ABSTRACT
Six southern Texas Eagle Ford cores were investigated to quantify mineralogical composition and total organic carbon (TOC). Machine learning of the x-ray fluorescence (XRF) data set was conducted using neural network analysis to predict mineralogies for L1, L2, and L3 and TOC for L1, L2, L3, Iona-1, Innes-1, and well “X.” Inees-1 and well “X” were used as blind tests to check the quality of the developed models. The online Neural Designer software was used to perform the training process and develop models. Quantitative laboratory-measured x-ray diffraction (XRD) mineralogies and TOC were used to conduct the training and develop high-resolution semiquantitative models, and the derived mineralogic and organic matter models were found to be promising. The modeled mineralogy and TOC represent continuous relative abundances, which are far more significant than scattered individual XRD and TOC point measurements. The significance of this study is that it allows for the use of relatively inexpensive and nondestructive XRF analysis that requires minimal sample preparation to construct high-resolution mineral abundances and TOC content. With modern advances in technology, XRF can now be measured on drill cuttings in real time while drilling is occurring, allowing operators to use the proposed method to construct semiquantitative mineralogical and TOC models for evaluating placement of laterals in prospective intervals and designing completion techniques accordingly.