Using atomic force microscopy to analyse the geomechanical properties of organic rich rocks

Abstract

Climate change will have a major impact on society in the 21st century and beyond, unless the right measures are taken in the next decade. These measures require a drastic decrease in carbon dioxide emissions to reduce the concentrations of CO2 in the atmosphere most likely through sequestration into geologic formations. Organic matter has a key role in two major types of carbon sequestration play; as a key component of a shale seal in many conventional reservoirs, and comprising the majority of coal reservoirs. As such recent research has focused on the mechanical properties of this organic component, with the Atomic Force Microscope and Nanoindentation used to measure Young’s modulus at the nanoscale. This research is expand upon by investigating the trends in organic matter Young’s modulus within marine shales, and compare an immature marine shale (Tarfaya) to a lacustrine equivalent (Green River) using the AFM. The results of this study indicate that there is a clear trend of marine shales exhibiting a bimodal distribution in modulus, with a soft phase centered around 5-9GPa and a stiffer phase centered around 18-24GPa. 13C NMR spectroscopy indicates that the increase in stiffness is tied to an increase in aromatic carbon, which could indicate increases in modulus across all organic matter with maturity. Here AFM is used on a suite of coal macerals from different depositional environments and maturities to assess if there are common trends. The results of this highlight that the modulus distribution of coal macerals is generally unimodal, and softer than that in shales, with all modal values <10GPa. There is however, a similar trend in terms of a stiffening with maturity, with all macerals stiffer in the mature Northumberland Coal than in the immature cannel or paper coals. Thermal modelling suggests that differential strain is more likely in immature coals, where there is a greater difference moduli of liptinite and inertinite macerals. This problem is reduced in the mature coal, with little difference between the maceral moduli, suggesting that deeper mature coal seams are better targets for CCUS than shallower less mature seams. Machine learning can be used to maximise already collected data by making inferences on samples where information is limited, using the trends from a larger dataset. Here the first attempt at using machine learning on SEM, EDX and AFM data is documented, using data collected from the Eagle Ford and Green River shales, with the goal of making mineralogic and geomechanical predictions. A variety of supervised and unsupervised machine learning methods were used, including; Multi-Layer Perceptron, KNN and Random Forest. The accuracies of these models on the test/training data is generally above 85%, and in the case of the KNN and Random Forest above 95%. However, when the model are used on an unrelated dataset, the accuracy decreases significantly. This research indicates that if machine learning is to be used, the training dataset and model should be selected with the end result in mind, whilst acquiring the datasets using a similar technique to a similar quality.