Data-driven optimization of wind pressure sensor placement on low-rise buildings using computational fluid dynamics and multi-resolution dynamic mode decomposition

This study presents a novel hybrid framework for optimal sensor placement to evaluate wind loads on low-rise buildings. Recognizing the challenges of deploying dense sensor arrays in turbulent atmospheric boundary layer wind tunnel tests, the proposed method integrates large eddy simulation with multi-resolution dynamic mode decomposition (mrDMD) to isolate spatiotemporally dominant flow features. Unlike traditional DMD-based approaches that capture global modes, the use of mrDMD enables scale-separated modal analysis, enhancing sensitivity to transient and localized flow dynamics. These modes guide a QR pivoting algorithm, which efficiently selects sensor locations that maximize information content. The framework demonstrates a sensor count reduction of over 80%, from 1426 candidates to just 182 sensors, while preserving high reconstruction accuracy (R > 90%) for both mean and fluctuating pressure fields. This distinction enables robust and cost-effective wind load assessment without compromising fidelity. The methodology is validated using wind tunnel experiments and is shown to be applicable for generalized wind scenarios through an angle-of-attack-unified sensor configuration. By combining modal decomposition with informed optimization, this framework advances state-of-the-art techniques in structural monitoring, offering practical utility in experimental and real-world applications.