Machine Learning for mechanical classification of organic-rich shale based on high-speed nanoindentation

In conjunction with scanning electron microscope (SEM) and energy-dispersive spectrometer (EDS), quasi-static nanoindentation has been widely used to investigate the mechanical properties of minerals and organic matters in shale at micro scale. However, due to the limited test efficiency of conventional nanoindentation measurement and the demand of mineralogical identification which is achieved by SEM observation and EDS analysis, the research scheme in previous works can be time-consuming and complicated. This work attempts to develop a new micromechanical research scheme with high test efficiency and automatic mineralogical identification. A newly-developed high-speed nanoindentaion technique is used to characterize the mechanical properties distribution of a shale sample from the Yanchang Formation in the Ordos Basin, China. Then, the mineralogical distribution in the corresponding areas is obtained by using MAPS Mineralogy. Finally, logistic regression is applied to link the mechanical properties distribution and mineralogical distribution, and to realize the automatic mineralogical identification based on nanoindentation results. In addition, to further investigate the influence of characterization experiments on machine learning results, the characterization abilities, including lateral spatial resolution, detection depth, and signal spacing, of the two experimental methods are compared. The detection depth of MAPS Mineralogy is markedly higher than that of nanoindentation, which means that the material volume detected by the two methods is different. The lateral range responding to applied force and incident electrons determines that the signal of data points at the boundary can be a mixture of two or more minerals. The influence of such detection depth difference and boundary effect is also discussed.