NETL is collaborating with Carnegie Mellon University to make faster and more accurate predictions on the properties of heat-resistant alloys and develop cost-effective, corrosion-resistant materials needed in flexible energy systems that will be highly efficient, produce fewer emissions and help meet the nation’s decarbonization goals while producing reliable supplies of electricity.
To produce durable alloys to manufacture turbine blades, pressure vessels, heat exchangers and other equipment, NETL is collaborating with CMU on a two-year project to further explore the “PSP connection” — a fundamental tenet of materials science that maintains Processing generates the microStructure that mediates material Properties.
The NETL-managed project, sponsored by the U.S. Department of Energy’s (DOE) Office of Fossil Energy High Performance Materials program, focuses on collecting microstructure image data and property metadata, and using computational tools to discover new PSP connections and design microstructures to achieve desired properties.
Microstructure refers to the substructures that form from the interaction between composition and processing of an alloy (a metal made by combining two or more metallic elements to induce hardness, toughness or other desired properties). Microstructural features in an alloy include grains, interfaces, precipitates, dislocations, voids and others. The amount, distribution and arrangement of these features govern materials properties, from mechanical response to corrosion to superconductivity.
In this project, CMU researchers are applying computer vision (CV) — technology in which a computer can extract, analyze and understand useful information from an individual image or a sequence of images — to create quantitative representations of microstructural images and apply machine learning (ML) methods to predict material properties.
Artificial intelligence techniques, including CV and ML, hold immense promise for extracting new knowledge from the rich, complex and multimodal data collected in materials science and engineering investigations. As the first application of these methods to heat-resistant alloy design, this project is expected to provide critical experience and insight for alloy development and could revolutionize microstructure design for performance.
ML models require extensive training data to minimize error and improve predictive accuracy. NETL’s Youhai Wen, Ph.D., a member of the Computational Science and Engineering Team, and Michael Gao, Ph.D., a member of the Structural Materials Team, are providing microstructural images to train CMU’s model.
Developing the CV/ML system to discover the PSP connections is proceeding in three stages. In the first stage, the team assembled a dataset of microstructural images and associated property metadata.
In the next stage, researchers will compare two CV image representation models to develop a CV approach to quantify the visual information contained in the microstructural images. Finally, the team will choose an ML method suitable for learning from the selected image representation.
Machine learning can rapidly accelerate materials research by automating performance predictions from microstructural image data. The project will provide critical insights to develop materials that can function under extreme temperature and pressure conditions and enable energy systems to increase their use of intermittent renewable energy resources and develop a new class of cost-effective materials for high-efficiency systems such as natural gas (or hydrogen-fired) combined cycles equipped with carbon capture and storage.
The U.S. Department of Energy’s National Energy Technology Laboratory develops and commercializes advanced technologies that provide clean energy while safeguarding the environment. NETL’s work supports DOE’s mission to ensure America’s security and prosperity by addressing its energy and environmental challenges through transformative science and technology solutions.