Reviewed by Lexie CornerDec 6 2024
Scripps Research, the Department of Energy’s Lawrence Berkeley National Laboratory (Berkeley Lab), and several other partnering institutions have successfully applied a machine-learning technique to expedite the discovery of materials for film capacitors—key components in electrification and renewable energy technologies. The study was published in Nature Energy.
This approach was used to screen a database of nearly 50,000 chemical structures, leading to the identification of a compound with record-breaking performance.
Other collaborators in the research include the University of Wisconsin–Madison, the University of California–Berkeley, and the University of Southern Mississippi.
The study underscores the increasing demand for film capacitors capable of withstanding high-temperature, high-power applications, such as electric vehicles, electric aviation, power electronics, and aerospace. Film capacitors are also critical components in inverters that convert solar and wind energy into alternating-current power for use in the electric grid.
This study builds upon previous work by K. Barry Sharpless, Ph.D., W.M. Keck Professor of Chemistry at Scripps Research; Peng Wu, Ph.D., Professor of Molecular and Cellular Biology at Scripps Research; and Yi Liu, Ph.D., Berkeley Lab Facility Director for Organic and Macromolecular Synthesis. In their earlier research, they discovered a new type of polysulfate compound for use in polymer film capacitors.
These thin polysulfate membranes protected the capacitors from damage caused by harsh conditions, including high operating temperatures and intense electric fields.
“As a chemist, our previous findings presented an existential challenge: How could a powerful electromagnetic energy wave from physics be tamed by passing through a thin polysulfate film?” notes Sharpless, co-senior author of the study.
Now, our new collaboration has enabled a significant advancement in this project, which seeks much better capacitor shields that could lead to crucial energy savings in common electric power applications. In short, our AI analysis quickly identified some key variables in the polymer design details that were predicted to add big improvements in the shielding properties of these polysulfate membranes. As reported in our new Nature Energy study, these earliest machine learning predictors for improving the capacitors are dramatically born-out by experiment.
K. Barry Sharpless, W.M. Keck Professor, Chemistry, Scripps Research
“For cost-effective, reliable renewable energy technologies, we need better performing capacitor materials than what are available today. This breakthrough screening technique will help us find these ‘needle-in-a-haystack’ materials,” added Liu, a co-senior author of the study.
Film Capacitors Require Heat-Resistant Materials
Batteries often receive significant attention as a key component in renewable energy applications, but electrostatic film capacitors also play a crucial role. These devices consist of an insulating material placed between two conductive metal sheets.
While batteries rely on chemical reactions to store and release energy over extended periods, capacitors store and discharge energy much more rapidly through applied electric fields.
Film capacitors are essential for regulating power quality in various power systems. For instance, they help prevent ripple currents and smooth voltage fluctuations, ensuring stable, safe, and reliable operations.
Polymers, which are large molecules composed of repeating chemical units, are ideal for use as the insulating material in film capacitors due to their lightweight nature, flexibility, and durability under applied electric fields.
However, polymers have a restricted capacity to withstand the high temperatures encountered in many power system applications. Extreme heat can diminish their insulating properties and lead to degradation.
Narrowing Down 49,700 Polymers to Three
Traditionally, researchers have relied on trial-and-error methods to identify high-performance polymers, synthesizing a limited number of candidates and characterizing their properties individually.
To expedite this process, the research team developed and trained feedforward neural network machine-learning models. These models screened a library of nearly 50,000 polymers to identify those with an optimal combination of properties, including high-temperature and strong electric field tolerance, high energy storage density, and ease of synthesis. The models highlighted three particularly promising polymers.
The three polymers identified by the machine-learning models had already been discovered in a prior study by Sharpless, Liu, and their team. They synthesized these compounds using a highly efficient technique called click chemistry, which rapidly links molecular building blocks into high-quality products. Sharpless, a co-recipient of the 2022 Nobel Prize in Chemistry, was recognized for his contributions to the development of the click-chemistry concept.
At Berkeley Lab’s Molecular Foundry, researchers fabricated film capacitors from these polymers and evaluated their properties along with the resulting capacitors. The analysis showed strong electrical and thermal performance. Capacitors made from one of the polymers demonstrated a record-breaking combination of heat resistance, insulating properties, energy density, and efficiency (high-efficiency capacitors minimize energy loss during charging and discharging). Additional testing confirmed the capacitors' superior material quality, operational stability, and durability.
Making Even Better Models
The research team is exploring several avenues for follow-up studies. One potential direction involves designing machine-learning models that offer deeper insights into how polymer structures influence their performance. Another promising area is the development of generative AI models capable of designing high-performance polymers directly, eliminating the need to screen extensive libraries.
The study was funded by the Department of Energy’s Office of Science (BES).
Journal Reference:
Li, H., et al. (2024) Machine learning-accelerated discovery of heat-resistant polysulfates for electrostatic energy storage. Nature Energy. doi.org/10.1038/s41560-024-01670-z.