Abstract
<jats:p>In recent years, the oil and gas industry has accumulated large amounts of drilling sensor data. However, much of the information contained in this data remains underutilized because raw time series data is often inherently unstructured, noisy, and is challenging to align with operational context (Payette et al., 2024). In this study, historical data from unconventional wells were analyzed, and a methodology for slip-to-slip segmentation (stand segmentation) of data was developed to consistently identify stand intervals in drilling operations. The resulting structured outputs enable applications such as MSE lithology transition modeling, extraction of performance metrics, and early detection of mechanical stuck pipe.</jats:p> <jats:p>Slip-to-slip segmentation is relevant because it allows systematic exploration of drilling data beyond simple time series inspection. By partitioning operations into consistent stand intervals enriched with metadata, machine learning models can be trained more reliably, and domain-specific business rules can be developed with higher robustness. This has important economic implications, as improved diagnostic tools and predictive models have the potential to contribute to reducing non-productive time (NPT), enhancing drilling efficiency, and lowering costs. The methodology aligns with industry trends emphasizing artificial intelligence (AI) for automation, optimization, and safety in drilling processes.</jats:p> <jats:p>Despite this relevance, segmentation into slip-to-slip intervals remains relatively underexplored. Most documented approaches rely on heuristic rules and thresholds applied to surface signals such as hookload and block position. These methods, while straightforward to implement, are sensitive to sensor quality, noise, sampling inconsistencies, and variability across rigs (Zhao et al., 2019; Isbell et al., 2022). To date, we are unaware of any broadly recognized AI-based framework specifically for stand boundary detection. In contrast, machine learning has been more widely applied to related problems. Studies have demonstrated its effectiveness in rig state classification (Coley, 2019; Ben, James, and Cao, 2019; Qiao et al., 2024), performance evaluation (Santos et al., 2024), and drilling condition analysis (Isbell et al., 2022). More recently, Santos et al. (2025) presented a data-driven operation classifier and stand counter tool, demonstrating the feasibility of integrating machine learning with structured segmentation for operational diagnostics.</jats:p>