Modelling the impact of a material’s mesostructure on device level performancetypically requires access to 3D image data containing all the relevantinformation to define the geometry of the simulation domain. This image datamust include sufficient contrast between phases, be of high enough resolutionto capture the key details, but also have a large enough 3D field-of-view tobe representative of the material in general. It is rarely possible to obtain datawith all of these properties from a single imaging technique. In this paper, wepresent a method for combining information from pairs of distinct but complementaryimaging techniques in order to accurately reconstruct the desiredmulti-phase, high-resolution, representative, 3D images. Specifically, the deepconvolutional generative adversarial networks to implement super-resolution,style-transfer and dimensionality expansion. It is believed that this data-drivenapproach is superior to previously reported statistical material reconstructionmethods, both in terms of its fidelity and ease of use. Furthermore, much ofthe data required to train this algorithm already exists in the literature, waitingto be combined. As such, our open-source code could precipitate a stepchange in the materials sciences by generating the desired high quality imagevolumes necessary to simulate behaviour at the mesoscale.
展开▼
