Home // SIMUL 2025, The Seventeenth International Conference on Advances in System Modeling and Simulation // View article

PAIRS: Physics-Enabled AI for Real-Time Simulations Surrogates

Authors:
Zeinab Alfaytarouni
Hamza Ben Ammar

Keywords: Physics-enabled AI; Physics-Informed Neural Networks; Ontology; Simulations.

Abstract:
The development and improvement of complex engineering systems increasingly depend on virtual and hybrid test benches for validating new designs or modifications to existing ones. Central to these test benches are simulation models, which are essential but time-consuming to develop due to their reliance on domain expertise. Full-fledged simulation models can also be slow, impeding the validation process that requires realtime simulation. Conversely, AI surrogate models, derived from sensor data, face constraints due to insufficient training data and potentially lacking physical sense. To address these challenges, we propose the use of physics-enabled AI models as surrogates, which strike a balance by integrating underlying physical laws through model equations, thereby requiring significantly less data for training. Once trained, these models operate in real-time, expediting the validation process. In this work, we introduce a Physics-enabled AI surrogate model development process that augments to the existing Machine Learning Operations (MLOps) workflow. Our approach employs an ontological framework to align user needs with a model template. We leverage Physics- Informed Neural Networks (PINNs) as the core building block for this template. Once a model structure is selected, the traditional MLOps process is applied to train and validate the AI surrogate. This methodology simplifies the model development process and hence accelerates the overall system development.

Pages: 43 to 49

Copyright: Copyright (c) IARIA, 2025

Publication date: September 28, 2025

Published in: conference

ISSN: 2308-4537

ISBN: 978-1-68558-300-2

Location: Lisbon, Portugal

Dates: from September 28, 2025 to October 2, 2025