NSF funds U of Kansas researcher developing machine learning technology to monitor and prevent thermal runaway in Li-ion batteries
Supported by a new five-year, $500,000-grant from the National Science Foundation, a researcher from the University of Kansas is developing machine learning technology to monitor and prevent overheating in lithium-ion batteries.
The research will develop a foundational framework for characterizing and monitoring LiB packs’ spatially and temporally distributed thermal behavior, building on a multi-disciplinary synthesis of ideas from first-principles modeling, machine learning, distributed estimation, and network systems.
The work is intended to drive new knowledge advancement in:
A hybrid modeling methodology that integrates first-principles-based and data-driven machine learning models;
Optimal estimation and machine learning theory based on hybrid models; and
Hybrid-model-based principles, algorithms and tools for temperature field reconstruction and thermal runaway detection.
The models and algorithms will be rigorously evaluated through a mix of theoretical analysis, software-based simulation, and experimental validation using a fully instrumented PEC SBT4050 battery tester.
Nowadays these lithium-ion batteries are everyplace in our society. They’re popular because they have many advantages, including high energy and power density, long cycle life, high voltage compared to other batteries and a low self-discharge rate. During the past decade, lithium batteries have become the most popular batteries for energy storage. But they’re vulnerable to thermal events. They can easily catch fire or have thermal explosions when ambient temperatures are high or when some internal failures occur. That’s because the lithium metal is highly reactive, and the commonly used electrolyte is flammable.—Huazhen Fang
Presently, most technologies to track the temperature of lithium-ion batteries are inadequate because sensors only can read the outside surface temperature of the batteries, according to the KU researcher.
Usually, the temperature on the surface is insufficient to tell us about the state of the cell. The internal temperature would tell us more about thermodynamics. But today there are few methods to place sensors inside a battery. However, using artificial intelligence and machine learning, we can predict the temperatures inside the cell that would give us the leverage to detect its behavior. The temperature at the surface would provide abundant data to be fed to a machine-learning approach and combined with mathematical models to predict what’s going on inside the cell.—Huazhen Fang
Rather than assuming a uniform temperature throughout a battery, as is the case with a present-day modeling approach called “lumped parameter models,” Fang said his computer-learning technique could predict variations in internal temperatures inside a battery—a more accurate and realistic means to calculate a battery’s potential to undergo thermal runaway.
When charging or discharging, the temperature distribution is uneven—usually higher inside near the electrodes—but the temperature outside on the surface is lower. Lumped models only consider even temperature distribution, but our method provides spatial-temporal reconstruction of the temperature.—Huazhen Fang
Fang said data from a lithium-ion battery fed into artificial intelligence to deduce internal temperatures could be processed in the device powered by the battery, or linked to cloud computing. If a battery undergoes thermal runaway, the device would be programmed to shut down or disconnect the battery before it becomes hot enough to catch fire or trigger an explosion.
With these innovations, lithium-ion batteries could be scaled up to more industrial levels via cells that bundle hundreds of batteries together. According to Fang, lithium-ion technology increasingly could be used in massive electrical grids to store and discharge electricity generated by sustainable technologies like solar and wind.
The problem is more pressing for large systems as they face higher vulnerabilities. In large systems, if one cell catches fire, then a domino effect will devastate the entire system. Nowadays, people across the industry are thinking of developing larger-scale energy storage based on lithium-ion systems. But the thermal-safety issue could slow the pace of the use of lithium-ion batteries for future grid energy systems. If successfully accomplished, our project can help address this challenge and widen the deployment of lithium-ion battery technology to make our society more sustainable.—Huazhen Fang
As part of the work under the new grant, several KU students will be trained in the battery modeling, data analysis and machine learning to develop the desired models and methods that will potentially push the thermal safety of batteries to new heights.