REAL-TIME RECONSTRUCTION SIMULATION BASED ON AUTONOMOUS BASIS FUNCTION SELECTION

Abstract

In recent years, the importance of improving the resilience of infrastructure facilities against unexpected large-scale disasters has been pointed out. This study is fundamental research for the development of a monitoring design method to contribute to the improvement of resilience performance from the structural design side. The basic framework of the sensor placement optimization method and the adaptive real-time prediction method is proposed for the case of an embankment on a liquefiable sand layer. A set of spatial basis functions is derived from the numerical analysis results, and a method that effectively utilizes the characteristics of the spatial basis functions is proposed. For the sensor placement optimization, we confirm that the greedy method with QR-pivoting provides valuable information from an engineering viewpoint. For adaptive real-time prediction, we proposed a method that uses the marginalized likelihood to immediately select the number of spatial basis functions for prediction (model selection) and then predicts the component coefficients using a Kalman filter. Finally, the numerical results of the input seismic ground motions for validation are compared with the adaptive real-time prediction by the proposed method. It was confirmed that the proposed method improves the prediction accuracy by selecting appropriate models in real-time according to the input seismic ground motion characteristics.