Nonlinear structural identification using extended kalman filter

Abstract

The performance of a weighted global iteration for the extended Kaiman filter was evaluated in this study. The problem is to identify the parameters of a hysteretic SDOF system defined by a Bouc-Wen model using a simulated earthquake input-response pair. When no noise was introduced, in most cases studied the weighted global iteration procedure converged to give reasonable estimates provided the ground shaking intensity was high enough to trigger significant yielding. For the few cases that diverged, a slight change in initial estimates, in weight, or in the noise covariance also resulted in meaningful identification. The presence of non-white noise in the observations did not appear to constitute a problem. But identification deteriorated as the noise-to-signal ratio increased. It was also found that the proper selection of observation variables was important. Overall, the extended Kaiman filter with weighted global iteration appears to be a stable tool for this class of nonlinear structural identification.