用神经网络确定梁上裂纹位置的研究

Locating a crack in a beam by neural networks

以间隙单元模拟裂纹 ,用有限单元法对带有裂纹的简支梁进行分析 ,计算了具有不同位置裂纹的梁在冲击荷载下的时间历程响应 .建立了人工神经网络模型 ,它的输出层单元用于裂纹端点的定位 ,输入数据为梁上结点的振动位移 .该网络用位移响应和裂纹始端坐标的数据进行训练 ,数据的正向计算和误差的反向传播交替进行 ,直至网络的误差收敛到一个极小的水平 .在输入梁的振动位移响应数据时 。

A simply supported beam with a crack was studied by finite element analysis methods. A crack was simulated by gap elements , and the impulsive time history responses of a set of beam models with different crack positions were calculated. In this research, a type of artificial neural network was built, in which, one node of the output layer was used to locate crack origin, and the displacement of the nodes in the beam as input data. The network was trained with the data of the time history displacement and the coordinate of the beginning point of the cracks. The data feedforward calculation and the error back propagation adjustment were performed alternatively, until the network error converged at an infinitesimal level. The trained network can give corresponding crack coordinate by recall, when put into the dynamic response data of a beam. Comparison with the accurate value showed that results given by the network were satisfied.