Automated assessment of wind damage to windows of buildings at a city scale based on oblique photography, deep learning and CFD. (15th July 2022)

• Record Type: Journal Article
• Authors: Gu, Donglian; Chen, Wang; Lu, Xinzheng
• Is Part Of: Journal of building engineering. Volume 52(2022)
• Publication Date: 2022-07-15
• Subjects: Wind damage -- Oblique photography -- Window -- Deep learning -- Automated assessment; Building -- Periodicals; 690.05
• DOI: 10.1016/j.jobe.2022.104355
• Languages: English
• ISSNs: 2352-7102

Abstract

Abstract: Windows are one of the non-structural components of buildings that are most vulnerable to wind damage. It is important to obtain rapid and accurate predictions of wind-induced window damage in urban areas. A novel automated method for simulating the wind damage to building windows at the city scale is proposed in this study based on oblique photography, deep learning, and computational fluid dynamics technologies. First, a method for extracting building façade pixels from oblique aerial images is proposed to provide the data basis for subsequent window recognition. Second, the Pix2Pix deep learning network is utilized to recognize windows on building façades. Finally, the window damage in building clusters is evaluated automatically based on the window failure model and time-varying wind pressure data obtained through computational fluid dynamics simulations. A real-world community in Shenzhen, China, is used as a case study to showcase the workflow and demonstrate the reliability and feasibility of the proposed method. The proposed automated method overcomes the limitation that existing methods are difficult to apply to real building clusters. The novel method and corresponding case study presented here can provide a reference for urban areas to mitigate the impacts of wind disasters. Highlights: A city-scale automated method for evaluating wind-induced window damages was proposed. An automated method for extracting building façade pixels from oblique aerial images was proposed. The deep-learning-based window recognition method achieves a satisfactory performance. The proposed assessment method is applicable to real building clusters.