GSA Connects 2022 meeting in Denver, Colorado

Paper No. 113-1
Presentation Time: 1:30 PM

SAND ANALYSIS AND GEOGRAPHIC PROFILING: FINDING THE CORRELATION OF COMPOSITION PROFILES USING THE COMBINATION OF XRF AND TRACE ELEMENT ANALYSIS


MAMEDOV, Sergey, HORIBA Instruments Inc., 20 Knightsbridge Rd, Piscataway, NJ 08854

This presentation aims to demonstrate the application of Data Fusion technology to classify sand from different sources. We will likewise show that Data Fusion Technology can improve the results of such classification. We will then compare and discuss the results of single source and data fused analysis. XRF is a valuable tool for identifying substances and confirming their identity without sample preparation. XRF emission lines, specific to particular elements, enable elemental and chemical identification. Statistical methods applied to a set of X-ray Fluorescence (XRF) spectra show the capability of detecting the slightest differences in the elemental composition of sand. Therefore, it plays a vital role in helping to determine the origin of sand. To illustrate the above, XRF spectra of sand will be highlighted in this presentation, and the ability of XRF to identify sand from different geographic locations, including the USA, Europe, and the Middle East.

The XGT-900 XRF analytical microscope was used in this study. Statistical data analysis such as Principal Component Analysis (PCA) or Partial Least Squares Discriminate Analysis (PLS-DA) takes these parameters into consideration.

We collected and analyzed XRF spectra of sand from different locations (USA, Europe, and the Middle East) (<400 spectra). These spectra were used to build a data set for classification. The results show that PCA allows one to differentiate samples with very similar spectra features. To improve the accuracy of classification, the method of Data Fusion was applied to two sets of data: major element concentration and trace element profiles. Examples of XRF spectra and trace element analysis of sand are presented, and the results are compared and discussed. Data shows that the location of an unknown sample of sand may be predicted using PCA or/and PLS-DA.

The combination of X-ray fluorescence and trace element analysis offers a powerful tool for the characterization and identification of sand. Preprocessing should be performed before concatenation of raw data. Classification of sand was improved using fused data. PCA model performed on fused data is more robust, and visual discrimination from class distributions is better than those results obtained by individual classification.