NOLINEAR DAMAGE IDENTIFICATION BASED ON MSVAR MODEL

Abstract

Damage in real engineering often exhibits nonlinear characteristics. For example, crack opening and closing can lead to bilinear stiffness of structures. Using simple linear model will miss the nonlinear information related to damage, and reduce the damage identification reliability. To solve this problem, a new nonlinear damage identification method based on the Markov-Switching Vector Auto-Regression (MSVAR) model is proposed. This method assumes that the damage will cause nonlinear characteristics in the structural responses. Since the hidden state smooth probability of the MSVAR model can reflect nonlinear changes in the data, the smooth probability-based information entropy is used as the damage warning indicator to monitor the damage state of the structure. For cracks, their locations can be infered using the damage location vector constructed by autoregressive coefficients of the MSVAR model. The crack width is represented as the interlayer displacement at the crack location (when the smooth probability switches from 0 to 100%). A numerical example and an experimental test are introduced to verify the effectiveness of the proposed method in the identification of crack damage.