Automatic control becomes increasingly important to assist drilling operations. Basically, the first stage of designing an automatic control for a simple system is by selecting a variable to be manipulated and a variable to be controlled in the system. A drilling rig has a large number of variables. This makes the design of automatic control more complex due to the number of variables. Here, we discuss three ways of dealing with ‘multivariable control’ design and evaluate them by simulation studies.
The first approach is by designing a control system to coordinate all different variables in a centralised way. The second approach is by dividing the system into subsystems, each of which has fewer variables. A control system can be designed independently for each subsystem provided that the interactions between different subsystems are small. The third one is combination between the first and the second approach.
The three different approaches of design are compared for coordination of variables in MPD. When drilling with narrow pressure margins, back pressure MPD is often used with the choke being manipulated to control the BHP. It is of interest to evaluate the possibility of having additional variables like pump rate and back pressure pump rate as manipulated variables. In particular, for the design using centralised approach we consider the model predictive control (MPC) while for the decentralised approach we consider a class of PI control. For the third approach we design internal model control (IMC). We focus on drill pipe connections where we need to control the BHP and the gel breaking simultaneously.
The design of MPC depends on a predictive model and the tuning of its parameters which can be challenging. However, MPC has an attractive feature where it is easy to include constraints in the computation. IMC has the advantage of simplicity and ease of tuning compared to MPC. The drawback of IMC is that we need to compute an inverse of a model representing our system which can be cumbersome when the dimension of multivariables increases. While the design of PI control is the simplest one, we have to observe that the interaction between the subsystems should not be critical for a certain period of time during drilling operations. The results improve our understanding of the possibilities and challenges of introducing automatic control to drilling operations.