Infrared image-analysis-based concrete inspection using machine learning

In recent years, due to the aging of concrete structures, deterioration such as flaking and delamination has occurred. Since such damage leads to spalling and may cause concrete pieces to fall, thus potentially injuring pedestrians or damaging vehicle passing below. Therefore, it is necessary to detect and repair structures quickly. In Japanese highway companies, hammering inspection is undertaken once every 5 years to detect structural damage. However, to conduct the hammering test, it is necessary to hammer the whole surface of the bridge, and arranging a bridge inspection vehicle is also necessary. Consequently, it is a cost- and labor-intensive undertaking, accounting for half of the highway bridge maintenance costs. In addition, traffic control is also required, which causes congestion and increases the risk of traffic accidents. This study aimed to solve these problems by utilizing infrared thermography. Infrared thermography is a nondestructive inspection technique used to detect regions of flaking and delamination by photographing the temperature inhomogeneity of concrete structure surfaces using an infrared camera. The cost and labor are drastically lower than that in the hammering test. Traffic regulation is also unnecessary, because the infrared thermography can be measured nondestructively from a distance. However, the damage detection accuracy based on the thermal image has been limited. This study reports the improvement in the detection performance by using deep learning, which ensures sufficient accuracy of the infrared method for practical use.