13th International Conference on Fracture June 16–21, 2013, Beijing, China -2- damage based on Bayesian model updating method using experimental modal data. Lastly, the stochastic simulation method is used to solve the Bayesian computation issue and generate the samples of the damage parameters identified. The following paper is organized as follows. Section 2 summaries the new damage detection strategy. Section 3 is the experimental verification section: firstly describes the experimental apparatus and procedures, and then presents the detection results of Step 1 and Step 2, respectively. A few conclusions are discussed in Section 4. 2. Theoretical Background 2.1. New Damage Identification Strategy The traditional damage detection methods for SHM can be classified into model-based method and non-model-based method [4]. Here, the “model” refers as the physical model of the real mechanical or civil structure. The non-model method usually directly uses the signal processing or statistical method to determine whether the damage occurs. This method is simple and straightforward, but helpless for quantifying the damage, such as the size, orientation and trends of the crack. Alternatively, the model-based method requires an accurate physical model and could quantify the damage but on the cost of intensive computation. Absorbing both advantages of the non-model method and the model-based method, a new damage detection strategy is proposed based on FBG and Bayesian model updating method. This process consists of the following two phases. (1) Roughly estimate damage area based on the distributed dynamic strain signal with the recognition accuracy of the gage lengths of FBG without a detailed analytical model. (2) Accurately identify the size, direction and depth of the damage with the recognition accuracy of the accuracy of the physical model based on Bayesian model updating method. The details of the process are showed in Fig. 1. Figure 1. Detailed process of the new damage detection strategy Based on the strain measurement, the dynamic strain response of FBG is more sensitive to the local small damage than the traditional displacement or acceleration measurements. But the environmental and operational variations, such as the change of temperature, usually disguise the signal variation induced by damage and cause the false-positive indication. So in Step 1, the dynamic strain signal is decomposed into the damage-sensitive signal component using the Hilbert-Huang Transform (HHT) method, the autoregressive (AR) model is then used to exact damage sensitive features, and lastly the Mahalanobos distance-based method is used to determine the approximate damage area, named as the damage suspicious region. But the identification accuracy of this region is low, because that is mainly affected by the grating length of FBG and the layout density of the sensor network.
RkJQdWJsaXNoZXIy MjM0NDE=