機械学習によるデジタル打音検査精度の高度化

抄録

When the integrity of structures such as a bridge or a tunnel confirms, a strain gauge or a hammering inspection are generally used. Although a measurement using a strain gauge is accurate, it takes much time to attach it and its reliability against rust and adhesion is regarded as a problem. Although a hammering inspection can be used easily and low cost, its accuracy is inferior to other inspection methods. As a result, there is a risk that it would lead to a major accident like the crash of the Sasago tunnel in December 2012 in Japan. Accordingly, it is considered that it is possible to make a machine judge not only the good or bad integrity but also the prediction of danger such as the position and the size of the defect by digitizing the information from a hammering inspection by using the AE sensor and machine learning. Therefore, in this study, it uses Convolutional Neural Network (CNN) which predicts from input features used in image recognition and speech recognition. CNN is more accurate than other neural networks by expanding data when there is only a small amount of data. In this study, FEM analyses are performed on a test body with a sheath hole made from concrete. And using machine learning from analysis results, we would like to investigate whether it is possible to predict the depth at which a defect exists, the diameter, the probability of having a defect.