlying laws of electrochemistry and thermodynamics.
“By combining them, we get the best of both worlds: a model that adaptively learns from data but is grounded in physical reality at all times,” Cengiz Ozkan said. “That not only improves the accuracy of the predictions but also their reliability.”
Experimental validation of the Neural ODE framework using real-world battery cycling data (NASA dataset). (CREDIT: iScience)
Learning From Real-World Batteries
To verify their system, the researchers trained and tested their model on large public NASA and Oxford University datasets. These included real charge and discharge cycles, temperature changes, current and voltage profiles, and long-term degradation trends.
Their model reduced prediction errors compared to conventional methods — by 0.018 volts for voltage, 1.37 degrees Celsius for temperature, and 2.42 percent for state of charge. In other words, that is, it can more precisely and reliably forecast a battery’s performance in different conditions.
Instead of simply saying a car’s battery is at 60 percent, SOM can forecast how that energy will perform in practice. It might warn that you’ll need to recharge halfway through a trip or that flying a drone under windy conditions isn’t safe. It essentially turns raw battery data into real, actionable advice.
It transforms raw battery data into decision-making, increasing safety, reliability, and planning for automobiles, drones, and any use case where energy must be matched to a real-world mission,” Mihri Ozkan stated.
SOM estimation for a short-range urban EV scenario. (CREDIT: iScience)
A Smarter Future for Energy Systems
The system is continual and active in progress. Computational complexity is a limitation at present — the model needs greater processing power than most battery management systems can currently provide. But the researchers estimate that as hardware becomes faster and more efficient, it will be feasible to integrate SOM into everyday devices.
The system has already shown promise for electric vehicles, unmanned aerial systems, and grid storage. The researchers plan to apply it to different battery types in the future, including sodium-ion, solid-state, and flow batteries.
“Our approach is generalizable,” Cengiz Ozkan said. “The same hybrid method can supply mission-aware prognostics to improve reliability, safety and efficiency in a wide range of energy technologies — from cars and drones to home battery systems and even space missions.”
Practical Implications of the Research
This innovation could revolutionize battery management, making energy systems safer and more reliable. Not only would future electric vehicles tell you how much charge you have left, but they would also warn you if your planned trip is outside safe limits.
SOM estimation framework applied to a long-haul EV case study. (CREDIT: iScience)
Drones would be capable of autonomously adjusting flight plans based on wind and battery health, and grid storage systems would be capable of optimizing energy balancing.
By merging machine learning and physics, SOM is taking energy technology into a new frontier of predictive management — one that foresees problems before they happen and helps people make smarter, safer choices about how they use power.
Research findings are available online in the journal iScience01854-1).
Related Stories
Like these kind of feel good stories? Get The Brighter Side of News’ newsletter.