In Stay Sharp Episode 91, Unpacking Digital Twins Series: Simulation, co-hosts Jonathan Scott and Juliann Grant welcome Shawn Freeman, VP, Technical, North America, for VIAS3D, and an expert on product simulation. Unlike recent episodes in the series, this discussion focuses on physical product twins, as well as what kinds of simulation is being done and how that technology fits in with the digital twin.
Where Simulation Meets Digital Twins
Shawn begins by explaining that traditional simulation, or computer-aided engineering (CAE), isn’t really part of a digital twin, but plays a supporting role. That’s because CAE typically consists of long-running simulations “tailored towards a specific use case in a specific configuration of the product,” he says. “They’re not looking at the entire feasible realm of the product or the product’s performance, which is really what you’d want in in a digital twin.”
The example he gives is vehicle performance and safety simulation. Typically that’s looking for the performance of the structure—how it absorbs energy and protects an occupant—which can be done within vehicle crash simulations. But to test a more qualitative measure, such as noise, vibration, and harshness (NVH), requires running many different simulations, each with specific criteria like temperature or atmospheric pressure plugged in. A digital twin, however, facilitates gathering a variety of criteria and plugging it into a collection of complex simulations—aka, a surrogate model.
Surrogate Modeling
Surrogate modeling is a way to create an approximation of a complex system, which allows for faster analysis, optimization, and exploration of different design scenarios. The result is a different kind of model, one we can run quickly and get answers from. “Your CAD geometry is a representation of something in the real world and your simulation model is a representation of that and that surrogate’s representation,” Shawn says. The point is to get a much faster response, so you can reduce time between iterations. He adds, “There are different forms of surrogates, but they all serve the same purpose, which is can I get a quicker run, a quicker response to my inputs to measure the outputs without having to go and wait until tomorrow to see my results.”
The Components of Simulation
Shawn describes four main components that all simulations will have: input, output, a model (aka, the representation of what you’re trying to create), and a solver. “The model will turn that CAD geometry into a discretized surface or collection of surfaces, along with the connections and the loads and the boundary conditions,” Shawn explains. “Then we will feed all that into a solver, which basically is going to take that set of equations based on the loads of boundary conditions and properties that you’ve given it, and run all the calculations to get to the result.”
The pedigree and trustworthiness of the information you’re feeding the model is critical, and that’s where the digital twin comes in. Shawn echoes the perennial “garbage in, garbage out” adage, noting “What’s the quality of that model and the quality of those results? That definitely plays into the digital twin. Because part of that whole pedigree is knowing, do I actually have good data? Do I have a good model here that I can trust when I’m making decisions?” Information pedigree requires traceability of the data, decisions, and history of components. Jonathan asks, “How can I trace back to what I did and make sure that it’s right? Or if there’s a problem, trace forward to correct it so that a digital twin somebody uses to make business decisions on is always the most up to date, the most accurate, the best it can be? Your point is you really have to think about that process.”
AI and Simulation
AI means a whole new approach to simulation. Whereas a traditional surrogate method simply connects inputs and outputs with polynomial equations, AI is different. Shawn says, “It’s actually going to try to understand what’s going on inside of the model, what is the stress at that location. With that kind of data, what it’s actually trying to do is create an accurate representation of the inner workings of the simulation as opposed to just the endpoints.” The additional AI advantage is the data is there, even if you’ve forgotten to specify a criterion for the simulation. Jonathan notes, “Our surrogates can get richer and richer. They’re still approximations, but we could get more answers out of them.”
Learn More About Simulation Possibilities
The full podcast contains more discussion and details on a variety of topics, including whether the digital twin drives simulation or vice versa; what kinds of questions a responsive digital twin can answer quickly to improve your bottom line; why real-world data is even more critical to digital twins than to your day-to-day work; and more.
Be sure to check out Stay Sharp Episode 91: Unpacking Digital Twins Series: Simulation and join us each week for a new podcast.
