Applying Offline Tools to Reduce On-Site Control Implementation Effort

Process control engineers are continuously challenged to reduce the effort to commission new advanced regulatory control strategies and/or linear/nonlinear multivariable control applications. Reducing on-site controller identification and tuning and the associated "interference" with process operations is especially desired – provided the commissioned application(s) meet(s) performance expectations. This paper will present a refining practitioner's perspective on applying tools to address this challenge.

For multivariable control applications on a number of refining units, it is becoming established practice to run automated closed-loop step testing tools to hone in on the control models that will enable good performance for the tuned/commissioned controller. However, this closed-loop step testing approach requires satisfactory “seed models” to enable control to be maintained during the step testing. For advanced regulatory control, similarly knowledge of approximate relevant process dynamics (effectively also “seed models”) – coupled with an appropriate tuning philosophy – enables good initial estimates for PID and other tuning parameters. As such the focus of this paper is on capturing seed models that can be used for either of these purposes. Approaches, referred to here as “Offline Tools”, that enable reduction of actual plant testing are reviewed.

A general resource in all instances is that (large) engineering organizations have the capacity to maintain, explicitly or implicitly, relevant data libraries – seed models for multivariable controllers or tuning parameters, possibly with associated identified process dynamics, for advanced regulatory control strategies.

For greenfield or significantly modified brownfield sites, operating data may not be available initially. However, operator training (or other (dynamic) process engineering) simulators may be available and their use for generating seed models and/or initial tuning for advanced regulatory control strategies will be reviewed. Process engineering model assumptions that impact the use of the models for these purposes will be highlighted. Of note is that with appropriate dynamic process engineering models, controller performance across a wide range of operations can conceptually be evaluated relatively quickly.

For actual operating plants, a range of operating data exists from the descriptive provided by operations and engineering staff through the very granular stored in historians. Steady state and/or dynamic models may also exist and the operating data can be used to identify the usefulness of such models, potentially initiating a discussion on making such models “current”. A range of approaches (“offline tools”) to utilize the available offline data to estimate seed models exists; it will be discussed that the optimal approach is typically some combination of these tools selected for the particular circumstances.