Baker Hughes (United Kingdom)

Abstract Carbon Capture, Utilization, and Storage (CCUS) processes involve multiple interdependent stages, from capturing CO2 emissions to transporting, utilizing, and permanently storing them. These operations are complex, dynamic, and sensitive to fluctuations, requiring robust monitoring systems to ensure efficiency, safety, and regulatory compliance. However, current monitoring frameworks typically rely on static thresholds and manually configured rules, which are inadequate for managing the variable conditions in CCUS environments. These legacy systems often generate excessive false alarms, fail to detect anomalies in real time, and delay response actions. Moreover, identifying the root causes of process deviations remains time-consuming and inconsistent, hindering operational effectiveness and regulatory adherence. There is a critical need for an intelligent, adaptive system that can monitor CCUS processes holistically, detect anomalies dynamically, and recommend corrective actions without relying on static configurations. We propose an AI-driven alerting and Root Cause Analysis (RCA) system tailored for CCUS operations. This solution uses a combination of unsupervised and reinforcement learning techniques to continuously analyze data across all stages of the CCUS value chain. Unsupervised models identify deviations from expected behavior without requiring labelled datasets, while reinforcement learning enables the system to improve detection accuracy over time through interaction with users and system feedback. A domain-specific knowledge graph captures relationships between historical incidents, enabling automated RCA and the suggestion of resolution strategies. The system integrates seamlessly with industrial IoT devices, SCADA systems, and cloud-based data platforms, creating a unified, real-time monitoring environment. The AI-driven system automates anomaly detection and RCA, removing the need for static, rule-based alerting. It categorizes alerts by severity and contextualizes them using historical data, operational standards, and regulatory criteria. When an alert is triggered, the system offers actionable insights and suggests resolution pathways, improving both speed and quality of decision-making. A feedback loop further enhances system accuracy and adaptability. Ultimately, this solution improves operational efficiency, reduces downtime, ensures environmental compliance, and minimizes the risk of human error. It empowers engineers and operators with a visual interface that delivers alerts, analytics, and recommendations in real time. The novelty of this approach lies in its self-learning capabilities and its integration of machine learning with domain-specific knowledge. Unlike traditional systems, it continuously adapts to changing operational patterns, eliminating the need for manual reconfiguration. The use of a knowledge graph for RCA and the integration of reinforcement learning distinguish this solution as a scalable, intelligent framework for sustainable CCUS operations.

Abstract When infill perforation intervention opportunities are evaluated, it is essential to confirm reservoir pressure. Without appraisal of reservoir pressure, perforation of depleted reservoirs can negatively impact the overall well performance and even introduce early water breakthrough. We successfully evaluated the reservoir pressures of unperforated high-pressure, high-temperature (HPHT) gas-condensate-bearing reservoir sands using pulsed-neutron well logging technology. This paper discusses multiyear time-lapse pulsed-neutron well logging in an HPHT gas-filled borehole using previous- and next-generation sleeved-multidetector instruments for through-casing reservoir pressure depletion monitoring. We employed a previous-generation sleeved-pulsed-neutron tool to acquire three-detector-based pulsed-neutron capture (PNC) data through a gas-filled wellbore to analyze reservoir pressure depletion in 2022. The challenge in the first logging run was a high-temperature environment close to the tool operating limit. Surveillance revealed that the upper target sand was not depleted, whereas the lower sands were depleted. The operator planned to perform time-lapse monitoring of the target sands in 2024 before the final perforation decision. The next-generation sleeved-pulsed-neutron tool was deployed in the 2024 run. PNC data were recorded twice as fast as the first run using the legacy tool. Crucial technological advancements in the next-generation pulsed-neutron tool enable time-efficient PNC data acquisition. First, the pulsed-neutron generator emits twice as many high-energy neutrons as the legacy tool. Second, the adoption of denser scintillation material for gamma-ray detectors allows for recording higher gamma-ray count rates with higher resolution. Finally, a high-speed digital electronics system processes gamma rays faster. The logging speed of the multidetector PNC mode with the new tool was twice that of the legacy tool at 20 ft/min. Benefiting from a faster logging speed, thus reducing the time the tool is in the hole, the heat in the borehole, approximately 350 degrees Fahrenheit, was not a significant concern. The data obtained using the next-generation pulsed-neutron tool demonstrated reliable coherence with the previous data acquired via the legacy tool, confirming a smooth transition from the old tool to the new tool and ensuring seamless time-lapse data analysis. Additionally, extensive Monte Carlo N-Particle (MCNP) models under various reservoir pressure conditions for legacy and new tools were integrated into the reservoir depletion analysis. Time-lapse analysis revealed that the upper sand remained under the virgin reservoir condition but that the lower sands unconnected to the upper sand underwent reservoir depletion. The target reservoir was subsequently perforated, confirming the virgin pressure status as predicted by the pulsed-neutron logging appraisal. The single-well opportunity increased productivity by approximately 15,000 barrels of oil equivalent per day (BOE/D). Technological advancements, resulting in increased logging speeds, extended the operator's operating window at high temperatures, allowing for a greater appraisal scope. Cased-hole depletion analysis presents a low-risk, cost-effective screening tool for in-fill production opportunities in late-life field production.

