Making Under Uncertainty - Edge Intelligent Decision for Drilling: From Data to Actions in Real Time Well Engineering

Abstract

Abstract Planning and learning are two primary approaches to intelligent decision making at the edge when the well is constructed. Planning enables us to take immediate actions far into the future, but it requires accurate well engineering models, which are often difficult to acquire in practice. The models are built on engineering assumptions which may not be valid all the time. But the data will provide additional support to suppress the assumption and improve the models. This paper presents the results that can be used under uncertainty through planning, through learning, most importantly by integrating planning and learning. Planning and learning are two primary approaches to intelligent decisions. When it comes to the edge decision whether it is manual or semi-automated or fully automated (Dupre, 2013) it requires tighter coupling of engineering models through microservices and reinforced learning from the data and feedback from the driller. Five major types of uncertainties are considered in calculating the drilling operational parameters are measurement uncertainty, data uncertainty, engineering model uncertainty, computational or algorithmic uncertainty and decision uncertainty. Several examples are presented as the well is steered and navigated with interactive tasks. Surface hookload and torque values serve as good indicators for some undesirable scenarios or anomalies during drilling, such as stuck pipe, buckling, and inadequate hole cleaning. However, to detect these risks, it requires drilling engineers to perform engineering model calibration manually and regularly, which costs more efforts and poses significant uncertainties on the detection. This paper describes how these problems are circumvented in addition by providing project ahead paths based on various constraints in real-time with predicted uncertainty zone. This option at the edge makes it possible for drilling engineers to monitor live drilling wells anywhere and anytime while enabling the rig personnel to make significant improvement to operations. Hence, the optimization is no longer static and becomes a dynamic function of depth. These conditions result in constantly varying constraints and, thereby, constantly varying optimized operating parameters to maximize the rate of penetration or minimize specific energy or well cost. This study presents a real-time optimization technique for rate of penetration with energy-based models. Based on the place and position of the bit, the updated data has been used to modify the proposed well design on the fly. The above underpinning methodology have been supported with practical field examples.