Research Of Distortion Data In Bridge Structure Healthy Monitoring Based On Support Vector Machine

In recent years, structural health problems of large bridges have been increasingly concerned about. Because of the natural factors and artificial factors, the structures of bridges are damaged during the long-term process of time in which bridges have been used. It will cause the serious accidents happening if the damages are not found opportunely and handled properly. So it is very import for bridges to be monitored in real time and evaluated accurately.Support Vector Machines has been widely used in pattern recognition and regression. In this paper, duo to the superior classification and prediction performance of SVM with small samples, Classification prediction model has been built with distortion data collected in Hangzhou Bay Bridge, and with which classification and prediction experiments are done. The model will be confirmed if the evaluation of bridge structure health is accurate duo to the model with experiments results.Bridge structure health monitoring data can be divided into t static data and dynamic data. Because of containing much bridge structure information, static data plays an important part in bridge structure evaluation. As one of the static data, distortion mainly displays the linear variety of main span and side span of bridge. The distortion data collected in Hangzhou Bay Bridge is sequential and non-linear, the data set is small sample data set and there is a multiple monitoring points rule found during the research. So based on the singal monitoring point and multiple monitorting points, two kinds of experience sample are built. Because of the character of distortion, SVM has been chosen for experience after the compariton between Gray theory and SVM. In this paper, two models are built called single monitoring point sample model and multiple monitoring points sample model. LS-SVM is chosen to be the research method and RBF kernel is chosen to be the research kernel function due to performance comparition result among kinds of SVM. Two models are verified with kinds of ways of classification and prediction experiments. A data analysis platform is designed and developed based on the experiments with which MATLAB and Visual Studio 2008 are been connected. The main functions of the platform are to research the experiment data, build the sample data, and invoke the algorithms.The results show that both of two classification and prediction models have good performance, and the multiple monitoring points sample model gets a better experiment results. It is sure that the state of bridge structure health can be evaluated well with the classification prediction model.