New regulations to double the use of sustainable renewable fuels in UK by 2020; 12.4% minimum blend by 2032
Japan Benex, Sumitomo Corp. to power plant with Nissan EVs, batteries

Scientists use machine learning and HiTp to speed discovery of metallic glasses by 200 times; approach applicable to other materials

Researchers at SLAC and their colleagues at Northwestern University and NIST have combined machine learning (ML) and high-throughput (HiTp) experiments to find a new system of metallic glasses in the Co-V-Zr ternary. In their approach, described in an open-access paper in Science Advances, the team trained a machine learning (ML) model on previously reported observations, parameters from physiochemical theories, and made it synthesis method–dependent.

In the work, they first constructed a paper begins by constructing an ML model based on known discoveries and predicted high likelihood of finding MG in the Co-V-Zr ternary; they subsequently validated this by HiTp experimentation. The results of HiTp experimentation then were used to train a greatly improved “second-generation” ML model.

In the paper, they demonstrate that this refined ML/HiTp model successfully predicts glass-forming ability (GFA) in three new ternary composition spaces, where no experimental observations exist. The paper illustrates how ML and HiTp experimentation can be used in an iterative/feedback loop to easily accelerate discoveries of new MG systems by more than two orders of magnitude as compared to traditional search approaches relied upon for the last 50 years.

IMG_0660
Schematic depiction of a paradigm for rapid and guided discovery of materials through iterative combination of ML with HiTp experimentation. Ren et al. Click to enlarge.

Some technologically important materials are kinetically stabilized. One such metastable class of materials is amorphous alloys of metals, namely, metallic glasses (MGs). The lack of crystalline order in MG significantly alters the properties of these materials, thus enabling novel and improved functionalities. For example, the absence of deformation pathways based on gliding dislocations leads to exceptional yield strength and wear resistance. Some MGs have enhanced corrosion resistance because of their ability to rapidly form protective films on the surface. MGs are therefore very promising candidates for structural applications, especially for high-cycle use in chemically harsh environments. Other MGs, such as the well-known Metglas system of alloys, exhibit high magnetic permeability, making them attractive for electromagnetic shielding.

Recent reports estimate upward of several million MGs, with a large fraction of them occurring in the multi-elemental composition space. However, less than a few thousand have been discovered in the last 50 years. The search for new MGs is challenging because they often contain three or more elements, and the nonequilibrium nature of the system implies that processing parameters (most commonly cooling rates, but in some cases enhanced surface diffusion) strongly influence formability. The vastness of the combined composition-processing space makes searches based on serial trial-and-error experimentation difficult and expensive; even rapid parallel synthesis combined with high-throughput characterization (HiTp experimentation) can stall without additional guidance. For instance, even an aggressive rate of synthesizing and fully characterizing one ternary per day would take more than 10 years of HiTp experimentations to search just the ternary combinatorial space encompassed by 30 common elements for MGs.

… Here, we demonstrate an alternate strategy that overcomes these limitations. We use machine learning (ML) iteratively with HiTp experiments to guide the search for new MGs. ML approaches are well suited to this problem because they can (i) begin with a heterogeneous and sparse data set, (ii) operate with less than perfect understanding of the underlying physics but still take physiochemical parameters to accelerate learning, and (iii) progressively improve by simply adding more observations to the training set. In particular, the ability of ML to find patterns in observed data makes it possible to model relationships that are as yet unexplained by physiochemical theories (PCTs) (for example, the relationship between synthesis method and GFA, as will be shown below).

… This paper highlights an emerging paradigm of data-driven discoveries for rapid and guided discovery of materials, whose functionality depends not only on chemical composition but also on synthesis.

—Ren et al.

It typically takes a decade or two to get a material from discovery to commercial use. This is a big step in trying to squeeze that time down. You could start out with nothing more than a list of properties you want in a material and, using AI, quickly narrow the huge field of potential materials to a few good candidates.

— Northwestern Professor Chris Wolverton, an early pioneer in using computation and AI to predict new materials and a co-author of the paper

The ultimate goal, Wolverton said, is to get to the point where a scientist could scan hundreds of sample materials, get almost immediate feedback from machine learning models and have another set of samples ready to test the next day—or even within the hour.

Over the past half century, scientists have investigated about 6,000 combinations of ingredients that form metallic glass, added paper co-author Apurva Mehta, a staff scientist at SLAC’s Stanford Synchrotron Radiation Lightsource (SSRL): “We were able to make and screen 20,000 in a single year.

While other groups have used machine learning to come up with predictions about where different kinds of metallic glass can be found, Mehta said, “The unique thing we have done is to rapidly verify our predictions with experimental measurements and then repeatedly cycle the results back into the next round of machine learning and experiments.

There’s plenty of room to make the process even speedier, he added, and eventually automate it to take people out of the loop altogether so scientists can concentrate on other aspects of their work that require human intuition and creativity. “This will have an impact not just on synchrotron users, but on the whole materials science and chemistry community,” Mehta said.

