Researchers at the Advanced Photon Source (APS) use coherent imaging techniques such as Bragg coherent diffraction imaging (CDI) and ptychography to study the response of mission-critical materials and technologies under ​‘live’ conditions. Some examples include imaging the dynamic response of battery electrodes during charge cycling, as well as phonon transport and mechanically induced defects in polycrystalline materials.

The typical image recovery algorithms on which Bragg CDI relies are iterative in nature and hence time consuming and computationally expensive, making real-time imaging a challenge. To address the shortcomings that limit the scope and speed of Bragg CDI, we built and trained 3D-CDI-NN, a deep convolutional neural network that has learned to invert raw diffraction data to the corresponding object structure and strain more than a hundred times faster.

In combination with the APS Upgrade project (APS-U), the development of 3D-CDI-NN will accelerate and enhance 3D nanoscale imaging techniques to enable scientists to better understand the response of materials under real-world conditions, which in turn will lead to better-designed materials for a variety of applications.