A representation learning framework for detection and characterization of dead versus strain localization zones from pre- to post-failure

Abstract Experiments have long shown that zones of near vanishing deformation, so-called “dead zones”, emerge and coexist with strain localization inside deforming granular media. To date, a method can disentangle these dynamically coupled structures from each other, pre- to post- failure, is lacking. Here we develop framework learns new representation the kinematic data, based on complexity grain’s neighborhood structure in kinematic-state-space, as measured by recently introduced metric called s -LID. Dead (DZ) are first distinguished (SZ) throughout loading history. Next dynamics DZ SZ characterized using range discriminative features representing: local nonaffine contact topology force transmission properties. Data came discrete element simulations biaxial compression tests. The deformation found be essentially dual nature. exhibit distinct yet dynamics, separation increasing lead up failure. Force congestion plastic mainly concentrate SZ. Although 3-core network highly prone damage SZ, it robust pre-failure microbands but decimated shearband, leaving fragmented at We also show how condition rolling resistance influence differently, thus casting light controls plasticity perspective emergent structures. Graphic abstract