����ֱ��

To predict which MPEA compositions to manufacture, APL partnered with and Maitreyee Sharma Priyadarshini at the .

With limited data, Clancy and Sharma Priyadarshini used their physics-informed Bayesian optimization algorithm to quickly sift through alloy possibilities and recommend MPEAs that would maximize hardness and provide the widest sample of data for future alloy development. Unlike with most deep-learning efforts, which require a large database, the team pushed to see whether few data points would still generate reasonable results. The algorithm only needed around a dozen data points to generate recommendations.

“We created a closed-loop cycle,” Clancy said. “We run APL’s data through PAL 2.0 to make predictions of compositions, APL fabricates the samples and tests the suggested alloys and analyzes the results, and then we restart the process based on the additional data collected. We’re turning the crank multiple times.”

The team continually builds off each test’s results. In the first round, the machine learning tool recommended alloy possibilities that were too expensive or unmanufacturable. The team is now in its third cycle, which is identifying alloys with more than twice the hardness.

After Clancy’s group provides recommendations on which alloys to test, the APL team creates the alloys and conducts arc melting, which sends an electronic current through the metal to melt it. Arc melting can be achieved with minimal material, making it a quick way to sample vastly different compositions.

“When the alloy is melted, it essentially creates dozens of different materials with different chemical compositions all within one sample,” said Eddie Gienger, a materials scientist at APL and lead author on the paper.

The pathway is created by integrating data from different sources: scanning electron microscopy (SEM) and energy-dispersive X-ray spectroscopy (EDS), which measure phase compositions, and nanoindentation, a process where a small triangular tip is pressed into tiny pieces of the alloy to measure hardness.

“With nanoindentation we can collect measurements from hundreds of locations automatically on the sample. We then map those indents to phase compositions identified using EDS and SEM to understand how each unique microstructure and composition will behave. Identifying regions that don’t correspond to known phases enables the discovery of new, potentially useful materials. Coupling that with mechanical property information allows us to identify promising compositions for future alloy development,” added Lisa Pogue, a materials scientist at APL.

That information, which included 17 unique MPEA compositions and over 7,000 unique data points, was compiled into a database for future alloy development.

“The process helps identify the possibility space for MPEAs,” Gienger said. “If a material needs to be designed to achieve certain mechanical properties, such as hardness, the pipeline could provide the information to engineer those properties.”

The technique is one of many APL is developing and testing to speed up and understanding. “The more tools we have to conduct high-throughput characterization, the more opportunities we have to apply it in spaces where new, stronger materials are needed,” Trexler said.