Damage Detection of Structures

    by Autoregressive Model and Threshold Value Determination

    Bashir, Muhamed Tahir



Muhammad Tahir Bashir Department of Civil and Environmental Engineering Seoul National University A structure can observe severe loading conditions during its life span which can cause damage to the structure. It has become very important to study the behavior of structures under normal and abnormal conditions to monitor the health of the structure. There is a lot of research going on the damage identification, localization and assessment. Autoregressive model is used to detect the damage in the structure. Covariance between autoregressive coefficients and residual of measured and calculated acceleration is proposed as a damage feature. Non Modal based Scheme is used for structural health monitoring and measured acceleration data from the sensors is utilized here. Initially data is in time domain which is then converted to frequency domain by applying the transfer function. Non Causal filter is designed to filter the lower frequencies which are resulted by perturbations, operational and over loading conditions of traffic.

   As we have the measured signals from the structure, which contains some noise and perturbation resulted by environmental effects, so to remove those perturbations time Windowing Technique is employed. Because those environmental changes occur in long duration of time, while the time window size is so small that, it remains constant in each time window. An algorithm has been developed to find the damage feature in real time. Two span truss bridge is analyzed for damage detection; a damage scenario is created by reducing the area of the truss members. Loading of the truck, car and bus is applied normally and then increasing the loading to study the damage. Algorithm captured both the damage location and timing. Extreme value distribution is utilized for an accurate selection of outliers because extreme value distribution is well established for tail distribution. Threshold value based on optimal sample size and significance level is determined. Some mistakes in previous study to determine the threshold value are also highlighted in this paper.

   A simplified approach is adopted to get the optimal sample size against each significance level. Observation and monitoring time is also studied to see the effects on the threshold value. Limited available data cannot predict the actual situation of the bridges, which are designed for hundred years. So at least one year observation period should be taken to predict the behavior of bridge which is designed for a century.


Key Word

Damage Detection, Time window, Autoregressive Model, Damage Feature, Non Causal Filter, Threshold value, Extreme value distribution.


[Full Text in Korean (PDF:1,138K)] [Prev] [List] [Next]