Abstract Carbon Capture and Storage is a key technology pillar to assist the globe with meeting net-zero commitments. For the world to meet its net-zero emissions commitments by 2050 large networks of CO2 pipeline infrastructure needs to be built in a short amount of time. A key component of such an extensive roll out is the safety of the infrastructure installed. Within CO2 pipelines a key additional risk is running ductile fractures or rapid crack propagation failures in pipelines which is exacerbated by the decompression behaviour of CO2. Spoolable composite pipes, including RTPs, could be an advantageous technology to roll out of CO2 pipelines. However, there is limited available literature on the susceptibility of such pipes to RCP and RDF failures. A literature study is presented detailing RDF in steel pipelines and their mitigation using composite crack arrestors. Additionally, literature available on RCP of thermoplastic pipes is shared. A comparison is made between spoolable composite pipes and composite crack arrestors, highlighting the similarity of the two. A full-scale RCP test following a modified ISO13477 test procedure was completed on a 6 Inch 1,500psi Glass Fiber reinforced RTP produced in Houston Texas, with CO2 fluid. The pipe was ruptured using a driven knife blade, after which the CO2 was released from the pipe over approximately 10.5 seconds. The crack surface was inspected following the burst, with the crack having only extended ~1 inch past the initiated crack. The results showed a high level of resistance of RTPs to RCP in CO2 pipelines.

Abstract Machine Learning (ML) involves the training of a model to make predictions based on data. A model might be described by several parameters, normally, weights and biases, where training is the process to determine the ideal parameters that comprise model, based on the dataset (a collection of raw data arranged in columns). ML can be divided in four main types, supervised learning, where training is based on labelled data; unsupervised learning, based on unlabelled data, where the goal is to find hidden patterns or structures within the data; semi-supervised learning, it uses a combination of labelled and unlabelled data; and finally, reinforcement learning, the learning is through trial and error, where the model receives reward by taking desirable actions and penalties for undesirable, the goal is to maximise the reward. There are a great variety of ML techniques for various tasks, such as dimensionality reduction, clustering, and regression, commonly applied for image recognition, speech recognition and predicting modelling. Scale inhibitor (SI) squeeze treatment is one of the most common techniques to inhibit the deposition of oilfield scale. It consists in bull-heading a scale inhibitor slug in producing wells, the chemical retains on the rock surface and is slowly released when the well is back in production, the well will be protected from scaling, if the produced chemical concentration is above a certain threshold, usually few ppm. Generally, process modelling is applied to predict the SI return concentration vs water produced, to determine the treatment design lifetime, based on the operating parameters, formation and inhibitor properties, particularly retention capability. Commonly, a historically matched field isotherm results in predictions with significant degree of accuracy, where the main disadvantage is that historical data for the producing well under consideration is necessary. The purpose of this manuscript is to develop a ML model, where the input is the treatment design, which consists of chemical slug volume, chemical concentration, and overflush volume, well formation characteristics (perforation heights, permeability and porosity), and the output the scale inhibitor return concentration vs. volume of water produced. The ML model was trained, tested and blind validated using squeeze treatment data for a field in the North Sea. The dataset consists of 141 field squeeze treatments, 9 for blind validation, and 131 for 80/20 training and testing split. The best ML model Random Decision Forrest (RDF), which resulted in significant accurate prediction over the 9 blind squeeze treatments. This model may be used as proxy for the process modelling of squeeze treatments.

Abstract Non-metallic composite downhole tubulars provide significant value including less carbon footprint and reduced total cost of ownership. However, matching pipe performance envelopes to combined loading in different scenarios cannot be performed using commercially available software, due to various challenges including anisotropic pipe behavior, variation of material properties across the depth and change of well conditions. This publication highlights multidisciplinary collaborative efforts to estimate such loading in composite tubulars and compare to existing use cases. Load cases and conditions were determined, and analytical calculation workflows were deduced for loads at various depths and conditions across the lifetime. Load cases and alternate loading sequences are listed, interactions between tubular & system components are reflected and proper boundary conditions are applied to estimate such combined loads. Validation was performed against finite element analysis methods and methodology was applied in specific well applications. Results using suggested analytical methodology show good matching with numerical results using finite element analysis. Loads were matched against pipe performance envelopes with suggested safety margins for different pipe designs. Whilst pipe design changes, variation of pipe effective properties, including stiffness, density, and dimensions, affect the total combined loads acting on the tubular system. Well construction parameters including deviation profile, depth, and geothermal gradient affect conditions inside the well, differential pressure and axial loads. Change of system components including pipe connections, completion, installation tools and ancillaries significantly affects such loading profiles. These system components can be optimized and custom components and installation procedures for non-metallic composite tubulars can be developed and qualified to reduce such loads on the pipe. Application of reel-deployed thermoplastic composite pipes is unique and novel in downhole applications and must be accompanied by customized design workflows and validation programs. A new analytical design approach was developed for estimating the thermoplastic composite pipe structural response. This publication is bringing an integrated workflow for estimating these loads on non-metallic downhole tubulars, which – to the best of the authors’ knowledge – has not been presented elsewhere.

