Search

This thesis evaluates the possibility of automated permeability estimation by utilizing the information obtained from whole core 2D and 3D CT-scan images of wells on the Norwegian continental shelf. To evaluate this possibility, end-to-end convolutional neural network (CNN) regression models were proposed. These models used two-dimensional image slices and three-dimensional sub-cube images of 3D whole core CT scan images to automatically predict continuous permeability at a millimeter scale resolution. More specifically, CNN regression models were trained to learn permeability obtained from routine core analysis (RCA) measurements. Initially, a suggested CNN regression model was trained on 2D image slices belonging to a subclass of data to learn the relationship between convolution-derived features and RCA-derived permeability values. In the next methodology applied, the CNN regression model was further trained on 2D images belonging to the entire well. In the final methodology, a model was trained on 3D sub-cube images belonging to the entire well by utilizing a classical and a modified CNN architecture. The preliminary results from all models utilized in the three aforementioned methodologies demonstrate degrees of deviation from the RCA permeability values. Therefore, a post-processing stage was executed to enhance the performance of the models and to decrease the discrepancy between the measured and predicted permeabilities. This is done by initially including image augmentation of the test set data and then ultimately averaging the predicted permeability values of those test set images. The results depict that the models are predominantly prone to be erratic prior to applying the post-processing stage. However, once the post-processing procedure was implemented, the results exhibit a relatively good correlation between the predicted permeability values obtained utilizing the proposed method and the core plug permeability measurements. To summarize, this thesis confirms that the models show consistent predictive results, and are able to identify a substantial portion of the variation in permeability measurements. It is indicative that the models are able to learn the relationship between the distribution of grey-level attenuations of the images and the permeability measurements. However, it is important to note that the proposed models were only trained on a subset of data from a single well. Ideally, we assume a model trained on the whole well and multiple wells would result in a more robust model with higher generalization capabilities. Furthermore, there are some limitations and uncertainties associated with the image artifacts in the training dataset, and image complexities that can negatively impact the training process and generalization capabilities of the proposed model. This thesis evaluates the possibility of automated permeability estimation by utilizing the information obtained from whole core 2D and 3D CT-scan images of wells on the Norwegian continental shelf. To evaluate this possibility, end-to-end convolutional neural network (CNN) regression models were proposed. These models used two-dimensional image slices and three-dimensional sub-cube images of 3D whole core CT scan images to automatically predict continuous permeability at a millimeter scale resolution. More specifically, CNN regression models were trained to learn permeability obtained from routine core analysis (RCA) measurements. Initially, a suggested CNN regression model was trained on 2D image slices belonging to a subclass of data to learn the relationship between convolution-derived features and RCA-derived permeability values. In the next methodology applied, the CNN regression model was further trained on 2D images belonging to the entire well. In the final methodology, a model was trained on 3D sub-cube images belonging to the entire well by utilizing a classical and a modified CNN architecture. The preliminary results from all models utilized in the three aforementioned methodologies demonstrate degrees of deviation from the RCA permeability values. Therefore, a post-processing stage was executed to enhance the performance of the models and to decrease the discrepancy between the measured and predicted permeabilities. This is done by initially including image augmentation of the test set data and then ultimately averaging the predicted permeability values of those test set images. The results depict that the models are predominantly prone to be erratic prior to applying the post-processing stage. However, once the post-processing procedure was implemented, the results exhibit a relatively good correlation between the predicted permeability values obtained utilizing the proposed method and the core plug permeability measurements. To summarize, this thesis confirms that the models show consistent predictive results, and are able to identify a substantial portion of the variation in permeability measurements. It is indicative that the models are able to learn the relationship between the distribution of grey-level attenuations of the images and the permeability measurements. However, it is important to note that the proposed models were only trained on a subset of data from a single well. Ideally, we assume a model trained on the whole well and multiple wells would result in a more robust model with higher generalization capabilities. Furthermore, there are some limitations and uncertainties associated with the image artifacts in the training dataset, and image complexities that can negatively impact the training process and generalization capabilities of the proposed model.

