Combining physics simulation, data and measurement together into AI can change the way engineers design and manufacture products.
Hexagon AB has big plans for its acquisition of CADLM, and its artificial intelligence (AI) software ODYSSEE, which can change the simulation world forever. Back in 2017, Hexagon entered the simulation space with the purchase of MSC Software. Since that acquisition, the company has focused on the Industry 4.0 space.
Roger Assaker, president of the Design and Engineering Software business unit at Hexagon, explained, “Our mission is to help the design engineering of products for sustainability, using smart manufacturing and industry 4.0. This is the business of Hexagon. We are pushing smart technologies.”
It could be argued that there is no smarter technology in the Industry 4.0 space than AI, so this acquisition makes perfect sense. But what does this all mean for the simulation world?
“In a nutshell, ODYSSEE AI allows you to learn from some finite element analysis (FEA),” Assaker said. “Today, we run physics-based analysis and solve the physics equation to predict the future. We have a boundary condition, loadings, geometry and materials, and we run an analysis to get an output based on these inputs. This is typically multiple hours to multiple days of analysis, depending on the physics you are solving.”
ODYSSEE, on the other hand, learns the relationship between a system’s inputs and outputs from past analysis. It then creates a data-based model that represents the physics-based model. This enables engineers to predict a response based on given conditions. In summary, what would take hours to days to assess using FEA can be done in seconds using an ODYSSEE model.
An overview of ODYSSEE technology.
Can ODYSSEE AI Be More Accurate Than Physics-based Simulations?
Instead of using AI to learn the road, or categorize images, ODYSSEE learns the relations between inputs and outputs. It then creates reduced-order models that enable engineers to work faster and smarter.
“It opens up the door for engineers to run faster analysis in almost real-time,” Assaker explained. “You leverage all the data you have, both numerical and physical, to move forward. It allows you to do design space explorations and account for uncertainty and variability.”
So, AI-based simulations are fast, but are they accurate? Well, it’s all about the data used to train the algorithm.
“With physics-based models, you can always have an approximation,” Assaker said. “Sometimes, AI allows you to complement the physics to get increased accuracy. Let’s say you have a model that is only 90 percent accurate on average. If you have 10 of these analyses and two to five measurements that are 99 percent accurate, you can take the set of measured data and analysis to train the AI model. The outcome of the AI could be more than 90 percent accurate. It could be 95 percent accurate.”
Engineers can also face scenarios that are complex to simulate in detail. Assaker referenced a tire rolling over a partially snowy road. The traction depends on the amount of snow, how much of the tire is in contact with snow, the temperature and much more.
Assaker explained that these multiphysics problems can be overwhelming, and AI can be used to discover unknown relationships in these systems without fully solving the physics. That is why he suggested using some real-world measurements to produce ODYSSEE data-based models.
“Physical models have some underlying limitations as you are not capturing everything,” he said.
By bringing measured data into the model, engineers can potentially capture more.
ODYSSEE AI’s Compatibility in the Simulation World
Simulation-based AI has many applications and can interface with various other smart technologies.
For instance, ODYSSEE works with LS-Dyna and other simulation software that is used for crash analysis. Since partnering with MSC, with the plan to acquire it, it’s now also connected to most of that portfolio of software, such as:
- Adams for system dynamics
- Cradle for CFD
- Digimat for materials
- Nastran for linear analysis
- Marc for nonlinear analysis
Assaker noted, “We want each [simulation] product in the center of excellence to provide a solution that is based on physics and a solution that is based on AI. What CADLM will do, in addition to being a stand-alone AI platform, will ensure these AI solutions. So, you can predict acoustics, structures, fluid and system dynamics with AI.”
As for how ODYSSEE connects to other simulation software, Assaker said, “It’s designed in such a way that developing an API to them would be a very simple exercise. It’s designed to consume data coming from any CAE product. We are not aiming to close ODYSSEE to only MSC Software.”
He added that ODYSSEE is also compatible with optimization tools and has its own optimization tools.
There are also clear digital twin applications for ODYSSEE.
“In a digital twin of a full system, that’s a lot of physics to capture and a lot of steps and feedback loops,” Assaker said. “If you do a local digital twin that is physics-based, it can take forever. The advantage of ODYSSEE is that it gives you a real-time response. It allows you to do a lot of feedback loops and get the relation from the input and output very fast. What we are trying to do is blur the boundaries between physics-based simulation, measurement and AI.”
How Many Data Points Does ODYSSEE Need? AI Fits in the World of Design Engineering
To summarize the ODYSSEE workflow, engineers analyze a system multiple times using simulation. They then follow that up with real-world measurements. Finally, they put all this data into the machine learning algorithm, and it will provide a 1D model that can predict the outputs of the system based on given inputs.
“Part of it is an RSM (response surface methodology) model,” Assaker said. “Some of the algorithms is a reduced-order model. Two key advantages to ODYSSEE, with respect to other tools like this, is that it can learn from a limited set of training data. You don’t need a lot of analysis. It can also take into account time-dependent behaviors.”
That begs the question, how many analyses does it take for ODYSSEE to get going?
“The idea is that maybe the expert will run 10 analyses to find the key points in a design space,” Assaker clarified. “With this, we build an AI model using ODYSSEE technology, then give it to the designer to run hundreds of explorations and test the wildest of ideas faster. We then come back and verify the last analysis with simulation again.”
But it isn’t always that easy. In most cases, 10 to 15 analyses will enable building an accurate model. Assaker does admit that other scenarios could take much more. He referenced a composite coupon strength analysis that needed 15,000 measurements to train the model. So, act like any other engineer working with any other model and don’t forget to verify and validate ODYSSEE models.
Where Does ODYSSEE AI Fit in the World of Design Engineering
“For us, ODYSSEE is a democratization tool, an improved accuracy tool, a digital twin tool and also a deployment tool,” Assaker said. “It provides prediction capabilities to other engineers, not only the detail analysts.”
With the acquisition complete, the plan is to integrate more MSC and CADLM into Hexagon’s measurement technology. These tools will become the prediction arm of the Hexagon Smart Manufacturing backbone. This backbone will enable systems to connect Hexagon’s design engineering, production and metrology technology into one digital thread that can be used to produce a digital twin.
“CADLM is already well integrated with a huge pipeline of customers to do AI-based simulation, but we believe in data and the merging of the digital world and the physical world together,” Assaker explained. “Quickly Hexagon, whose business is to measure things, will look at the physical data in addition to simulation data to influence production, quality and service maintenance.”
Do you think AI will revolutionize simulation? Comment below.
