Girder Bridge's Damage Identification Based On Support Vector Machines

Existed bridge structures often suffer different kinds of damages, the accumulation of such damages may results in disastrous accidents. Therefore, as the core of bridge health monitoring systems, structural damage detection theories and methods have become the studied focuses of many scholars and engineers to work out. After outlined the state-of-art of the structural damage identification, basic theory and methods of Support Vector Machines (SVM) are described. As a sort of statistical learning method, SVM is applied to damage identification of girder bridges from the viewpiont of pattern recognition.Then some valueable results are obtained. The main research work is summarized as follows:1. The basic concepts of bridge health monitoring and damage identification are presented. The contemporary methods of bridge damage identification are summarized and reviewed.The major difficulties of damage identification are also analysed briefly.2. SVM's basic theory and methods are described briefly and systematically.3. A sort of method based on the theory of SVM has been proposed in the application of bridge damage identification. The in-depth discussions have been focused on feature selection and extraction, construction of feature vectors, selection of SVM types and kernel's parameters, and the evaluation of SVM methods.4. Numerical simulation calculation and analysis have been applied on a reinforced concrete simply-supported girder bridge.Dynamic fingerprints before and after damage have been extracted by self-programmed software.According the need of damage identification, feature vectors have been established.Then sample sets according to the need of the damage detection are obtained.The sample set have been trained by Support Vector Classification (SVC) and the damage location is achieved. Researches also have been done on the influences of the selection of feature vectors and parameters of SVC.Thereafter, several dregree of damages at one certain damage location have been carried out by the aid of Support Vector Regression (SVR) .Results of average level for the identification of damaged extent are presented.At last; the change of identification and location of damage has been discussed to two kinds of feature sample sets under the influence of noise factors.Finally, the main conclusions are figured out and some damage detection problems are also discussed.