Estimation and control for autonomous directional drilling with rotary steerable systems




Keller, Alexander M.

Journal Title

Journal ISSN

Volume Title



Accessing difficult to reach hydrocarbon reservoirs while simultaneously reducing risk and increasing efficiency has driven demand for precise autonomous directional control of rotary steerable systems (RSS). Despite advances in RSS technology, precise control remains difficult primarily because the tool is located deep underground which introduces challenges that require advanced methods to overcome. In the deep subsurface environment, many parameters are difficult or impossible to measure. One consequence of this fact is model mismatch is often encountered which degrades control performance. Another consequence, resulting from the fact that the position of the tool cannot be measured underground, is that trajectory control is implemented on the surface imposing large communication delays on the control loop. This dissertation investigates methods to overcome these challenges and improve the accuracy and reliability of automated directional drilling. A Markov Chain Monte Carlo (MCMC) based method is proposed to estimate time-varying model parameters in real-time using only measurements commonly obtained while drilling. The method is evaluated on historical field data and in closed-loop simulation. Also, the utility of estimation for human-in-the-loop operation is explored through the design of an early warning system. In order to incorporate information of the model uncertainty into the control framework, a stochastic model predictive controller (SMPC) is designed. As part of this development, a series of approximations are made to a first-principles model of directional drilling that significantly reduces the computational complexity of the controller. The SMPC is posed as a convex second-order cone optimization problem that is solved efficiently and enables real-time directional control. In open-loop, the controller is optimal and satisfies state and input constraints probabilistically. In closed-loop, it satisfies input constraints, is approximately optimal, and has minimal violation of state constraints as demonstrated by simulation-based experiments. Lastly, to address the challenge of unobservable position downhole, a machine learning (ML) based estimation method is developed, and several algorithms are compared in terms of accuracy and computational demand, which is critical to practical implementation. The impact of estimation errors on control performance is analyzed, and the results are used as metrics in the evaluation of the models on historical drilling data.


LCSH Subject Headings