From Alliant to Forge

02.28.11 -

by Allison Copenbarger and Barbara Jewett

NCSA supercomputers support years of steel research

For many people, a first summer job may be a memory best forgotten. But Brian Thomas' first job at a steel company proved to be both memorable and life altering.

When Thomas was a 20-year-old undergraduate student majoring in engineering, he worked at a large Canadian steel producer and became fascinated by the process of continuous casting steel—the process that turns molten steel into sheets and that is used to produce 92 percent of the world's steel. He decided he would like to use his engineering skills to refine the process.

Thomas went on to earn a PhD in metallurgical process engineering at the University of British Columbia and took a position on the mechanical engineering faculty at the University of Illinois. Now, 26 years later, Thomas' research is regarded as some of the top in his field and he is the director of the Continuous Casting Consortium (CCC), a cooperative research effort between his research group at the university and the steel industry.

With annual global steel production at almost 1.5 billion tons (100 million tons in the United States, 96 percent of that through continuous casting), steel production accounts for an important fraction of the total energy consumed and greenhouse gases produced in the world. Even small improvements to this process, says Thomas, can have a profound benefit to society.

"For me the idea is to take this process which looks to outsiders kind of Dark Ages—crude and as empirical as you can get—and combine that with high technology," he says.

HPC changes an entire industry

High technology is where NCSA comes in. Thomas began his career at Illinois around the time NCSA opened its doors, so his research has been linked to the center since its beginning. "We've sort of grown up together," he says, noting he's used every computer from Alliant (and Cray X-MP) to the new Forge.

And the center has provided Thomas more than just compute cycles. NCSA research scientist Seid Koric has worked closely with Thomas for over 10 years to develop better algorithms for modeling thermal stress in steel continuous casting. After joining NCSA, Koric began work on an engineering PhD and became a member of Thomas' research group; the two have collaborated ever since. This unique research relationship with an NCSA collaborator has produced many important tools and findings.

Many steel processes, such as continuous casting, involve multiple coupled phenomena, including fluid flow, heat transfer, solidification, distortion, and stress generation. The computational complexity of the phenomena is so high that it challenges the capabilities of even the best numerical methods and computer hardware. Consequently, there is a growing need to use high-performance computing resources and better algorithms to make these simulations feasible. To enable realistic quantitative predictions of the formation of defects in these commercial solidification processes, high-fidelity simulations are needed, and with today's compute resources they are now starting to become possible, says Thomas.

Koric and Thomas have developed several ground-breaking numerical methods for solving highly nonlinear, complex multiphysics problems that they've applied to their simulations to gain new insights into the continuous casting process. For instance, an algorithm for constitutive-equation integration was implemented into the commercial finite-element code Abaqus, which improved the code's performance in solving solidification stress problems on HPC machines by more than an order of magnitude. For the first time, it demonstrated and verified the significant advantages in scale-up for large three-dimensional problems on parallel computers.

And a new enhanced latent heat method was recently developed by Koric.

The steel-making process

Though the process of making steel "has been around for millennia," says Thomas, it is very intricate and complicated to continuously cast steel into sheets.

The steel is first produced in batches and delivered in ladles. The steel flows from the ladle into the tundish via a nozzle to protect the metal from oxidizing in the air. The tundish, a pouring box that holds the molten steel and then continuously delivers it to the mold, serves to remove most of the harmful inclusions.

In the mold, the molten steel freezes against water-cooled copper walls. The mold oscillates vertically to prevent sticking. The machine also withdraws the solidified metal from the mold through the bottom at the same rate that the molten metal flows into the mold—sensors control this perfect rate.

After leaving the mold, the metal is solid only on the outside. It is transported between support rolls and sprayed with water until the center is completely cooled. Then the steel is cut to the desired size. Thomas explains that a slight change in any of these stages could create vast differences and defects in the final product. For example, changing the geometry of one nozzle can make a defect appear or disappear.

Or if the fluid flow and solidification are not perfectly set, there can be a "breakout," in which molten steel pours through a crack in the partly-solidified casting, covering the area with superheated molten steel, which is very dangerous for the workers and costly for the steel plant.

Computer modeling aids industry

Thomas and his team have modeled all aspects of the continuous casting process: turbulent fluid flow, transfer by nozzles, heat transfer, stresses and strains, and solidification. According to Thomas, one important success has been the pioneering work of understanding the mixing of different steel grades.

Higher-quality grades of steel are very expensive, so contaminating them with a different grade would produce a costly, low-quality product. But stopping and restarting the process is also costly. Thus, casting different grades together in sequences benefits from the predictions of computational models of the mixing. Steel production is a high-volume, low-profit-margin industry. Improved efficiency and consistent steel quality, relative to low-cost alternatives, means producers realize more profit. Better understanding of how defects form, explains Thomas, allows steel continuous casters to produce steel with fewer defects and operate at a higher casting speed, which in turn enables higher production rates (and accompanying efficiencies), and also yields better energy efficiency, such as less furnace time needed after casting to reheat the slab back up to rolling temperatures.

"What most people don't realize," says Koric, "is that Brian Thomas is THE go-to guru for the steel industry. Everybody in the world comes to Brian and the University of Illinois to solve their problems, not just the producers in the United States. And he really is probably the only person who can help them."

Another ongoing project is computational fluid-dynamics analysis of the turbulent flowing and solidifying steel to predict defects such as internal inclusions. The process includes using magnetic fields to manipulate the steel flow. This technique for controlling the flow of the molten steel also may improve its quality.

"Everyone knows if you properly apply magnetic fields then molten steel will change its flow direction, and may improve its quality," says Thomas, "but it's extremely difficult to do measurements of molten metal under magnetic fields to optimize this process—it's a very harsh environment to physically venture into with sensors."

Casting steel with fewer defects makes safer steel products. Thomas says with the improved modeling methodologies, and the application results obtained with those models, his team has provided a better understanding of how defects such as internal inclusions and cracks form during the casting process (including the formation of breakouts) and improved the design of key casting variables, such as nozzle geometry and mold taper, which enable windows of casting parameters to produce steel with fewer defects.

The results obtained using supercomputers have been described in over 100 technical publications and have been implemented by the member steel companies of the CCC. In addition, Thomas teaches short courses on the continuous casting process to industry. Koric was given an adjunct faculty affiliation with the Mechanical Science and Engineering Department in 2010 for his extensive research collaborations and HPC support for Thomas and other MechSE faculty members. Thomas also teaches about the continuous casting process and computational models in courses to his students.

Decreasing the material scrapped due to defects such as cracks, even by a small percentage, results in a large net savings to steel manufacturers and customers. Based on the roughly 100 million tons of steel produced each year in the United States and approximately $400 per ton net cost of scrapping, a 1 percent reduction in yield loss would save about $400 million per year. Increasing casting speed and decreasing spray cooling to conserve just 10 percent more of the internal energy of the strand would produce energy savings during reheating of $350 million per year (based on $0.06 per kilwatt hour) and an associated decrease in emissions.

"We want to find out ways to make the steel as safe and high quality as we can, and at the same time be very cost competitive," Thomas says. "My expertise is on the computational modeling side, but I want to make a real world impact."

This project was funded by the Continuous Casting Consortium members and the National Science Foundation.

For more information, go to: http://ccc.illinois.edu/.

Team members:
Joseph Bentsman
Seid Koric
Brian Thomas
Pratap Vanka

Student team members:
Rajneesh Chaudhary
Seong-Mook Cho
Lance Hibbeler
Chuanbo Ji
Yonghui Li
Rui Liu
Bryan Petrus
Matt Rowan
Kun Xu