(Nanowork NewsLike people, materials evolve over time. They also behave differently when stressed than when relaxed. Scientists wanting to measure the dynamics of how materials change have developed a new technique that leverages X-ray photon correlation spectroscopy (XPCS), artificial intelligence (AI), and machine learning.
The technique creates “fingerprints” of different materials that can be read and analyzed by neural networks, which are computer models that make decisions in a similar way to the human brain.
In a new study (“AI-NERD: Uncovering Beyond Equilibrium Relaxation Dynamics with AI-Informed X-ray Photon Correlation Spectroscopy”) by researchers at the U.S. Department of Energy (DOE) Argonne National Laboratory's Advanced Photon Source (APS) and Center for Nanoscale Materials (CNM), the scientists combined XPCS with an unsupervised machine learning algorithm, a type of neural network that does not require expert training. The algorithm teaches itself to recognize hidden patterns in sequences of X-rays scattered by colloids (groups of particles suspended in solution). APS and CNM are DOE Office of Science User Facilities.
“To understand how materials behave and change over time, we need to collect X-ray scattering data,” said James (Jay) Horwath, a postdoctoral researcher at Argonne National Laboratory and first author of the study.
These patterns are too complex for scientists to detect without the help of AI: “When you shine an X-ray beam on something, the patterns become so diverse and complex that even experts have a hard time making sense of them,” Horwath said.
To gain a deeper understanding of their subject, researchers need to condense all the data into a fingerprint that contains only the most important information about the sample. “It's like having the genome of the sample, so you have all the information you need to reconstruct the whole picture,” Horwath says.
The project is called Artificial Intelligence for Non-Equilibrium Relaxation Dynamics, or AI-NERD. The fingerprints are created using a technique called an autoencoder, a type of neural network that converts the original image data into a fingerprint (which scientists call a latent representation), and also includes a decoder algorithm used to convert the latent representation back into the full image.
The researchers' goal was to collect fingerprints with similar characteristics into neighborhoods, creating a fingerprint map of the material. By examining the characteristics of the different fingerprint neighborhoods on the map together, the researchers were able to gain a deeper understanding of how the material was structured and how it changed over time as it was stressed and relaxed.
Put simply, AI has great general pattern recognition capabilities, allowing it to efficiently classify and sort the various X-ray images into a map. “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,” Horwath said. “AI is an expert at pattern recognition.”
Using AI to make sense of the scattering data will be especially important once the upgraded APS begins operation. The improved facility produces an X-ray beam that's 500 times brighter than the original APS. “We'll need the power of AI to organize the data that's coming out of the upgraded APS,” Horwath says.
CNM’s theory group is collaborating with a computational group in Argonne National Laboratory’s X-ray Sciences Division to perform molecular simulations of the polymer dynamics demonstrated by XPCS and, going forward, synthetically generate data to train AI workflows such as AI-NERD.
