Deep-Learning-Based Automated Sedimentary Geometry Characterization From Borehole Images

Sedimentary geometry on borehole images usually summarizes the arrangement of bed boundaries, erosive surfaces, crossbedding, sedimentary dip, and/or deformed beds. The interpretation, very often manual, requires a good level of expertise, is time consuming, can suffer from user bias, and becomes very challenging when dealing with highly deviated wells. Bedform geometry interpretation from crossbed data is rarely completed from a borehole image. The purpose of this study is to develop an automated method to interpret sedimentary structures, including the bedform geometry resulting from the change in flow direction from borehole images. Automation is achieved in this unique interpretation methodology using deep learning (DL). The first task comprised the creation of a training data set of 2D borehole images. This library of images was then used to train deep neural network models. Testing different architectures of convolutional neural networks (CNN) showed the ResNet architecture to give the best performance for the classification of the different sedimentary structures. The validation accuracy was very high, in the range of 93 to 96%. To test the developed method, additional logs of synthetic data were created as sequences of different sedimentary structures (i.e., classes) associated with different well deviations, with the addition of gaps. The model was able to predict the proper class in these composite logs and highlight the transitions accurately.

Enhanced Assessment of Fluid Saturation in the Wolfcamp Formation of the Permian Basin

Conventional resistivity models often overestimate water saturation in organic-rich mudrocks and require extensive calibration efforts. Conventional resistivity-porosity-saturation models assume brine in the formation as the only conductive component contributing to resistivity measurements. They also do not reliably assimilate the spatial distribution of the clay network and pore structure. Moreover, they do not incorporate other conductive minerals and organic matter, impacting the resistivity measurements and leading to uncertainty in water saturation assessment. We recently introduced a resistivity-based model that quantitatively assimilates the type and spatial distribution of all rock constituents to improve reserves evaluation in organic-rich mudrocks using electrical resistivity measurements. This paper aims to expand the application of this model for well-log-based assessment of water/hydrocarbon saturation and to verify the reliability of the introduced method in the Wolfcamp Formation of the Permian Basin. Our recently introduced resistivity model uses pore combination modeling to incorporate conductive (clay, pyrite, kerogen, brine) and nonconductive (grains, hydrocarbon) components in estimating effective resistivity. The inputs to the model are volumetric concentrations of minerals, conductivity of rock components, and porosity obtained from laboratory measurements or interpretation of well logs. Geometric model parameters are also critical inputs to the model. To simultaneously estimate the geometric model parameters and water saturation, we developed an inversion algorithm with two objectives: (a) to estimate the geometric model parameters as inputs to the new resistivity model and (b) to estimate the water saturation. The geometric model parameters are determined for each rock type or formation by minimizing the difference between the measured resistivity and the resistivity estimated from pore combination modeling. We applied the new method to two wells drilled in the Wolfcamp Formation of the Permian Basin. The formation-based inversion showed variation in geometric model parameters, which improved the assessment of water saturation. Results demonstrated that the new method improved water saturation estimates by 24.1% and 32.4% compared to Archie’s and Waxman-Smits models, respectively, in the Wolfcamp Formation. The most considerable improvement was observed in the Middle and the Lower Wolfcamp Formations, where the average clay concentration was relatively higher than the other zones. There was an additional 70,000 bbl/acre of hydrocarbon reserve using the proposed method compared to when water saturation was quantified using Archie’s model in the Permian Basin, which is a 33% relative improvement. It should be highlighted that the new method did not require any calibration effort using core water saturation measurements, which is a unique contribution of this rock-physics-based workflow.

NanoTags for Improved Cuttings Depth Correlation

During a drilling operation, rock cuttings are often sampled off a shale shaker for lithology and petrophysical characterization. These analyses play an important role in describing the subsurface, and it is important that the depth origin of the cuttings be accurately determined. Traditionally, mud loggers determine the depth origin of the sampled cuttings by calculating the lag time required for the cuttings to travel from the bit to the surface. These calculations, however, can contain inaccuracies in the depth correlation due to the shuffling and settling of cuttings as they travel with drilling fluid to the surface, due to unplanned conditions like drilling an overgauge hole, and due to other unforeseen drilling events, especially critical in horizontal sections. We, therefore, aimed to remedy these inaccuracies by developing a series of styrene-based nanoparticles that tagged the cuttings as they were generated at the drill bit. These “NanoTags” were tested while drilling in Q4 2019, and the results indicated that the NanoTags did, in fact, have the potential to identify some systematic errors compared with traditional mud-logging calculations.

