Optimal Design About Concrete Filled Circular Steel Tubular Members Based On Artificial Neural Network

This paper analyses the behavior of circular CFT (Concrete Filled Circular Steel Tubular Structures) in the load of axial compression, eccentric compression, tension and bending, and the influence of parameters on the ultimate bearing capacity. On the basis of these conclusions, respective artificial neural network models are made by using the characteristics of artificial neural network such as adaptability, fault toleration and fuzziness. For the column in loads of axial compression, five important parameters including ratio of L/D, radius-thickness ratio, confinement, yield strength of steel tubular and yield strength of concrete are selected as input parameters, after trained by 104 groups of data, a five-layer back-propagation network model is established. The test results by 29 groups of data prove the model is available. A universal formula for compression strength is promoted and compared with existing specifications. For the columns in loads of eccentric compression, five important parameters including ratio of L/D, eccentricity, eccentric ratio, yield strength of steel tubular and yield strength of concrete are selected as input parameters, after trained by 52 groups of data, a five-layer back-propagation network model is established. The test results by 13 groups of data prove the model is available. A simple formula used in calculation of compressive capacity is promoted. For the columns of bending structure, five important parameters including radius-thickness ratio, steel ratio, confinement, yield strength of steel tubular and yield strength of concrete are selected as input parameters, trained by 22 groups of data, a five-layer back-propagation network model is established. For the seismic properties of High Strength Concrete filled CFT Column, five important parameters including axial pressure ratio, ratio of steel tubular strength and concrete strength, ratio of steel tubular area, reinforcement ratio, velum steel ratio are selected as input parameters, after trained by 30 groups of data, a five-layer back-propagation network model is established. The test results by 8 groups of data prove themodel is available. The formulas are verified by test data and the results show that the promoted formulas are more accurate, concision and explicit than other existed specifications and it can be used for practical engineering.On the basis of artificial neural network models, two optimal design programs are put forward by MATLAB language to optimize the section of circular CFT columns under axial compression and eccentric compression. It can select an optimum section quickly on the promise of keeping the bearing capacity and economic. By this method, a lot of shortages including local optimum, uncertainties and expert experience in the traditional mathematical model can be dealt with well.