Quality assessment of peak pressure estimates in high-rise building using large eddy simulation

• Quantifies numerical and modelling errors in LES-predicted peak pressures. • Applies systematic grid and model variation to map error estimates. • Uses 96-cells-per-breadth resolution to achieve low grid-induced uncertainty. • Shows grid-resolution sensitivity exceeds that of the sub-grid-scale model constant. • Confirms 95 % of peak pressures match wind-tunnel data within ±30 % tolerance. Analysis of peak surface pressures is essential for specifying wind-resistant cladding of tall buildings. Large eddy simulation (LES) holds promise for predicting peak wind action, yet its reliability is challenged by uncertainties arising from extreme value analysis and grid-controlled scale filtering. This study postulates that quantifying numerical and modelling errors in the LES pressure solutions improves design reliability and guides grid resolution specifications for high-fidelity peak pressure prediction. Thus, the key objectives were to ( i ) map magnitudes of numerical (truncation) and modelling errors along facades—achieved here through the novel application of the systematic grid and model variation method—and ( ii ) assess the accuracy of low-uncertainty peak pressure solutions against experimental measurements. Simulations reproduced a benchmark from the Tokyo Polytechnic University, modelling a square-based building with height-to-breadth ratio of 5:1 subjected to orthogonal and oblique wind incidences. Peak pressure coefficients at 500 locations were extracted from a Gumbel distribution of extremes, recorded during 30 min of equivalent full-scale exposure to 100-year return wind speeds. Using an isotropic wall-tangent resolution of 96 cells per building breadth, the LES peak pressure results showed generally low uncertainty, with area-averaged numerical and modelling errors of 7.3 % and 4.1 %, respectively. This configuration demonstrated respectable accuracy: 55 %, 85 %, and 95 % of peak pressure predictions fell within ±10 %, ±20 %, and ±30 % of the experimental references. Ultimately, this work presents a systematic LES framework for reliable peak pressure prediction, well-suited for third-party reproduction.