I’m excited to highlight some progress GE Research has made in modeling the formation of ice from water droplets in contact with cold surfaces. For several years, a multi-disciplinary team of researchers at GE Global Research has been developing “icephobic” surfaces. We have observed that certain types of surfaces hinder ice formation, but the exact mechanism was unknown. We use simulations as a means to gain insight into the conditions under which ice can be suppressed. Many industrial systems that operate in cold environments stand to benefit from resisting ice including wind turbines and offshore Oil & Gas drilling and production rigs operating in extremely cold environments. I sat down with Dr. Masako Yamada who works in the Advanced Computing Lab I lead and discussed recent results she has collected in computationally modeling the formation of ice (also called “ice nucleation” – as solid water is a crystal that grows from a “nucleus” that forms out of the liquid.)
We are running these simulations on the Titan Cray XK7 supercomputer at Oak Ridge National Lab – why do we need so large a machine?
The computational technique we use, molecular dynamics, is notoriously time-consuming. “Molecular” means we track the position of every single water molecule. “Dynamics” means we calculate very short slices of time (specifically: femtosecond slices – [Wikipedia: a femtosecond is to a second what a second is to about 31.7 million years].) It’s analogous to creating a high-speed video using an atomic microscope. Titan is one of the few resources in the world that can handle our need. We have won 80 million CPU hours through the Department of Energy ASCR Leadership Computing Challenge to run these calculations. We are constantly pursuing higher performance. We recently achieved 5x speedup by converting our code to run not only on the CPUs but also the GPUs (graphics processing units) as “accelerators.” Even so, we can only model water droplets that are about 50 nanometers in size (far smaller than real world droplets) and we still cannot run our models to simulate as long a time period as we would like.
Simulation of ice spreading through a water droplet. This is not just an animation/cartoon. It’s a real scientific model that’s being developed on Titan, the #1 ranked supercomputer in the USA. Video credits: Mike Matheson (Oak Ridge National Lab)
How are icephobic surfaces typically developed?
This is a field of tremendous interest across industry, academia and the government labs. At present, experiments drive most of the science. Usually, many candidate surfaces are fabricated and the effectiveness of the surfaces is evaluated by placing water in contact with the surface and measuring the depression of freezing temperature, delay in onset of freezing, or reduction in adhesion strength of ice. The experimental setup can be as complex as using microdroplet generators, high speed cameras, thermocouples, wind tunnels and centrifuges… or as simple as putting different surfaces outside in the winter and visually comparing ice buildup.
How do we make sure a computer model gives us a good result compared to a physical test?
Nobody trusts modeling data on its own. We would typically model a “dummy” system that is simpler than the target system and compare against experimental results. We might also change just a few variables (such as temperature and pressure) to confirm that the trends are in sync with experiment. Only then can models be used for prediction. Nature is very complex and results from the models might not exactly match experimental results. However, highly complex simulations that were previously inaccessible are now being executed thanks to the growth in supercomputing power.
360 view of the ice crystal inside the water droplet. Video credits: Mike Matheson (Oak Ridge National Lab)
What advantages are there to virtual engineering test over physical tests?
In the virtual world, we can monitor the position of every single molecule in femtosecond increments. We can see exactly how the water molecules interact with the surfaces. This is simply impossible using any physical test. In addition, in the virtual world, the results are not impacted by dirt, defects and other random sources of noise. These imperfections certainly need to be accounted for in real life, but in the research stage, it’s helpful to be able to develop the surfaces in a perfect environment.
How would icephobic surfaces impact everyday sorts of problems like ice cream scoops or car windshields?
Icephobic surfaces can alter cold water contacting a cold surface by: 1) lowering freezing temperature, 2) delaying onset of freezing, 3) reducing adhesion (stickiness) between ice and surface, and/or 4) bouncing water droplets off before they can freeze. Surface composition and features may impact one, some or all of these effects. For the ice cream scooper, since the ice cream is already frozen and it’s not moving, the primary function would likely be (3). (Although there could also be aspects of (1) that might be part of the process, as depressing the freezing temp by even a few degrees would be helpful. For the car windshield, depending on whether the car is moving or still, and whether the water is liquid or ice/snow, and ambient temperature, any one of the four functions would be involved.