Determining representative mechanical parameters of clay matrix in mudstones using nanoindentation mapping and machine learning data analysis: a novel top seal characterization approach

Mudstones and shales are fine-grained sedimentary rocks that can serve as top seals of geological reservoirs in various geoenergy applications. Apart from traditional oil and gas exploration, the urgent need for underground storage of energy carriers (e.g., H₂) and climate-relevant gases (e.g., CO₂) facilitated extensive research on pore structural and mechanical parameters and their influence on the seal capacity of these rocks. The fracture behaviour of mudstone seal rocks controls the risk of seal failure due to microfracturing as a response to various geological processes (e.g., buoyancy pressure from the reservoir). In this contribution, the high-speed nanoindentation mapping approach was carried out for a proven mudstone top seal sample (~1629 m; quartz 31%, clay mineral 39%) from a Vienna Basin oil field. The nanoindentation results were then post-processed with machine learning-based tools to obtain representative mechanical parameters of the clay matrix. k-means clustering analysis was performed using three input features including hardness (H), reduced elastic modulus (E_r), and the elastic-plastic deformation ratio based on the obtained load-displacement curves. In addition, broad ion beam-scanning electron microscopy (BIB-SEM) maps were taken before and after the nanoindentation to correlate the indentation results with direct imaging information and to verify the k-means clustering results. A total of 8 indentation map arrays (7 × 7 indents) were placed to test the sensitivity of different tips to indentation depth and load rate. The comparison of BIB-SEM image data and k-means clustering showed that decisions on phase assignment can be significantly improved and performed in a shorter time by k-means clustering analysis, still showing an overall good agreement with manual selections. For the studied mudstone sample, the resulting average E_r and H values of the clay matrix range at 17.58 ± 6.89 GPa and 0.63 ± 0.76 GPa (n=30), respectively for the Berkovich tip and at 15.03 ± 4.79 GPa and 0.38 ± 0.23 GPa (n=62), respectively for the Cube Corner tip. The testing with both tips shows that despite the strongly heterogeneous microstructure of the indented clay matrix the obtained mechanical parameters are not sensitive to indentation depths and hence representative values can be determined from minimum volumes with statistical significance. Nevertheless, compared with the Berkovich tip, the Cube Corner tip sampled deeper depths (58-443 nm for the Berkovich tip and 412-1747 nm for the Cube Corner tip) and introduced more surface damage. By increasing the load rate from 1000 to 6000 μN s⁻¹, the indentation testing tended to be unstable and the surface showed strong damage at the highest load rate. In conclusion, this contribution represents an important methodological step towards the implementation of combined high-speed nanoindentation mapping and machine learning data analysis as a feasible high throughput tool for the mechanical characterization of mudstones and similar fine-grained sedimentary rocks. The presented approach is planned to be applied to an extensive set of mudstone samples from the Vienna Basin with the purpose to link mechanical property changes to burial diagenesis.