An Integrated Petrophysical Characterization of a Siliciclastic Tight Gas Reservoir in Neuquén Basin, Western Argentina

The Río Neuquén Field is located between Neuquén and Río Negro provinces, Argentina. Historically, it has been a conventional oil producer, but it was converted to a tight gas producer from deeper reservoirs. The targeted geological formations are Lajas, which is already a known tight gas producer, and the less-known overlaying Punta Rosada Formation, which is the main objective of the current work. Punta Rosada presents a diverse lithology, including shaly intervals separating multiple stacked reservoirs that grade from fine-grained sandstones to conglomerates. The reservoir pressure can change from the normal hydrostatic gradient to up to 50% of overpressure. There is little evidence of movable water. The key well in this study has a comprehensive set of openhole logs, including pulsed-neutron spectroscopy data, and is supported by a full core study over 597 ft. Additionally, data from several offset wells were used, containing sidewall cores and complete sets of electrical logs. This allowed the development of rock-calibrated mineral models, adjusting the clay volume with X-ray diffraction data, porosity, and permeability with core measurements, and linking the log interpretation to dominant pore-throat radius models from MICP Purcell tests. Several water saturation models were tested, attempting to adjust the irreducible water saturation with NMR and Purcell tests at reservoir conditions. As a result, three hydraulic units were defined and characterized, identifying a strong correlation with lithofacies observed in cores and image logs. A cluster analysis model allowed the propagation of the facies to the rest of the wells (50). Finally, lithofacies were distributed in a full-field 3D model, guided by an elastic seismic inversion. In the main key well, in addition to the openhole logs and core data, a casedhole pulsed-neutron log (PNL) was also acquired, which was used to develop algorithms to generate synthetic pseudo-openhole logs such as bulk density and resistivity, integrated with the spectroscopy mineralogical information and other PNL data, to perform the petrophysical evaluation. This enables the option to evaluate wells in contingency situations where openhole logs are not possible or too risky, and also in planned situations to replace the openhole data in infill wells, saving considerable drilling rig time during this field development phase. Additionally, the calibrated casedhole model can be used in old wells. This paper explores the integration of different core and log measurements and explains the development of rock-calibrated petrophysical and rock type models addressing the characterization challenges found in tight gas sand reservoirs. The results of this study will be crucial to optimize the field development.

Identification of Breakout Behind Casing: Methodology to Obtain Openhole-Equivalent Caliper Measurements Through Slotted Liner Using the Density Tool

Growth in the coal seam gas industry in Queensland, Australia, has been rapid over the past 15 years, with greater than USD 70 billion invested in three liquified natural gas export projects supplied by produced coal seam gas. Annual production is of the order of 40 Bscm or 1,500 PJ, with approximately 80% of this coming from the Jurassic Walloon Coal Measures of the Surat Basin and 20% from the Permian Coal Measures of the Bowen Basin. The Walloon Coal Measures are characterized by multiple thin coal seams making up approximately 10% of the total thickness of the unit. A typical well intersects 10 to 20 m of net coal over a 200- to 300-m interval, interbedded with lithic-rich sandstones, siltstones, and carbonaceous mudstones. The presence of such a significant section of lithic interburden within the primary production section has led to a somewhat unusual completion strategy. To maximize connection to the gas-bearing coals, uncemented slotted liners are used; however, this leaves fluid-sensitive interburden exposed to drilling, completion, and produced formation fluids over the life of a well. External swellable packers and blank joints are therefore used to isolate larger intervals of interburden and hence minimize fines production. Despite these efforts, significant fines production still occurs, which leads to the failure of artificial lift systems and the need for expensive workovers or lost wells. Fines production has major economic implications, with anecdotal reports suggesting up to 40% of progressive cavity pump artificial lift systems in Walloon Coal Measures producers may be down at any one time. The first step in solving this problem is to identify the extent and distribution of fines production. The wellbore completion strategy above, however, precludes the use of mechanical calipers to identify fines-production-related wellbore enlargement. A new caliper-behind-liner technique has therefore been developed using a multiple-detector density tool. Data from the shorter-spacing detectors are used to characterize the properties of the liner as well as the density of the annular material. This is particularly important to evaluate as the annulus fill varies between gas, formation water, drilling and completion fluids, and accumulated fines. The longer-spacing detector measurements are then used in conjunction with pre-existing openhole formation density measurements to determine the thickness of the annulus, and hence hole size, compensating for liner and annulus properties.

