Open Collections

Wong, Jeremy

University of British Columbia

As HVAC systems become more prevalent in densified residential buildings, natural ventilation and its benefits have been pushed out the window. These benefits, including energy reductions and air quality improvements, can be harnessed most effectively if building designers can estimate the natural ventilation rate. Since all ventilation models rely heavily on the window opening angle, some form of a sensor is necessary. Prior works have attempted this with traditional algorithms with limited success. Tangentially, a lot of effort has been invested in using machine learning to locate windows on facades. This work combines both of these realms, using photographic techniques and machine learning to not just locate windows but also predict their opening state. This thesis can be broken down into three main parts: the data collection, the preliminary machine learning approach, and the hybrid dataset machine learning approach. The results show the machine learning model can predict window opening state with over 90% accuracy; most windows perform similarly, with the exception of windows with a more head-on angle. Furthermore, using 3D modelling and photorealistic renders to supplement the real images of the dataset proves to be a promising avenue for continued work in this domain. The combination of real and synthetic images, which creates a hybrid dataset, improves model performance in some situations. Aside from these results, these datasets will be made public for others to approach the problem; this includes both the real and synthetic images. The publication of this dataset can facilitate further research into this topic; possible routes that can be taken are presented while also considering the implications and limitations of the research.

Text

2024-01-11

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/87211

Mechanical Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/