In this study we explore experimentally the effects of fault heterogeneity on the rupture nucleation. We conducted friction experiments between two polycarbonate plates, with a periodically heterogeneous friction interface. The rupture propagation is monitored with an ultra‐fast video camera by taking advantage of the photo‐elastic properties of the material used. We show that the nucleation process does not always consists of a monotonic growth of the rupture velocity. Instead, the rupture front advances with an alternation of slow and fast episodes that accelerates until it reaches a point at which fast propagation dominates. This complex nucleation process is compared to the results predicted by previous numerical studies on heterogeneous interfaces. Finally, we test whether it is possible to describe this complex nucleation process as a homogenized nucleation. We also point out a large variability in the rupture process due to the uncontrolled stress heterogeneity that occurs during the loading process.
The numerical modelling of the behaviour of materials at the microstructural scale has been greatly developed over the last two decades. Unfortunately, conventional resolution methods cannot simulate polycrystalline aggregates beyond tens of loading cycles, and they do not remain quantitative due to the plasticity behaviour. This work presents the development of a numerical solver for the resolution of the Finite Element modelling of polycrystalline aggregates subjected to cyclic mechanical loading. The method is based on two concepts. The first one consists in maintaining a constant stiffness matrix. The second uses a time/space model reduction method. In order to analyse the applicability and the performance of the use of a space–time separated representation, the simulations are carried out on a three-dimensional polycrystalline aggregate under cyclic loading. Different numbers of elements per grain and two time increments per cycle are investigated. The results show a significant CPU time saving while maintaining good precision. Moreover, increasing the number of elements and the number of time increments per cycle, the model reduction method is faster than the standard solver.