At the Advanced Photon Source, Bragg Coherent Diffraction Imaging (BCDI) provides a powerful tool for obtaining high-resolution images from nanocrystalline materials. The raw diffraction data produced by BCDI experiments commonly includes spurious data referred to as ​“alien signal.” To process the BCDI data and produce accurate images, the experimenter must carefully examine the raw data and erase any alien signals. This data cleaning process—which depends on human intuition and judgment—is too slow to handle the massive volumes of data that will be produced after the APS Upgrade. Scientists in the X-Ray Science Division are using artificial intelligence to replace human intuition for a faster and more rigorous approach to data cleaning.

Experimenters can intuitively recognize alien signals in BCDI partly based on the human intuition for patterns; People have an intrinsic ability to recognize a difference between a symmetric shape and an asymmetric shape. In BCDI measurements, clean diffraction data of an unstrained crystal has perfect symmetry. In uncleaned BCDI data, the ​“good” data shows this expected symmetry, while the alien signals don’t follow the same pattern. Thus, the alien signals naturally appear different to the person examining the data.

To mimic the human ability to recognize spurious data, we used unsupervised machine learning to recognize separate clusters of data that correspond to separate diffraction signals. The algorithm then calculates the size and the symmetry properties of each cluster of data, deleting clusters that are very small or don’t fit the symmetry of the rest of the data. The computational algorithm is completed in a few seconds, while the traditional human approach can require many tens of minutes to clean a single dataset.

This AI-based approach will allow BCDI data processing capabilities to keep pace with the technological advances of fourth-generation synchrotron light sources.