For hundreds of years, new materials were discovered through trial and error, or luck and chance. Now, scientists are using artificial intelligence to accelerate the process.
Recently, researchers at Northwestern University used AI to discover how to create new metal and glass hybrids 200 times faster than they would in the laboratory. Other scientists are creating databases of thousands of compounds so that algorithms can predict which ones combine to form interesting new materials. Others are still using AI to extract published articles on "recipes" to make these materials.
In the past, scientists and builders mixed materials to see what was being formed. This is how cement was discovered, for example. Over time, they learned the physical properties of various compounds, but much of the knowledge was still based on intuition. "If I asked why watered Japanese steel was better for making knives, I do not think anyone could have said it," says James Warren, director of the Materials Genome Initiative at the National Institute of Standards and Technology. "They simply had an artisan's understanding of the relationship between that internal structure and the genius."
Now, instead of using the artisan's knowledge, we can use databases and calculations to quickly trace exactly what makes a material much stronger or lighter, and that has the potential to revolutionize the industry after the industry, according to Warren. The time elapsed between the discovery of a material and its integration into a product such as a battery can be more than 20 years, he adds, and accelerating the process will lead to better batteries and glass for cell phones, better rocket alloys and better sensors. for health devices. "Anything made of matter," Warren says, "we can improve."
. To understand how new materials are made, it is useful to think of a materials scientist as a cook, according to Warren. Let's say you have eggs, and you're in the mood for something hard and firm. Those are the properties of the dish you want, but how do you get there? To create a structure where both white and yolk are solid, you need a recipe that includes step-by-step instructions for process the egg – harden it – only the way you want it. Materials science uses these same concepts: if a scientist wants certain material properties (for example, light and difficult to fracture), he will look for the physical and chemical structures that would create these properties and the processes, such as melting or beating metal, that would create these structures.
Databases and calculations can help find answers. "We perform calculations of materials at a mechanical level, calculations sophisticated enough to predict the properties of a possible new material in a computer before they are manufactured in a laboratory," says Chris Wolverton, materials scientist at Northwestern. University that manages the database of open materials of Quantum. (Other important databases include the Materials Project and the Materials Cloud.) The databases are not complete, but they are growing, and they are already giving us exciting discoveries.
Nicola Marzari, researcher at the École Polytechnique Fédérale de Lausanne in Switzerland, used databases to find three-dimensional materials that can be peeled off to create single-layer 2D materials. An example of this is highly publicized graphene, consisting of a single sheet of graphite, the material in a pencil. Like graphene, these 2D materials could have extraordinary properties, such as strength, that they do not have in their three-dimensional form.
Marzari's team had an algorithm that filtered information from several databases. From more than 100,000 materials, the algorithm finally found about 2,000 materials that could be peeled off in one layer, according to the article published by Marzari last month in Nature Nanotechnology . Marzari, who manages Materials Cloud, says that these materials are a "hidden treasure" because many have properties that could improve electronics. Some conduct electricity very well, some can convert heat into water, others absorb energy from the Sun: they could be useful for semiconductors in computers or batteries, so the next step is to investigate these possible properties more closely.
Marzari's work is an example of how scientists are using databases to predict which compounds could create new and exciting materials. Those predictions, however, have yet to be confirmed in a laboratory. And Marzari still had to tell his algorithm to follow certain rules, such as looking for weak chemical bonds. Artificial intelligence can create a shortcut: instead of programming specific rules, scientists can tell AI what they want to create, as a gifted material, and AI will tell scientists what is the best experiment to execute the new material.
This is how Wolverton and his team at Northwestern used AI for an article published this month in Science Advances . The researchers were interested in making new metallic glasses, which are more resistant and less rigid than metal or glass and that could someday improve phones and the spacecraft.
The AI method they used is similar to the ways in which people learn a new language, says study co-author Apurva Mehta, a scientist at the National SLAC Accelerator Laboratory at Stanford University. One way to learn a language is to sit down and memorize all the rules of grammar. "But another way of learning is simply with experience and listening to someone else speak," says Mehta. His approach was a combination. First, the researchers examined the published articles to find as much data as possible on how different types of metal glasses have been manufactured. Then, they fed these "rules of grammar" into an automatic learning algorithm. Then, the algorithm learned to make its own predictions about what combination of elements would create a new form of metallic glass, similar to how someone can improve their French by going to France instead of infinitely memorizing conjugation tables. Mehta's team then tested the system's suggestions in laboratory experiments.
Scientists can synthesize and test thousands of materials at once. But even at that speed, it would be a waste of time to blindly test all possible combinations. "They can not just throw the whole periodic table to their team," says Wolverton, so the role of the AI is to "suggest some places to start." The process was not perfect, and some suggestions: like the exact proportion of necessary elements, they were deactivated, but the scientists were able to form new metallic glasses. In addition, doing the experiments means that they now have even more data to feed back into the algorithm, so that each time it becomes smarter and smarter.
Another way to use AI is to create a "recipe book" or a collection of material recipes. In two articles published late last year, MIT scientists developed an automatic learning system that scans academic documents to discover which ones include instructions for making certain materials. He could detect with 99 percent accuracy what paragraphs of an article included the "recipe" and with 86 percent accuracy the exact words of that paragraph.
The MIT team is now training the AI to be even more precise. They want to create a database of these recipes for the scientific community in general, but they need to work with the publisher of these academic documents to make sure that their collection does not violate any agreement. Eventually, the team also wants to teach the system to read documents and then create new recipes by itself.
"One of the objectives is to discover more efficient and profitable ways of manufacturing the materials we already manufacture," says MIA study co-author and materials specialist Elsa Olivetti. "Another is, here is the compound that the science of computational materials predicted can we suggest a better series of ways to do it?"
The future of materials science and AI looks promising, but the challenges remain. First, computers simply can not predict everything. "Predictions themselves have errors and often work in a simplified model of materials that does not take into account the real world," says Marzari of EPFL. There are all kinds of environmental factors, such as temperature and humidity, that affect the behavior of the compounds. And most models can not take them into account.
Another problem is that we still do not have enough data about each compound, according to Wolverton, and the lack of data means that the algorithms are not very intelligent. Having said that, he and Mehta are now interested in using their method in other types of materials besides metallic glass. And they hope that one day, you do not need a human to do experiments, it will only be AI and robots. "We can really create a completely autonomous system," says Wolverton, "without any human being involved."