Metin Sitti and graduate student Zhou Ye (now at Intuitive Surgical Inc.) of Carnegie Mellon University have been designing microrobots that can do simple tasks such as creating tiny current flow in a liquid and picking up and dropping microscopic objects. Most of their work had concentrated on creating microrobots and testing them under a microscope. But they had also begun simulating their robots on a computer, to enable them to accelerate testing without expanding their cost and manpower needs. However, the “finite-element method” computational approach they were using was taking days of calculations to do even a simple simulation on the computer in their lab.
“The boundary-element method cut simulation times from days to minutes, even on their lab machine. With the supercomputer, they can do ultra-high-resolution or complex simulations.”—Anirban Jana, Pittsburgh Supercomputing Center
How PSC and XSEDE’s ECSS helped
To speed their simulations, the scientists turned to Anirban Jana, an XSEDE Extended Collaborative Support Service (ECSS) expert at the Pittsburgh Supercomputing Center (PSC). Jana looked at their computational approach and realized that a different method, called the “boundary-element method,” could greatly simplify their calculations and provide more accuracy in addition. By happy coincidence, Carlos Rosales-Fernandez of XSEDE-member the Texas Advanced Computing Center had years earlier written an open-source code for a very similar boundary-element simulation of objects in anarrow, fluid-filled channel. Jana rewrote the code to simulate a more general fluid-filled cavity, helped the CMU team develop workflows to make the calculations run smoothly and “parallelized” the code to make it run on a supercomputer. He also helped them write an XSEDE proposal to get computational time on PSC’s Blacklight system.
The change to the boundary-element method allowed the computations to run in minutes rather than days, even on the scientists’ own computer. When they tested the new code on Blacklight, the program ran successfully in an environment that will allow them to simulate vastly more complicated—and more realistic—systems and environments. A paper on their initial work is now in review in a peer-reviewed scientific journal, with Jana a co-author.