Building AI-Guided Pathways to Materials for Extreme Energy Environments

Ian McCue connects machine learning with rapid fabrication and testing to accelerate the discovery and qualification of materials for fusion energy, geothermal systems, and other demanding applications.

The promise of technologies such as fusion energy and enhanced geothermal power depends partly on materials that can withstand punishing conditions: intense heat, radiation, corrosive fluids, and extreme mechanical stress. Finding those materials through conventional research can be painstakingly slow. Researchers must choose among enormous numbers of possible compositions, fabricate samples, test their performance, and determine whether they can operate safely for years. Ian McCue is developing a faster, more systematic approach.

McCue, assistant professor of materials science and engineering and the Morris E. Fine Junior Professor in Materials and Manufacturing at Northwestern Engineering, combines artificial intelligence with high-throughput experiments and advanced manufacturing to navigate vast materials-design spaces. His work focuses on creating and qualifying materials for extreme environments, with applications including fusion reactors, geothermal systems, aerospace vehicles, and advanced manufacturing.

“We broadly study extreme environments in my group,” McCue said. “This can refer to corrosion, which is important for geothermal, as well as the high heat fluxes and temperatures relevant to fission and fusion."

Rather than focusing primarily on inventing new AI algorithms, McCue is pioneering research workflows that connect computational predictions with material fabrication, characterization, and testing. The goal is to create a continuous learning process in which each experiment improves the model’s ability to choose the next promising material.

Closing the loop between AI and experiments

The number of possible materials grows rapidly as researchers combine multiple chemical elements in different proportions. An alloy system containing five or six elements can produce hundreds of thousands—or even millions—of possible compositions and processing conditions. Fabricating and testing each one would be prohibitively expensive and time-consuming.

McCue uses methods such as Bayesian optimization to make the search manageable. The AI model draws on existing data and results from earlier experiments to identify which candidates would be most informative or most likely to meet the desired performance targets. Researchers fabricate and test those candidates, feed the results back into the model, and repeat the cycle.

“We might start with 100,000 different alloy compositions,” McCue said. “Then we go through screening processes and iterative loops with fabrication and testing to pin down perhaps one to five samples to make and evaluate in detail.”

McCue was part of a team from Johns Hopkins University and its Applied Physics Laboratory that used this closed-loop approach to search for new superconductors, materials that conduct electricity without resistance below a critical temperature. The team trained machine-learning models on databases of known materials, tested the models' predictions experimentally, and fed the results back to improve the next round of recommendations. Over four cycles, the effort identified a previously unknown superconductor in the zirconium-indium-nickel system and more than doubled the success rate for discovering superconductors, according to the study published in npj Computational Materials. McCue cites the project as evidence that AI predictions become substantially more useful when they are embedded in an experimental feedback loop

“I think oftentimes people are approaching AI to optimize a particular problem,” McCue said. “We’re using it for discovery.”

Designing materials for fusion energy

McCue is now applying that systems-level perspective to one of energy research’s most difficult materials challenges: developing components that can withstand the harsh conditions inside fusion reactors.

Fusion devices expose their interior surfaces to extraordinarily high temperatures, heat flows, radiation, and energetic particles. McCue’s group studies materials for plasma-facing components, including the reactor’s first wall and divertor—the structures that must directly withstand conditions created by the hot plasma.

Many leading candidates are metals, but McCue sees promise in ultra-high-temperature ceramics, which can operate beyond the limits of metals yet remain poorly characterized under fusion-relevant conditions. Closing that gap is a new aim his group is preparing with collaborators at and beyond Northwestern. Combining rapid ceramic synthesis and property screening within a Bayesian optimization loop would allow a model to choose the most informative candidates for each round of experiments, speeding exploration of a vast space of compositions and processing routes. Processing matters because it sets the microstructure that governs how a ceramic conducts heat, resists cracking, or responds to radiation.

"The first step is building a workflow for high-throughput fabrication, screening, and evaluation of these ceramics," McCue said. The goal is materials that combine thermal conductivity, fracture resistance, high-temperature strength, and resistance to radiation-induced swelling.

Accelerating the path from discovery to deployment

Identifying a promising material is only the beginning: before it enters a nuclear-energy system, researchers and regulators need extensive evidence of reliable performance, a qualification process that has historically taken decades. Fusion is harder still, since no existing facility reproduces a commercial reactor's full environment.

McCue sees room for AI to integrate results across experiments and simulations, identify evidence gaps, and prioritize the most valuable tests. "There is a significant push to accelerate qualification for fission and to establish qualification pathways for fusion," he said. "That is a ripe area for AI."

His long-term ambition is to help shrink qualification from 30 years to less than ten (and ideally less than five), which will require coordinated advances in experiments, computation, data infrastructure, and regulation. That deployment focus runs through his work. "I always aim to do application-inspired research," he said. "The goal is for it to transition to a company or government application on roughly a 10-year horizon."

Building the materials laboratory of the future

McCue’s work also contributes to a larger shift toward more integrated and autonomous materials laboratories. In a 2025 MRS Bulletin article, he joined researchers from across the country in outlining a vision for future laboratories that combine automated experiments, advanced characterization, simulations, AI, and shared data infrastructure. Such systems could accelerate innovation while expanding researchers’ access to sophisticated discovery tools.

At Northwestern, McCue’s group is already bringing several elements of that vision together. His team develops rapid experimental methods, computational tools, and algorithms to design materials while accounting for real-world manufacturing constraints.

In one example, McCue and collaborators created an on-the-fly computational method for designing compositionally graded materials, in which the chemical makeup changes gradually across a component. These gradients can create strong bonds between dissimilar materials and eliminate bulky mechanical fasteners. The framework uses concepts from robotic path planning to search complex composition spaces, reducing calculations that previously took weeks or months to a matter of hours.

Although initially developed for aerospace applications, the methods are broadly applicable to energy systems in which components must join materials with different thermal, mechanical, or corrosion-resistant properties.

McCue is also beginning to extend his extreme-environments research into geothermal energy. Equipment used in deep geothermal wells must withstand high temperatures, pressures, and corrosive brines for extended periods. AI-guided materials design could help researchers explore new alloys, coatings, and processing strategies suited to those combined conditions.

Across these applications, McCue’s distinctive contribution is connecting AI to the physical realities of making, measuring, and deploying materials. By closing the loop between computational intelligence and experimentation, McCue is helping move materials research beyond isolated predictions toward a faster path from discovery to energy technology applications.

Selected publications

Pogue, E. A., New, A., McElroy, K., et al. “Closed-loop superconducting materials discovery.” npj Computational Materials 9, 181 (2023). https://doi.org/10.1038/s41524-023-01131-3

Price, S., Cao, Z., and McCue, I. “On-the-fly path planning for the design of compositional gradients in high dimensions.” Materials & Design 256, 113983 (2025). https://doi.org/10.1016/j.matdes.2025.113983

Banerjee, S., Meng, Y. S., Minor, A. M., et al. “Materials laboratories of the future for alloys, amorphous, and composite materials.” MRS Bulletin 50, 190–207 (2025). https://doi.org/10.1557/s43577-024-00846-y

Abd-Elaziem, W., Reidy, J., and McCue, I. D. “In-Situ Material Characterization of Steel Through High-Speed Imaging During Orthogonal Cutting.” In TMS 2026 155th Annual Meeting & Exhibition Supplemental Proceedings, 217–225 (Springer, 2026). https://doi.org/10.1007/978-3-032-13828-6_18