Case Studies on Multistring Isolation Evaluation in P&A Operations

Cement sheath quality assessment is a critical initial step in plug and abandonment (P&A) operations during oil and gas well decommissioning. However, the technologies commonly used require unimpeded access to the casing annuli, thus enforcing the need for production tubing pulling or inner casing milling. Cement integrity or isolation evaluation through multilayered well casing strings will provide the opportunity to significantly reduce operational time and costs and to greatly simplify the traditional P&A process. As desired by the industry for years, recent advancements in isolation evaluation have proven the feasibility to assess cement sheath quality without the removal of production tubing or inner casing. The new development, consisting of a sophisticated logging apparatus with a novel processing methodology, led to a groundbreaking technology evaluating zonal isolation through multiple casing strings in wells. The logging tool is deployed in the borehole using E-line, slickline, or coiled tubing. Then, the acoustic energy that is emitted and received by the tool travels through the tubing and surrounding annulus to reach the isolation barrier behind the casing. A proprietary frequency-domain processing algorithm successfully identifies the desired signal by discriminating it from overwhelming undesired signals such as tubing arrivals. The latest development stage further enables the segmentation of the measurements, providing an improved sensitivity to detect the azimuthal variations in the cement sheath quality. Case histories of applying omnidirectional and segmented multistring isolation evaluation technology in field trials in the North Sea are presented in the paper. The measurement accuracy has been verified through side-by-side comparisons with industry-standard cement bond log (CBL) and ultrasonic logs recorded after the tubing was removed. Additionally, the technology has been proven applicable to various casing or tubing weight and size combinations with tubing eccentric inside the casing. Thus, it is practicable in actual well configurations and suitable for the deviated well sections as well. In conclusion, this innovative technology that exhibits quantitative assessments of bonding or isolation conditions of wells in multistring configurations provides a cost-effective solution during P&A and further demonstrates a great potential to accelerate along the path to a rigless P&A operation.

Enhanced Mineral Quantification and Uncertainty Analysis from Downhole Spectroscopy Logs Using Variational Autoencoders

This paper describes an innovative machine-learning application, based on variational autoencoder frameworks, to quantify the concentrations and associated uncertainties of common minerals in sedimentary formations using the measurement of atomic element concentrations from geochemical spectroscopy logs as inputs. The algorithm comprises an input(s), encoder, decoder, output(s), and a novel cost function to optimize the model coefficients during training. The input to the algorithm is a set of dry-weight concentrations of atomic elements with their associated uncertainty. The first output is a set of dry-weight fractions of 14 minerals, and the second output is a set of reconstructed dry-weight concentrations of the original elements. Both sets of outputs include estimates of uncertainty on their predictions. The encoder and decoder are multilayer feed-forward artificial neural networks (ANN), with their coefficients (weights) optimized during calibration (training). The cost function simultaneously minimizes error (accuracy metric) and variance (precision or robustness metric) on the mineral and reconstructed elemental outputs. Training of the weights is done using a set of several-thousand core samples with independent, high-fidelity elemental and mineral (quartz, potassium-feldspar, plagioclase-feldspar, illite, smectite, kaolinite, chlorite, mica, calcite, dolomite, ankerite, siderite, pyrite, and anhydrite) data. The algorithm provides notable advantages over existing methods to estimate formation lithology or mineralogy relying on simple linear, empirical, or nearest-neighbor functions. The ANN numerically capture the multidimensional and nonlinear geochemical relationship (mapping) between elements and minerals that is insufficiently described by prior methods. Training is iterative via backpropagation and samples from Gaussian distributions on each of the elemental inputs, rather than single values, for every sample at each iteration (epoch). These Gaussian distributions are chosen to specifically represent the unique statistical uncertainty of the dry-weight elements in the logging measurements. Sampling from Gaussian distributions during training reduces the potential for overfitting, provides robustness for log interpretations, and further enables a calibrated estimate of uncertainty on the mineral and reconstructed elemental outputs, all of which are lacking in prior methods. The framework of the algorithm is purposefully generalizable so that it can be adapted across geochemical spectroscopy tools. The algorithm reasonably approximates a “global-average” model that requires neither different calibrations nor expert parameterization or intervention for interpreting common oilfield sedimentary formations, although the framework is again purposefully generalizable so it can be optimized for local environments where desirable. The paper showcases a field application of the method for estimating mineral type and abundance in oilfield formations from wellbore-logging measurements.

