Surrogate models simplify complex physics tasks

Physics simulations have a problem – the engineers who need the results of those simulations often don’t have time to wait. Add in real-world settings with many independent calculations required (for example, separate thermal, mechanical, and electromagnetic elements in the system), and, and “Multiphysics” computations It may seem like an either-or proposition that it is both realistic and real-time.

Coders and Modelers gathering in Burlington, Mass. New inroads will be discovered this week COMSOL simulation software environmentOver three days of keynotes, workshops, and demos, COMSOL users will weigh new approaches to solving the simulation time crunch.

,surrogate model There is an interesting new technique where you take your completely multigenerational model and compress it into a compact format that is quick to evaluate using machine learning,” says Björn Sjödin, senior vice president of product management at the Stockholm-based parent company of COMSOL.

The challenge is broader than Comsol alone. according to a Review Published earlier this year in the journal sequential computer scienceA range of industries face simulation bottlenecks, where, the authors say, “performing high-fidelity simulations can take even weeks per design.”

surrogate model, the custom The authors note, the equations below are included for simplified versions of larger simulation environments. In other words, surrogates capture the essential behaviors of specific systems, but without so much computational overhead. Often this trimming-down process may involve strategically sampling the original complex model at critical points, and then training a faster approximation that can predict outcomes for new scenarios.

“You can evaluate these models immediately,” Sjodin says of Comsol’s surrogate modeling system. “Whereas if you solve the full model with unknown inputs, it might take you 15 minutes. And people are very impatient.”

According to Sjodin, European automotive manufacturers are now using COMSOL’s surrogate models to rapidly simulate entire electric vehicle battery packs, enabling real-time decisions that once required managers and engineers on a coffee break or long wait. Meanwhile, Sjodin says, a Swiss institute has deployed the COMSOL Surrogate System as an app for Indian farmers to predict food spoilage in cold storage. Surrogate simulations, the institute found, enabled farmers to reduce food spoilage by 20 percent.

COMSOL’s full numerical simulations predict the performance of an antenna surface (right region), while its streamlined surrogate model (left region) comes to almost identical results with almost less running time.Comsol

Creating Multiphysics in an App

Sjodin says that COMSOL is intended to transform users of simulation systems into something closer to software developers in their own right.

“You can compile those apps into standalone executables that you can distribute around the world without paying any kind of license,” Sjodin says.

The company’s surrogate models, he says, are capable of running as standalone applications, which can work on a laptop or smartphone.

“If you want to give it to someone on the factory floor, these surrogate models are really useful because it allows you to quickly evaluate and get results,” Sjodin says. Models run quickly compared to full multiphysics simulations Because the app version of the thermal performance and chemical composition of a specific battery pack comes pre-loaded. The simulation is fast, because it is already based on pre-calculated parameters specific to the physical environment to be simulated—and only the environment to be simulated.

In addition to AI smarts that speed up computing time for each run, Comsol also relies on other tricks. what modelers say “Low Order” Model (ROMS) includes optimizations such as mathematical pattern recognition and strips down some of the more complex equations in the calculations. ,Neural networks come into play there, but also other technologies, more traditional lower order modeling technologies,” he says.

For example, in 2024 Industry-wide reviewResearchers from Trieste, Italy International School for Advanced Study A range of ROM techniques have been described that are based on more than just AI or neural networks.

“ROMs are divided into two large families: intrusive methods, in which one directly manipulates the governing equations, and non-intrusive methods, in which only simulation data are considered,” the researchers wrote. The paper shows that a mixture of neural nets and more traditional mathematical ROM tools can achieve computational speedups up to 100,000 times faster than models without ROM smarts added.