Abstract by Ryan McCrary
Machine Learning and Full-Pattern Fitting Methods For Geochemical Analyses
The purpose of this research is to improve current techniques for quantitative phase analysis using powder X-ray diffraction data (QXRD) by incorporating machine-learning into our MATLAB-based full-pattern fitting program. In this method of full-pattern fitting, the user selects candidate phases from a library of experimentally determined standards that he/she believes will best fit the pattern observed for the sample. Some of the library standards have been identified as not pure samples of the intended mineral. To solve this problem, we will implement the Rietveld method to identify which standards need to be corrected. This is critical to further work on this project, which will involve training our full-pattern fitting program to recommend which phases to try for any given sample phase pattern.