Engineers at the University of Waterloo have developed a new machine learning-based solution to monitor heat loss in buildings. The system combines advanced technologies like thermal camera-based imaging, deep learning, and mathematical models to study and analyze patterns of heat loss in constructions. Thermal anomalies due to thermal bridging and façade defects like cracks, air leaks, moisture damage, and infiltration can lead to almost 20 – 34% energy loss. According to the researchers in the Mechanical and Mechatronics Engineering Department and the School of Architecture at the University of Waterloo, early detection of thermal anomalies followed by building repairs can help save up to 50% energy consumption for heating and cooling buildings and up to 42% energy savings in residential buildings.
The study was concluded with an on-site test at a multi-unit residential building in the extreme cold climate of Winnipeg, Canada. The research aimed at building upon existing technologies like thermography to create an automated model for heat loss detection which is more accurate, efficient, and quicker than traditional tools. The study utilized findings from machine learning, mathematical modeling, infrared thermography, imaging techniques, etc. to develop an integrated thermography solution using advanced deep learning with YOLOv7.
• Engineers at the University of Waterloo have developed a new machine learning-based solution combining thermal camera-based imaging, deep learning, and mathematical models to monitor heat loss in buildings.
• The new machine learning-based system approaches thermal anomaly detection comprehensively and automates the process of data collection and analysis entirely.
• YOLOv7, a deep learning-based CNN model was successfully used to analyze large datasets of thermal images and accurately detect anomalies even in complex and challenging natural conditions.
• Early detection of thermal anomalies using the model followed by building repairs can help save up to 50% energy consumption for heating and cooling buildings and up to 42% energy savings in residential buildings.
Traditional thermography in buildings relies on manual inspections and obsolete techniques of data analysis. The method is labor-intensive and time-consuming. Moreover, the manual process is largely ineffective in detecting and calculating heat losses due to complex environmental factors. On the other hand, the new machine learning-based system approaches thermal anomaly detection comprehensively. Apart from automating the process of data collection and analysis, the system factors in often overlooked phenomena like inconsistencies in insulation or material properties that significantly add to heat losses in buildings.
The new system deploys YOLOv7 algorithm to detect areas with thermal anomalies through image processing. The system automates the analysis of data collected by infrared thermography (IRT) used for identifying thermal bridges by visualizing variations in temperatures. Further, the system identifies leakages and estimates the rate of heat loss through the use of computer vision-based algorithms. The novelty of the project stems from the fact that it combines these diverse mechanisms to achieve maximum accuracy in the detection of thermal anomalies. Machine learning algorithms and architectures using convolutional neural networks (CNN) enable the system to overcome the challenges associated with existing technologies
The research team at the University of Waterloo comprising of Ali Waqas and Mohamad T. Araji used YOLOv7, a deep learning-based CNN model for image processing and object detection in real environments. The model was trained and deployed using PyTorch, a Python-based deep learning framework. The model successfully analyzed large datasets of thermal images and accurately detected anomalies even in complex and challenging natural conditions. The model also proved to be ideal for immediate on-site inspections as it can speedily process data in real time. YOLOv7 emerged as the most apt model after a series of tests conducted using other ML models. Apart from major improvements in existing thermography technologies and YOLO models, the research also explored the impact of factors like the size of heat loss areas, the direction of building façades, the magnitude of heat loss, surface temperature, wind speed, and impact of unintended objects like cars or trees.
The research consisted of four phases namely, establishing boundary conditions, data-collection through thermal imaging, data analysis through YOLOv7 deep learning model, and calculating the rates of heat loss through a mathematical equation. Thermography was conducted using handheld or UAV-mounted thermal cameras which captured data by analyzing the infrared radiation emitted by objects. For the next step, three datasets were created using the captured data to train and validate the deep learning model. The third dataset was tested using YOLOv7 after annotation and augmentation of the training dataset. This was followed by the conclusions generated by the deep learning model identifying areas and range of heat loss. Finally, the results were subjected to a mathematical model to calculate the rate of heat loss.
The Waterloo researchers successfully tested the machine learning-aided thermography in a multi-unit residential building in extremely cold temperatures. Almost 28 areas were identified which were responsible for heat loss in the building. Surface temperatures across the building façade were accurately measured using thermal imaging and the deep learning model. Wall intersections and windows were found to be the major heat loss regions. Based on precise calculations, the model provided a focused intervention plan that could result in a 70% reduction in heat loss.
The YOLOv7-based thermal loss detection in buildings can prove decisive for ConTech professionals across the world committed to climate-friendly innovations. Apart from the model’s robustness in collecting and analyzing data, the final round of quantification is critical for planning and implementing targeted building repairs. Knowing the areas, locations, size, and the exact extent of heat loss can enable maintenance teams to strategically minimize thermal anomalies in buildings. Moreover, the before and after heat loss figures can help assess the effectiveness of repairs. ConTech innovations like machine learning-aided thermography can kick off an era of informed decision-making and pave the way for climate-friendly smart investments in the construction industry.
If you are a proptech company and want to promote your products for free, go to proptechbuzz.com and submit your products. For investors or proptech buyers, sign up on our platform to stay informed about exciting updates and trends in the Proptech Ecosystem.
Explore more Proptech news at proptechbuzz.com/news, for news tips and promotions, reach out to marketing@proptechbuzz.com
By Proptechbuzz
By Ravi Kumar
