With today's advanced technologies, it might seem that engineers design ship propellers, jet turbines, and scramjet engines using precise engineering methods that always yield perfectly efficient designs. But in fact the turbulent flows that occur in these devices are so complex that engineers are still forced to depend on trial and error experience and use approximate parameters in tasks from designing jet engines to predicting pollutant dispersal from industrial smokestacks.

Now, Krishnan Mahesh, associate professor in the Aerospace Engineering and Mechanics Department of the University of Minnesota, and his group are applying the enormous power of SDSC's DataStar and TeraGrid resources to conduct simulations of unprecedented realism. “Traditionally, such high-fidelity simulation methods have been restricted only to fairly simple geometries,” said Mahesh. “But massive parallel computing platforms such as DataStar at SDSC have now made it possible to simulate complex flows that would have been inconceivable a decade ago.”

What is novel about Mahesh's work is that the researchers have developed numerical methods and turbulence models that are flexible enough to handle real-world engineering geometries without compromising the accuracy needed to reliably simulate the complicated details of turbulence.

These simulations open the door to understanding the intricate details of turbulent flows in a range of real-world applications from ship propellers to jet engines and exotic hypersonic scramjet aircraft engines. This understanding can help engineers design better devices, as well as guiding researchers to new experiments that work in tandem with these high quality simulations to build greater understanding of these important engineering problems.

Propeller Crashback

Imagine a large ship steaming at full speed. Suddenly, the ship needs to stop or turn, and the captain throws the propeller into full reverse. This dramatic reversal produces large fluctuating forces that can sometimes break a propeller blade or impair the ability to steer the vessel.

While current engineering methods are adequate to simulate a propeller operating under normal design conditions, they are inadequate to model the unsteadiness in blade forces encountered in experiments and the real-world emergency crashback maneuver. To give engineers a simulation approach that can capture the full complexity of propeller crashback and help them predict required blade strength, Mahesh and graduate student Martin Vysohlid, supported by the Office of Naval Research, are performing Large-Eddy Simulations (LES) of the complex geometry and the 3-D flow around a reversing propeller. The simulations show low-frequency unsteadiness similar to what is observed in experiments, and torque and thrust coefficients significantly closer to experimental values than previous methods.

This project demonstrates the potential of Mahesh's simulation methodology to predict off-design conditions in marine environments, and the researchers plan to extend this work to investigating another hard-to-predict problem, blade cavitation, the disruptive bubbles that can form and collapse on propellers under extreme conditions.

Simulating Scramjets

As the Space Shuttle ages and the nation searches for cost-effective and reliable access to space, one promising technology is known as the scramjet, a supersonic combustion ramjet that can fly ten or more times the speed of sound at very high altitudes on the way to space. NASA successfully tested a scramjet in 2004.

But the design of such exotic engines is challenging, involving compressible turbulence in super- and hypersonic combustion and the interaction of shock waves with turbulent boundary layers, problems that defy current simulation capabilities. To give engineers tools to model these extreme flow regimes, Mahesh and graduate students Yucheng Hou and Jeffery Doom have developed a novel computational algorithm that can handle the special challenges of these extreme flows. This work is performed as part of the AFOSR-supported Center for Hypersonics at the University of Minnesota.

Jet in a Crossflow

Another important NSF-supported simulation, which Mahesh has run for more than 45,000 hours on SDSC's DataStar, is a jet emerging into a crossflow, for example, a plume of smoke rising from a smokestack. The researchers want to understand this in order to be able to reliably predict such problems as how far pollutants will spread from a smokestack and to help engineers design better gas-turbine combustors and fuel injectors.

Graduate student Suman Muppidi has used exact solutions of the flow equations, without approximations, known as known as Direct Numerical Simulations or DNS, to study how jets mix with crossflows. A key result the researchers have found is a significantly better scaling law for the jet trajectory, that is, they can better predict how the jet will curve as if moves out into the crossflow, in relation to such factors as the jet's speed and size and the speed of the crossflow.

Massive Computers

“SDSC supercomputers and staff have been central to our group's ability to achieve these important new research results,” said Mahesh. In the last year, his group has used 88,439 hours on DataStar, running on up to 1,024 processors; nearly 30,000 hours on the TeraGrid IBM P690 nodes at SDSC; and more than 100,000 hours on the distributed TeraGrid facility. “The fast I/O capability of DataStar is a very important feature in efficient execution of our large-scale simulations,” said Mahesh. “In addition, all our simulation results are archived in the multi-petabyte archiving facility that SDSC provides.”

In their simulations, analysis, and modeling of turbulent flows, Mahesh and his group focus on fundamental advances in both numerical algorithms and greater understanding of the flow physics. Computational methods include Large-Eddy Simulations (LES), which model the effect of small eddies on the larger-scale flow, and Direct Numerical Simulations (DNS), which solve the full Navier-Stokes equations describing the flow without approximations, using unstructured grids on massively parallel computing platforms.

In addition to the above projects, other simulations Mahesh's group and collaborators have conducted at SDSC include a first-time simulation of the turbulent flow inside a commercial gas-turbine combustor in which they developed their simulation methodology, which is now being adopted by industry. The success of these simulations is encouraging Mahesh and his group to proceed with further research, and in addition to using DataStar and the TeraGrid they also plan to conduct simulations using a one million hour allocation on SDSC's new Blue Gene supercomputer.