Access to the mechanical response of materials subject to external loads is provided by modern optical methods such as digital image correlation (DIC). Experiments generate large amounts of data, which have to be analyzed automatically using classical algorithms, suitable models, or, most recently, machine learning (ML). However, experiments are expensive and manual labeling of data is time-consuming and tedious. This is why unsupervised learning and generative methods are moving into focus.

In this work, we use Generative Adversarial Networks (GANs) to generate DIC-like physical displacement fields of fracture mechanical problems. The models are trained with DIC-measurements from real fatigue crack growth experiments. We show how to generate fake DIC data from corresponding finite element simulations using cycle-consistent GANs. With a trained generator we are able to sample simulation-consistent DIC displacement data without any real experiments. Finally, these synthetic datasets can be used to train machine learning models in cases without real experimental data, for example aircraft component tests.