Perovskites are a family of materials that are currently the leading contenders for potentially replacing today’s silicon-based solar photovoltaics. They hold the promise of very thin and lightweight panels that can be fabricated at room temperature instead of hundreds of degrees with ultra-high-throughput, which is cheaper and easier to transport and install. But getting these materials from controlled laboratory experiments into a product that can be manufactured competitively has been a long struggle.
Manufacturing perovskite-based solar cells involves optimizing at least a dozen or so variables at once, even within a particular manufacturing approach, among many possibilities. But a new system based on a new approach to machine learning could accelerate the development of customized production methods and help make next-generation solar power a reality.
The system, developed by researchers at MIT and Stanford University over the past few years, makes it possible to integrate data and information from prior experiments based on personal observations by workers experienced in the machine learning process. This makes the results more accurate and has already led to the creation of perovskite cells with an energy conversion efficiency of 18.5 percent, a level competitive for today’s market.
The research is reported in the journal jouleIn a paper by MIT professor of mechanical engineering Tonio Buonasci, Stanford professor of materials science and engineering Reinhold Dauskard, recent MIT research assistant Jay Liu, Stanford doctoral graduate Nicholas Rolston, and three others.
Perovskites are a group of layered crystalline compounds defined by the configuration of atoms in their crystal lattice. There are thousands of possible combinations and many different ways to make them. While spin-coating technology is used in most lab-scale development of perovskite materials, it is not practical for large-scale manufacturing, so companies and laboratories around the world are translating these laboratory materials into a practical, buildable product. Looking for ways to do it.
“When you’re trying to take a lab-scale process and then move it to something like a startup or a manufacturing line, there’s always a big challenge,” says Rolston, an assistant professor at Arizona State University. The team looked at a process they thought had the most significant potential, Rapid Spray Plasma Processing, or RSPP.
The manufacturing process would involve a moving roll-to-roll surface, or series of sheets, on which the precursor solution for the perovskite compound would be sprayed or ink-jetted in sheet form. Rolston says the material will then proceed to a curing stage, providing “a rapid and continuous output with greater throughput than any other photovoltaic technology.”
“The real success with this platform is that it will allow us to scale so that no other content has allowed us to do this,” he says. “Even materials like silicon require a much longer time frame due to processing. While you might think [this approach as more] Like spray painting. ,
Within that process, at least a dozen variables may affect the outcome, some more controlled than others. These include the composition of the starting material, temperature, humidity, speed of the processing path, the distance of the nozzle used to spray the material onto a substrate, and the curing method. Many of these factors can interact with each other, and humidity, for example, can become uncontrollable if the process is in the open air. It is impossible to evaluate all possible combinations of these variables through experimentation, so machine learning was needed to help guide the experimental process.
But while most machine-learning systems use raw data such as measurements of the electrical and other properties of test samples, they generally do not include human experiences such as qualitative observations made by experimenters of visual and test samples. Other properties of, or information from, various experiments reported by other researchers. Therefore, the team found a way to incorporate such extraneous information into a machine learning model, using a probability factor based on a mathematical technique called Bayesian optimization.
Using the system, he says, “having a model that comes from experimental data; we can detect trends we weren’t able to see before.” For example, they initially had trouble adjusting to uncontrolled variations in humidity in their ambient setting. But the model showed them “we can overcome our humidity challenges, for example, by changing the temperature and turning a few other knobs.”
The system now allows users to more quickly direct their process to optimize it for a given set or required results. In their experiments, the team focused on optimizing power output. Still, the system can also incorporate other criteria, such as cost and durability — together some team members continue to work, Buonasi says.
The researchers were encouraged by the Department of Energy, which sponsored the work, to commercialize the technology, and they are currently focusing on technology transfer to existing perovskite manufacturers. “We’re reaching out to companies now,” Buonasi says, and the code they developed is made freely available through an open-source server. “It’s on GitHub now, anyone can download it, anyone can play it,” he says. “We’re happy to help companies use our code.”
Already, several companies are gearing up to produce perovskite-based solar panels, even though they are still working out how to produce them, says Liu, who is now at Northwestern Polytechnical University in Xi’an, China. are in. He says companies out there aren’t building on a large scale yet, but are instead starting out with more modest, high-value applications such as building-integrated solar tiles where presence is essential. Three of these companies are “on track or being prompted by investors to build 1 meter by 2-meter rectangular modules”. [comparable to today’s most common solar panels]Within two years,” he says.
“There’s no consensus on which manufacturing technology to use,” says Liu. The RSPP methodology, developed at Stanford, “still has a good chance” to be competitive. And the machine learning system developed by the team could prove critical in guiding the optimization of whatever process is being used.
“The primary goal was to speed up the process so it requires less time, less experimentation, and fewer man hours to develop something that could be used for free, for industry,” they said. it is said.
“The current work on machine-learning-driven perovskite PV fabrication has largely focused on spin-coating, a lab-scale technique,” says University Professor Ted Sargent at the University of Toronto. He was not involved with this work, “demonstrating a workflow that is readily adapted to the deposition techniques that dominate the thin-film industry. Only a small number of groups have the engineering to drive such progress.” and there is an accompanying expertise in computation.” Sargent adds that this approach “could be an exciting advance for manufacturing a broad family of materials” that include LEDs, other PV technologies and graphene, “In short, any Also industry that uses some form of vapor or vacuum deposition.”
David L. written by chandler
Source: Massachusetts Institute of Technology