Fenomenet med borevæske som slipper ut fra brønnen og inn i formasjonen under boring av en oljebrønn er kjent som tapt sirkulasjon (lost circulation). Ved siden av økonomiske tap på grunn av tapt tid ved boring, er det også mulighet for å skade formasjonen. En potensiell løsning for å hindre tap av borevæske er å tilsette partikler til borevæsken, kjent som Lost Circulation Materials (LCM). Disse kan tette brudd og permeable soner. I denne oppgaven er formasjonen forenklet til en kule-pakke. Den numeriske simuleringen utføres med en Dense Discrete Phase Model (DDPM), som løser væskestrøm ligningene og partikkel-væskeinteraksjonene, koblet til en Discrete Element Method (DEM), som løser partikkel interaskjoner; kontakter mellom partikler og mellom partikler og pore-vegger. Karakteriseringen av væske-partikkel strømmen oppnås ved å variere trykkforskjellen over modellen. Væskehastighet, partikkelmassestrøm, permeabilitet og volumetrisk strømningshastighet på væsken overvåkes under tilstoppingsprosessen. Økningen i trykkforskjellen på innløpsoverflaten resulterer i en høyere væskehastighet ved innløpsoverflaten og en høyere massehastighet for faste partikler. Det reduserer også den volumetriske strømningshastigheten ved utløpet, noe som resulterer i en mer effektiv tetting av det porøse mediet. During the drilling of an oil well, the phenomenon of drilling fluid escaping from the annular region into the formation is known as lost circulation. Besides the financial and time losses due to mitigation procedures, there is also the possibility of damaging the formation. A potential solution to the lost circulation problem is adding solid particles to the drilling fluid, known as Lost Circulation Materials (LCM). These particles can seal fractures and highly permeable zones. In this work, the rock formation is simplified to a spheres pack. The numerical simulation is performed via the Dense Discrete Phase Model (DDPM), which solves the fluid flow equations and the particle-fluid interactions, coupled to the Discrete Element Method (DEM), which solves the interaction between particles and between particles and walls. Characterization of the liquid-solid flow is obtained by varying the pressure difference in the numerical domain. Fluid velocity, solid particle flow rate, and normalized fluid flow rate are monitored throughout the clogging process. The effectiveness of changing some parameters is studied in this work. Increasing particle diameters shows to be a good solution to reduce fluid loss. The variation in particle-fluid density ratio on the other hand presented little change in the final fluid flow rate. Injecting particles of different sizes can be effective if all particles are larger that the pore throat. The increase in the pressure difference on the domain results in a lower fluid flow rate, but also in fewer particles trapped in the porous medium.

Forskningsspørsmålet i dette arbeidet var effekten av vannkvalitet på spontan oppsuging (SI) i karbonatkjerner. Målet var å endre fra oljefuktende til vannfuktende tilstand kjerner for å oppnå mer spontan oppsuging av vann og forbedret produksjon av olje. Hovedmålet var å forstå hvordan den ioniske sammensetningen av oppsugd vann dikterer fuktighetsendringen. For å gjennomføre disse eksperimentene var det nødvendig å bygge en rigg designet for å operere ved 96°C for en uavbrutt SI. Vi brukte kjerner fra to steder; fra Ainsa i Spania og fra Angola. Disse kjernene ble ansett som representative for et Equinor-operert oljefelt utenfor Brasil. Basert på petrofysiske egenskaper var kjernematerialet fra Angola mer heterogent enn det fra Ainsa. X-ray diffraction (XRD) og back-scatter electron imaging (SEM) resultater klassifiserte kjernematerialene som kalkstein uten anhydrid. Dette prosjektet sammenligner også to aldringsprosesser, og design for den mest effektive prosessen. Effekten av ionisk sammensetning på fukting ble studert ved å analysere utvinning av olje fra SI av formasjonsvann (FW), syntetisk sjøvann (SSW) eller modifiserte versjoner av syntetisk vann ved 96°C. Den ioniske sammensetningen av SSW ble endret ved å øke magnesium (Mg2+) og sulfat (SO2−4) ion konsentrasjonene. Tre forskjellige modifiserte syntetiske sjøvann; en med to ganger sulfatkonsentrasjon og fire ganger magnesiumkonsentrasjon (SSW-2S4Mg), et annet vann med to ganger sulfatkonsentrasjon og åtte ganger magnesiumkonsentrasjon (SSW-2S8Mg), og til slutt et med lavt saltinnhold der NaCl-konsentrasjonen ble redusert til 10% av SSW-konsentrasjon (0,1-NaCl-SSW). For reproduserbarhet av resultatene ble to kjerner brukt i hver runde av SI-eksperimentene. Kontaktvinkelmålinger utført ved romtemperatur på polerte steinprøver eldet ved både 25°C (romtemperatur) og ved 96°C ble også utført ved bruk av samme salt-vann løsninger for å supplere resultatene av spontan oppsuging. Zeta-potensiale målinger ble utført på Angola 8μm stein-partikler suspendert i hver av test-vannene for å studere stein-vann overflaten ved 25°C og 70°C. En forbedret SI-rigg som tillater en uavbrutt SI-prosess ble bygget, og fungerte jevnt uten lekkasjer, trykk opp-bygging og uten å koke vannet. Fra SI-eksperiment fikk vi økt oljeproduksjon på 6-10% med SSW etter FW, noe som viser at SSW er en EOR-væske i kalksteinkjerner. Utvinning på 4-6% og 4-10% med SSW-2S4Mg og SSW-2S8Mg, viste henholdsvis at berikelse av SSW med Mg2+ og SO2−4 ioner gir økt vannfukt i kalksteinkjerner, noe som forårsaker mer spontan oppsuging. Eksperimentet med Ainsa-kjerner ga en ekstra utvinning på 5% med 0,1-NaCl-SSW etter SSW, noe som viser virkningen av selektiv fortynning. Ytterligere utvinning av 13% med SSW-2S4Mg