This page contains a Flash digital edition of a book.
HPC APPLICATIONS: MULTIPHYSICS


Converging on reality


Multiphysics software can account for multiple interacting effects at the same time and lead to more accurate models, but this approach increases the computational demands. Thus there’s a strong effort to adapt such software to HPC systems. Paul Schreier investigates progress towards this goal


t’s easy to make simple models of real- world systems that focus just on basic concepts. But the real world adds many other aspects that an accurate model can’t ignore. That’s the basic concept behind multiphysics modelling software.


I


A perfect example A transformer is a perfect example, explains Glen Hansen, head of the Multiphysics Methods Group at the Idaho National Laboratory, the Department of Energy’s lead nuclear and energy R&D facility. Physically, there are two sets of windings around a common core to boost or reduce voltage levels. To model an ideal transformer you simply apply Maxwell’s equations at a specific frequency and watch the effect of varying core size and number of turns in the windings. However, when you actually build such a transformer, you’ll find that the operating parameters vary from those the model predicts, because an approximation based only on the simple transformer equation neglects heating, losses and the expansion of materials. In operation, the windings and core heat up and their mechanical properties change, including thermal expansion, and this can have a marked effect on electromechanical performance. This in turn modifies the current and the amount of heat generated. This interaction back and forth between heat and current continues until the model reaches equilibrium.


Another easily comprehensible example of multiphysics concerns fluid-structure interaction (FSI). When a flexible object is placed in a flow of air or liquid, it presents a certain shape and thus resistance to the flow. The object bends, resulting in less


26


surface area blocking the flow, so there is now less force to bend it. The model must account for both effects – structural mechanics and fluid flow – and how they interact before it can converge on an accurate solution.


These simplistic examples are far from what engineers and scientists study today; now we have software that makes it possible to solve a problem that might involve three, four, five or even more phenomena to get a truly accurate model. But consider that even a single-physics problem might take hours or days to solve depending on the problem as well as the resolution and accuracy you need, so you can appreciate that running several physics at once will extend the solution time. Here HPC can make a large contribution to cutting solution times, from days to hours – or even minutes.


Multiphysics from the ground up


When looking at multiphysics software, you want to make sure that the package can couple the physics you need. Many codes can handle common physics such as fluid flow, heat transfer and structural mechanics, but if your problem is more involved, you might have to look further. One of the most versatile multiphysics packages comes from Comsol. Designed from the ground up to handle multiphysics applications, it can couple any of the many types of physics it supports. In addition, it has taken common couplings and prepackaged them into modules to make it easier for users to work with them; among the modules are those for structural engineering, heat transfer, acoustics, MEMS, RF, earth sciences, batteries and


SCIENTIFIC COMPUTING WORLD OCTOBER/NOVEMBER 2010


fuel cells. In addition, users can enter equations for any physics that are not built in. This, says senior VP of marketing Bernt Nilsson, is in contrast to most other companies, which have taken a number of single-physics codes and combined them afterwards into a multiphysics environment.


While multithreading support has been available in Comsol Multiphysics for several years, with the 4.0 release it has added cluster support for Microsoft HPC Server 2008 and Linux. It does so in two ways. First is the use of distributed direct solvers, and the second is in parametric sweeps where a new parameter is assigned to another node. As for performance gains, consider this example: an analysis of a patch antenna running a distributed solver on a Microsoft HPC 2008 cluster with four nodes, each with eight cores, shows a speedup of 12x compared to running on a single node. A final interesting note concerns pricing: users of a floating network license have free, unlimited access to cluster power, because any simulation job can be deployed to any number of clustered computers at no additional cost. The largest supplier of engineering simulation software, Ansys, has gained much of its technology through acquisitions; the best known is likely when it purchased Fluent for its CFD code and more recently Ansoft, which specialised in electromagnetics, circuit and system simulation. Does that mean that codes from different sources won’t work smoothly in a multiphysics environment? Not according to Barbara Hutchings, director of strategic partnerships, who says ‘the core of the Ansys vision has been to grow by acquisition, but with a consistent focus to


www.scientific-computing.com


Page 1  |  Page 2  |  Page 3  |  Page 4  |  Page 5  |  Page 6  |  Page 7  |  Page 8  |  Page 9  |  Page 10  |  Page 11  |  Page 12  |  Page 13  |  Page 14  |  Page 15  |  Page 16  |  Page 17  |  Page 18  |  Page 19  |  Page 20  |  Page 21  |  Page 22  |  Page 23  |  Page 24  |  Page 25  |  Page 26  |  Page 27  |  Page 28  |  Page 29  |  Page 30  |  Page 31  |  Page 32  |  Page 33  |  Page 34  |  Page 35  |  Page 36  |  Page 37  |  Page 38  |  Page 39  |  Page 40  |  Page 41  |  Page 42  |  Page 43  |  Page 44  |  Page 45  |  Page 46  |  Page 47  |  Page 48  |  Page 49  |  Page 50  |  Page 51  |  Page 52  |  Page 53  |  Page 54  |  Page 55  |  Page 56
Produced with Yudu - www.yudu.com