An Instance Segmentation and Clustering Model for Energy Audit Assessments in Built Environments: A Multi-Stage...

Area of Science:

• Building Science
• Artificial Intelligence
• Thermal Engineering

Background:

• Current heat loss quantification (HLQ) methods are often qualitative or rely on limited data, hindering automated energy audits.
• A need exists for quantitative HLQ solutions suitable for large-scale building energy assessments.

Purpose of the Study:

• To develop and validate a novel deep learning framework for automated heat loss quantification using extensive thermal imaging data.
• To segment building envelope components from thermal images and estimate their thermal performance.

Main Methods:

• Utilized a deep learning approach, including Mask R-CNN and Faster R-CNN, for object detection and instance segmentation of building components from over 100,000 thermal images acquired via unmanned aerial systems (UAS).
• Employed K-means and threshold-based clustering (TBC) to estimate surface temperatures and calculated the overall heat transfer coefficient (U-value).
• Applied the model to eleven academic campuses and compared results with American Society of Heating, Refrigerating, and Air-conditioning Engineers (ASHRAE) standards.

Main Results:

• Mask R-CNN demonstrated superior performance in segmenting building components, achieving high mean Intersection over Union (mIOU) scores for facades (73%), roofs (67%), and windows (55%).
• Threshold-based clustering (TBC) provided more consistent surface temperature estimations compared to K-means across different times of day.
• Building thermal efficiency was found to be influenced by image acquisition, building geometry, and environmental factors like temperature and wind speed.

Conclusions:

• The proposed deep learning framework offers a robust and scalable solution for automated heat loss quantification in buildings.
• Accurate HLQ requires integrating thermal imaging data with building geometry and dynamic environmental parameters.
• The methodology provides valuable data for energy audits and compliance with building energy standards.