In this paper, based on structural tests of soil specimens, a neural network (NN) soil model is incorporated into the finite element method (FEM), and the stability of an excavated soil slope is evaluated by means of finite element-neural network (FE-NN) hybrid algorithms, and some significant conclusions are drawn out.


An important recent interest in neural networks (NNs) within soil mechanics and geotechnique is the question of its coupling with other numerical methods. In the majority of reported cases, the application of NNs leads to a stand alone system. In some cases it is difficult to establish exactly how these new tools fit together with the existing tools (numerical methods). The key to maximizing the benefits of this new NN technology is in its integration into our existing tools, for example FEM, thereby endowing the later with increased capability. This paper targets this issue specifically, evaluating stability for an excavated soil slope employing FE-NN hybrid model, and the numerical results are quite good.


Modeling is a fundamental method in research of engineering problems. Modeling of the observed phenomena, such as the constitutive behavior of material, enables the understanding of phenomena and makes most engineering analysis possible.NNs (Fig. 1) offer a new and fundamentally different method of approaching the task of constitutive modeling. First, the material behavior, as represented in the experimental data, is examined in order to identify the major features of material behavior. Once the material behavior is sufficiently well understood, and its main features identified, then a mathematical model is developed to simulate this behavior. The artificial neurons in the input layer may represent the state of stresses, the state of strains and the stress increments. The units in the output layer may represent the strain increments.

This content is only available via PDF.
You can access this article if you purchase or spend a download.