Structural analysis of the isotropic composites by combining limit analysis with artificial neural networks

Abstract

In this paper, we describe an approach for the structural analysis of isotropic composite structures by combining the reduced order model of limit analysis with artificial neural networks (ANN). At first, the ANN is introduced in order to represent the effective strength surface of isotropic composites implicitly in the macro principal stress space. The input neurons are the macro principal stress components, while ANN output neuron is the limit load factor. In order to estimate the limit load factor, the reduced order model of limit analysis for the representative volume element is used. Then, the structural analysis can be easily implemented in the computational framework of the FEM. Macro stress is evaluated by using the elastic analysis of the homogenized composite structure, and the safety is estimated via the effective strength surface represented by ANN. As a result, structural analysis of composites can be reduced into that of common homogeneous materials. Numerical examples show that the proposed method is an efficient approach of the structural analysis of composites structures.