Innovative tools for decision-making in drilling
Navigating through complex geological formations is a constant challenge within drilling. Ressi Bonti Muhammad has developed new tools to help geosteering professionals make more informed decisions and improve the chances of success.
– A drilling operation and geosteering is like navigating through a pitch-black road with only the car dashboard as your guide. You have some information about where you’ve been, but the path ahead is shrouded in uncertainty, says Ressi Bonti Muhammad.
He has just finalized his PhD – the first from our research centre DigiWells.
Geosteering is about making decisions to direct and keep the well within the target layer based on real-time information from drilling. The overall goal is to bridge the gap between decision theory and advisory in practical drilling operations.
– I've worked on the development of new tools and techniques that can utilize all available information, explains Muhammad.
Muhammad and DigiWells director Erlend Vefring
Improving the decision support system
Muhammad used reinforcement learning algorithms to develop and improve a decision support system that provides optimization-based decisions under geological uncertainty during geosteering.
– Previous research in this area has typically relied on traditional decision-making algorithms. We sought to employ a novel approach that has shown considerable promise in addressing sequential decision-making problems. Reinforcement learning algorithms has outperformed human decision-making on various problem domains, for example when beating expert gamers at classic ATARI games, which is all about making the right decisions. The transfer value to geosteering is great, explains Muhammad.
– Ressi initially tested reinforcement learning in simple test environments tailored for theory-based mathematical models. The results were better than we expected, and eventually we paired the reinforcement learning with general subsurface state estimation, explains senior research Sergey Alyaev at NORCE and one of Muhammad’s supervisors.
Close collaboration within the centre
DigiWells is a NORCE-led research centre, working together with industry on digitalization, automation and autonomous drilling over the course of 8 years.
– Working together with the research communities from NORCE and the University of Stavanger, and at the same time getting insight from the industry partners at DigiWells, was of great importance to me throughout my work, says Muhammad.
As part of DigiWells, Muhammad collaborated with the other PhD-students connected to the research centre.
– We became a tied group of PhD-students that shared our progress with each other and pushed each other to cross the finish line. It was both helpful and motivational, says Muhammad.
Reidar Bratvold (UiS), Ressi Bonti Muhammad, Alejandro Escalona (UiS) and Sergey Alyaev (NORCE)
Continuing the work
Muhammad’s research inspired a reinforcement-learning line of research in a spin-off research project called “Distinguish” (https://geosteering.no/distinguish). The main purpose of the project is to use machine learning to “learn” geology and build an artificial-intelligence system for decision support in complex operations on the Norwegian Continental Shelf.
– As part of the project we have extended Ressi’s work and connected it to a cloud platform used in many geosteering operations worldwide, says Sergey Alyaev.
