Only multiphysics meets future requirements

Share this on social media:

We must move beyond software designed to solve single physics problems, according to Lars Langemyr

We have long known about the complex interactions among various areas of physics, and these interactions are becoming more complex. For instance, a structural-mechanics model in the area of MEMS (microelectromechanical systems) must take into account thermal strains, fluid interaction, and forces induced from electromagnetic phenomena, all of which typically interact in a nonlinear way. Only recently have we developed tools to model such complex interactions accurately. The required multiphysics modelling software, and the affordable computing power that makes it accessible to any scientist, continue to evolve.

Scientific-modelling software has made great strides in recent years, an important fact given that some packages based on differential equations go back two or three decades. Such legacy packages were written to address a single area of physics such as structural mechanics, electromagnetics, or fluid mechanics. Further, they're written in programming languages such as FORTRAN and thus represent a technological dead end - they're extremely difficult to update to take advantage of modern methods and techniques. We believe this new generation of scientific-modelling software brings the ease of use associated with a sophisticated graphics user interface.

Concentration distribution in a turbulent mixer. This tracer injection simulation couples the k-epsilon turbulence model to a transient analysis of the convection diffusion equation

Furthermore, these general-purpose packages simultaneously solve multiple areas of physics. Emerging fields such as MEMS, nanotechnology, and bioengineering, as well as virtually every established field of engineering, need software that can couple various types of physics in an equation-based modelling environment.

Legacy analysis packages for a single physics - whether structural mechanics, fluid mechanics, electromagnetics or any other field - were optimised for that one area. Their problem-setup mechanisms and solvers are pre-programmed and offer little, if any, flexibility. It might be possible to partially solve multiphysics problems with legacy software, but the task is cumbersome. The scientist must perform a number of iterations, solving one physics model at a time and then using that solution as the input to the next model. This tedious job is prone to errors and by no means guarantees success. It's also difficult to modify a model to account for differing interactions among the physics phenomena.

We have designed the latest generation of modelling packages from scratch for multiphysics analysis. Developers take advantage not only of the latest algorithms, solvers and numerical techniques, they also apply programming languages and tools to make their software faster and more memory-efficient than ever.

Even though legacy software might be highly optimised with numerical methods for a particular physics, we are confident that today's multiphysics packages incur no performance sacrifices. Independent benchmark studies show that equation-based multiphysics modelling packages such as Comsol's FEMLAB are just as capable as single-physics software. That's no surprise because specialist packages solve the same basic equations with state-of-the-art solvers, most of which arise from academic research. So a general-purpose package that uses the same algorithms for heavy number crunching should be as fast and memory-efficient.

Memory efficiency is important because realistic multiphysics problems are getting large. Fortunately, developments in the computer industry make it possible to solve them with affordable hardware. For instance, 64-bit computing platforms have dropped in price and are available for typical desktop environments such as Windows, the Mac OS, and Linux.

We've proven that 'general-purpose' doesn't necessarily mean that multiphysics packages must be difficult to set up for a particular problem. Advanced packages come with model libraries, whose entries are preconfigured for common problems. Further, comprehensive packages supply physics-specific libraries or modules that have key equations already programmed in, but in a way that someone can change the equations and other parameters as necessary. Candidates for such modules include structural engineering, chemical engineering, electromagnetics, heat transfer, geophysics, fluid dynamics, acoustics and even the newest branches of study including MEMS.

Another essential tool advanced packages bring is full scripting, which enables the automation of parametric studies and optimisations. While legacy programs have command languages, they generally lack subroutines and other programming structures found in a full-blown programming language, whereas advanced scripting tools include these features.

We're only starting to see multiphysics modelling packages scratch the surface. We'll soon see them perform full-system simulations as would be useful in the process industries. Users will be able to model separate processes, each with a completely different set of multiphysics, into one meta model for the factory. Clearly these tools are the future of scientific modelling.

Lars Langemyr is VP of development at Comsol AB