Fifth EAGE Digitalization Conference & Exhibition

This study explores the integration of data Lakehouse platforms, Apache Spark, and BI(Business Intelligence) tools to enhance the efficiency of historical drilling data analysis for well planning. Advanced downhole equipment generates terabytes of data at high frequencies, necessitating robust storage and processing solutions. By leveraging a corporate data lake and cloud-based PySpark, the study organizes and interconnects hundreds of well datasets, integrating them into a dedicated database for seamless visualization via BI tools.

Key analyses included Dogleg Severity (DLS) computations, machine learning (ML)-based bit performance evaluation using Logging While Drilling (LWD) data. Challenges such as data preprocessing, outlier removal, and code validation were addressed through iterative development and condition enhancements. The study highlights the automation of offset wells analysis, significantly reducing time and effort compared to manual approaches.

Novel contributions enabling Drilling Engineers to adopt roles of Data Engineers and Scientists. The findings underscore the potential of ML in automating analytical workflows and extracting actionable insights from extensive datasets, driving efficiency and innovation in drilling operations.

In this paper we present findings from a study to explore the possibilities of using pulsed electromagnetic (EM) waves for CO2-storage monitoring purposes, to help mitigate the release of large quantities of CO2 in the atmosphere. The study focused on the geological setting of the Mississippi Basin, targeting the detection of CO2 leaks at depths ranging from 350 to 3000 m. The primary objective was to assess the effectiveness of surface-based monitoring technologies, such as pulsed EM detection, that eliminate the need for wellbore access. A sensitivity study was performed to determine the minimum CO2 concentration and leak size detectable using this technology by utilizing finite-difference and machine learning-based computational analysis.

The simulation was run for different D (diameter of CO2 cube) values. The CO2 scan is most sensitively detected in the data using our in-house anomaly detector. A neural network is trained to reproduce the time measured time series across many windows with minimal error. The simulations suggest CO2 leaks of the order of 2–4% volumetric concentration can be reliably detected from the surface at a depth of 350m. Simulation results indicate 3000m leak detection is difficult but theoretically possible for a concentration of 20% and up.

This abstract presents a new proprietary solution adopted worldwide by Eni since 2024. The Platform is designed to address critical challenges in discrete real-time data management for rig operations, such as data unavailability, lack of standardization, and time-consuming processes. Utilizing a cloud-based platform, the proprietary solution for discrete data collection enables near real-time data upload and validation, significantly improving data operational efficiency and decision-making. By centralizing data management and monitoring, the new solution delivers measurable enhancements in data quality, streamlines operational workflows, and supports informed decision-making regarding data. This study underscores the Platform’s transformative role in tackling key operational challenges related to data management and driving digital innovation in rig operations.

The findings conclude that the solution accelerates the digitalization and automation of business processes, resulting in enhanced productivity and more effective data management. By simplifying data collection, validation, and distribution, the platform ensures the availability of high-quality data for analysis. Its real-time data management capabilities enable continuous monitoring and instant feedback, fostering adaptive strategies and operational excellence. These features empower faster, data-driven decision-making and more agile responses to data management related operational challenges, positioning the new platform as a pivotal tool for advancing efficiency and innovation.

This study explores the use of large language models (LLMs) in coding agents to enable geoscientists to converse with subsurface databases through natural language queries. By utilizing a ReAct (reasoning and action) framework, the agents can dynamically plan and execute SQL queries based on user input and adapt to errors or intermediate results. The study tests the performance of the agents on a complex SQL database of Petrel data and metadata from over 1200 projects. Initial challenges included the agents’ difficulties in understanding table relationships and correctly formulating queries, which were mitigated by providing database descriptions and adding specific ReAct loop strategies to help solve the task. The agents demonstrated improved accuracy, particularly in complex queries requiring joining data from multiple tables, while also reducing response time and resource costs. Results indicate that users can effectively interact with their data without needing SQL expertise, revealing the potential benefits of coding agents in enabling new subsurface workflows.

