Northeastern scientists introduce a new cost- and energy-efficient computational model for designing alloys that factors in material defects.
Humans began creating alloys around 5,000 years ago by combining copper and tin to produce bronze. Since then, alloy design has advanced dramatically, says Moneesh Upmanyu, professor of mechanical and industrial engineering at Northeastern University.
“Now, it’s definitely a science [and] less of an art because we have the periodic table and we know the properties of all these elements that we are mixing together,” he says.
The Journal of Applied Physics recently selected Upmanyu’s new research paper on alloy design as an Editor’s Pick.
The paper introduces a new computational model that offers strategies for alloy design of real materials in seconds. Compared to traditional lab experiments and AI-based approaches, the model offers greater speed, cost efficiency and accuracy.
The work was conducted in collaboration with Changjian Wang, a former Northeastern graduate student.
Previous computational tools — including those based on machine learning and artificial intelligence — often failed to account for a critical factor, Upmanyu says: real-life crystalline materials, such as metals and ceramics, contain defects.
In materials science, defects are irregularities or imperfections in a crystal’s atomic structure. While they may sound like flaws, defects are often intentionally introduced to enhance properties such as strength, conductivity and corrosion resistance.
The new model takes into account an important class of material defects (grain boundaries) and the tendency of the mixed solutes to gather — or segregate — around the structural imperfections during alloy formation.
“You are dealing with these defective materials by default, and all these alloy design techniques ignore this,” Upmanyu says. “They just can’t factor that in because it’s a very complex system with all these defects in place.”
One well-known example of such material defect that has been studied extensively over the past century, Upmanyu says, is a dislocation. It occurs when an entire atomic plane is missing from a crystal’s structure. Despite this imperfection, dislocations allow for plastic deformation of a material without breaking by letting the defect move through the crystal lattice.
When alloys are formed by mixing with solutes, or dissolved substances, the dislocations act as preferred locations for the solutes. The solutes attach to dislocation threads like a swarm of bees, making it harder for dislocations to move. By engineering these defects and behaviors of solutes in alloys, Upmanyu says, humans can make stronger, cost-effective materials.
His research focuses on another key defect: grain boundaries. These occur in polycrystalline materials — such as copper — at the interfaces where differently oriented crystal grains meet. Unlike dislocations, these defects run along surfaces within the material.
“For a crystal small enough to hold between your fingers, conventional alloys with micron-sized grains have a grain boundary area as large as a basketball court,” Upmanyu says.
That is a vast area for solutes to attach to, he says, which affects the entire mixing strategy when alloys are being made as well as their mechanical, electrical and magnetic properties.
Material engineers often manipulate these boundaries to control, for example, the direction of electricity conduction, by orienting the grains in the crystals along one direction.
“The motion of grain boundaries with the solutes is completely ignored in current general alloying theory,” Upmanyu says.
His model examines how solutes affect that motion.
“If I look under a microscope at finite temperature (non-zero absolute temperature that affects the energy state of a system), these grain boundaries and these defects are not static, they are dancing around, they are moving,” Upmanyu says. “And we exploit these fluctuations of the grain boundaries with solute segregated to them.”
The model tracks how much and when solute segregates, and how it impacts the motion of grain boundaries.
“Which is a first step toward understanding how the properties of the material are modified by these solutes at the grain boundaries,” he says.
The paper focuses on steel, an alloy of iron and carbon. However, Upmanyu notes the model applies broadly — not just to metals but also to ceramics like metal oxides.
“We feel it’s general enough because it’s based on fluctuations of interfaces and grain boundaries. All these interfaces fluctuate at finite temperature,” he says. “There is always segregation of solutes. It’s universal.”
To reflect this broader scope, the researchers use the term “interface” rather than “grain boundary” to include non-crystalline materials.
The model realistically simulates how solutes interact with both defects and each other.
“If I take a snapshot of what we actually have simulated and use it as an input to actually extract the alloy properties, it’s identical to what you see in an experiment,” Upmanyu says.
The model works with two or more base materials and can be extended to predict thermal, electrical and magnetic properties of resulting alloys.
Another advantage: it delivers accurate predictions using very short simulation times.
“We’re looking at a computational investigation of this fluctuation over nanoseconds,” Upmanyu says. “You’re taking a very brief snapshot of how this thing fluctuates, and coming up with the modified behavior based on that.”