Accelerating Microstructural Analytics with Dask for Volumetric X-ray Images

While X-ray microtomography has become indispensable in 3D inspections of materials, efficient processing of such volumetric datasets continues to be a challenge. This paper describes a computational environment for HPC to facilitate parallelization of algorithms in computer vision and machine learning needed for microstructure characterization and interpretation. The contribution is to accelerate microstructural analytics by employing Dask high-level parallel abstractions, which scales Numpy workflows to enable multi-dimensional image analysis of diverse specimens. We illustrate our results using an example from materials sciences, emphasizing the benefits of parallel execution of image-dependent tasks. Preliminary results show that the proposed environment configuration and scientific software stack deployed using JupyterLab at NERSC Cori enables near-real time analyses of complex, high-resolution experiments.