A dual domain decomposition method for finite element digital image correlation

The computational burden associated to finite element based digital image correlation methods is mostly due to the inversion of finite element systems and to image interpolations. A non-overlapping dual domain decomposition method is here proposed to rationalize the computational cost of high resolution finite element digital image correlation measurements when dealing with large images. It consists in splitting the global mesh into submeshes and the reference and deformed states images into subset images. Classic finite element digital image correlation formulations are first written in each subdomain independently. The displacement continuity at the interfaces is enforced by introducing a set of Lagrange multipliers. The problem is then condensed on the interface and solved by a conjuguate gradient algorithm. Three different preconditionners are proposed to accelerate its convergence. The proposed domain decomposition method is here exemplified with real high resolution images. It is shown to combine the metrological performances of finite element based digital image correlation and the parallelisation ability of subset based methods. keywords parallel computing, global DIC, high performance computing, substructuring, finite elements, domain decomposition