应用BP神经网络修正有限元模型中的刚度矩阵

Using Backpropagation Neural Networks to Correct the Stiffness Matrix in a Finite Element Model

在有限元素法中，用理论分析建立的有限元模型很难完全与实际相符，这便带来了如何寻找有限元模型的误差源和怎样对它进行修正的问题。用静力试验修正刚度矩阵，再以修正后的刚度矩阵为基准修正其它矩阵是方法之一。目前存在的问题是该方法的修正结果对静位移中各元素相对误差的不一致具有高度敏感性。本文应用神经网络修正有限元模型中的刚度矩阵，既消除了上述敏感性，同时具有可修正性。

It is important how to find the sources of error in a finite element model and how to correct the error because the model established by analysis can not correspond to reality completely. One practical method is to modify the stiffness matrix by a static test and then modify other matrices according to the modified stiffness matrix. The problem with this method is that the result is highly sensitive to the difference of relative errors in the static displacements. In this paper,the backpropagation neural networks was used to solve the problem and satisfactory results were obtained.