Laying the Groundwork for Efficient Drilling: A Centralized Data Approach to ROP and Subsurface Modelling

Introduction

Optimizing the Rate of Penetration (ROP) in drilling requires a shift from traditional methods to a data-driven strategy that merges real-time insights with subsurface data. It’s evolved from simply drilling faster and cutting costs to a unified, holistic approach, where safety, efficiency, teamwork, and achieving the well’s objectives come together to get it right the first time. Sonic logs have historically been indispensable for pore pressure and geomechanical modelling, offering critical subsurface detail to guide operations. In their absence, traditional approaches, such as empirical formulas, rock physics models, or signal processing techniques, often fall short. Machine learning is transforming this landscape, enabling precise sonic log predictions from diverse data sources and creating a centralized foundation for optimizing ROP.

Central to this approach is a deep understanding of the interaction between drilling parameters and the rock being drilled. Machine learning algorithms and advanced subsurface modelling empower teams to refine decisions about drill bit selection, weight on bit (WOB), rotary speed (RPM), and mud weight, aligning operational parameters with real-time subsurface conditions. This data-driven insight fosters smarter, safer, and more effective drilling operations.

The innovative aspect of this approach lies in its holistic integration of data. By combining real-time subsurface modelling, machine learning, and a unified data platform, ROP is not only optimized but also safety, operational efficiency, and interdisciplinary collaboration are significantly enhanced. While not entirely new, this methodology distinguishes itself through its comprehensive inclusion of geomechanical parameters, advanced data integration, and the sophistication of its machine learning models. By addressing the shortcomings of surface-only techniques and earlier machine learning solutions, a more adaptive, unified, and predictive framework is established that advances industry practices.

Limitations of Traditional ROP Optimization

Traditional methods for ROP optimization primarily rely on surface parameters such as WOB, RPM, and mud flow rates. While these metrics are convenient and readily available in real-time, they fail to capture the geological realities beneath the surface. Subsurface factors like lithology, rock strength, and pore pressure variations play a critical role in determining ROP. Without accounting for these variables, traditional approaches remain incomplete, assumption-driven, and thus error-prone.

For instance, when drilling through formations with interbedded lithologies, unaccounted rock strength variations often lead to unexpected rate changes, increasing costs and tool wear. Incorporating both surface and subsurface data enables drilling teams to align parameters dynamically with changing geological conditions, ensuring more efficient and reliable operations.

A Pragmatic Approach to Sonic Prediction with Machine Learning

To overcome the limitations of traditional sonic data derivation methods, we implemented a machine learning-driven framework to enhance the accuracy of sonic log predictions. This innovation integrates data from diverse sources, including surface drilling parameters, mudlogs, geological logs, and image-based cuttings analysis, into a unified data foundation accessible to all users.

Key components of the framework:

• Data ingestion and preparation: Standardized loading of drilling data and cuttings images ensures consistent, reliable, depth-correlated inputs.
• Feature engineering: By incorporating image-based features such as colour, texture, and shape of cuttings, geological insights are enhanced. Subtle lithological variations detected in real-time cuttings images can prompt immediate adjustments to ROP, improving efficiency.
• Model training and evaluation: We leveraged machine learning algorithms such as XGBoost, which excels at identifying complex, non-linear relationships in high-dimensional datasets. Initial trials showed a reduction in prediction deviations by 10% compared to traditional methods, aligning predicted and observed sonic values within acceptable industry thresholds.

This unified approach establishes a robust foundation for ROP optimization, aligning drilling strategies with real-time geological insights and improving ROP models through high-quality, data-driven inputs.

Figure 1: Comparison of GR log (green) and cutting positions (orange circles) alongside DTCO (blue and red) and DTSM (orange and purple) predictions. The difference plot highlights deviations between predicted and observed sonic values, illustrating model accuracy across varying depths.

