Brining geometric deep learning technology in Siemens’ Simcenter STAR-CCM+ software to accelerate AI-powered design exploration
Siemens is expanding its Simcenter advanced engineering simulation and test solutions portfolio with the addition of Simcenter PhysicsAI software, which can enable engineers to evaluate bold design concepts and variations almost instantly
Simcenter PhysicsAI brings geometric deep learning technology in Siemens’ Simcenter STAR-CCM+ software to advance AI-powered design exploration. The new capabilities help engineers to create efficient AI-reduced-order models (ROMs) from computational fluid dynamics (CFD) simulation data and conduct what-if exploration faster with fewer computational resources than traditional workflows.
Sam Mahalingam, executive vice president, Simulation, HPC and AI, Siemens Digital Industries Software, said, “Engineering simulation isn’t limited by physics – it’s limited by how quickly we can explore possibilities. Our strategy is to help customers learn from past work, accelerate exploration, and use AI as a prescriptive engine for innovation. We’re not replacing deterministic truth – we’re making it scalable and immediate.
“We’re harnessing AI to help engineers rapidly explore and pinpoint the best solutions to their problems. Simcenter PhysicAI allows customers to target challenges with unprecedented speed and intelligence, frees teams from repetitive tasks and fragmented data and empowers engineers to focus on what matters most: driving creativity and accelerating innovation.”
Simcenter PhysicalAI enables engineers to train and validate AI surrogate models using simulation results, run near-instant predictions on new geometries in their simulation-integrated environment, and maintain easy access to high-fidelity CFD simulation as the validation reference. This approach helps engineering teams to reduce turnaround time, explore more designs with less computing power, and accelerate decision-making.
Simcenter Physics AI can streamline the development and use of AI ROMs to support:
- Evaluate performance across thousands of design variants in minutes. Train predictive models using an advanced transformer neural network architecture optimised for geometric data.
- Train AI models using historical data and newly generated results, including prior Design of Experiments (DOE) studies, reducing the need to rerun CFD simulations.
- Shift early-stage screening from solver runs to AI inference. Integrate AI ROMS into optimisation studies to explore hundreds of design variants in hours, not weeks.
- Quantify prediction accuracy with error metrics and validation tools, ensuring AI models capture correct performance trends to enable confident engineering decisions.
- Simcenter PhysicsAI utilises GPU acceleration for optimal performance, with predictions running up to 100x faster on the GPU than on the CPU.
