Integration of Physics Models and ML Algorithms Enhances Automated Sliding Performance

_ This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 227896, “Enhancing Automated Sliding Performance Through the Integration of Physics Models and Machine-Learning Algorithms,” by Matthew Summersgill, SPE, Jefferson L. Xu, SPE, and Angus L. Jamieson, SPE, Helmerich & Payne, et al. The paper has not been peer-reviewed. _ Automated sliding has existed in the industry for years. In early deployments, simply meeting steering requirements with nearly 100% automation, even at the cost of drilling performance, was considered a technical success. Now, algorithms are expected to meet or exceed performance standards set by the best drillers. A multifaceted approach leveraging precise rig control, physics models, and machine-learning techniques aims to deliver consistent high-level performance in a scalable manner. Introduction Once tools run through the rotary table, only two aspects of the drilling process have a meaningful effect on the outcome: rate of penetration (ROP) and tool-face control. A single metric combining both these aspects, effective ROP, serves as a benchmark that characterizes the performance of both a human driller and an automated slide-control system fairly. To evaluate progress toward full automation, a second metric, interventions per foot drilled, is useful. Many modern automated sliding systems with a remote-control interface might be considered either a Level 3 or 4 system, where the controller can perform all “driving” functions under specific circumstances, but the driver must be prepared and has the option to take control at any time. Definitions and consequences for both steering success and failure in the context of directional drilling and driving differ greatly. In drilling, the distinction grows blurry between partial automation and a Level 5 fully automated system, where the driver’s attention or interaction is not required. When the frequency of interventions dwarfs the pace at which the engineering team can investigate, interventions to address breakdowns of steering logic become lost in a sea of interventions made to marginally improve performance. To confront this challenge, a deliberate design decision may be made to direct steering interventions when testing new steering algorithms, requiring a remote driller instead to disable the controller completely and switch to a different user interface. Removing this human safety net enables evaluation of an automated control system’s true performance. A mature directional-drilling-automation platform is key to enabling testing, both in early-stage testing and when scaling up to multirig trials. Equipment and Processes Test Infrastructure. Development and testing of this model-based slide controller relied heavily on a mature directional-drilling software and rig-automation platform. Development cycles that might have taken months in the recent past can be reduced to days or hours. Investments in high-bandwidth networking, data replication and backhaul systems, and automated build and deployment of software to rigsite servers enhance productivity of an engineering team. To test changes without a team member on site, an uninterrupted high-speed connection to the rig is required to monitor and troubleshoot. Development of automated build, testing, and deployment tools for software also pays dividends for algorithm-development efforts.