Machine learning-aided thermography for autonomous heat loss detection in buildings

Abstract

Efficient inspection of heat loss from building façades is an active research area, driven by the need for improved auditing and monitoring solutions. Thermography is a valuable tool for identifying anomalous heat patterns. This paper presents a novel method for autonomous inspection of heat loss from building façades by combining thermal camera-based imaging, advanced deep learning with YOLOv7 for anomaly detection, and a mathematical model for heat flow quantification. The methodology achieved a mean average precision (mAP@0.5) of 0.770 for anomaly detection when tested on thermal data collected from a multi-unit residential building in an extremely cold climate. The highest heat loss measured at 1533.73 W was from the south façade, while the largest area of heat loss, covering 15.65 m2, was identified on the southeast façade. The study identified 28 regions of thermal anomaly on the building façades, and heat loss from these regions was quantified for inspection.