Aerospace and Defense

Machine Learning and Informatics Accelerate New Material Discoveries

Using adaptive design to reveal targeted properties in a shape-memory alloy.

listen to this story

The potential applications for machine learning and informatics seem limitless.

From a wide variety of applications in medicine and healthcare to saving the whales, these areas of artificial intelligence are solving problems in new and inherently innovative ways.

Most recently, the machine learning-informatics power couple has been improving the process of discovering new materials based on targeted characteristics.

Feedback from experiments: an augmented dataset with four new alloys. (Image courtesy of Los Alamos National Laboratory.)

An initial alloy experimental dataset with known thermal dissipation and features or materials descriptors serves as input to the inference model. The model is then trained and cross-validated with the initial alloy data. A data set of unexplored alloys defines the total search space of probable candidates. The trained model in is applied to all the alloys to predict their thermal dissipation. The design chooses the ‘best’ four candidates for synthesis and characterization and the new alloys then augment the initial data set to further improve the inference and design.  (Image courtesy of Los Alamos National Laboratory.)

Although the new strategy was developed using nickel-titanium-based shape-memory alloys, the results can be applied to any materials class (e.g., polymers, ceramics or nanomaterials) or target properties (e.g., dielectric response, piezoelectric coefficients and band gaps).

The system can also be adapted for a variety of other applications, such as optimizing process conditions in advanced manufacturing, and can be generalized to optimize multiple properties. For example, in the case of the nickel-titanium alloy the target properties were low dissipation and a transition temperature several degrees above room temperature.

Research Assistance from Machine Learning

As the chemical complexity of materials increases, using the traditional method of trial and error is insufficient. Even without accounting for defects, solid solutions and multi-component compounds, running thousands of quantum mechanical calculations on macro, micro and chemical characteristics of new materials is an almost insurmountable task for human researchers, but an ideal one for machine learning.

Machine learning is a method of data analysis which uses algorithms that iteratively learn from data to automate analytical model building. As a result, computers with machine learning capabilities are able to discover insights without users having to program them with explicit instructions. Informatics can then utilize the resulting data to generate meaningful results for human beings.

The adaptive design framework, (Image courtesy of Los Alamos National Laboratory.)

The adaptive design framework. Prior knowledge from previous experiments and physical models is used to make predictions that include error estimates. The results feed into a database that provides input for the next iteration of the design loop. The green arrows represent the step-wise approach of the state-of-art using experiments or calculations. The red star shows that although sample number 3 is not the best predicted choice relative to sample 4, the ‘expected improvement’ by selecting it is greater than other choices due to the large uncertainty. (Image courtesy of Los Alamos National Laboratory.)

In this case, the researchers developed a framework that uses uncertainties to iteratively guide the next experiments to be performed in search of a shape-memory alloy with very low thermal hysteresis. These alloys are critical to improving fatigue life in engineering applications.

“The goal is to cut in half the time and cost of bringing materials to market,” said Turab Lookman, physicist and materials scientist at Los Alamos National Laboratory. “What we have demonstrated is a data-driven framework built on the foundations of machine learning and design that can lead to discovering new materials with targeted properties much faster than before.”

The research for this project was carried out at Los Alamos National Laboratory and the State Key Laboratory for Mechanical Behaviour of Materials in China.

The results were published under the title “Accelerated search for materials with targeted properties by adaptive design” in the journal Nature Communications.