Quantile autoregressive modeling for non-linear change detection in vibrating structural systems

Abstract

Abstract This article presents quantile autoregressive modeling (QAR), a new tool in relation to structural health monitoring addressing the problems associated with sudden changes in linear stiffness, transitions from linear to non-linear structural states and contiguous non-linear changes of states. Acceleration data is modeled using QAR process which involves fitting autoregressive (AR) coefficients at various pre-selected quantiles. The necessity of such a modeling stems from the inability of traditional AR models to account for non-stationary variance and the presence of non-linearity. Percentage changes in the QAR coefficients at different quantiles is proposed as a conditional indicator to detect changes in state of the structural systems. Numerical simulations on a hysteretic system subjected to El-Centro ground motion and a laboratory-scale experiment on a two-storied shear building model undergoing real time mass-loss demonstrate the robustness of the proposed methodology.