
Optimising Ventilation Systems in Above-Ground Buildings Using Coupled CFD_NSGA-II Analysis



Intro
We combine advanced simulation, data-driven modelling and compliance expertise to help clients meet increasingly stringent building regulations while improving safety and cost efficiency.
The Challenge
Traditional smoke and ventilation designs for high-rise and above-ground buildings are often based on rule-of-thumb or experience. This can lead to over- or under-engineering designs, higher operational costs, and uncertainty over compliance with evolving standards – especially the new tenability criteria for evacuation lift lobbies in BS 9991 (2024). We needed a robust, objective way to design mechanical smoke ventilation systems that would:
- Keep escape stairs and lift lobbies smoke-free during fire and evacuation phases
- Meet all relevant British Standards and guidance
- Minimise fan energy and capital costs
- Reduce design iteration time and stakeholder uncertainty
The Solution / The Turning Point
We developed and applied a hybrid optimisation framework coupling Computational Fluid Dynamics (CFD) tools (Fire Dynamics Simulator) with the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) – an AI evolutionary algorithm. This approach automatically tests combinations of key variables (fan flow rate, damper free area) to identify optimal trade-offs between performance and cost while respecting regulatory constraints.


How It Works
- NSGA-II generates potential configurations using smart mutation and crossover.
- Each configuration is evaluated in FDS for smoke visibility, temperature, pressure and velocity.
- Any configuration violating tenability limits (visibility <10 m, temperature >60 °C, pressure < 60 Pa, velocity >5 m/s) receives an automatic penalty.
- Over successive generations, the algorithm converges to a Pareto front of optimal solutions.
This allowed us to explore a large design space quickly, using only industry-feasible values for fans and dampers, and achieve convergence in as few as ten generations (five FDS simulations per generation).

Life After AI
By replacing manual iteration with AI-driven optimisation, we can now:
- Offer evidence-based, defensible ventilation designs that demonstrate compliance under worst-case firefighting conditions.
- Deliver clear, visual outputs (contour plots, heat maps) that give stakeholders and approving authorities confidence in the design.
- Reduce simulation runs and design time compared to exhaustive testing.
Impact Summary & Metrics
In our case study:
- Visibility improved by up to 27.3% in the lobby while keeping lift lobby and stair completely smoke-free.
- Fan duty reduced by ~36.3% compared to the initial design (3.0 m³/s vs. 5.5 m³/s typical industry practice), directly lowering capital and operating costs.
- All tenability criteria (temperature, pressure, velocity) remained within accepted limits throughout the firefighting phase.
- Convergence achieved after only 10 generations, avoiding 7 extra generations (35 FDS runs) compared to unconstrained search.
Looking Forward
Future work includes extending this methodology to account for high-hazard fire loads such as electric vehicle (EV) battery fires ( 8+ MW HRR), integrating additional AI tools for predictive scenario testing, and coupling with BIM to automate ventilation design from early project stages. This will further enhance our ability to protect vulnerable occupants and optimise smoke control systems for next-generation building designs