Estimation study of stress concentration factor of crack structure based on BP neural networks

Stress concentration factor is an important parameter to describe the stress concentration phenomenon and crack propagation process of structure. In order to respond the fatigue characteristic of crack structure, quantitative prediction model of stress concentration factor is build. The stress of box structure is analyzed by using the FEM. The stress state used for the calculation of stress concentration factor is extracted from the path between roots of crack and the stress release area according to the FEM result. Then the stress concentration factor is solved according to the extracted stress state. The prediction model of stress concentration factor based on BP neural networks is achieved by nonlinear training the data, the parameter of crack is the input of BP neural networks and the stress concentration factor is the output. The average prediction accuracy up to 91.4% is achieved by using the nonlinear mapping network, which makes the precise estimation study of relationship between stress concentration factor and crack parameter come true.