Engineers are always looking for materials with very specific properties for their projects. Unfortunately, there are too many options for researchers to just guess and check until they find what they are looking for. Even if they had to model the materials, instead of testing them in the laboratory, it would take too long to find the relevant material.
Fortunately, researchers have created algorithms using artificial intelligence that can find the right material for any project. In a recent article, a team of researchers at Carnegie Mellon University and the University of Calgary refined one of these algorithms, allowing researchers to quickly and accurately find materials with the desired properties.
“Because the space of materials is so huge, it is very difficult to experimentally and computationally characterize the properties of the material,”; said Amir Barati Farimani, associate professor of mechanical engineering at the Cabinet of Ministers. “So we create algorithms or models that can quickly predict the properties of a material.”
To use artificial intelligence, or AI, researchers must first teach the algorithm using known data. The algorithm then learns to extrapolate new ideas from this information. Barati Farimani and his team learned an algorithm for data on the chemical composition of materials. In particular, they included information on the role of electrons in determining the properties of the material. These chemical data created a new material descriptor for the algorithm, according to Barati Farimani.
Because this algorithm can predict the properties of a wide range of materials, it has many applications. For example, the algorithm could find a material with thermal properties suitable for solar panels. He could also identify materials for drugs and batteries. To use this algorithm, the researcher can simply achieve pre-trained models of deep learning, find the property they are looking at.
These algorithms are improved as they become faster and more accurate. If the algorithm is not accurate enough, the results will be unusable. If the algorithm is too slow, researchers will never be able to access the results. Currently, the team has found that their algorithm is better than other leading algorithms.
“You can use this algorithm and teach a model of deep learning and predict them in a split second,” said Barati Farimani. “The point is to prove that it provides different types of materials with high accuracy – then every industry can use it.”
Their article was published in Physical review materials. CMU PhD student Mohammadreza Karamad, PhD student Rishikesh Magar and researcher Yutin Shi were also cited as co-authors. Other authors include Samira Siarostami and Ian D. Gates of the University of Calgary.
The algorithm involves the composition of new materials
Mohammadreza Karamad et al. Orbital graph of the convolutional neural network for predicting material properties, Physical review materials (2020). DOI: 10.1103 / PhysRevMaterials.4.093801
Provides mechanical engineering at Carnegie Mellon University
Citation: Order! AI finds the required material (2020, October 16), received on October 16, 2020 from https://phys.org/news/2020-10-ai-material.html
This document is subject to copyright. Except for any bona fide relationship for the purpose of private study or research, no part may be reproduced without written permission. The content is provided for informational purposes only.