Carrying Capality Assessment Of Double-curved Arch Bridge Based On Finite Element Model Updating Of RBF Neural Network

Double arch bridge is a kind of bridge which is unique in China with ethnic flavor andcharacteristic, it is also a kind of arch bridge built mostly in60~70’s of last century. Becauseof the modular assembling structural pattern and low reinforcement structure used in doublearch bridge, the structural integrity of it is congenitally deficient, in addition to the defect inthe design and construction, in-service double arch bridges have different degrees of damageunder natural environment and overload traffic condition. In order to ensure the smoothnessof traffic and understand the actual working status of bridges (damage condition, actualbearing capacity etc.), scientific evaluation of the actual working status of existing doublearch bridge should be carried out. In this paper, the initial finite element model of double archbridge is modified based on RBF neural network, finite element model that reflects the actualworking status of in-service double arch bridges is established, and on the basis of themodified finite element model, evaluation of the bearing capacity factor of the whole bridgecontrol section and the ultimate bearing capacity of naked arch is conducted.Based on the background of Tang Gou Bridge, RBF neural network is used to modify theinitial finite element model, and the modified finite element model is used as the criterion toevaluate the bearing capacity of bridge. The main work is as follows:1. Investigate the appearance of double arch bridges, comprehensive evaluation of thebridges is carried out according to "Standards for technical condition evaluation of highwaybridges " and "Specification for evaluation of load-bearing capacity of highway bridges". Theactual arch axis and the diseases that have influence on the bearing capacity of double archbridge are fully considered in finite element model, so as to achieve the purpose of modelmodification. Static load experiment is carried out to bridges in field, reasonable experimentalconditions and experimental sections are extracted for the determination of static optimizationsamples of posterior neural network.2. Sensitivity of parameters is analyzed, and design parameters that have significantinfluence on structural static response characteristics (deflection) are selected as themodifying design parameters. Determine the optimal space of the selected parameters; neuralnetwork training samples are reasonably selected on the basis of uniform design theory forneural network training. Based on the trained network, generalization properties of RBFneural network are used to calculate the target value of design parameters, namely the actualvalue of modifying design parameters. In order to verify the modification property of RBFneural network, first-order optimization algorithm in ANSYS is used to conduct the finiteelement model modification, compare these two results and verify the feasibility and utility ofRBF neural network. 3. By using the modified finite element model as the criterion, bearing capacity factor ofbridges is calculated from three aspects, namely the real strength of section, effects of deadand live loads and structural damage. Bearing capacity factor of structure control section iscalculated by considering reduction of arch rib elastic modulus, reduction of arch rib effectivearea and overload to comprehensively evaluate the bearing capacity of bridges. Ultimatebearing capacity of double arch bridge under various kinds of load combination is checkedbased on method of ultimate bearing capacity.