Development of ML algorithm to improve in situ measurement of the thermal properties of a building

This study presents a method to enhance the accuracy of the procedure for in-situ measurement of the wall thermal transmittance and the overall performance of the building envelope based on IoT infrared sensors. The sensor used is the Comfort Eye, consisting of an IR node to acquire thermographic maps and a desktop node for environmental parameters. This will be achieved by applying a Machine Learning (ML) algorithm to detect and identify the various elements present within the wall, including windows, which exhibit different emissivities that are required for thermal transmittance measurement. Accounting for these differences is crucial to improve measurement accuracy and simplify the calculation of thermal transmittance. In particular, the thermographic images are processed with a detection algorithm, You Only Look Once (YOLO-v5) trained with a personalized dataset and accuracy for detecting elements (such as windows). Results show that the metric values of precision, recall, F1 for the implemented algorithm to detect windows are 0.79, 0.84, 0.81 respectively. Moreover, the identification of the different elements of the wall, having a thermographic map and therefore a punctual measurement of the transmittance value of the entire wall, allows the presence of thermal bridges to be correctly identified.