GSA Annual Meeting in Phoenix, Arizona, USA - 2019

Paper No. 6-8
Presentation Time: 10:45 AM

DATA-DRIVEN QUANTITATIVE PHASE ANALYSIS WITH ROCKJOCKML


BICKMORE, Barry R., Department of Geological Sciences, Brigham Young University, S389 ESC, Provo, UT 84602

X-Ray powder diffraction (XRPD) is a standard technique for quantitative phase analysis (QPA) of geological materials, but while there are limitations to the accuracy of the technique itself, in practice the most important limiting factor is user experience. The more phases there are in a material, the more non-unique its XRPD pattern becomes, to the point that simple search-match routines, involving large databases of reference materials, cannot carry the analysis to an accurate conclusion. Instead, the analyst must make a number of choices during the analysis process, e.g., which phases to include in the analysis and which measures of quality to weight more heavily.

My group is developing a new program, named RockJockML, for QPA of XRPD patterns of whole rocks and sediments, and our design attempts to minimize user error and automate the analysis as much as possible. The numerical engine and database is based on that of RockJock, a full-pattern fitting program by Dennis Eberl, which was programmed in Microsoft Excel. RockJockML is programmed in MATLAB, and is much faster, with optimization times typically < 1 s, rather than 10-30 min for equivalent calculations in the original Excel versions. This substantial speed increase allows the analyst to try many more variations of the analysis than would otherwise be possible. The program also uses the crowd-sourced database of mineral localities at mindat.org to calculate relative abundance of phases and probabilities of different phases occurring together, in order to suggest phases to include in the analysis.

We are now beginning the process of using machine learning techniques to more fully automate the analysis process. In its present form, the program is very user friendly, and is available for MATLAB users, and in standalone versions for Windows 10 and MacOS.