Scientists are trying to find new semiconductor materials that can promote the efficiency of solar cells and other electronics. But the speed of innovation is a bottleneck at the speed on which researchers can manually measure important physical properties.
A fully autonomous robot system developed by MIT researchers can speed up things.
Their system uses a robotic probe to measure an important electrical property known as photoconduction, how electrically responsible for the presence of light.
Researchers injected material-science-domain knowledge from human experts in machine-learning models that guide the decision of robot. This enables robots to identify the best places to contact a material with probes to get most information about its photoconduction, while a particular plan finds the fastest way to move between the process contact points.
During the 24-hour test, a fully autonomous robot probe took more than 125 unique measures per hour, with more accurate and reliability than other artificial intelligence-based methods.
Dramatically by increasing the speed on which scientists can characterize the important properties of new semiconductor materials, this method can increase the growth of solar panels that produce more electricity.
“I find this letter incredibly exciting because it provides a passage for autonomous, contact-based characteristics methods. Each important property of a material cannot be measured in a contactless manner. If you need to contact your sample, you want it to be fast and you want to maximize the amount of that information that you receive,” Mechanical Engineering’s Professor, Professor, Proster, Proster, Prosterous Proposal.
His co-writers prominent writer Alexander (Alex) Semen, a graduate student; Postdox Basita Das and Kangu ji; And graduate student Fong Sheng. Work appears in today Science progress,
contact
Since 2018, researchers in Buonassisi’s laboratory are fully working towards the Discovery Laboratory. He has recently focused on the discovery of new perovesites, which is a square of semiconductor materials used in solar panels such as photovoltaic.
In pre -work, he developed techniques to rapidly synthesize and print unique combinations of perovicite materials. He also designed imaging-based methods to determine some important physical properties.
But photoconduction is characterized by the most accurately keeping an probe on the content, shining a light and measuring the electrical reaction.
“To allow our experimental laboratory to operate as quickly as possible and accurately, we have to come up with a solution that will produce the best measurement while reducing the time it takes to run the entire process,” is called Siemen.
To do this, an autonomous system requires the integration of machine learning, robotics and material science.
To begin with, the robot system uses its onboard camera so that to take the image of a slide with the perovsite content printed on it.
It then uses a computer vision to cut the image into the segment, which is fed in a nervous network model especially designed to include domain expertise from chemists and material scientists.
“These robots can improve the recurrence and accuracy of our operations, but it is still important to be a human in the loop. If we do not have a good way to apply knowledge -rich knowledge from these chemical experts to our robot, we are not able to search for new materials,” said Semenamene.
The model uses this domain knowledge to determine the optimal points for investigation to contact based on the size of the sample and its content structure. These contact points are fed to a path planner who finds the most efficient way to investigate all points.
The adaptability of this machine-learning approach is particularly important because printed samples have unique shapes, from spherical drops to structures.
“It is almost like measuring snowflakes – it is difficult to achieve two that are similar,” BoonCC is called.
Once the traveler finds the smallest route, it sends a signal to the motors of the robot, which manipulates the investigation and rapidly measures at each contact point in succession.
The key to the speed of this approach is the self-preserved nature of the nerve network model. The model determines the optimal contact point directly on a sample image – without the need for label training data.
Researchers also intensified the system by increasing the path plan process. They found that adding small amounts of noise in the algorithm, or adding randomism helps to find the smallest path.
“As we progress in this era of autonomous laboratories, you are really coming together to be able to get together in the same team – the understanding of these three expertise – hardware buildings, software, and material science.”
Rich data, fast results
Once they built the system from the ground, the researchers tested each component. Their results showed that the nerve network model found better contact points over time than seven other AI-based methods. In addition, the path plan algorithm found less path plans than other methods.
When they place all the pieces together for 24-hours a completely autonomous experiment, the robotic system organized more than 3,000 unique photoconduction measurements at a rate of more than 125 per hour.
In addition, the level of expansion provided by this accurate measurement approach enables researchers to identify the hotspots with high photoconductions as well as materials fall.
“Being capable of collecting such rich data, which can be captured at such fast rates without the need for human guidance, the new high performance begins to open doors to search and develop semiconductors, especially for stability applications such as solar panels,” semen says.
Researchers want to continue construction on this robotic system as they try to create a completely autonomous laboratory for material search.
This work is supported, in part, MIT Energy Initiative, Mathworks, Acceleration Consortium of the University of Toronto, the US Department of Energy and the US National Science Foundation first by solar, ENI.
