Machine Learning Automated Analysis of Enormous Synchrotron X-ray Diffraction Datasets

X-raydiffraction (XRD) data analysis can be a time-consumingandlaborious task. Deep neural network (DNN) based models trained withsynthetic XRD patterns have been proven to be a highly efficient,accurate, and automated method for analyzing common XRD data collectedfrom solid samples in ambient environments. However, it remains unclearwhether synthetic XRD-based models can be effective in solving micro(& mu;)-XRDmapping data for in situ experiments involving liquid phases, whichalways have lower quality and significant artifacts. In this study,we collected & mu;-XRD mapping data from a LaCl3-calcitehydrothermal fluid system and trained two categories of models toanalyze the experimental XRD patterns. The models trained solely withsynthetic XRD patterns showed low accuracy (as low as 64%) when solvingexperimental & mu;-XRD mapping data. However, the accuracy of theDNN models significantly improved (90% or above) when we trained themwith a data set containing both synthetic and a small number of labeledexperimental & mu;-XRD patterns. This study highlights the importanceof labeled experimental patterns in training DNN models to solve & mu;-XRDmapping data from in situ experiments involving liquid phases.