The AI-NERD model learns to generate a unique “fingerprint” for each sample of XPCS data, allowing it to identify trends and repeating patterns. [Argonne National Lab]
What if scientific advances could happen without scientists having to perform all the experiments? New AI developments at Argonne National Laboratory have taken a concrete step towards that vision, “taking a concrete step towards autonomous materials discovery,” says a recent Nature Communications paper. The field of materials science faces something of a conundrum as the demand for new, higher-performing materials in sectors ranging from renewable energy to aerospace outstrips researchers' ability to discover and characterize them the traditional way.
Traditionally, understanding how matter changes at the atomic level has relied on time-consuming, labor-intensive experiments. Observing such microscopic dynamics has proven difficult. Techniques such as X-ray Photon Correlation Spectroscopy (XPCS) can provide a glimpse into atomic behavior, but the data generated is incredibly complex.
Introducing AI-NERD
Enter AI-NERD (Artificial Intelligence for Non-Equilibrium Relaxation Dynamics), a novel technology from Argonne National Laboratory that promises to accelerate materials research. This approach leverages unsupervised deep learning, specifically autoencoder neural networks, to analyze complex X-ray data and reveal hidden “fingerprints” of material behavior. Without expert training, AI-NERD teaches itself to recognize patterns in XPCS data, creating condensed material “fingerprints” from complex X-ray scattering patterns. This groundbreaking technology allows researchers to map and analyze material behavior in ways never before possible.
“The goal of the AI is to treat the scattering pattern as a normal image or photograph and understand it to figure out what the repeating pattern is. AI is an expert at pattern recognition,” James (Jay) Horwath of Argonne National Laboratory, first author of the study, said in a press release.
Decoding the material genome
These fingerprints are more than just patterns. They are condensed representations of a material's structure and behavior, distilling vast amounts of XPCS data into essential features. “It's like the genome of the material,” Horwath explains. “It contains all the information we need to reconstruct the complete picture.”
What sets AI-NERD apart is its ability to learn and identify patterns without expert guidance. This unsupervised learning approach allows the system to discover hidden relationships and trends in material behavior that scientists might not see. By processing and classifying X-ray scattering images, AI-NERD creates a comprehensive map of material dynamics, giving researchers a new lens through which to view atoms and molecules.
In this study, scientists used a technique called Uniform Manifold approximation and projection (UMAP) to convert a complex dataset into a simple two-dimensional image, as shown in the image above. UMAP is similar to another popular method called t-distributed stochastic neighbor embedding (tSNE). Both of these methods attempt to preserve the relationships between data points while reducing dimensionality. For an overview of word embeddings, see the article What are embeddings and how to explore them in R&D.
The broad field of AI in materials science has been gaining momentum in recent years.
This development in AI-assisted materials research at Argonne National Laboratory is part of a broader trend of artificial intelligence revolutionizing materials science. In recent years, several notable advances have paved the way for the development of AI-NERD.
In 2023, researchers made great strides in using AI for materials synthesis and characterization. For example, one team developed an AI system that can extract “recipes” for manufacturing materials from scientific papers. The system can identify correlations between raw chemicals and the resulting crystal structures, streamlining the materials discovery process.
Another group developed an AI system that recognizes patterns in blends of different ingredients, an innovation that could enable the AI to suggest alternative blends of known ingredients, opening up new avenues of synthesis.
While machine learning continues to be a trending topic, its use in materials science is not new: In 2018, researchers at Virginia Tech developed a machine learning framework that trains “on the fly” to accelerate the development of computational models for materials design.
Similarly, an AI system called ARTIST, developed by Aalto University and the Technical University of Denmark, can instantly determine how molecules respond to light, potentially accelerating the development of technologies such as flexible electronics.
In nuclear materials research, the University of Wisconsin-Madison and Oak Ridge National Laboratory in 2018 unveiled an AI system trained to detect and analyze microscopic radiation damage in potential nuclear reactor materials. The system outperformed human experts in both accuracy and speed.
The potential of AI-NERD
In particular, the development of AI-NERD is expected to particularly facilitate the analysis of XPCS data. Once the upgraded APS is up and running and produces an X-ray beam 500 times brighter than its predecessor, the need for efficient data processing will become even more critical. “We'll need the power of AI to sift through the data we get from the upgraded APS,” Horwath emphasizes. AI-NERD will be able to create “fingerprints” of materials and identify patterns in large datasets, potentially improving researchers' understanding of materials dynamics in a variety of applications.
