Based On Neural Network Bridge Fatigue Damage Assessment Study

At present, the large numbers of railway steel bridges constructed early are presenting their aging status in china. Because constructing techniques were limited and the bridges were designed according to a lower load criterion at that time, the fatigue problems were not fully considered even not considered. In addition, due to the increment of railway traffic loading and raised service speed, , the fatigue damage of bridges becomes more and more serious, and probably results in unsafe service for these steel bridges. How to appropriately evaluate the state of these steel bridges and analyze their fatigue damage and remaining fatigue life has become very important.As the part of "The research on threshold values classification of health monitoring and fatigue life evaluation of the Nanjing Yangtze River Bridge (NYRB)", the work in this thesis is to study on the time history of stress and fatigue damage of NYRB's main members under train loading, and to use BP neural networks to evaluate the fatigue damage of NYRB's main members with the help of finite element analysis and MATLAB. The main contents are detailed as following:1. Comprehensively review state-of-arts of fatigue evaluation of bridges, summarize some problems and development trends of fatigue evaluation of bridges, and introduce the basic concepts used in the analysis of fatigue damage.2. Systematically introduce the fatigue evaluation methods, the steps for evaluating the fatigue damage and remaining fatigue life. The basic theories of artificial neural networks, BP neural networks, BP classic algorithm and LM algorithm in detail are also introduced and analyzed. The basic principle for the damage identification based on BP neural networks is discussed and the way to fatigue evaluation for NYRB is designed.3. Build the 3D finite element model of NYRB. Transient dynamic analysis is completed in the whole course of train's passing through, after programming the train axle loads measured by NYRB monitoring system to the nodes of longitudinal girder as a set of moving concentration loads with constant velocity, and the train axle weight as mass element on the corresponding nodes.4. Calculate the stress time histories of the main members by using the finite element model of NYRB. The stress level of main members is low under the action of running trains, and most of the stress is lower than 20MPa. Based on the platform of MATLAB, the program of rain-flow counting method is compiled to treat the stress time histories, and the fatigue stress spectra are obtained after that. The fatigue damage of the main members is calculated with reference to the empirical fatigue life formulas of BS5400, and the calculation results show that the fatigue damage is very low.5. The train axle load and axle base are used as input samples, and the fatigue damage of the main members is used as output samples. The network structure of nonlinear mapping between the input samples and output samples is built by BP network structure, and the generalization ability of that is tested by test samples. The results show that the network using for fatigue damage evaluation of the main members of NYRB is effective.