Detection of Abrupt Changes of Structure using Regularized Autoregressive Model Joo Sung Kang
ABSTRACT This paper presents a non structural model based damage detection algorithm using meas-ured acceleration data. Damage is defined as an abrupt change which means a change of system parameter of a target structure occurs either instantaneously or at least very fast with respect to the sampling rate of the measurements. The autoregressive model is employed to analyze the characteristics of measured acceleration data statistically. A time windowing technique is employed to make time varying system in order to de-tect abrupt change sensitively. In the time windowing technique, the autoregressive model is estimated sequentially using measured data within a finite time period which called time window. The perturbations due to environmental changes are commonly varied gradually during long time period and the time window size is relatively very smaller than a period of environmental perturbation so it can be assumed that the perturbation of environmental changes can be neglected within the time window. The covariance between residuals and autoregressive coefficients is proposed as a new damage feature in order to complement drawbacks of both of residuals and autoregressive coefficients and to draw more reliable results. A regularization technique is adopted to alle-viate the ill-posedness and to stabilize the instabilities arise from estimating autoregressive coefficients using least squared error method. The extreme value distribution is utilized to detect outliers because damage information almost lies in the tail of distribution and the extreme value distribution is well estab-lished for the tail distribution. The generalized extreme value distribution(GEV) is utilized for simplicity. The weighted average method with normalization of damage feature is adopted to draw integrated decision using damage detection results from each sensor whether the target structure is sound or not. The validity and accuracy of the proposed algorithm will be verified through numeri-cal simulation studies under normal operational condition and severe event such as earth-quake excitation. The moving vehicle load is simulated in case of operational condition and Kobe earthquake and Elcentro earthquake excitation is simulated in case of severe event. A numerical simulation studies will be demonstrated on 2 span continuous truss, 3 span con-tinuous beam and 5 story shear building.
Key Word Detection of abrupt change, autoregressive model, time windowing technique, damage fea-ture, outlier detection, integrated decision
|