Estimation of Seismic Displacement Response Using a Kalman Filter with Data-Driven State-Space Model Identification

EasyChair Preprint 10083

Abstract

This paper proposes a system identification method combining a data-driven state-space model and the unscented Kalman filter (UKF) to estimate displacement time histories of systems with complex material nonlinearities under seismic excitations. In this method, the state-space equations are first constructed based on a series of polynomial functions using training data which are the dynamic responses generated from a finite element (FE) model with complex hysteresis behaviors. Specifically, the state-space equations are trained separately for the linear and nonlinear regions of the responses. Subsequently, the trained state-space equations are employed in the UKF to estimate the system displacements. In this stage, only the accelerations of the test data are regarded as observations. In the UKF, the time-variant process noise covariance matrix and the time-invariant measurement noise variance are inferred by the Robbins-Monro algorithm and the Markov chain Monte Carlo method, respectively. The proposed method is demonstrated on an FE bridge pier model using different input ground motions, and the results show that the proposed approach enables to accurately estimate the system displacements.

Keyphrases: Displacement estimation, State-space model identification, Unscented Kalman Filter, data-driven approach, seismic response

@booklet{EasyChair:10083,
  author    = {Yuki Kakiuchi and Yaohua Yang and Masaru Kitahara and Tomonori Nagayama},
  title     = {Estimation of Seismic Displacement Response Using a Kalman Filter with Data-Driven State-Space Model Identification},
  howpublished = {EasyChair Preprint 10083},
  year      = {EasyChair, 2023}}