Meet a RC Researcher- Peiyuan Liu


From left: Kevin Kellogg, Casey LaMarch and Peiyuan Liu

Peiyuan Liu, Casey LaMarch, and Kevin Kellogg are entering their fourth year of research on cohesive particle flows at the University of Colorado, Boulder. Liu began this research when he completed his Ph.D. with a background in particle cohesion; LaMarch earned her Ph.D. in granular flow, but focused on friction rather than cohesion as a graduate student. Kellogg, who is currently a graduate student, gained his experience in computational research at the Colorado School of Mines.

            The team’s goal is to develop a model of cohesive particle flows. LaMarch explains, “Right now, a lot of the particles used in industry are cohesive, like flour. A lot of industry relies on empirical correlations to figure out how materials will react to storage and transportation, and sometimes that correlation comes from a different system and a different material than yours.” Pharmaceutical companies, oil refineries, chemical companies, and energy companies using particles for solar heat transfer operate at approximately 30% efficiency in their work with particles due to unknowns about particle behavior. “It’s pretty staggering,” La March says. Once created, the model will allow for enormous improvements in industrial efficiency.

            Liu’s focus is the use of the Discrete Element Method (DEM) model on the Janus supercomputer. This model can simulate the behavior of one million individual particles, balancing their forces and tracking their positions and interactions. One of Liu’s simulations requires two months of continuous run-time on the supercomputer, using 100 central processing units. LaMarch performs actual experiments with particles, and her results help the team evaluate and adjust the model. LaMarch uses ten trillion particles in each of her experiments, creating some discrepancies between her results and Liu’s; limits in computational power prevent Liu from increasing the size of the DEM model. To bridge this gap in size, Liu and Kellogg are working to develop a continuum model, which would use average values from the detailed DEM model to simulate particles as a continuous material, rather than as individuals. Once its code is written, the continuum model will also run on Janus.

            The model is very accurate in simulating particle cohesion when it is able to include all relevant inputs. Tiny variations in characteristics such as surface roughness, particle shape and humidity impact particles’ cohesive properties. Kellogg says, “A lot of time is put into making sure all these inputs are correct. Peiyuan has been very careful to construct a model that takes all of them into account.” Feedback between LaMarch’s experimental results and Liu’s models allows both parties to improve continuously, determining whether inconsistencies result from improper experimental measurement or missing inputs in the model.

            Although Liu’s research group has access to a smaller supercomputer called Heisenberg, Liu prefers to use Janus because it gives him access to a proprietary compiler, which speeds up the process of translating his code into a language the computer can use to execute his commands. Using Heisenberg’s open-source compiler is far more time-consuming. Using Janus also gives Liu access to the Research Computing staff. He says, “When we have issues, we ask Research Computing for help. If we cannot compile our code, or we need software to be installed, we call them in.” Liu is working on another project with Thomas Hauser, the director of Research Computing, in which he is attempting to improve the speed and efficiency of the DEM model. Liu continues, “We have had a very good experience with the Research Computing team. It’s very helpful and it’s free, and that’s very, very good.”