Implementation of online model updating with ANN method in substructure pseudo-dynamic hybrid simulation

Abstract

Substructure pseudo-dynamic hybrid simulation (SPDHS) is an advanced structural seismic testing method which combines physical experiment and numerical simulation. Generally, the key components which display nonlinearity first are taken as experimental substructures for actual test, and the remaining parts are modeled in simulation. Model updating techniques can be effectively applied to enhance the model precision of nonlinear numerical elements. Specifically, the constitutive model of the experimental substructure is identified online by the instantaneously-measured data, and the corresponding numerical elements with similar hysteretic behaviors are updated synchronously. Artificial neural network (ANN) can recognize the system which cannot be represented by definite numerical model, and thus avoids the structural response distortion caused by the inherent numerical model defects. In this study, a framework for online model updating in SPDHS with ANN method is expanded to implement actual test validation. Moreover, the effectiveness of ANN method is demonstrated by practical tests of a two-story frame model with bending dampers. Additionally, the unscented Kalman filter technique and offline ANN identification approach are both examined in the test validation. The experimental results show that, under the identical loading history, the online ANN method can significantly reduce the model errors and improve the accuracy of SPDHS.