Data-driven adaptive mesh refinement for enhanced FE-DIC in the presence of cracks

Global Digital Image Correlation (DIC) allows for accurate characterization of fracture phenomena through robust kinematic registrations. The displacement discontinuity can be trustfully highlighted using a finite element mesh, whereby the identification of complex crack patterns becomes possible. However, this operation is computationally expensive at the pixel level if no mesh optimization is performed. In this work, an adaptive meshing framework for Finite Element-based DIC equipped with a data-driven damage classification step is proposed. Benefiting from a multi-variate dataset provided by DIC results (i.e., strain and registration error), a Gaussian mixture model fitting algorithm classifies the specimen into different regions (i.e., damaged and undamaged ones). The FE mesh for the damaged zone is progressively refined with a conforming mesh generation, which significantly improves the image registration. Different benchmarks with varying complexity of crack patterns are investigated as proof of concept. It is shown that the proposed framework successfully obtains extremely refined representations of crack networks at the pixel level. The numerical implementation is minimal and non-intrusive within an existing FE-DIC code.