A Basic Study on Abnormality Detection of Dams by Analyzing Seismic Data Using an Autoencoder

Abstract

We conducted a basic study to establish a method for detecting dam abnormalities using an Autoencoder based on seismic data acquired at dams during earthquakes, with the aim of efficiently assessing dam damage when dams are affected over a wide area due to a large-scale earthquake. Specifically, we conducted abnormality detection using Autoencoder on seismic data from two dams of different types that had temporarily decreased their horizontal primary natural frequencies after large earthquakes in the past, and evaluated the accuracy and analyzed the causes of misjudgment. Furthermore, based on the results of these investigations, we have discussed the challenges and directions for practical application.