Introduction of the combination of thermal fundamentals and Deep Learning for the automatic thermographic inspection of thermal bridges and water-related problems in infrastructures

ABSTRACT

Infrastructure inspection is fundamental to keep its service performance at the highest level. For that, special attention should be paid to the most severe defects in order to be able to subsequently mitigate or even eliminate them. Therefore, this paper introduces the combination of an automatic thermogram pre-processing algorithm and a Deep Learning (DL) model, Mask R-CNN, applied to thermal images acquired from different infrastructures (buildings, heritage sites and civil infrastructures) with water-related problems and thermal bridges. The pre-processing algorithm developed is based on thermal fundamentals. As an output, the thermal contrast between defect and defect-free areas is increased in each image. Then, Mask R-CNN is trained using the pre-processing algorithm outputs as input dataset to automatically detect, segment and classify each defect area. The training process of Mask R-CNN is improved by the prior application of the proposed pre-processing algorithm in terms of time. This shows the capacity of thermal fundamentals to improve the performance of the DL models for their application to the InfraRed Thermography (IRT) field. In addition, DL models are introduced for the first time in the thermographic inspection of water-related problems and thermal bridges when inspecting an infrastructure.

KEYWORDS:

• Infrared thermography
• deep learning
• thermal principles
• mask R-CNN
• infrastructure inspection
• water-related problems
• thermal bridges
• automatic processing

Acknowledgments

This work has used computational resources from the Supercomputing Centre of Galicia (CESGA).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was funded by the Ministerio de Ciencia, Innovación y Universidades (Gobierno de España) through programs for human resources (Ministerio de Ciencia, Innovación y Universidades FPU16/03950), and by the Cátedra Iberdrola VIII Centenario – University of Salamanca (Spain). This project has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No 769255. This document reflects only the author’s view, and the European Commission is not responsible for any use that may be made of the information it contains.