How robots and AI are helping develop better batteries

Around the beginning of the year, Carnegie Mellon researchers used robotic systems to run dozens more experiments to create electrolytes that would allow lithium-ion batteries charge faster. This was a major obstacle to widespread adoption of electric cars.

The Clio system, which included automated pumps, valves and instruments, mixed various chemicals together and then measured how it performed against critical battery benchmarks. These results were then fed into Dragonfly, a machine-learning program that proposes different chemical combinations that might be even more effective.

The system produced six electrolytes that outperformed the standard one when they were placed into small test cells by the Carnegie researchers, according to a paper published in Nature Communications. The best one showed a 13% improvement over the top-performing baseline battery cell.

Developing better electrolytes is essential for improving the performance, safety and cost of batteries. Because they reduce the inconvenience of long waits at charging stations, faster-charging batteries can make electric cars and trucks more attractive.

In recent years, research laboratories have combined automated systems with machine learning software. This software identifies data patterns and helps to improve at specific tasks. This allows them to create materials that are ideally suited for particular applications. Scientists have tapped into these methods to identify promising materials for solid state electrolytes, solar photovoltaic cells, and electrochemical catalysts. Several startups have emerged to commercialize the approach as well, including Chemify and Aionics.

Materials discovery has been a process that relied on intuition, informed speculation, trial-by-error, and trial-and-error. It can be time-consuming and difficult because of the many possible combinations and substances. Researchers may end up on many wrong paths.

In the case of electrolyte components, “you can mix them in billions of different ways,” says Venkat Viswanathan (associate professor at Carnegie Mellon), a coauthor of Nature Communications’ paper and cofounder and chief scientist of Aionics. Jay Whitacre, director at the university’s Wilton E. Scott Institute for Energy Innovation, was his co-principal investigator. They collaborated with other Carnegie researchers to investigate how robotics, machine learning, and machine learning can help.

The promise of a system such as Clio or Dragonfly is that it can quickly work through a wider range of possibilities than human researchers and then apply what it has learned in a systematic manner.

Dragonfly doesn’t have any information about chemistry and batteries so it isn’t biased in its suggestions. Viswanathan said that the researchers choose the first mixture. It then explores a variety of combinations, including mild refinements to the original and completely new suggestions. Finally, it identifies the best combination of ingredients that achieves its goal.

In the case of battery research, the Carnegie Mellon team sought an electrolyte to speed up the charging time. The electrolyte helps to shuttle ions (or atoms that have a net charge due the loss or gain an electron) between the electrodes of a battery. During discharge, lithiumions are formed at the anode (or negative electrode) and flow through the electrolyte solution to the cathode (or positive electrode), where they gain electrons. This process is reversed during charging.

One of the key metrics Clio measured and sought to optimize is “ionic conductivity,” or how readily ions flow through the solution, which directly affects how quickly a battery can recharge.

But commercial electrolytes must be able to perform well across a range of measures. This includes total life cycles, power output and safety. Improvements in one area may often lead to problems in the others.

In the future, the Carnegie Mellon researchers plan to improve the machine-learning tools and accelerate robotic experiments. They also hope to run experiments with multiple objectives rather than one performance goal. The grand hope is that machine learning and automation will allow for faster discovery of new breakthrough materials. This will help the world reduce its climate-related emissions.