The new data collection method at North Carolina State University (NC) is said to reduce the cost and environmental impact of material discovery. Details of the work are provided Natural Chemical Engineering.
The Autonomous Driving Laboratory is a robotics platform that combines machine learning, automation, and chemistry and materials science to discover materials more quickly.
The automated process allows machine learning algorithms to use the data from each experiment in predicting the next experiment to run to achieve the targets programmed into the system.
“Imagine whether scientists can discover groundbreaking materials for clean energy, new electronics, or sustainable chemicals in days rather than years. “This work brings us closer to that future.”
Until now, autonomous labs using continuous flow reactors have relied on steady-state flow experiments. In these experiments, various precursors are mixed and chemical reactions occur, but flows continuously in microchannels. The resulting product is characterized by a suite of sensors once the reaction is complete.
“This established approach to autonomous driving labs has had a dramatic impact on material discovery,” Abolhasani said. “This allows us to identify promising material candidates for a particular application in months or weeks rather than years or weeks, reducing both cost and environmental impact. However, there was still room for improvement.”
In steady-state flow experiments, the system is idle while the reaction is occurring, as the automated lab needs to wait for a chemical reaction to occur before characterizing the resulting material.
“We have now created an autonomous driving lab that utilizes dynamic flow experiments where chemical mixtures are constantly changing through the system and are monitored in real time,” Abolhasani said. In other words, I created a system that basically doesn't stop running, rather than running individual samples through the system and testing them one by one after reaching steady state. Samples are constantly moving through the system. The system does not stop characterizing samples, allowing you to capture data about what samples are being done every half second.
“For example, instead of creating one data point for what an experiment generates after a 10-second reaction time, there are 20 data points. One is one of the reaction times, after a 1-second reaction time, etc. Instead of waiting for each experiment to finish, the system is constantly learning.”
According to the team, collecting this much of the additional data will have a major impact on the performance of autonomous driving labs.
“The most important part of an autonomous driving lab is the machine learning algorithms used by systems. “This streaming data approach allows the machine learning brain of an autonomous driving lab to make smarter and faster decisions in just a few seconds.
In this work, researchers found an autonomous driving lab that used steady-state flow experiments over the same period and incorporated a dynamic flow system that generated at least 10 times more data than an autonomous driving lab that could identify the best material candidates in the first attempt after training.
