Extraction of Dynamic Soil Properties Using Extended Kalman Filter

Abstract

A procedure for nonlinear system identification of soils using strong‐motion records obtained in a downhole seismograph array is presented. This study uses, in conducting the identification, the extended Kalman filter, which is a sequential parameter estimation scheme. The Bouc‐Wen model is used in characterizing the nonlinear backbone curve of soils, and a generalized Masing's rule is implemented to model stress reversals. The stress‐reversal history obtained from earthquake data may be ambiguous. To avoid using a faulty stress‐reversal history, only a fraction of the earthquake data is used. The ground is modeled as a three‐degree‐of‐freedom hysteretic structure. Identification is first carried out based on soil hysteresis loops, using a weighted global iteration scheme. These results are then incorporated into the identification of the ground. The parameters identified yield shear wave velocities that are in good agreement with those from cross‐hole measurements. Similar agreement is also seen between the calculated accelerations and the actual records.