From hypersonic aircraft to nuclear-powered submarines, many of today’s most advanced defense systems rely on a special class of materials known as refractory alloys. This class refers to metals that do not melt or weaken easily, even in extreme heat.
An alloy is a material made by combining two or more metallic elements to achieve properties no single metal can offer on its own – greater strength, for example, or better resistance to corrosion. Refractory alloys are based on elements such as tungsten, niobium and molybdenum, which have some of the highest melting points of any metals.
Their atoms are held together by strong chemical bonds and arranged in a stable crystal structure that resists deforming, even at extreme temperatures. Where conventional alloys begin to soften and slowly deform under constant stress, refractory alloys retain their strength, making them essential for components exposed to extreme heat, stress and radiation.
Most refractory alloys in service today were designed decades ago. They predate modern 3D printing of metal parts, also called additive manufacturing, and artificial intelligence.
To execute metal 3D printing, a laser or electron beam melts successive thin layers of metal powder.
This builds up a 3D part directly from a computer model by adding material layer by layer, rather than using molds or removing material from a solid block. 3D printing allows shapes that are impossible with traditional manufacturing methods. However, many current refractory alloys are difficult or impossible to manufacture reliably using these techniques.
This mismatch can slow the domestic production of new parts. To help address these manufacturing and supply-chain challenges, our team of materials researchers at Arizona State University and UNSW Sydney has formed a new international collaboration to redesign high-temperature alloys.
Old alloys in a new manufacturing world
Additive manufacturing allows defense and aerospace manufacturers to produce complex components locally, on demand and with far less material waste. In principle, it is ideal for producing replacement parts for aircraft, spacecraft and naval systems.
In practice, many refractory alloys crack, warp or develop internal defects when 3D-printed. Their compositions were optimized for casting or forging, not for the rapid melting and solidification involved in laser-based printing. In 3D printing, a laser melts and resolidifies metal thousands of times in quick succession, creating steep temperature gradients that generate enormous internal stresses. Several key refractory metals are brittle at room temperature and cannot absorb those stresses without cracking.
brightstars/Photographer’s Choice RF via Getty Images
Redesigning these alloys using traditional trial-and-error methods would take decades.
Teaching computers to design new metals
Our alternative approach uses reinforcement learning, a form of artificial intelligence best known for training computers to master games such as Go or chess.
Designing a new alloy is a bit like mixing ingredients for a recipe, but at the atomic level. Instead of planning moves on a board, the AI system explores thousands of possible alloy recipes – for example, different combinations of chemical elements. Even tiny changes in the ingredients can completely change how the final material behaves.
The AI evaluates each candidate virtually against multiple criteria, including strength at temperatures above 1,800 degrees Fahrenheit (1,000 degrees Celsius) and resistance to damage caused by reacting with oxygen at high heat, as well as weight, cost and, crucially, whether it can be reliably 3D-printed.
Vitor Rielli
Alloys that should perform well are rewarded, while those that fail are discarded. Over repeated cycles, the system learns which chemical combinations work best.
We can then manufacture and test the most promising AI-designed alloys in the laboratory. Their real-world performance feeds back into the model, steadily improving its predictions.
Strategic benefits beyond the laboratory
The implications of our research extend beyond the lab.
For defense agencies, faster materials development means quicker deployment for next-generation engines, hypersonic vehicles and systems that protect against heat. AI-designed alloys can be optimized for strength, heat resistance and manufacturability. For example, NASA’s GRX-810 alloy, designed with computational methods and 3D-printed, is 1,000 times more durable at high temperatures compared with traditional alloys.
Traditional manufacturing of refractory metals wastes up to 95% of the raw material through machining – removing unwanted material to create the precise shape – but 3D printing can bring that figure close to zero.
Our work is an international collaboration. At Arizona State University, the focus is on AI-driven computational design. UNSW Sydney’s facilities allow for high-temperature testing by looking at the metal’s microstructure and conducting additive manufacturing under realistic conditions.
Challenges still ahead
This approach is not without hurdles. One of the biggest is data scarcity: AI models learn from existing experimental results, and for refractory alloys, that data is limited. Far fewer alloys in this class have been systematically tested, compared with more common materials like steel or aluminum.
There are also practical constraints. Refractory metal powders suitable for 3D printing are expensive and difficult to source, and scaling up from small laboratory samples to full-sized components is difficult. An alloy that performs well as a thumbnail-sized test sample may behave very differently when printed as a large, complex part.
Finally, AI predictions must always be validated experimentally – and those experiments are costly and time-consuming. The system does not eliminate the need for rigorous physical testing.
A new model for defense-focused research
Our collaboration is in its early stages. We are currently building the AI model and assembling the experimental databases it will learn from. Later this year, the first candidate alloy compositions will be selected for 3D printing and laboratory testing. The results will feed back into the model.
We are also working with defense research agencies to ensure our work aligns with real-world needs and to lay the groundwork for larger-scale programs.
In an era where technological advantage increasingly depends on speed and adaptability, reimagining how we design the metals behind defense systems can improve the systems themselves.
Houlong Zhuang receives funding from Security and Defence PluS Alliance.
Vitor Rielli receives funding from Security and Defence PluS Alliance