Abstract Coiled tubing operations become ever more complex to meet the demands within the industry. A worldwide challenge is a smaller group of highly qualified field personnel to meet these expectations and without such can lead to a potential compromised service delivery. To alleviate this our focus is on the increased level of unit automation providing the unit operator with greater monitoring, feedback, technical advice and, if needed, intervention on the unit to prevent failures. The combined solution is derived from three main areas: automatically update the CT force limits, continuous monitoring of key parameters to provide operational advice, and if required to prevent a potential failure, automatically intervene with control of the appropriate injector functions. Core to the system has been taking a highly complex CT flow and force model and transferring it to run automatically with only minimal intervention from field operators or specialists. The paper will detail the multiple system capabilities and the associated historic service delivery issues that created the need for the automated system. For example, the injector control system utilizes five different limit measurements to auto stop the injector to prevent failure. This includes the ability to auto monitor downhole force measurements during logging operations, which liberates wireline personnel to attend to other tasks during CT deployment and recovery. System enhancements such as the ability to easily recalibrate force modeling by adjusting friction and surface loads will be detailed. The system has further enhanced the ability for field crews to track expected vs actual loads and take preventative action without the need for a field engineer to be present. The system is being utilized across the service provider's fleet of units. Statistics on reduction of incident rates will be presented. Many operating companies are fully engaged and supportive of the journey to operational automation/remote monitoring and service delivery improvements. The majority of the features covered above have never been presented in a technical paper. This paper would provide valuable industry insight into the current and future expectations of technology that will aid the operating companies’ goals.

The necessity of forecasting recovery from the gas hydrate reservoirs is restricted due to the limiting data availability for reservoir properties, such as porosity and hydrate saturation, used in reservoir models. Predicting recovery becomes essential to determine the feasibility and viability of a reservoir. Therefore, it is paramount to understand uncertainty propagation from reservoir characteristics to reservoir performance by using reservoir simulations. In this work, we explore the uncertainty in reservoir performance due to the uncertainty in reservoir characterization of porosity and saturation. We employ the Monte Carlo method to analyze the impact of uncertainty on gas and water production from a class 2, confined, oceanic methane hydrate reservoirs. We observe that uncertainty in porosity leads to a substantial uncertainty in our prediction of gas recovery until a critical porosity value. If the porosity is higher than the critical porosity value, then the gas production using depressurization does not change significantly. This is attributable to the thermal nature of gas hydrate dissociation in which high porosity means less heat content from the sediments for hydrate dissociation. For the same reason, at the same porosity, variation in hydrate saturation does not lead to a high variation in gas recovery. The uncertainty in porosity leads to a 24% coefficient of variation in cumulative gas production, uncertainty in initial hydrate saturation leads to 3% and their combined uncertainty leads to 20% coefficient of variation in cumulative gas production. Using the cumulative density function of the gas production, we show that our P90, P50, and P10 estimates will vary depending on the production time considered for estimation.

Abstract Flexible pipelines are often subjected to cover requirements on the seabed due to protection or stability issues. These requirements may be satisfied using a range of techniques, e.g. covering the pipe with rocks or soil. Due to pressure and thermal expansion in operation the pipe may be susceptible to lateral or upheaval buckling (UHB) out of the cover which may violate the cover requirements and could ultimately be detrimental to the flexible pipe. The present paper is concerned with the design of reliable and cost-efficient cover solutions for unbonded flexible pipes. A new DNV guideline will introduce “Allowing Irreversible Displacement” design approach which allow for the pipe to move within the cover provided that the end-of-life limit states for pipe and cover are satisfied. The paper includes a description of how a model following that approach may be defined, including considerations on load case matrix and a discussion on relevant sensitivity cases to ensure conservatism in the cover design. A case study is included to show how the principles of the guideline may be applied in UHB design.

Abstract One of the critical approaches for scale control is the proper selection and use of scale inhibitors. Laboratory tests help to select the appropriate scale inhibitor with the most common testing methods, including static bottle test and dynamic scaling loop test. Recently, Kinetic Turbidity Test (KTT) has gained increased recognition as a new testing method for scale inhibitor evaluation due to short testing time, simple sample preparation, and good reproducibility. There has been a good deal of research and studies on KTT as a technique for multifunctional scales, including calcium carbonate and barium sulfate (especially low scaling brines), halite and silicates. In addition, KTT can examine dispersant effects of polymers and surfactants on scale and other solids and can work under anaerobic conditions to give mechanistic understanding in the presence of iron. This paper discusses an alternative and efficient scale inhibitor testing method and gives insight for scale treatment chemistry and dosage by comparing and contrasting the different evaluation methods for scale inhibitors.