Clustering well-log data into electrofacies using machine learning is a well established problem in the geophysical and data-scientific communities and has become of major relevance in the last decade. Many of the publications in the field, however, have limitations pertaining to the size of the learning dataset, the choice of the right number of clusters, and the management of noisy output (particularly stemming from the underlying spatial autocorrelation of the electrofacies). In this work, we propose an automated workflow which responds to these shortcomings and apply it on a case study from offshore wells in Australia. The approach lies in three main steps, which are the pretreatment of data via principal-component analysis, the clustering of the data via a Gaussian mixture model and the optimization of the number of clusters using silhouette analysis, and the homogenization of the clustering results using the stochastic-relaxation algorithm. The approach shows promising results on the test case by being coherent with, albeit less nuanced than, the ground truth given by an expert’s classification.

The occurrence of geomechanical events (e.g. drag, stall, kick) might hinder production and incur in losses. In this context, manual classification of daily drilling reports is a daunting task. In this work, we propose a method -grounded in the In-Context Learning paradigm - that leverages a commercial Large Language Model to classify daily drilling reports so that we enrich the prompt fed to the LLM to boost its performance. Experimental results attest the effectiveness of our approach comapred to other three variants. This work might help the oil industry in extracting valuable information from large amount of data in order to mitigate losses and to support data driven decision making.

Oil and gas companies invest heavily in acquiring surveillance datasets over producing fields. Quickly integrating and understanding these datasets is crucial for maximizing asset value and visualizing them in one place is an important step in this process. However, there are still significant challenges in co-visualizing live production and surveillance data with geological maps, seismic products and well data in 2D or 3D. Field teams often spend significant time manually creating ‘moment in time’ maps or models, which are quickly outdated. Subsurface practitioners need an easy-to-use application, accessible to all disciplines, which automates the visualization and integration of critical field static and dynamic data. This paper describes bp’s journey to creating a global, multidisciplinary web-based tool designed to help field teams collaborate more effectively and rapidly to make informed reservoir management decisions and to maximize the value of their surveillance data. It explores the initial requirements, challenges in developing and deploying the solution, and highlights the opportunities this tool can unlock for field teams.

Unstructured and semi-structured data like reports, logs, and other related records and documents are abundant in the oil and gas industry but challenging to analyze due to lack of standardization and complexity. Traditionally, custom data parsers had to be written to aggregate and extract insights from these data sources. However, generative AI (Gen AI) can now accelerate this process. Leveraging large language models (LLMs) and Gen AI techniques, it now possible to use Gen AI to help create data parsing templates by specifying sections of interest on the sample data files. Gen AI can then scale and apply these templates to a broader set of data and automatically adjust the template selection based on the file or document type. This approach streamlines and simplifies pipelines to aggregate diverse unstructured and semi-structured data sources into centralized enterprise data repositories, unlocking faster insights and data-driven decision making. As data volumes grow, Gen AI provides a powerful way to tame unstructured data complexity in the energy industry.

This study presents an innovative deep learning-based workflow for the semantic segmentation of sedimentary components in core images. The approach focuses on identifying and quantifying major components of reef systems (corals, coralline algae, microbialites, bioclastic sands...), from cores collected during IODP Expedition 389 – Hawaiian Drowned Reefs.

High-resolution images of the cores were obtained with a linescan camera, resulting in continuous images. A U-Net architecture, widely recognized for its efficiency in image segmentation, was employed using a weighted Dice-Sorensen coefficient to address class imbalances. The model achieved a segmentation accuracy of 81.1% after post-processing with Conditional Random Fields (CRF), improving the quality of segmentation results. Secondly, the model also quantifies the relative proportions of sedimentary components along core depths, aiming at facilitating the interpretation of paleoenvironmental variations recorded by reef systems.

This workflow provides a robust, automated solution for core analysis, reducing the time and expertise required. It holds significant potential for industrial applications like reservoir characterization and societal studies, such as paleoenvironmental reconstructions. Perspectives includes technical challenges (color normalization, benchmark the loss weight...) and methodological improvements (model refinement for unusual or complex sedimentary patterns, integration of complementary datasets to develop a multiple data source approach.

The energy industry is in the process of undergoing a profound digital transformation. This transformation is using advanced technologies to enhance efficiency, safety, and sustainability. User Experience (UX) design plays a pivotal role in this evolution, ensuring that sophisticated tools and data are not only accessible but also useable for professionals in diverse operational contexts. This paper explores the strategic importance of UX in enabling the digital transformation of the energy industry. We discuss examples where UX-driven solutions have empowered decision-making, leading to measurable improvements in performance and quality. By considering the role of the UX in the digital strategy, we will show how we aim to bridge the gap between the technology and the people who rely on it.