Taming the Thunder Horse With Axes and Vectors

The Thunder Horse Field targets Middle Miocene deepwater turbiditic reservoirs. Despite being prolific, the mapping of the ~180 m thick, partly amalgamated reservoir sandstones is challenging. Seismic quality is reduced by the presence of salt structures. The salt overburden and high formation pressure require the use of heavy mud weights and oil-based drilling fluids, which limit the resolution and interpretation potential of borehole image logs (BHI). Halokinetic movements caused significant post-depositional deformation of the already complex gravity-driven sediment stack, and the reservoir beds drape against an E-W oriented salt wall. Consequently, the assessment and removal of the structural dip component are not trivial, and the evaluation of paleo-transport directions is considerably more complicated compared to undisturbed deepwater reservoirs. This paper illustrates the potential of eigenvector methods to BHI from Ruehlicke et al. (2019) for reconstructing the depositional slope and the architecture of mass transport complexes in the case of chaotic depositional settings and uncertain structural dip. Figures from Henry et al. (2018) are used wherein part axial analysis was performed on data from a group of Thunder Horse wells and presented in more detail.

Data Quality Considerations for Petrophysical Machine-Learning Models

Decades of subsurface exploration and characterization have led to the collation and storage of large volumes of well-related data. The amount of data gathered daily continues to grow rapidly as technology and recording methods improve. With the increasing adoption of machine-learning techniques in the subsurface domain, it is essential that the quality of the input data is carefully considered when working with these tools. If the input data are of poor quality, the impact on precision and accuracy of the prediction can be significant. Consequently, this can impact key decisions about the future of a well or a field. This study focuses on well-log data, which can be highly multidimensional, diverse, and stored in a variety of file formats. Well-log data exhibits key characteristics of big data: volume, variety, velocity, veracity, and value. Well data can include numeric values, text values, waveform data, image arrays, maps, and volumes. All of which can be indexed by time or depth in a regular or irregular way. A significant portion of time can be spent gathering data and quality checking it prior to carrying out petrophysical interpretations and applying machine-learning models. Well-log data can be affected by numerous issues causing a degradation in data quality. These include missing data ranging from single data points to entire curves, noisy data from tool-related issues, borehole washout, processing issues, incorrect environmental corrections, and mislabeled data. Having vast quantities of data does not mean it can all be passed into a machine-learning algorithm with the expectation that the resultant prediction is fit for purpose. It is essential that the most important and relevant data are passed into the model through appropriate feature selection techniques. Not only does this improve the quality of the prediction, but it also reduces computational time and can provide a better understanding of how the models reach their conclusion. This paper reviews data quality issues typically faced by petrophysicists when working with well-log data and deploying machine-learning models. This is achieved by first providing an overview of machine learning and big data within the petrophysical domain, followed by a review of the common well-log data issues, their impact on machine-learning algorithms, and methods for mitigating their influence.

The Impact of Overbalanced Drilling From Exploration/Appraisal Wells to Field Development Plan

During well planning, drillers and petrophysicists have different principle objectives. The petrophysicist’s aim is to acquire critical well data, but this can lead to increased operational risk. The driller is focused on optimizing the well design, which can result in compromised data quality. In extreme cases, the impact of well design on petrophysical data can lead to erroneous post-well results that impact the entire value-chain assessment and decision making toward field development. This paper presents a case study from an Upper Jurassic reservoir in the Norwegian Sea where well design significantly impacted reservoir characterization. Three wells (exploration, appraisal, and geopilot) are compared to demonstrate the impact of overbalanced drilling on both log and core data. Implications for reservoir quality assessment and volume estimates are discussed. Extensive data collection was initially carried out in both exploration and appraisal wells, including full sets of logging while drilling (LWD), wireline logging, fluid sampling, and extensive coring. Both wells were drilled with considerable overbalanced mud weights due to the risk of overpressured reservoirs in the region. The log data were subsequently corrected for significant mud-filtration and fines invasion, with calibration to core measurements guiding the interpretation. A thorough investigation of core material raised suspicion that there could also be significant adverse effects on core properties resulting from overbalanced drilling. The implications were so significant for the reservoir volume that a strategic decision was made to drill a geopilot well close to the initial exploration well prior to field development drilling. The well was drilled 6 years after the initial exploration phase with considerably lower overbalance. Extensive well data, including one core, were acquired. The recovered core was crucial in order to compare the reservoir properties for comparable facies between all three wells. The results from the core demonstrate distinctly different rock quality characteristics, especially at the high end of the reservoir quality spectrum. Results of the core study confirmed the initial hypothesis that overbalanced drilling had significantly impacted the properties of the core and well logs. This study shows how well design adversely affected petrophysical measurements and how errors in these data compromised geological and reservoir models, leading to a suboptimal field development plan that eroded significant value. This example provides a case study that can be used to improve well designs so that petrophysicists and drillers can both be part of the same value creation result.