Towards an Automated Method for Detecting and Quantifying Window Air Leakage

Air leakage through windows, as the weakest components of building envelopes, significantly affects energy consumption and occupant comfort. Evaluating air leakage is essential for targeted retrofit interventions given their high cost. Existing methods like blower door test are notably intrusive, and non-intrusive methods such as infrared thermography can detect air leakage but lack the ability to quantify leakage. This study addresses these limitations by investigating a non-intrusive approach to quantifying air leakage. A YOLO-based machine learning model was developed to automatically detect leakage in IR images, paired with an image-based metric to quantify leakage. The approach was validated using a custom dataset comprising laboratory-based and on-site images. Results demonstrated that this approach accurately identifies and quantifies air leakage, enabling systematic and targeted retrofit decisions. The proposed methodology extends beyond windows, with potential applications to other building envelopes, also serving as an objective tool to evaluate installation quality and workmanship.