An X-ray computed micro-tomography dataset for oil removal from carbonate porous media

This study reveals the pore-scale details of oil mobilisation and recovery from a carbonate rock upon injection of aqueous nanoparticle (NP) suspensions. X-ray computed micro-tomography (μCT), which is a non-destructive imaging technique, was used to acquire a dataset which includes: (i) 3D images of the sample collected at the end of fluid injection steps, and (ii) 2D radiogram series collected during fluid injections. The latter allows monitoring fluid flow dynamics at time resolutions down to a few seconds using a laboratory-based μCT scanner. By making this dataset publicly available we enable (i) new image reconstruction algorithms to be tested on large images, (ii) further development of image segmentation algorithms based on machine learning, and (iii) new models for multi-phase fluid displacements in porous media to be evaluated using images of a dynamic process in a naturally occurring and complex material. This dataset is comprehensive in that it offers a series of images that were captured before/during/and after the immiscible fluid injections.


Methods
The rock sample (D = 12.8 mm, L = 34.5 mm) was encapsulated in epoxy resin and cut to appropriate dimensions to fit in the flow cell used in this study. The flow cell used here is highly X-ray transparent, for design details see Pak et al. 14 . The epoxy-encapsulated core was vacuum saturated with deionised water. The core flooding system was assembled and installed on the rotary table of a laboratory-based μCT machine.
The experimental set-up is shown in Figure 1. The fluid injections were designed to occur under the capillary dominant flow regime (capillary number~10 − 7 ) 1 to represent the processes relevant to fieldscale operations. A back pressure regulator was used at the outlet end of the core to avoid air bubble generation during the experiments. This back pressure was kept at 518 kPa during the nanofluid injections. Fluid injections were performed from the bottom of the core. For this rock the pore volume (PV) was calculated 850 μL (porosity = 19.1%). The X-ray attenuation of water is negligible compared to the rock (dolomite mineral). Therefore, a 3D scan of the water-saturated rock captures the pore-structure of the rock. From this point the experiment is composed of: (i) fluid injections in the core, (ii) collection of 2D radiograms during the injection steps, and (iii) collection of 3D images at the end of each injection step. The details of the fluids and nanoparticles used in this study is summarised in Table 1.
Oil injection was carried out (more than 10 PVs), subsequent to the initial water saturation, which achieved an initial oil saturation of 78%. Next, deionised water was injected in the core, this represented the waterflooding process that is a common improved oil recovery technique. The subsequent two steps involved injection of nanofluids at 0.06 wt% and 0.12 wt% concentration. After each fluid injection step we closed the valves and started the imaging within 30 min. Table 2 summarises the details of the injections steps.
To increase the X-ray attenuation of the oil phase, 1-iododecane was used as a dopant. The selected oil mixture, therefore, enhanced image contrast while also matching the density of the fluids. The density of 50:50 vol% 1-iododecane/dodecane oil (ρ o ) is 1.005 g/cm 3 which is very close to water density (ρ w ) of 1 g/cm 3 and that of the two nanofluids (1.0006, 1.0012 g/cm 3 ). This eliminated the potential for gravitydriven fluid redistribution during 3D image acquisition.
Three series of 2D radiograms were collected during fluid injections (3000-5000 radiograms, every 2 s). It should be noted that a 2D radiogram of a 3D object represents the cumulative attenuation of the X-ray beam projected on the detector's plane. The fluid movements within the core were captured on these 2D radiograms. As explained in , re-slicing these stacks of radiograms in the Z (time) dimension presents an effective means for direct comparison of fast fluid movements between the different injection steps. 2D radiograms are shared using the .raw format. We note that a .raw image can be automatically converted to a stack of 2D images using imageJ 21 or other image processing software packages, e.g. Avizo 22 .
The μCT scanner used in this study is a cone-beam system built at the University of Edinburgh. It comprises a 10-160 kV Feinfocus dual transmission and reflection source (reflection used in this study), a Micos UPR-160-air rotary table and a Perkin Elmer 0822 amorphous silicon flat panel X-ray detector with a terbium doped gadolinium oxysulfide scintillator. The collected μCT images captured a field of view of 11 mm side-length, 2-3 mm away from the inlet face of the core. Hence, the images do not show the inlet and outlet of the core. This section was chosen to ensure that capillary end effects do not influence the volume investigated. Each 3D volume is comprised of 2000 radiograms (exposure time: 2 s) captured during a full 360°rotation. A 0.8 mm aluminium filter was used during the scans to reduce the measurement noise and beam hardening effect. The source-sample and sample-detector distances were 37 mm and 549.5 mm, respectively. The X-ray source voltage and current were 120 kV and 195 μA, respectively. The X-ray spot size is 5 μm. Reconstructed 3D images have the voxel resolution of 13.25 μm. The shared 3D images are of 8 bit format. These images were rescaled (after reconstruction in Octopus 23 ) with offset and slope of 0 and 7.84 × 10 −3 , respectively. Figure 2 summarises the steps taken to post-process and analyse the collected 3D images. We have used the non-local means filter 24 (applied on XY planes, search window:21, local neighbourhood window: 5, and similarity value: 0.4) followed by the unsharp mask filter 25 (3D, edge size: 5, and edge contrast: 0.7) both implemented in Avizo 22 software (version 9).
To ensure our analysis stays consistent within the entire dataset we created a rock mask from the dry rock image using watershed segmentation method 26 . This mask is included with the shared data and is labelled as "Rock-Mask". Applying this mask on different images within the sequence removed the rock phase which in turn facilitated the segmentation of oil and water phases using simple thresholding. Labelling 27 was used to identify connected voxels and classify objects in each image.