The team said the method will be useful in all kinds of experiments, especially in searches for materials such as metallic glass and catalysts whose performance is strongly influenced by the way they’re manufactured, and those where scientists don’t have theories to guide their search. With machine learning, no previous understanding is needed. The algorithms make connections and draw conclusions on their own, and this can steer research in unexpected directions.

One of the more exciting aspects of this is that we can make predictions so quickly and turn experiments around so rapidly that we can afford to investigate materials that don’t follow our normal rules of thumb about whether a material will form a glass or not. AI is going to shift the landscape of how materials science is done, and this is the first step.

—paper co-author Jason Hattrick-Simpers, a materials research engineer at NIST

The paper is the first scientific result associated with a DOE-funded pilot project in which SLAC is working with Citrine Informatics to transform the way new materials are discovered and make the tools for doing that available to scientists everywhere.

Founded by former graduate students from Northwestern and Stanford University, Citrine has created a materials science data platform where data that had been locked away in published papers, spreadsheets and lab notebooks is stored in a consistent format so it can be analyzed with AI specifically designed for materials.

Until recently, thinking up, making and assessing new materials was painfully slow. For example, the authors of the metallic glass paper calculated that even if you could cook up and examine five potential types of metallic glass a day, every day of the year, it would take more than a thousand years to plow through every possible combination of metals. When they do discover a metallic glass, researchers struggle to overcome problems that hold these materials back. Some have toxic or expensive ingredients, and all of them share glass’s brittle, shatter-prone nature.

Over the past decade, scientists at SSRL and elsewhere have developed ways to automate experiments so they can create and study more novel materials in less time. Today, some SSRL users can get a preliminary analysis of their data almost as soon as it comes out with AI software developed by SSRL in conjunction with Citrine and the CAMERA (Center for Advanced Mathematics for Energy Research Applications) project at DOE’s Lawrence Berkeley National Laboratory.

In the metallic glass study, the research team investigated thousands of alloys that each contain three cheap, nontoxic metals. They started with a trove of materials data dating back more than 50 years, including the results of 6,000 experiments that searched for metallic glass. The team combed through the data with advanced machine learning algorithms developed by Wolverton and graduate student Logan Ward at Northwestern.

Based on what the algorithms learned in this first round, the scientists crafted two sets of sample alloys using two different methods, allowing them to test how manufacturing methods affect whether an alloy morphs into a glass.

Both sets of alloys were scanned by an SSRL X-ray beam, the data fed into the Citrine database, and new machine learning results generated, which were used to prepare new samples that underwent another round of scanning and machine learning.

By the experiment’s third and final round, said co-author Apurva Mehta, a staff scientist at SSRL, the group’s success rate for finding metallic glass had increased from one out of 300 or 400 samples tested to one out of two or three samples tested. The metallic glass samples they identified represented three different combinations of ingredients, two of which had never been used to make metallic glass before.

… we show that iterative application of ML modeling and HiTp experimentation is a very powerful paradigm to find new MGs. We used it here to discover three new ternaries with large GFRs. We further show that even when the first-generation ML predictions are not significantly more accurate than predictions of PCTs, they rapidly surpass them with additional observations and can capture nuances that are harder to predict otherwise. We believe that this paradigm of data-driven discovery can be easily extended to accelerate the search for a wide range of technologically important materials, from high-entropy alloys to catalysts. It is particularly attractive for materials for which fully predictive PCTs have yet to be developed, and synthesis methods and other complex parameters (such as morphology, microstructure, and substrate interaction) play a large role in determining their properties and performance.

—Ren et al.

SSRL is a DOE Office of Science user facility. In addition to SLAC, NIST and Northwestern, scientists contributing to this study came from the University of Chicago’s Computation Institute, the University of South Carolina and the University of New South Wales in Australia. The SLAC pilot project with Citrine is funded by the Advanced Manufacturing Office of DOE’s Office of Energy Efficiency and Renewable Energy, and includes collaborating scientists from NIST, DOE’s National Renewable Energy Laboratory and the Colorado School of Mines. The CAMERA project at Berkeley Lab is supported by the DOE Office of Science.

Resources

  • Fang Ren, Logan Ward, Travis Williams, Kevin J. Laws, Christopher Wolverton, Jason Hattrick-Simpers, Apurva Mehta (2018) “Accelerated discovery of metallic glasses through iteration of machine learning and high-throughput experiments” Science Advances doi: 10.1126/sciadv.aaq1566

Comments

Verify your Comment

Previewing your Comment

This is only a preview. Your comment has not yet been posted.

Working...
Your comment could not be posted. Error type:
Your comment has been posted. Post another comment

The letters and numbers you entered did not match the image. Please try again.

As a final step before posting your comment, enter the letters and numbers you see in the image below. This prevents automated programs from posting comments.

Having trouble reading this image? View an alternate.

Working...

Post a comment

Your Information

(Name is required. Email address will not be displayed with the comment.)