13th International Conference on Fracture June 16–21, 2013, Beijing, China -6- in which the blue bar indicates that no damage occurred, the red bar means the damage, and the green horizontal line is the threshold which is calculated assuming that the distribution of the square of the Mahalanobis distance is chi-square distribution with the DOFs equal to the AR model order. Figure 4. Results of damage detection and location in channel 19, 26, 27 and 28 There is no damage occurred when the number of blue-bar is larger than the red-bar in each state, otherwise, the damage appears. Observing the amount of the red bars, the number of the red bars in the channel 19, 27, and 28 is large, so the conclusion is that the approximate location of the damage is near the position of the channel 19, 27, and 28, but lacking more detailed information, such as the direction, size and severity about the damage. So next the Bayesian model updating method is used to identify the specific parameters of the damage. 3.3. Detection Results of Step 2 In order to identify the damage parameters, the plate structure is divided into the 100 rectangular elements of 10 by 10. Then the non-dimensional parameter, representing the contributions of the element stiffness to the system stiffness matrix, is introduced to model the damage. Based on the detection results of Step 1, only the six parameters near the damage approximate area, θi (i = 1, 2, …, 6), are updated in Bayesian model updating method of Step 2. 3.3.1. FE Model Refinement A precise model is required before applying Bayesian model updating method for damage identification, so the FE model is refined to minimize the model error by conducting a series of parametric analysis.
RkJQdWJsaXNoZXIy MjM0NDE=