2023 Volume 79 Issue 13 Article ID: 22-13013
The purpose of this study is to perform immediate damage detection of the instrumented structure during a large earthquake based on the machine learning of the linear response recordings due to the frequently observed small and medium earthquakes. Preparing a model of the structure that considers nonlinearity with trilinear model, both linear and nonlinear responses are calculated by giving input ground motions of different strength. Then, the autoencoder is trained with linear response records, and weights of the network are determined. By applying the autoencoder to several response recordings of different nonlinear levels considering large earthquakes, the value of the obtained reconstruction error as an anomaly index showed variation according to the damage level of the structure. It was found that immediate detection of structural damage caused by a large earthquake may be possible by using nonlinear response recordings.
