Leveraging AI modelling for FDS with Simvue: monitor and optimise for more sustainable simulations

TL;DR

The paper tackles the high cost of fire safety simulations by integrating data-driven surrogates, model-guided optimisation, and a provenance-focused platform. It demonstrates that ML surrogates (including U-net and Gaussian Processes) can predict heat and smoke dynamics orders of magnitude faster than full CFD, while optSim enables efficient search for optimal design parameters or worst-case fire locations. Key contributions include Simvue with strong data lineage, real-time monitoring, and collaboration features, plus quantified speedups and reduced resource use (e.g., ~$10^{4}$× compute efficiency and significant reductions in simulation counts). The work has practical impact by enabling sustainable, scalable fire-safety analyses and facilitating rapid iteration across campaigns and teams, with carbon-footprint awareness via ecoClient.

Abstract

There is high demand on fire simulations, in both scale and quantity. We present a multi-pronged approach to improving the time and energy required to meet these demands. We show the ability of a custom machine learning surrogate model to predict the dynamics of heat propagation orders of magnitude faster than state-of-the-art CFD software for this application. We also demonstrate how a guided optimisation procedure can decrease the number of simulations required to meet an objective; using lightweight models to decide which simulations to run, we see a tenfold reduction when locating the most dangerous location for a fire to occur within a building based on the impact of smoke on visibility. Finally we present a framework and product, Simvue, through which we access these tools along with a host of automatic organisational and tracking features which enables future reuse of data and more savings through better management of simulations and combating redundancy.

Figures (8)

• Figure 1: Examples of various fire locations at different times simulated with .
• Figure 2: The Simvue dashboard, showing a collection of simulations from a sprinkler design optimisation campaign. A parallel coordinates plot has been created using the metadata and metrics which are automatically tracked by Simvue.
• Figure 3: Diagrams showing data traceability with Simvue. The dependencies graph (bottom left) shows a series of runs which all depend on the same code file and each produce their own cloud of output artifacts. The lineage graph shows the files and runs required to generate a selected Smokeview file.
• Figure 4: Real-time monitoring of a live simulation.
• Figure 5: Alert notification and runs terminated when visibility distance drops below 3m.