Integrating Geomechanical Parameters for ROP Optimization

Building on the sonic prediction framework, we integrated key geomechanical parameters into the ROP optimization process. Traditional models often rely on surface parameters alone, supplemented occasionally by gamma ray (GR) logs. However, the primary driver of drilling efficiency lies in the rock itself.

By incorporating parameters such as Unconfined Compressive Strength (UCS), GR, WOB, and torque (Trq), data which are all routinely acquired during drilling or derived through subsurface workflows, our model dynamically adapts to geological variations across intervals. UCS, for example, reflects rock strength, while WOB and Trq measure alignment with that strength. Although UCS serves as an example, the approach is versatile enough to optimize any elastic property (e.g., shear modulus, bulk modulus, Poisson's ratio, Young's modulus) or mechanical property (e.g., UCS, friction angle, cohesion, tensile strength) to enhance performance.

Using a Random Forest model, which dynamically adjusts predictions based on interval-specific data, we achieved robust prediction across varying geological conditions. Notably, this model adapts to new inputs in real-time, continuously refining ROP predictions, even with limited pre-drill data.

Figure 2: An integrated display of well logs includes Gamma Ray, Unconfined Compressive Strength (UCS), Weight on Bit (WoB), torque, and ROP measurements. Predicted and actual ROP are compared, with highlighted depth intervals showcasing predictive accuracy. The next log presents optimized (red) and raw ROP, while the final log outlines thresholds for bit usage in forward predictions to deeper depths. Data is presented from left to right.

Unified Data and Enhanced Communication Across Teams

A unified data foundation achieves more than simply enhancing data availability, it bridges the divide between geological and engineering teams, fostering collaboration and aligning workflows. Historically, data from cuttings analysis, geological logs, and drilling metrics existed in isolated software tools, limiting opportunities for cross-disciplinary insights. Geologists often analysed cuttings within specialized systems, while engineers worked independently with drilling metrics, creating silos that hindered communication.

By centralizing these diverse data sources, including image-based cuttings analysis, sonic predictions, and real-time drilling parameters, this approach eliminates barriers, providing parallel access to critical information for all teams. This shared platform enhances communication, ensuring that the most accurate and current data informs decisions, promoting a comprehensive understanding of subsurface conditions.

Operations geologists analysing real-time cuttings data, for example, might identify a lithological change that requires engineers to adjust drilling parameters. With unified data, such insights can be immediately communicated and acted upon, minimizing delays and optimizing operations. Rigorous data governance practices, such as quality control checks and automated validations, further ensure data reliability and resolve discrepancies between geological logs and real-time drilling metrics.

This ROP optimization framework is not just an improvement in speed or efficiency, it represents a fundamental evolution in data integration and accessibility. By moving beyond isolated workflows, this cohesive, data-driven approach deepens subsurface understanding and establishes a new benchmark for safer, more efficient, and sustainable drilling practices.

Figure 3: Workflow diagram illustrating the data integration process from databases, data lakes, and network drives to Curate’s centralized platform via API and autoloaders, facilitating seamless data flow and distribution for enhanced drilling analysis.

Conclusion

Our approach to ROP optimization, built on sonic prediction, geomechanical integration, and unified data, marks a step forward in how drilling teams navigate complex subsurface environments while seamlessly connecting subsurface and operations teams. By adopting a cohesive, data-enriched platform, we enable smarter, real-time decision-making that redefines the future of drilling operations. As the industry evolves, robust investments in data infrastructure and integration will become indispensable for addressing the growing challenges of modern subsurface exploration and drilling. This unified workflow establishes a new benchmark for operational efficiency, safety, and collaboration, elevating the standards for excellence across all facets of drilling.

Acknowledgments

Thanks to our service teams whose expertise in geomechanics, real-time pore pressure analysis, and data integration has been instrumental in shaping this approach. The data used in this blog are synthetic, designed to demonstrate our methodology and showcase its capabilities while protecting proprietary information.