Electric aircraft could change the way people travel over relatively short distances, from connecting communities to providing new options for urban transportation and emergency response. But making these aircraft practical requires batteries that can store a large amount of energy without becoming excessively heavy, while also performing robustly throughout a flight. Battery energy density and robustness therefore remain important challenges for electric aviation.
One complication is that an aircraft never operates under exactly the conditions assumed during design. The power required during takeoff may differ from one flight to another, temperatures change, and the duration of different phases of a mission can vary. At the same time, batteries coming off a manufacturing line are not perfectly identical: small variations in their material properties can affect how they perform.
This raises an important question: should we design the battery that performs best under one ideal set of conditions, or one that performs reliably across many possible conditions?
Our research at UC San Diego explores the second approach.
We use computer models that reproduce the electrochemical processes occurring inside lithium-ion batteries. Detailed models can provide valuable information about battery behavior, but they are computationally expensive. Studying uncertainty makes this problem even harder because the model may need to be evaluated thousands of times to determine how a battery performs across many possible operating conditions.
A common way to study this uncertainty is the Monte Carlo method. In simple terms, we repeatedly run the battery model while randomly varying uncertain quantities, such as operating conditions or material properties, and use the results to estimate the range of possible battery performance. The challenge is that doing this thousands of times with a detailed battery model can quickly become computationally expensive.
To make this more practical, we use a multi-fidelity strategy. Instead of relying only on the most detailed battery model, we combine it with faster, simplified models. The detailed simulations keep the predictions anchored to high-fidelity physics, while the less expensive models allow us to explore many more possible conditions. In one of our examples, this approach reduced the total number of model evaluations by about 68% compared with conventional Monte Carlo while targeting the same statistical accuracy.
We then use these predictions as part of an optimization process. Rather than simply searching for the battery with the highest possible energy density under nominal conditions, the algorithm also considers how much the battery's performance changes when operating conditions and material properties vary. That distinction matters. In our constant-discharge test case, the design optimized only for nominal conditions achieved the highest average performance but also showed the greatest variability. Our uncertainty-aware designs gave up a modest amount of average energy density in exchange for more consistent performance across uncertain conditions.
We also tested the method using a simplified electric vertical takeoff and landing, or eVTOL, mission consisting of takeoff/climb, cruise, and descent/landing. We allowed the required power, the duration of the flight phases, ambient temperature, and battery material properties to vary. Our optimized designs again reduced the spread in battery performance compared with designs created only for nominal conditions.
The broader idea is not limited to one particular battery or aircraft. As electric aviation develops, engineers will need to design systems that work not only in an ideal computer simulation, but across the uncertainties encountered in manufacturing and real operation.
By combining detailed physics with faster computational models, we can begin accounting for that variability during the design process itself. Rather than designing an electric-aircraft battery for one perfect flight, the goal is to design one for the range of flights it may actually experience.