Application of neural networks to the correction of a stiffness matrix in a finite element model

Abstract

It is important how to find the sources of error in a finite element model and how to correct them because the model established by analysis could not correspond to the real structure completely. One practical method is to correct the stiffness matrix by a static test first, then correct other matrixes according to the corrected stiffness matrix. The problem with this method is that the result is often highly sensitive to differences in relative errors in static displacements. The neural networks were applied in this study and preliminary results were obtained. The new approach is able to get rid of the high sensitivity and can modify stiffness matrix well. © 2011 Springer-Verlag.