Structural nonlinear damage identification based on Bayesian optimization GNAR/GARCH model and its experimental study

Abstract

Cracks and bolt looseness are common damages in a steel structure. Under the time-history loading, nonlinear characteristics of variable stiffness exist in cracks, and bolt looseness has complex nonlinear characteristics of stiffness and damping. To effectively solve the problem of structural nonlinear damage identification, an output-only damage identification method based on the GNAR/GARCH (general expression for linear and nonlinear autoregressive/general autoregressive conditional heteroskedasticity) model with Bayesian optimization is proposed. First, the stationarity test and continuous difference processing was carried out on the measured acceleration data, and an order determination method based on Bayesian optimization was proposed to automatically select the optimal model order. Then, a model structure optimization method based on Bayesian optimization was proposed to select the optimal model structure of a GNAR/GARCH model, and a GNAR/GARCH model was established based on the measured acceleration data for damage identification. Finally, a new damage indicator PIDI (probabilistic Itakura distance indicator) was proposed to judge the structural healthy status. The nonlinear damage experiment of three-story frame and relay tower was used to verify the effectiveness of the proposed method. In addition, in order to illustrate the efficiency of the proposed method for nonlinear damage identification, a comparative study was carried out using other two damage identification methods. The results show that the nonlinear damage identification method based on the GNAR/GARCH model with Bayesian optimization can easily identify the nonlinear damage of the frame and relay tower structure, and it can effectively identify the structure damage with single or multiple nonlinear damage sources.