bekreftet at reduksjon av NaCl-innholdet i SSW forbedrer aktiviteten til potensielt bestemmende ioner i fuktigendrings-prosessen i kalksteinskjerner. Dette ble også bekreftet med kontaktvinkel målinger på vannvåte prøver etter aldring i 0,1-NaCl-SSW i 3 dager. Ulike aldrings-metoder ble studert ved å elde to Ainsa-kjerner med enten dynamisk eller statisk elding. For denne sammenligningen ble et nytt eksperimentelt oppsett for dynamisk aldring designet og bygget. Dette oppsettet kan takle høye temperaturer og trykk under 35 bar. Spontane oppsuging resultater viste at dynamisk aldring er en bedre aldringsprosess for å forandre fuktighet enn statisk elding. En produksjon på 12% ble registrert i kjernen eldet dynamisk, mot 4% for statisk elding, da SSW ble introdusert i tertiær modus. The research topic in this work was the effect of water quality on spontaneous imbibition (SI)in carbonate cores. The target was to alter the oil wetting state in outcrop cores to a more water wet state in order to have spontaneous imbibition of brine and improved crude oil recovery. The main aim was to understand how the ionic composition of imbibing brine dictates the wettability change. In order to carry out these experiments a rig designed to operate at 96°C for an uninterrupted SI was needed. Outcrop cores used were of two origins; outcrop cores from Ainsa in Spain and from Angola, and these outcrop cores were considered representative of an Equinor operated oil field off-shore Brazil. Based on petrophysical properties, the core material sourced from Angola was more heterogeneous than that from Ainsa. X-ray diffraction (XRD) and back-scatter electron imaging (SEM) results classified the rock materials as limestone without anhydrite. The research also compares two aging processes, and design for the most efficient process. The effect of ionic composition on wettability alteration was studied by analysing oil recoveries due to SI of formation water (FW), synthetic seawater (SSW) or modified versions of synthetic seawater at 96°C . The ionic composition of SSW was altered by increasing the magnesium (Mg2+) and sulfate (SO2−4) ion concentrations. Three different modified synthetic sea waters; one with two times sulfate concentration and four times magnesium concentration (SSW-2S4Mg), a second water with two times sulfate concentration and eight times magnesium concentration (SSW-2S8Mg), and finally a low salinity water where the NaCl concentration is reduced to 10% of the SSW concentration (0.1-NaCl-SSW). For reproducibility of results, two cores were used in each run of SI experiment. Contact angle measurements conducted at room temperature on polished rock chips aged both at 25°C (room temperature) and at 96°C were also done using the same set of brines to supplement results of spontaneous imbibition. Zeta potential measurements were conducted on Angola 8μm rock particles suspended in each of the test brines to study the rock-brine interface at 25°C and 70°C. An improved SI rig that allows an uninterrupted SI process was set up and worked smoothly without leakages, pressure build up and water boiling. From SI experiment, improved oil recoveries of 6-10% with SSW after FW, showed that SSW is an EOR fluid in limestone cores. Recoveries of 4-6% and 4-10% with SSW-2S4Mg and SSW-2S8Mg, respectively, showed that enriching SSW with Mg2+and SO2−4ions improves the water wetting state in limestone cores and causes more spontaneous imbibition. The experiment with Ainsa cores gave a recovery of 5% with 0.1-NaCl-SSW after SSW showed the impact of selective dilution. Additional recovery of 13% with SSW-2S4Mg confirmed that reducing the NaCl content in SSW improves the activity of potential determining ions in the wettability alteration process in limestone cores. This was also confirmed with contact angles on water wet samples after aging in 0.1-NaCl-SSW for 3 days Different aging methods were studied by aging two Ainsa cores with either dynamic or static aging. For this comparison, a new experimental set up for dynamic aging was designed and built. This setup could handle high temperatures and pressures below 35 bars. Spontaneous imbibition results showed that dynamic aging is a better wettability alteration process than static aging. A recovery of 12% was recorded in the core aged dynamically and 4% for the statically aged when SSW was introduced in a tertiary mode

This experimental study investigates the osmosis and the spontaneous emulsification as oil mobilization mechanisms during the low salinity water flooding by measuring the connate water expansion area in an oil-wet microfluidic system. Two types of synthetical oil, heptane and dodecane, and different salinity brines are chosen to use in the experiments. Meanwhile, surfactants are added into the oil phase to calculate the water transportation rate due to the spontaneous emulsification, comparing with the water transportation rate due to the osmosis. Dynamic light scattering and pendant drop experiments were conducted with the same fluids to study the spontaneous emulsification process. The HSW expansion due to the osmosis was observed at the microfluidic experiments with both heptane and dodecane under a salinity gradient. The water transportation rate for the osmosis is as high as the rate for the spontaneous emulsification with enough amounts of surfactants added into the heptane. Insufficient surfactants added may inhibit the water transportation in the synthetic oil. The dodecane shows weaker water transportability in the same condition. It is still hard to give any conclusion on the effect of the oil viscosity and water solubility.