13th International Conference on Fracture June 16–21, 2013, Beijing, China -1- A New Damage Identification Strategy for SHM based on FBGs and Bayesian Model Updating Method Yanhui Zhang1, Wenyu Yang1,* 1 State Key Lab of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan, Hubei, 430074, China * Corresponding author: mewyang@mail.hust.edu.cn Abstract One of the difficulties of the vibration-based damage identification methods is the nonuniqueness of the results of damage identification. The different damage locations and severity may cause the identical response signal, which is even more severe for detection of the multiple damage. This paper proposes a new strategy for damage detection to avoid this nonuniqueness. This strategy firstly determines the approximates damage area based on the statistical pattern recognition method using the dynamic strain signal measured by the distributed fiber Bragg grating, and then accurately evaluates the damage information based on the Bayesian model updating method using the experimental modal data. The stochastic simulation method is then used to compute the high-dimensional integral in the Bayesian problem. Finally, an experiment of the plate structure, simulating one part of mechanical structure, is used to verify the effectiveness of this approach. Keywords structural health monitoring, damage detection, fiber Bragg grating, Bayesian model updating, stochastic simulation 1. Introduction The nonuniqueness of the results of damage identification is the one of the difficulties of the vibration-based damage identification methods. The different damage locations and severity may cause identical response signal. This problem is even more severe for the detection of the multiple damage. The reason is that, the number of the measured points in real application is limited and only the limited modes could be estimated. Furthermore, the modelling error and the measurement noise is usually inevitable, some erroneous modes could have modal parameters closer to the estimated modal parameters than the model with the correct damage locations and amount [1]. By explicitly considering the modelling error and the measurement noise, Bayesian model updating approach is an excellent way to model prediction error and provide the uncertainty information of the damage identification results [2]. Based on the Bayesian formula, Bayesian model updating approach could incorporate the engineering judgments, the mathematical models and the measured data together to make robust identification for damage. The results of damage identification are expressed though the post probability density function (PDF), rather than pinpointing a single solution in the traditional deterministic approach. The post PDF quantifies the confidence level of the identified results, which usually provides an important reference for maintenance decision. The fiber Bragg grating (FBG), considered to be a promising technology, has been increasingly applied to the SHM process. FBG has several advantages, such as immunity to electromagnetic interference, high sensitivity, light weight, and so on. The excellent multiplexing capability of the FBG facilitates its use as a distributed sensor system, which not only monitors the local key parts of the structure but also captures the overall dynamic information. Panopoulou et al. [3] developed a complete damage detection system using FBGs. The dynamic strain response data from the FBG is first measured, then the feature indices are extracted by various signal processing methods, and finally an artificial neural network is utilized to detect and locate damage. This system has been demonstrated by a thin composite panel and a honeycomb structure and is planned for use in a future application of an antenna reflector. This paper uses the distributed FBG as the sensor network and proposes a new strategy for damage detection though two steps, which firstly estimate the approximate damage area from the dynamic strain signal using the statistical pattern recognition method, and then accurately evaluates the
RkJQdWJsaXNoZXIy MjM0NDE=