Researchers in California have developed a new diagnostic measure that can reportedly predict whether a battery can successfully perform a specific task. The proposed model could be used in electric vehicles, unmanned aerial systems and network storage applications.
Scientists of the University of California, Riversidehave developed a new diagnostic measure for electric vehicles (EVs) that determines whether they can complete an upcoming journey.
It’s called the State of Mission (SOM) and uses both battery data and environmental factors, such as traffic patterns, elevation changes or ambient temperature, to generate real-time, task-specific predictions. Additionally, the team has developed mathematical and computational frameworks to calculate the SOM.
“It is a mission-aware measure that combines data and physics to predict whether the battery can complete a planned task under real-world conditions,” co-author Mihri Ozkan said in a statement. “Our approach is designed to be generalizable. The same hybrid methodology can deliver mission-aware predictions that improve reliability, safety and efficiency across a wide range of energy technologies, from cars and drones to home battery systems and even space missions.”
To calculate the SOM, the new model uses three input classes related to the mission profile, environmental conditions and battery dynamics. It starts by processing historical time series data to estimate the initial internal state vector of the battery. Then, neural ordinary differential equations (neural ODEs) simulate the continuous time evolution of electrochemical, thermal, and degradation states. By using physics-informed neural networks (PINNs), the model adheres to the results based on physical laws. Ultimately, using sequential learning architectures provides a coherent, end-to-end battery health estimation system.
The new model produces three results: the first is a binary SOM, which indicates whether a battery can complete the mission. The next is a quantitative SOM, which indicates how easily and safely the battery can complete the mission. Finally, it also provides a probabilistic SOM, which represents the probability that the mission will be successful. The group used data from the Oxford battery degradation dataset and the NASA PCoE battery aging dataset to train the model. Part of the data was ultimately also used for testing.
Image: University of California, Riverside, iScience, CC BY 4.0
“The model learns from how batteries charge, discharge and heat up over time, but also respects the laws of electrochemistry and thermodynamics. This dual intelligence allows it to make reliable predictions even under stress, such as a sudden drop in temperature or a steep uphill climb,” says co-author Cengiz Ozkan. “By combining them we get the best of both worlds: a model that learns flexibly from data, but always remains rooted in physical reality. This makes the predictions not only more accurate, but also more reliable.”
Using a computational framework implemented in Python, the group simulated two case studies to investigate their SOM model. The first involved a passenger car, traveling a 23 km urban round-trip route, with ambient temperatures ranging from 18 to 32 C. The initial battery state of charge (SOC) was 58%, the initial state of health (SOH) was 87%, the resistance state (SOR) was approximately 12% and the mean cell temperature (SOT) was 26 C. The model determined that the mission is feasible, with a quantitative SOM score of 92.4%.
Image: University of California, Riverside, iScience, CC BY 4.0
The second mission involved an electric long-range cargo vehicle, covering a mixed route of 275 km of which 110 km in mountainous conditions, with an ambient temperature range of 26-42 C. The SOC in this case was 87%, the SOH was 78% and the SOT was 33.6 C. The model also found this mission to be feasible, with a quantitative SOM of 73.5%. “Across the evaluated dataset, the model achieves root-mean-square errors (RMSEs) of 0.018 V for voltage, 1.37 C for temperature and 2.42% for SOC, reflecting strong agreement with empirical data,” the team added.
“Right now the main limitation is the computational complexity,” says Mihri Ozkan. “The framework requires more processing power than current lightweight, embedded battery management systems typically provide.” However, she emphasized that she is optimistic and that the model could soon be applied to electric vehicles, unmanned aerial systems, network storage applications and other areas.
The new system was introduced in “State of the art: battery management with neural networks and electrochemical AI”, published in iScience.
This content is copyrighted and may not be reused. If you would like to collaborate with us and reuse some of our content, please contact: editors@pv-magazine.com.
Popular content
