Researchers from Carnegie Mellon employed a robotic system to conduct dozens of experiments at the beginning of the year to create electrolytes that could speed up the charging of lithium-ion batteries, overcoming one of the main challenges to the mainstream usage of electric vehicles.
The Clio system of automated pumps, valves, and instruments mixed a variety of solvents, salts, and other chemicals together before evaluating the performance of the resulting solution against important battery standards. These outcomes were then fed into the Dragonfly machine learning system, which used the information to suggest various chemical combinations that would function even better.
According to a recent study in Nature Communications, the system ultimately produced six electrolyte solutions that performed better than a typical one when the Carnegie researchers put them into miniature test cells. The most effective one demonstrated a 13% improvement over the best baseline battery cell.
For batteries to perform better, be safer, and cost less, better electrolytes must be created. Faster batteries help reduce the aggravation of lengthy wait times at charging stations, which is crucial for making electric automobiles and trucks more appealing.
In order to create materials that are best suited for specific purposes, research labs have recently been increasingly combining automated systems with machine-learning software, which discovers data patterns to improve at defined jobs. These techniques have been used by researchers to find potential materials for solar photovoltaic cells, solid state electrolytes, and electrochemical catalysts. Chemify and Aionics are two startups that have emerged to commercialise the methodology.
In the past, experts in the field of materials discovery have developed and evaluated choices using a combination of intuition, reasoned speculation, and trial-and-error. However, the sheer number of compounds and possible combinations makes it a challenging and time-consuming process that might lead researchers in a number of incorrect directions.
According to Venkat Viswanathan, an associate professor at Carnegie Mellon, co-author of the Nature Communications research, and co-founder and chief scientist at Aionics, “you can mix and match them in billions of ways” when it comes to electrolyte components. To investigate how robotics and machine learning could assist, he worked with Jay Whitacre, head of the university’s Wilton E. Scott Institute for Energy Innovation and the project’s co-principal investigator, as well as other Carnegie academics.
Dragonfly doesn’t have any knowledge of chemistry or batteries, so its recommendations are unbiased only in the sense that the researchers chose the first mixture, according to Viswanathan. From then, it tries a wide range of combinations, ranging from slight modifications of the original to wholly novel ideas, before zeroing in on a combination of elements that consistently outperforms its predetermined objectives.
The Carnegie Mellon team was searching for an electrolyte to shorten the battery recharge time in the context of battery research. In a battery, the electrolyte solution facilitates the movement of ions—atoms with a net charge as a result of the loss or gain of an electron—between the two electrodes. Lithium ions are produced at the anode, the negative electrode, during discharge, and they move through the liquid toward the cathode, the positive electrode, where they pick up electrons. This happens in the opposite direction while charging.
Ionic conductivity, or how easily ions move through a solution, is one of the primary metrics Clio measured and attempted to enhance. Ionic conductivity has a direct impact on how rapidly a battery can recharge.
Commercial electrolytes face the additional problem of having to perform well across a range of parameters, such as total life cycles, power output, and safety, and advancements in one area can come at the expense of others.
The Carnegie Mellon researchers plan to speed up the robotic experiments in their upcoming work, improve the machine-learning tools, and conduct tests with several goals rather than just one performance target.
In the effort to reduce greenhouse gas emissions, it is hoped that automation and machine learning can advance the identification of the next generation of ground-breaking materials, resulting in better batteries, more effective photovoltaics, and other advancements.
