For Karthik Duraisamy, it began with a pair of pictures in a textbook on computational methods.
One was a photograph of a missile that had crashed into a slab of concrete. Its nose was a crumpled mess. The other was a similar image based on a computation of the event, with a resulting crumple that looked remarkably similar to the original.
Duraisamy, then an engineering undergraduate in India, had always liked mathematics, but he realized with sudden clarity what one could accomplish just by crunching the right numbers the right way.
A computational engineer was born.
“In high school we learn about F=ma. You throw a stone and you can calculate the trajectory of that stone. That was cool, but this was just amazing – that you can take such unbelievably complex problems and write some equations and solve them using a computer program. Even more fascinating? the equations are but different versions of F=ma”
Duraisamy’s Computational Aerosciences Lab focuses on computational modeling of complex aerospace problems, with a specific focus on data-driven modeling. In data-driven modeling, real world data is transformed into modeling information, which in turn is processed to develop computer models of engineering systems that would otherwise be prohibitively expensive to build and test.
His analytical nature and penchant for mathematics could have landed him in any number of fields but Duraisamy chose aerospace for the same reason some might look elsewhere – because challenges are more difficult and the stakes are high.
“More than in any other field, the need for precision is immense,” he said. “An airplane is a hyper-efficient machine and one that has to be almost 100% reliable while consisting of more than a million parts. So our designs tend to be cutting-edge and that requires detailed modeling, and that is where some of our research comes in. In fact, many computational techniques for fluid flows that are used in all fields of science and engineering were invented by the aerospace community.. and it happened because those methods were necessary for technological progress in the aerospace industry.”
As part of a NASA grant, Duraisamy is using vast databases of air flow measurements from flight tests, simulations and experiments in wind tunnels to develop a more accurate picture of turbulence around airplane wings.
The work has implications not just in aerospace but in any area of fluid flow. It could just as easily apply to meteorology or blood flow – the generality of the procedure is as important as specific applications to Duraisamy. The students in his lab share a similar enthusiasm and have developed individual expertise in areas that include numerical methods, physical modeling, machine learning and aerospace applications.
He applies the same deliberate, analytical process to any problem he encounters. Understand the problem, consider the possible sensitivities, envision the potential outcomes; consider the probability of each; act accordingly. It doesn’t always help at home, where his wife jokes that, for an engineer, he’s not very practical. It turns out that just because one can consider and weigh every potential outcome doesn’t mean one should.
“Maybe I ask too many questions of the task at hand and I don’t solve it soon enough,” he said. “She ends up fixing it herself, but perhaps I could’ve simulated it better”
But Duraisamy has never outgrown the amazement he first felt when he realized that he could solve a real-world problem by working through an equation on a piece of paper.
“It gives me a sense of joy whenever I solve a problem, even in class. After all these years, it is still uplifting to see how real world problems can be so elegantly solved.. be it the spacing of sand ripples on a beach or airflow patterns on a wing.”
After earning an aerospace engineering master’s degree from the Indian Institute of Science, he found a PhD advisor, James Baeder, at the University of Maryland who had a similar combination of interests in computing, math and aerospace.
He left Maryland with two degrees, an aerospace engineering PhD and a masters in applied mathematics that he picked up along the way. The two fields overlap in his work with numerical methods, modeling and uncertainty quantification.
“Anyone can give you an answer to a scientific problem based on a computation, but as computers are getting more powerful and models are evolving, it is time to ask, ‘OK, but how confident are you in your answer?’” said Duraisamy, who began exploring the question as a research professor at Stanford before moving to Michigan in 2013. “There are some factors that you know well about a physical problem, some that you are not certain about, and then there are some things you don’t know about at all. Somehow you need to account for all of these things in your computational model and still provide predictions and place bounds on your answer. That’s the central issue, and in some sense, the holy grail of computational modeling.”
In his free time, he enjoys traveling, hiking, and going crazy about Maryland Terrapin basketball (which he predicts to be the number one team in college basketball next year, with no uncertainty in his answer.)