Estimating surface temperature from thermal imagery of buildings for accurate thermal transmittance (U-value): A machine learning perspective

Abstract

Thermal performance assessment of building(s) is an essential process for optimal energy management, heat-loss evaluation, and energy audit applications. Such an assessment can help foresee the requirements for future intervention(s) and aid in benchmarking energy performance. This paper provides a review of several thermal performance assessment techniques and a broad classification based on measurement types, methods, and applications. Moreover, the article provides a comprehensive survey of various quantitative indices utilized for practical heat-loss assessment of building elements. This paper’s unique contribution is the proposed three-layer framework that details the handling and processing of UAS-based thermal imagery for heat loss quantification. Primarily, the novelty of this work lies in the application of an instance segmentation technique (Mask R–CNN) to compute the thermal transmittance values (e.g., U-values) for various objects (e.g., doors, walls, windows, and facades). To the best of our knowledge, this research work is first-of-its-kind using a sizeable thermal data repository (e.g. 100,000 augmented images). Multiple standard U-values are analyzed for windows and walls and compared with The American Society of Heating, Refrigerating, and Air-conditioning Engineers (ASHRAE) building standards. The preliminary results of Mask-RCNN from over 100,000 trained (including augmented) images from multiple campus buildings yield the following performance metrics: 1) provides an Average Precision (AP) of 0.67 (windows) and 0.46 (facades); and 2) Intersection of Union (IoU) of 0.05 (windows) and 0.5 (facades) respectively. Moreover, the U-values are consistently close enough to the ASHRAE standards in distinguishing window types (e.g. 0.77 for single-pane windows and 0.38 for double-pane windows).