Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : September 11, 2022 - September 14, 2022
This research is about the diagnostic method of anomaly of bridge measurement using machine learning. In this study, diagnostic method of the condition of bridges via convolutional neural network (CNN) based on the acceleration responses is proposed. Spectrogram images of acceleration response are applied to image classification using a CNN. The condition of bridges is identified or detected by classifying spectrograms for each condition of the bridge. The number of learning generations is able to increased by employing a CNN model that incorporates a dropout layer. The optimal number of learning generations is determined from the learning curve using the loss function. Result of the damage detection for both of the bearing section (damage level high) and slab section (damage level low) shows high accuracy. Above identification accuracy, bearing section shows higher accuracy. Therefore, it be able to say that proposed damage diagnostic method using CNN is effective for condition diagnosis of bridges.