Electromagnetic radar has been used for detecting internal defect in concrete structure, since the presence of a defect, such as a crack, causes changes in the scattered waves. However, visually identifying scattering difference due to a defect requires a high level of skill, and it takes a lot of effort when there is a large amount of observation data. Therefore, the development of automatic identification technology using machine learning (ML) is underway. However, the training data collection method and the performance of air-gap detection using ML has not been explored in detail. In this paper, firstly, we investigate the detection of an infinite air-gap parallel to the surface of a concrete slab using artificial neural network (ANN) identification technology in layered medium models, and discuss how to train high performance ANN with less training data. Secondly, we study the characteristic of ANN detection for an oblique air-gap using 2D homogeneous medium models.
抄録全体を表示