Host: Science Council of Japan
Co-host: Japan Society for Safety Engineering, The Japanese Geotechnical Society, Japan Society of Civil Engineers, The Japan Society of Mechanical Engineers, Architectural Institute of Japan, The Japan Society for Aeronautical and Space Sciences, The Society of Materials Science, Japan, The Japan Society of Naval Architects and Ocean Engineers
Name : The 10th Japan Conference on Structural Safety and Reliability
Number : 10
Location : [in Japanese]
Date : October 25, 2023 - October 27, 2023
This study concerns a method for diagnosing bridge anomalies using machine learning. In this study, bridge abnormality diagnosis method using convolutional neural network (CNN) from acceleration responses is proposed. Spectrogram images of acceleration responses are applied to image classification by CNN. By classifying spectrograms for each bridge condition, bridge anomalies are detected and the condition is identified. In order to improve the accuracy of damage classification, a two-step classification method is used: first, a classification is performed in the major damage category, and then the sub-categories are identified within the major category. The two-step classification improved the accuracy of damage identification in the sub-category of girder damage, indicating that the two-step classification is effective in improving the identification accuracy of this method.