Automated damage detection of bridges sub-surface defects from infrared images using machine learning

According to the American Road and Transportation Builders Association (ARTBA), 46,052 of America's 616,087 bridges are rated “structurally deficient” and need urgent repairs. The detection of damages through conventional methods, such as visual inspection and hammer tests are expensive, time-consuming, and challenging to perform without interfering with traffic operations. In the last years, different Non-Destructive Evaluation (NDE) techniques such as computer-vision-based crack detection, impact echo, ultrasonic surface waves, electrical resistivity, ground-penetrating radar, and infrared thermography (IRT) have been developed to inspect aging structures. Among all, IRT has shown the capabilities of detecting defects resulting in different temperature distribution. It can be useful to identify sub-surface damages as delamination and water infiltration, hardly detectable using other traditional methods. In this paper, an algorithm to automatically detect damages in bridges from IR images is proposed. The algorithm exploits the temperature difference between damaged and undamaged parts through machine learning and computer vision techniques to highlight the location of flaws in the structure. Laboratory experiments and real-world analysis on in-service bridges are described in this research to validate the proposed method's accuracy. This study aims to automate the damage detection phases on large-scale structures.