Improved Safety and Efficiency of Tubular Installations through Real-Time Optimization, Machine Learning, and Physical Modeling

Summary Casings are a vital part of well construction, effectively isolating zones of different pore and fracture pressures as well as preventing contamination of the surrounding formation. When installing a casing, the primary concern is it becoming stuck, which could result in potentially losing all lowered equipment and having to redrill the section. Loss of kilometers of pipe is costly, and the motivation is high to avoid this. Traditionally, an operator manually produces a so-called broomstick plot and adds hookloads that are recorded in real time. This process is tiresome, difficult due to rig movement, and not reproducible, as interpretation of real-time hookloads is highly opinionated. In this paper, we present a real-time algorithm that automates this workflow. Using artificial intelligence in the form of a transformer encoder, real-time data are converted to weights recorded while lowering. An augmented soft-string physical model provides the simulated up- and downweights, which are fitted to the observations with the use of numerical optimization. The discrepancy between the observations and simulation is interpreted by a transformer encoder that raises alarms when the risk of getting stuck is imminent. This algorithm has been operational on the Norwegian Continental Shelf, UK Continental Shelf, and Continental Shelf of Brazil since 2022, allowing automated surveillance of hundreds of casing runs.