Developing a hierarchical Digital Core Analysis workflow for petro-physical characterisation of cross-laminated reservoir rocks at pore scales
The study presented in this thesis addresses a critical shortfall of Digital Core Analysis (DCA) in the petro-physical characterisation of natural rock samples that exhibit complex and spatially non-separable multi-scale arrangements of mineral grains of variable sizes and chemical compositions. Since those multi-scale arrangements define non-stationary (heterogeneous) grain-pore distributions, they must be characterised on sufficiently large sample volumes for sample reconstruction at pore and grain scales, to simulate the fluid flow processes for predicting petro-physical properties. Such characterisations can only be done in a multi-stage manner at multi-scales and multi-locations on a sample, due to the mutual constraints of the imaging field of view and resolution. Typical multi-scale approaches fuse images from various sources based on shared features (e.g. disproportionately large grains) as cross-scale references, but have limited application when the shared features are less obvious (non-separable) across multi-scale images. For common cross-lamination in sandstone rocks that exhibit hierarchical structures controlling grain-pore distribution and are demarcated by physical features such as inter/intra-lamina (set) bounding surfaces, a multi-scale DCA workflow would be ideally suited for the petro-physical characterisation but would be challenging to apply when the visibility of cross-scale shared features (bounding surfaces) is weak. This work argues that the workflow needs to have the following elements: 1) a multiscale spatial sampling/characterisation procedure; 2) a way to maintain correct cross-scale spatial correlation among individual sampled datasets through shared features, as reference to the same sample, and 3) a stochastic reconstruction procedure that honours the multiscale structures identified in 1) and correlated in 2), and reconstructs the poregrain-scale non-stationarity they expressed across the correlated datasets. The work presented in this thesis focuses on cross-lamination in sandstone rocks to identify challenges arising to multi-scale characterisation and to develop a suitable hierarchical DCA workflow. For this purpose, an outcrop tight Aeolian cross-laminated sandstone sample of 6x6x6 cm3 that includes representative hierarchical structures was chosen. The key achievements are: 1) a suitable imaging-based exploratory and recursive sampling and characterisation procedure to maximise correlation between datasets. Analysis on the resultant multi-scale datasets showed that sharp and gradual grain compositions change between adjacent laminasets (inter-laminaset) and laminae (intra-laminaset), respectively, to form thin but finite-thickness bounding surfaces, which are ideally suited as shared physical features. However, challenges are identified including a) the inter-laminaset bounding surfaces are obscured by “artefacts” and cannot be enhanced by existing techniques and b) intra-laminaset bounding surfaces are hard to be even observed because of gradual composition variation. 2) To identify the two types of bounding surfaces as shared features to help registration, two new techniques were developed by exploring geological knowledge on structural morphology and mineral compositions, respectively, and were shown to be able to successfully identify and extract shared inter- and intra-laminaset bounding surfaces. 3) For a pair of realisations of sampled inter- and intra-laminaset bounding surfaces and estimated grain-pore distribution information, a non-stationary index map of grain texture (size and orientation) on the whole sample domain can be constructed to define the mean of affinity transformations for stochastically populating stationary (homogeneous) grain and pore structures by a reconstruction method which is implemented based on Multiple Points Statistics (MPS). One reconstructed pore-grain model was obtained in this work and contains 11900x11900x11900 voxels at a voxel resolution of 4 µm. This model was verified to capture key pore-grain variations associated with the characterised lamination structures.