Study On Seismic Behavior Of Abnormal Joints Of Reinforced Concrete Frame Based On Neural Networks

This study investigates the feasibility of using neural networks to analyze the seismic behavior of abnormal joints of reinforced concrete frame; the progress and basic properties of the neural networks are introduced. Several problems, involving collection of samples, construction of neural networks topology, determination of the number of neural cells and choice of training method, are discussed.The research mainly includes following contents:1. According to the mechanism of Feed forward Neural Network, some issues about Neural Networks-based modeling of hysteretic behavior of reinforced concrete member are discussed. Back Propagation(BP) Neural Networks model is build to predict the hysteretic behavior of the reinforced concrete abnormal joints under low reversed cyclic loading.2. Neural Networks technology is applied to build the complex non-linear relationship between the anti-shear capacities of the reinforced concrete abnormal Joints under low reversed cyclic loading and the main factors. Back Propagation Neural Networks model are build to predict the anti-crack capacity and bearing capacity.3. According to the mechanical character and failure model of reinforced concrete abnormal joints under low reversed cyclic loading, a damage analysis method with sub-structure is developed, and a new damage model based on conventional methods and neural networks are put forward.According to the compared with the test data, satisfactory results are achieved. Therefore, it is proven that the neurocomputing is a practical and promising tool for the analysis on seismic behavior and mechanical properties of abnormal joints of reinforced concrete frame.