AI Meets Simulation at NAFEMS Americas

At the 2024 NAFEMS Americas conference in Louisville, KY in July, simulation users, software providers, integrators, and academics converged to discuss the latest developments in engineering simulation. 

NAFEMS is the international association for engineering modeling analysis and simulation, with chapters around the globe. To kick off the conference, Tim Morris, NAFEMS CEO, outlined the agenda for the three-day event, and highlighted some of the major developments in the industry – with artificial intelligence (AI) at the top of the list.

“AI is particularly relevant to our world in simulation,” he said. “We are learning a lot about how it can be applied in our environments.” Other notable developments included the growth of systems simulation and model-based systems engineering; increased access to high-performance computing; CAE vendor consolidation; and simulation certification efforts.

But AI came up again and again across a number of sessions. Keynote speaker Anthony Petrella, director of the Computational Biomechanics Group at the Colorado School of MINES, was asked about AI in the context of discussion of the future of engineering education.

“Using AI is something that every student is doing,” he said. “I think we should avoid viewing it as a threat. We should view it as a tool that we should leverage to its maximum extent. We are all still mid-stream in discovering the right ways to do that.”

Accelerated Computing Enables AI-based Simulation

Ian Pegler, Global Business Development Manager for CAE at NVIDIA, led a standing-room-only session on Demystifying AI for CAE, during which he explained basic AI concepts.

According to Pegler, AI can potentially help solve some of the speed and computational demand challenges faced by companies that want to expand their use of simulation. “Companies want to use simulation to work faster and make better products,” he said. “They want to run more simulations and do it a lot quicker, but with limited resources. We can use AI to get faster feedback early in the design process.”

In most of the emerging products offered by simulation software vendors, AI models are trained on existing test and simulation data to help provide faster analysis of a high number of design iterations to help engineers narrow their options before running more comprehensive simulation studies. The use of NVIDIA RTX™GPUs provides the speed and computational horsepower to do this efficiently, both at the workstation level and in the data center.

During the presentation, Pegler explained the history and structure of the neural networks that power AI solutions, as well as new frameworks like TensorFlow and PyTorch that help developers build models without having to code everything from scratch. NVIDIA also offers the Modulus framework, which is an open-source library that helps companies build AI models.

He also explained the role of the GPU in accelerating AI adoption because of their ability to rapidly perform computations like mixed-precision matrix multiplication that are critical for AI applications. The fact that NVIDIA GPUs excel at such mixed-precision operations, with stronger integer and floating point performance, makes the GPU an ideal choice for these applications. Combined with the NVIDIA software toolchain and features for training and developing AI models, this puts the NVIDIA GPUs at an advantage, compared to AI PCs that rely on neural processing units (NPUs),

NVIDIA was also part of a workshop, led by Mustafa Kaddoura, Senior Application Engineer at Ansys, titled Leveraging AI to Boost Computation-Intensive Simulations. 

The workshop explored how AI is applied to engineering simulations across industries, and the challenges posed by high computational demands. The workshop also included a CFD demonstration using Ansys SimAI, the company's new cloud-based generative AI platform. 

Kaddoura said that the AI models can be efficiently trained on NVIDIA RTX GPUs with large datasets to provide faster insights to engineers. 

“AI can be very powerful for repeated simulations where there are lots of design iterations,” Kaddoura said. 

However, because the application of AI in these workflows is relatively new, most customers will need to investigate the size of the data set needed to train these models for their specific use cases. Kaddoura said Ansys recommends testing the model against an actual Design of Experiments (DOE) to ensure accuracy. 

Earlier this year at its GTC event, NVIDIA announced its new Blackwell chip for AI workloads. The company was also involved in the Dell Technologies World 2024 event, where Dell and NVIDIA discussed AI-enabling hardware innovations. NVIDIA RTX GPUs are also part of the Dell AI-ready Precision engineering workstations. Dell and NVIDIA have also collaborated on the Dell AI Factory with NVIDIA solution for enterprise AI deployments.

Ford also presented a use case that leverages AI for creating surrogate models for CFD simulations during the GTC conference. You can watch a replay here. 

For more information on GPU-accelerated CAE, download the AI-Powered Engineering white paper, and check out the NVIDIA Accelerating and Advancing CAE E-Book Series. 

More NVIDIA Coverage