Research On Structure Damage Identification Method Based On Support Vector Machine And Vibration

In this dissertation, the current state of the researches and the developmental trend about structural damage identification are simply depicted, as well as the rationale of support vector machine (SVM) and the arithmetic is also introduced. The method based on SVM and vibration, which contains the structure modal parameter identification and pattern recognition, is presented in the dissertation. By the analysis of the numerical simulation example and the damage experiment of steel beam, the results prove that it improves the traditional method to increase the precision, systematic, efficiency and robustness of the identification.The basic theory of structure damage identification by using the methods of dynamics fingerprint, such as modal flexibility and modal curvature, are detailed. Based on the numerical simulation of the simply-supported beam, not only are the diagnosis process of the methods introduced, but also the diagnosis ability and anti-noise ability of damage identification are discussed. It is the same with the damage experiment of steel beam. The experiment modal data of the damage are collected. When the damage identification is obtained by using the index of modal flexibility and mode curvature, the sensitivity of the index to damage is verified. The experimental results prove that both the two indexes have better sensitivity for the characteristics of damage structure. Because of the poor anti-noise property of the modal curvature index, it has difficulties for the practical engineering of structure damage identification; on the other hand, the index of modal flexibility has high anti-noise ability for the identification of damage's location of structure, so this index is feasibility to the structure damage identification.The SVM which is based on the statistical learning theory is one of the principles of pattern recognition method. The inherent defects of neural network are overcome by using the classification method of SVM, such as the problems that network structure is difficult to ascertain, over fitting, under fitting and local minimum, also it has high classification rate under the small samples situation. The simulation results of the simply supported beam prove that the network of SVM are trained by using the samples of modal flexibility index, and has high identification rate for the damage's location and damage's severity of beam. According to inputting some random noise to the sample, the anti-noise ability of SVM has been discussed, and the results prove that the SVM has nice anti-noise ability. When the results of previous research are applied to the actual project, through the damage identification of the test model of the steel beam, in a small noise levels, the damage identification results are consistent with the test phenomenon. Consequently, the results show that this approach is a promising method for damage identification.At last, the main contributions and conclusions of this dissertation are summarized and some problems which need further research are set forward.