Research Framework and Institutions
FireModel-AI is developed within the BESKID research framework in collaboration with the University of Wuppertal and Forschungszentrum Jülich.
AI-driven reduced-order modeling framework for transient fire prediction based on POD and ANN surrogate models trained on high-fidelity CFD datasets.
The framework combines Proper Orthogonal Decomposition (POD/SVD) with Artificial Neural Networks (ANNs) for rapid prediction of transient Heat Release Rate (HRR) evolution.
High-dimensional CFD snapshots generated using the Fire Dynamics Simulator (FDS) are assembled into a snapshot tensor and compressed into a low-dimensional reduced basis using Proper Orthogonal Decomposition (POD).
Only the dominant POD modes are retained. Instead of solving the governing Navier–Stokes equations during prediction, an Artificial Neural Network (ANN) learns the nonlinear mapping between the material parameter vector μ and the reduced coefficients a.
The transient fire response is reconstructed as:
This enables near-instantaneous prediction of HRR time-curves while retaining the dominant physical behavior extracted from the original CFD simulations.
FireModel-AI is developed within the BESKID research framework in collaboration with the University of Wuppertal and Forschungszentrum Jülich.