An innovative AI approach to battery research
Table of Contents
Engineers at the University of Michigan have developed a new artificial intelligence (AI) tool that can predict the life of lithium-ion (Li-ion) batteries significantly faster than traditional methods. This AI tool, using data from previous research, is able to predict the durability of new battery concepts after only about 50 charge-discharge cycles, instead of thousands as was the norm. This allows researchers to save months, even years, of research work, as well as reduce the power consumption needed to test prototypes.
The developers estimate that predicting the lifetime of new battery designs can save up to 95% of electricity and 98% of time compared to current testing methods.
AI research methods and benefits
“By studying historical battery designs, we use the principles of physics to establish a generalized relationship between early tests and lifetime. Thus, we can minimize experimental effort and achieve high-precision predictions for new battery designs,” explains Jiu Song, a scientist in the Department of Electrical and Computer Engineering at the University of Michigan and lead author of the study.
The research was supported by Farasis Energy US, a California battery manufacturing company. The company also provided battery samples and design/test data to evaluate the model’s effectiveness. Interestingly, the model was trained only on freely available, public data.
The model is based on a pedagogical approach known as ‘learning through discovery’ or ‘learning by doing’. This approach assumes that the model receives tasks and resources to independently search for solutions based on its experience and acquired knowledge. After successfully solving many tasks, the model is able to solve similar tasks even without additional resources.
“Discovery learning is a general machine learning technique that could be applied to other fields of science and engineering,” emphasizes Jiawei Zhang, the first author of the study.
How the AI system works
The MI system predicts battery life based on its design and cyclic operating conditions such as temperature and amperage. The system selects battery variants to fill its knowledge gaps and tests them for approximately 50 cycles. The results of the experiments are fed to an “interpreter” that accesses historical data and performs calculations using a physical battery simulator. Based on these data and calculations, the system predicts the lifetime of the experimental batteries. A further learning model combines the new information with previous predictions to estimate the lifetime of a new battery design. Even in the process of conducting experiments, this system significantly saves time and energy, and there is potential for further improvement.
Although the next generation of Li-ion batteries differ from previous ones in chemistry, structure and materials, the researchers say there are parallels between them that can help predict the performance of the latest designs. The “Interpreter” does not use simple statistical characteristics of the current and voltage signals, but the physical properties underlying all designs to highlight commonalities between different batteries. Using this information, MI analyzes the batteries in two aspects: their internal properties (physical and chemical properties) and operating conditions. For example, at high temperatures, certain chemical processes can cause battery cells to degrade, but at lower temperatures, this mechanism is less pronounced.
The researchers tested their AI model using data and battery samples from Farasis Energy. After training on a data set containing only cylindrical cells similar to AA batteries, the model successfully predicted the performance of larger cells. While full-scale tests can last up to thousands of cycles and take months or even years, 50-cycle tests can take only a few days or weeks. This resulted in 95% electricity savings due to the need for fewer batteries and fewer cycles.
Future prospects and past achievements
The researchers plan to use this same approach to predict other aspects of performance, such as safety and charging speed. Previously, South Korean scientists discovered a new degradation mechanism for lithium-ion batteries, as well as European researchers developed an innovative technology for monitoring the condition of batteries that could improve their safety and service life.
The results of the study were published in the journal Nature.
date:2026-02-09 16:27:00
Keep reading