Adaptive estimation of downhole pressure for Managed Pressure Drilling operations

Abstract

The objective of Managed Pressure Drilling (MPD) is to enable precise control of the downhole pressure during drilling and completion operations. An essential part of an automated MPD control system is the downhole pressure estimator, usually referred to as the hydraulic model, which in most cases is the limiting factor for achievable accuracy of the system. In this paper we present a new design of an adaptive observer for online estimation of the downhole pressure in the well. Unlike conventional hydraulic models, which are based on complex partial differential equations, the adaptive observer is based on a simplified lumped model of ordinary differential equations, combined with online adaptation that tunes uncertain model parameters utilizing topside measurements. We demonstrate that the novell adaptive observer design captures the important hydraulics of MPD operations with high accuracy, and fundamentally improves the convergence rate of the parameter estimaties by using delayed observers. The results are demonstrated on field data from an offshore MPD operation.

Challenges of Modeling Drilling Systems For the Purposes of Automation and Control

“…Managed pressure drilling (MPD) is providing a rich source of problems and challenges for the control systems community, Breyholtz (2010), Fredericks (2008), Geehan (2010) and Stamnes (2011). The base-line objective is to control the bottom-hole pressure (BHP) in the face of all Copyright held by the International Federation of Automatic Control disturbances.…”

Adaptive approximation-based estimation of downhole pressure in managed pressure drilling

“…The downhole pressure is usually measured, but with conventional equipment this measurement has low bandwidth, is delayed, and is unreliable. Good pressure control therefore requires pressure estimation using also topside measurements [7], [3].…”

“…These measurements typically include measurements of the flow into, and possibly out of, the well, so it is not an unreasonable assumption to make. The reason for making this assumption here is that it simplifies analysis considerably (compare the work in [7], which estimates friction not using a multi-model approach, but where the analysis does not rely on knowledge of flow), and it facilitates PE-insights in the multi-model case. It is the authors' opinion that the PE analysis of the approach considered herein can give insight into PE requirements of other schemes for friction estimation in drilling, also schemes using methods not based on 'local learning'.…”