• A full CFD-integrated NSGA-II framework is developed to optimise smoke-ventilation design under multi-objective, multi-constraint conditions. • The optimisation treats visibility and fan duty as competing objectives while embedding tenability requirements via hard penalty functions for temperature, pressure, and velocity. • A practical, industry-aligned constrained search-space methodology is introduced, reducing CFD computational demand without sacrificing convergence accuracy. • Adaptive mutation strategies are implemented within NSGA-II to preserve population diversity and prevent stagnation in a discretised design variable space. • Comparative evaluation with an unconstrained search demonstrates that the filtered search-space approach reliably converges to the same optimal region, validating its robustness. • Firefighting-phase CFD assessments confirm that NSGA-II–optimised designs maintain full compliance with stair and lift-lobby tenability requirements, fulfilling new BS 9991 expectations. This study presents a hybrid optimisation technique comprised of non-dominated sorting genetic algorithm-II (NSGA-II) technique and computational fluid dynamics (CFD) to systematically configure buildings ventilation systems. The objective functions are to minimise the fans duty simultaneously adhering to British standards. Such that the visibility is maximised in the lobby by supplying the minimum flow rate to achieve perfect standards compliance. The NSGA-II algorithm then dynamically alters the fans flowrates and damper free area as variables to achieve the objectives. The results revealed an optimum configuration for which the visibility was improved up to 26.6%, by using 37.1% less fan duty. The optimisation also incorporates the new tenability requirements for lift lobby, as such temperature, pressure and velocity values remained within acceptable range of ≤60°C, ≥-60Pa, ≤5m/s for the optimal configuration. To avoid exhaustive search for impractical variables combinations, the search space was constrained to only test configurations which are practically feasible to accommodate in industry, which helped ascertaining the best solution with commensurately lower computational effort (7 generations less, 35 FDS simulations in total). As such, convergence occurred after 10 generations with standard deviation of lower than 0.1 for both objective functions. Re-evaluating the optimisation technique with unconstrained search space led to nearly similar results as the inputs converged onto a value with <∼4% relative different, and the objective value remained within ∼4% neighbouring of the value obtained for constrained search space. Comparison with other robust optimisation techniques, i.e. NSGA‑III and Multi-Objective Evolutionary Algorithm based on Decomposition, are also discussed.
Zandie et al. (Wed,) studied this question.