Experimental Study Of Damage Identification For Long-span Self-anchored Suspension Bridge

As the long-span suspension and cable-stayed bridges are widely used in the world, the bridge safety and maintenance costs become the one of the most attention problem for the bridge owners and government. Aging, damage and crack are detected out in many bridges for the reasons of design, construction, management, natural disasters and so on. If there is any serious bridge damage, national economic, social stability, people's lives and properties will be affected profoundly. In order to ensure bridge structure safety, it is essential to assess the bridge health condition. Bridge structural health monitoring system can obtain the real-time status of the bridge operational condition, so it has important theoretical and practical value to study the bridge structure health monitoring.As one algorithm of the machine learning based on the statistical learning theory, Support Vector Machine (SVM) is specifically to the small samples learning case. SVM is one powerful tool in pattern recognition with the high accuracy and good generalization capability, and SVM pattern classification model is set up on the ground of the deflection change under different structural damage locations and degrees to identify the bridge damage. Basing on the damage experiment of one certain long-span self-anchored suspension bridge, the application of SVM to this damage experiment is studied. The results show that the proposal method can accurately locate the damage. The specific studies are listed as following:1. The overview of bridge structural health monitoring and damage identification are briefly introduced, and the difficulty of bridge damage identification are presented by inductive analysis.2. The basic theory and methods of the statistical learning and SVM are systematically introduced, the construction method of the structural damage index is presented also.3. The damage experiment of one certain long-span self-anchored suspension bridge model is taken as the study object, the bridge structural deflection under different load case is considered as the damage feature vectors, and the classification function of SVM is adopted to locate the damage of this bridge model. The training samples are consisted of the theoretical calculated results of finite element model, and the test samples are composed of the experimental measured bridge reflections, and the SVM is adopted to locate the damage aided by the sub-regional and multi-step damage identification schedule. The influences on the identification accuracy caused by different kernel functions and the corresponding parameters are compared and analyzed. Some improved testing methods for the bridge damage identification are put forward on the base of recognition results.Finally, the main conclusions are figured out and some damage identification problems are also discussed.