In the past scientists and engineers have mixed the materials together to see what the result is. So was discovered, for example, cement. Over time, they learned the physical properties of different compounds, but most of the knowledge is still based on intuition. "If you ask why Japanese steel was better than others for the manufacture of blades, I don't think anyone could answer you," says James Warren, a scientist from the National Institute of standards and technology. "They just had a craft understanding of the relationship between the amazing inner structure of the steel and its properties."
The tamahagane knives from Japanese steel.
According to Warren, instead of having to use a craft experience, we can use a database and calculations to quickly and accurately determine what makes the material so strong or easy, and it can potentially lead to a revolution in the industry. The time between detection of the material and its use in the final product — for example, the battery may be over 20 years, he adds, and the acceleration of the process will inevitably lead us to a better battery and glass for mobile phones, best alloys for missiles and better sensors for medical devices. "Everything is made of matter, we can improve," says Warren.
According to Warren, for understanding how new materials are produced, it is useful to think about the material not as a scholar but as a cook. Let's say that you have the egg and you are determined to make something edible and firm. These are the properties of dish you want, but how do you make it? To create a structure where the protein and the yolk is solid, you need a recipe that includes step-by-step instructions for processing eggs — in our case, it is necessary to cook a certain time. Science uses these concepts: if a scientist wants from the material certain properties (for example, lightness and resistance to fracture), he will look for physical and chemical structures that would create these properties, and processes such as smelting or forging of metal — which will help in their creation.
The database and calculations can help to find answers. "We do a rather complex quantum-mechanical calculations to predict the properties of possible new material on the computer before it is ever created in a laboratory," says Chris Wolverton, a scientist from northwestern University. "The database is not complete, but they grow and have given us exciting discoveries".
Nikola Marzari, a researcher from the Swiss Polytechnic Academy of Lausanne, has used databases to search for 3D content, which can be separated from each other to create 2D materials with a thickness of only one layer. One example of such substances is highly touted graphene, consisting of a single layer of graphite the material in pencil. Like graphene, these 2D materials can have unusual properties, such as the fortress in the gap that they have in their three-dimensional form.
The aerogel of graphene is one of the lightest substances in the world.
The team is Marzari algorithm "sifting" of information from multiple databases. According to an earlier study published last month in Nature Nanotechnology, from more than 100 000 materials the algorithm found around 2000 that can be used as a single layer. Marzari, who directs the project, Materials Cloud, says these materials are "treasure", because many of them have properties that can improve the electronics. Some of them are very well conduct electricity, some well accumulate solar energy: all this can be useful for creating new types of semiconductors in computers or for new batteries, so the next step is a more thorough study of possible properties of these materials.
Work Marzari is one example of how scientists use databases to predict, and what connections can create new and interesting materials. However, these predictions still need to be confirmed in the laboratory. And Marzari still had to tell him his algorithm is forced to follow certain rules — for example, look for the weakest chemical bond. Artificial intelligence can create "shortcut": instead of programming according to certain rules, scientists can tell the AI what they want to create — for example, ultrastrong material — and the AI will suggest the best scientists experiment to create a new material.
So Woolverton and his team at northwestern University used AI in their work, published in Science Advances. The researchers were interested in the creation of new metallic glasses, which are stronger and less rigid than a conventional metal or glass, and one day can start to be used in phones and spacecraft.
The method that they used is the same as people learning a new language, says study co-author Apurva Mehta, a scientist from Stanford University. One way of learning a language is trivial memorizing all the grammar rules. "But there is another way of learning is to listen and memorize a speech someone else," says Mehta. Their approach was a combination of the two. First, the researchers reviewed published articles to find as much data as possible about how you made different types of metallic glasses. Then they made these "rules of grammar" in machine learning algorithms. After that algorithms taught to make their own predictions about what combination of elements will create a new form of metallic glass — similar to how someone can improve their French by going to France, rather than endlessly memorizing declensions and cases. The team then checked Mehta proposed AI materials in laboratory experiments.
Scientists can synthesize and test thousands of materials at a time. But even with such speed banal overkill would be a waste of time and most importantly resources. "They can't just throw the entire periodic table in their equipment," says Wolverton, therefore, the role of AI is to "suggest some experiments with the highest chance of success." The process was not perfect, and some features — for example, the precise ratio required elements have been disabled to simplify, but scientists were still able to form a new metallic glass. In addition, the experiments indicate that scientists now have more data to return to the algorithm and make it smarter and smarter every time.
The unusual properties of a metallic glass under tension.
Another way to use AI is to create a "cookbook" or collection of "recipes" of different materials. In two papers published late last year, scientists from the Massachusetts Institute of technology (MIT) have developed a machine learning system that scans of academic documents to find out which of them contain instructions for creating certain materials. She was able with 99% accuracy to detect which of the paragraphs of article included a "recipe", and with 86% accuracy — the exact right words in this paragraph.
The team at MIT now teaches AI to be even more accurate. They would like to create a database of such "recipes" for the entire scientific community, but they need to coordinate the work with the authors of all these scientific documents to ensure that their collection does not violate any agreements. In the end, they also want to teach the system to read the article, and then to invent new "recipes".
"One of the goals is to find more efficient and economical methods for the manufacture of materials we already produce," says co-author and an expert on materials at MIT Elsa Olivetti. "Another matter is the material predicted by the AI. And how best to create?"
The future of AI and material science looks promising, but challenges remain. First, computers just can't predict everything. "The predictions have errors, and we often work with a simplified model that does not account for the real world," says Marzari. "There are all kinds of environmental factors such as temperature and humidity, which affect the behavior of the compounds. Most of the models cannot take this into account".
Another problem is that we still do not have enough data on each connection, and the lack of algorithms is that they are not too smart. However, Wolverton and Mehta are now interested in using his method to search other types of materials similar to metallic glass properties. And they hope that someday we will no longer need people to conduct experiments — this will deal with AI and robots. "We can really create a completely Autonomous system, without human intervention," says Wolverton.
- 16-05-2020Ten technologies that will shake the world in 2020
- 31-03-2020How to design the city based robotic transport
- 23-03-2020The birth and life of computers
- 09-02-20202020: trends and forecasts
- 05-01-2020Nine major technological breakthroughs 2019
- 23-08-2013Earth vs. The sun, or the change of magnetic poles
- 30-11--0001Physics saw the beginning of times and confirmed the expansion of the Universe
- 30-09-2010From Russia suggest that you escape as soon as possible