We asked four colleagues here at COMSOL to forecast what the year ahead might look like in terms of modeling and simulation trends. While they each had their own specific ideas and reasoning, three main themes surfaced across their answers: an extension of the current use of simulation, faster answers via surrogate models and simulation apps, and a desire to build larger models. Read on to see what they said about 2025 and beyond.
Prediction #1: Back to Basics — but Better
David Kan offered up an analogy between simulation and fashion, of all things. “Trends are funny things. Trends are things that not only look forward, but they also look back. Like in fashion, often what’s old becomes new again: the retro style, the vintage style… What’s happening in simulation is very similar, where we’re going back to the basics and doing harder, better, and more challenging problems with the tools that are in place. We’re getting back to the fundamentals and doing those models in a more rigorous way, because we have better computational tools now,” said David.
Does that mean he thinks that modeling and simulation is stuck in the past? No, not at all. David clarified that thanks to increased computing power and GPU technology available today, it is possible to iterate faster than before. “It’s worth remembering that we have this firm foundation of core fundamental computational technology that we’re using to do that,” he added.
Hardware and software have both come a long way since the early 2000s.
Along those same lines, Andrew Strikwerda shared his take: “The biggest trend in modeling and simulation that I’ve seen is simply a proliferation of simulation to more and more areas. More people are interested in getting simulation results and more people are realizing the value of those results on what it is that they’re working on,” Drew said.
Ruud Böger echoed that sentiment. “In 2025, I think modeling and simulation will evolve in various ways. We see that there are large investments in the semiconductor industry where the major world powers want to localize their supply chain. We also see that there is a lot of demand for simulation in renewable energy right now,” he said. Ruud also predicts that the simulation software itself will become easier to use, which should help make it more prevalent.
Prediction #2: Faster Answers Through Surrogate Models and Apps
There’s no denying that today’s society expects everything to happen quickly. We crave same-day delivery, high-speed internet, and instant gratification. It’s only natural that the same expectation would be applied to simulation software.
Andrew brought up the concept of simulation apps and how COMSOL Multiphysics software users already build easy-to-use apps and distribute them for anyone to use. He went on to say that “We’re going to see more apps that are powered by surrogate models. These are models that you can train on underlying datasets so that you can get very, very fast results through the power of neural networks that you can train in COMSOL Multiphysics on data generated by COMSOL Multiphysics.”
An app that delivers lightning fast results thanks to a surrogate model.
Ed Fontes agreed and expanded on that. “What I see is the biggest trend right now is the ability to create high-accuracy surrogate models. Like what we’ve introduced in COMSOL Multiphysics now: deep-neural-network-based surrogate models. And those are used in so many instances: you can use them for system modeling, uncertainty quantification, and design of experiments. These types of studies require a huge amount of simulations, and full 3D models would be too heavy to use. So what you do is basically invest in running these 3D models to train a surrogate model or a lumped model, and then you use that for the heavy-duty for system modeling to implement it into an app,” he shared.
The general workflow for training a surrogate model based on a full 3D model.
“If you think about apps, right, you want the answer immediately. It has to be almost interactive,” Ed continued, “With a surrogate model, trained on the full 3D multiphysics model, you can actually get the answer in milliseconds. And so everybody that has to spend ten minutes waiting for the results, that’s not what they expect, you know, in this age. That’s the biggest trend: even though you’re dealing with huge 3D models, you get the answer immediately.”
In support of apps, Ruud said that “More and more people are making simulation apps. This makes it possible for people who don’t have a background in numerical simulation to still get the benefits of simulation. I think that more apps will be deployed for people to use simulations [in 2025].”
Prediction #3: Larger Models
Ruud is also predicting that folks will be building larger models. He shared that whereas “People used to do simulation of a single component, recently, they do multicomponent simulations or large parts of, let’s say, an equipment manufacturing machine or even a full hearing aid, for example.”
An acoustics model of a complete hearing aid.
In a similar vein, David indicated that he thinks that “this is going to be a matter of continuing the theme of more computational power, better computational algorithms, producing more, giving the engineers more ability to do many, many more iterations of their simulations. And through that, they are able to discover more facts.” He continued to say that “People use COMSOL for many reasons. One is to understand new physics. One is to, use it for production, for R&D and design. And as we go on, we need more and more computations to be done. And that’s the biggest evolution that will continue to happen in 2025.”
Plus: Looking Beyond 2025
We also asked them to look beyond 2025, into the long-term future. Ruud and Ed both considered how the broader population might use simulation in the future.
“Well, the long-term future of modeling and simulation, of course, is difficult to predict,” responded Ruud. “I do see that the usage of simulation apps is growing in terms of how they’re being applied. So right now, we see that design engineers and manufacturing engineers use them. I think consumers will use them more and more as well. Right now, if you, let’s say, paint your house, you use sort of a rule of thumb: every five years, you paint the walls that are exposed to the sun. But maybe in the future when you buy paint, you also get a simulation app, which allows you to take a picture of your house and then does a simulation in the background to see which areas are exposed to what degree, then run a chemical simulation to see how long does this paint last and its conditions and its location,” explained Ruud.
Ruud continued to say that “There are many examples like that. For example, when you buy a router, what is going to be the electromagnetic field in every room in your house? How many do you need? Where do you need them? It’s similar for, let’s say, acoustics, for speakers. I think there will be a lot of consumer applications where, right now, we don’t really know that it’s possible, but there is a big need for it. And I think that demand will be met in the future.”
Ed suggested that the long-term future of modeling and simulation would be along the same lines as the 2025 trends. “Those new methods, those new abilities to actually obtain a reply immediately, that will make a simulation more useful,” he said. Referencing an article about farmers using simulation apps to forecast the freshness of fruit and produce, he went on to say that “Nobody in their right mind would expect to use modeling and simulation for that kind of application ten years ago. It was too heavy duty. Who would run that on a supercomputer? Now you can just run it on a server and it doesn’t require a lot of power.”
“Well, with these new technologies and with surrogate models and so on, this will spread much more into consumer applications where you want to know things in advance, where you want to predict things in advance, and that’s where modeling and simulation can do a good job if you have validating models. We will not develop these type of apps, but our customers will do that. So it’s secondary app users. And that’s something that we foresaw already twenty years ago.”
Comments (0)