Study of static and dynamic damage identification in multifield materials with artificial intelligence, BEM and X-FEM

Abstract

The research proposed in this Thesis approaches the multifield materials by two main branches: the study of the fracture mechanics problem (direct problem), and the development of a damage identification methodology (inverse problem). Numerical methods have been studied for the direct problem and techniques have been proposed so new problems can be treated, notably a new formulation for enrichment function in extended finite element method, and a far field fundamental solution to be used conjointly with the boundary element method. With respect to the inverse problem, artificial intelligence techniques have been combined in a hybrid damage identification scheme, using supervised (neural networks) and unsupervised (self-organizing algorithms) learning techniques, providing excellent identification results despite the presence of high levels of external interference in the measured response. Experimental damage assessment was also investigated in this Thesis, and a methodology using data fusion and a Gaussian mapping has proved to provide good identification results.