Machine Learning Leads to the Development of Ultra-Tough Materials

Artificial intelligence is making plenty of headlines lately. Indeed, from smart classrooms to Atlantic voyages, this technology is radically transforming a wide range of engineering and scientific domains. Now developers at MIT are exploring a new field of application, i.e., materials engineering. More specifically, they are using machine learning, a technique that trains the abilities of an AI system to solve problems autonomously by analyzing millions of combinations.

The quest for new materials had so far required computationally intensive simulations. As each variation had to prove its toughness and behavior on an atomic scale, including the trajectory of each atom, the optimal combination required hours or even days for calculations. The new machine-learning-driven approach can carry out the same process in a matter of milliseconds.  

The goal of the research team was to assess the way cracks propagated throughout the molecular structure of a material. Unlike former methods, where the fracture point was established by analyzing each combination, machine learning allows artificial intelligence to detect the relationship between combinations, which are the typical patterns of the most robust and most fragile materials.  

This innovative technology project carried atomistic simulations of layered coatings made of crystalline materials and found the hardest structure almost instantaneously. The goal of the project was to create new coatings for the aerospace industry, such as the ceramic plates that protect space shuttles. However, the applications could cover a large number of areas, from body prostheses to buildings.

Super-compressible materials too

While MIT has leveraged artificial intelligence to develop ultra-tough materials, other research teams are exploring different properties. An example of this would be the work carried out by Miguel Bessa, assistant professor at TU Delft in the Netherlands. The way satellites can open long solar sails from an exceedingly small package inspired him to develop a highly compressible yet durable material. The resulting material could allow manufacturing pocket-sized bicycles or umbrellas. Just like his colleagues at MIT, Bessa understood that the traditional approach was computationally expensive and that a trial and error would not cut it.

Thus, his team decided to go down the machine learning path, which reduced the need for physical experiments. By using this software, they focused on a set of brittle polymers that are highly compressible on a macroscopic scale, while tough and resistant on a microscopic one.

Technically, these algorithms will be able to help in the development of new materials despite using incomplete data sets. Just by having enough accurate data available, the platform can carry out the calculations in an autonomous way and find the best combinations.

Source: MIT, Phys