Data Records
For each 3D image collected at the end of the injection steps here we share the (i) 2D radiograms, (ii) reconstructed images, (iii) filtered images, and (iv) segmented images (oil and water phases). This dataset also presents three sets of 2D radiograms collected during the injection steps, details of the shared images are summarised in Tables 3, 4, 5 (Data Citation 38). The dolomite rock of this dataset is a Silurian Dolomite from an outcrop (Thornton formation, located near Chicago, Illinois, in US). This rock has been studied by us since 2011. Readers can find relevant data/ images/analysis of this rock in our previous publications including Pak et al. 16 , Pak et al. 28 , and Pak et al. 11 . Additionally, detailed discussions on two-phase flow experiments and wettability (initial and alteration) of this rock in presented in Pak 29 . Data of special interest include: • Mercury porosimetry (including capillary pressure vs. saturation (Pc-Sat) curves and the extracted pore-throat size distributions) 28 • Porosity and permeability measurement 30 • Scanning electron microscopy images 29 • XRD (X-ray Diffraction) data for this rock examining the rock mineralogy which, in summary, suggests this rock is made of more than 99% dolomite 29 . Table 6 summarises the outputs of this experiment. Figure 3 lists the parameters used for data viewing in ImageJ. Figure 4 shows example slices of the data for reference.

Technical Validation
Prior to the injection of oil into the core (i.e. before it becomes a two-phase flow problem), we examined the saturation state of the core for a single fluid phase. The goal was to ensure the initial water saturation was successful, i.e. a system with no trapped air in the rock. A brine phase (KI, 2.5 M) was injected in the core and the sample was scanned. We specifically used the KI salt as iodine is highly X-ray attenuating, this provides a good contrast between the dolomite/brine/potential air phases on the acquired μCT image. Figure 5 shows an example slice of the core. It shows a successful initial saturation which is essential for a reliable image segmentation. In cases were air is present in the core from the beginning image segmentation may result in assigning the air phase to a fluid with close attenuation (here, water due to low contrast). Following this step, the brine was displaced out by water injection (more than ten PVs). Images were post-processed to remove the measurement noise with the aim of producing more reliable segmentations.
The sequences of radiograms collected during the injection steps enabled monitoring of the fluid injections closely, and in particular to assess whether a steady state was being approached or whether fluid mobilisation was still in progress. Also, use of transparent tubing in the injection and production

Injection Step
Flow rate (μL/min) Pore Volume Followed by collection of a 3D image labelled as:    lines allowed us to monitor the injected fluids to ensure no air bubbles were injected in the core. Added to these was the use of back pressure regulator which assisted in running an air-free experiment.

Usage Notes
Suggested software to process the presented images is ImageJ 21 (version 1.52 or later) which is an open-source software. In addition to ImageJ we also used Avizo 22 software (version 9) in our related publication 17 . For filtering the measurement noise we suggest using non-local means 24 filter, for segmentation we suggest watershed segmentation 26    The 2D radiograms collected during each 3D scan is of great value for developers of reconstruction algorithms. The 3D reconstructed images are useful for developers of segmentation algorithms and noise filters. In addition, these images can be used to test algorithms developed for artefact correction, e.g. ring artefacts. Although example μCT images can be found in public domain, in most cases smaller and lower resolution images are shared which are not useful in examining the capabilities of algorithms that aim to process larger and higher resolution images. An example platform that enables researchers to share μCT data (and related measurements) of porous material is the "Digital Rock Portal" which was made public in 2015 36 . At this point in time, this platform has close to 20 projects directly related to imaging of dry rocks or fluid flow processes in rocks. Some research groups have made μCT images available through their university webpages such as the data shared by the Imperial College London via their "Petroleum Engineering & Rock Mechanics Group" webpage 37 . These are mostly images of dry rocks and pore-network models extracted from these rocks for the purpose of pore-scale flow simulations.
The 2D radiograms collected during each injection step are very useful in comparing the flow dynamics at pore-scale for the different injections. In Pak et al. 17 we have used this data to show injection of silica nanoparticles in a carbonate rock can remobilise the trapped oil droplets. In this publication we used a re-sliced form of these 2D radiograms (in the time dimension). Our analysis was, however, limited to qualitative examination of the re-sliced images as the segmentation of these images were challenging. Quantification of the extent of oil remobilisation in this experiment, therefore, remains an open question. New segmentation algorithms may provide reliable segmentation of these data.
Similar to other measurement techniques, μCT imaging is limited by its resolution. Therefore, we believe that acquisition of other independent data such as static and dynamic properties of the rock is useful in understanding the captured images and the improvements offered by the new image processing/analysis algorithms. As such, we refer the readers to our previous publications 14,16,28,29 in which this rock has been studied from different perspectives and using a range of experimental and modelling techniques. Quantitative information extracted from the above-mentioned images Data Table 6. Summary of the output of this experiment.  In summary, the presented data can be used for optimising and testing reconstruction, segmentation, and other image processing algorithms. We foresee potential for these data as a teaching resource as well as a research resource. Currently, no "reference material" exists for μCT, especially within the geosciences context. Given the broad range of independent characterisation of this rock presented in our other publications, we consider this to be a well characterised rock compared to other available datasets.