Multivariable Predictive Control

What are the advantages of using Multivariable Predictive Control technology over conventional regulatory schemes ?

The MVPC technology is more suited for processes involving many variables, multiple interactions and significant response delays between inputs and outputs. The MVPC controller incorporates both feedback and feedforward type control actions in controlling both the present and predicted trajectories of the controlled variables. Constraint pushing and cost optimization can also be easily added to the control matrix. While it is possible to configure several DCS blocks with feedforwards, ratios, etc., MVPC provides an elegant approach that can reduce lifetime maintenance costs and generate higher benefits. 

How does GMAXC compare with other Multivariable Predictive Controllers ?

The MVPC technology is over 20 years old. But, despite this long history and a wide base of implementation, MVPC technology typically takes about six months to implement and can cost several hundred thousand dollars per installation. With the recent advances in hardware and software, GMAXC offers the MVPC technology at a commodity level for rapid implementation by in-house or third party personnel. The main advantages are in the areas of model identification, DCS interface and overall ease of usage. In some cases, the GMAXC controller has been implemented in weeks, and these automated plants operate unmanned during week-ends and night shifts (see "Advance Process Control in Record Time", Shawn Wilhelm and David Seiver, CONTROL, May 1999).

The main features of GMAXC over other MVPCs are:

Script Add-on	VBA Script allowing adaptve and non-linear MVPC control
Process Database/ Historian	Access type relational database included as a part of the controller. Data can be used for performance monitoring reports
HYSYS Dynamic Simulator Interface - Virtual Plant Testing	Seamless (OLE Automation) interface with HYSYS Dynamic simulator for model identification, controller tuning and controller testing.  Plant testing and commissioning time reduced to a minimum
DCS Interface	OPC and DDE Interfaces available for most DCS Systems (ABB, Bailey, Foxboro, Siemens Moore, Yokogawa Centum).  Other available interfaces are with Wonderware and OSI Software PI database
Lab Analysis Integration	Time Delayed Lab analysis can be used to retroactively update the internal prediction models.
FDA Requirements	Extensively tested to meet US FDA requirements at multiple plant sites
Redundant/Backup Mode	Configuration for Stand-by mode on a back-up controller is included as an integral part of GMAXC

How long does it take to implement GMAXC/MVPC for a typical process unit ?

For a typical process unit, we can offer a complete online GMAXC/MVPC solution in about three months time, subject to plant testing availability for dynamic models identification.

What is Plant Testing and how long does it take ?

In plant testing, the independent process variables (e.g. feed rate, reboiler heat) are perturbed by about ± 3% to study their interactions with dependent variables (e.g., product qualities and tower temperatures). Typically, it takes about 12 hours for each set of 5 independent variables, and the total plant testing is expected to take about 24 to 36 hours.

Can plant testing be avoided ?

Yes, in some cases the historical data can be sufficient to identify dynamic models. Or, if an HYSYS dynamic simulation model is available, the GMAXC-HYSYS interface can be used to design and test a controller.

What are the minimum Hardware and Software investments ?

Assuming the availability of a DCS system, the hardware requirement is an average Pentium PC with Ethernet connectivity to the DCS system. The software requirements are GMAXC, OPC/DCS DDE Server and GMAXC_OPC/DDE (GMAXC OPC/DDE Client).

New Generation Intelligent Control Technology

What are the limitations of conventional Multivariable Predictive Controllers?

Typical Multivariable Predictive Controllers (MVPCs) are static in formulation:

Their models do not change as the process shifts to different operating modes. Even though all MVPCs use dynamic models, these dynamic models are usually based on the most prevalent operating mode. 

When the process mode changes significantly because of both planned and unplanned events, the dynamic models do not change to reflect the new dynamic behavior of the process. The tuning factors, such as response speed, prioritization between several control variables, move limits, etc. are also normally based on adjustments required during the normal operating mode.

This limits the conventional MVPCs from being able to properly handle large sudden disturbances or to control more than one mode of process operation optimally.

How do these limitations affect the control of Processes Changing Modes During Normal Operations ?

During the process mode change period, the existing dynamic models with their long range steady state predictions do not represent the present or the near future behavior of the process. Short-term behavior models, with different interaction gains between variables are required, along with changes in tuning to allow aggressive changes in manipulated variables. Also, when a mode change is completed, the new process mode will require its own set of dynamic models and tuning constants. But, the static formulation of conventional MVPCs does not allow this adaptability. And, in most cases, this results in below par control performance during the change-over period and in the new operating mode.

How would I-GMAXC overcome these limitations for Process Mode Changes ?

In I-GMAXC, the MVPC controller is reformulated at each execution. Hence, if a Mode Switch is detected, different tuning parameters and different dynamic models are used during this transient period. Also, during this period, more emphasis is given to short term recovery than to long range behavior (as the process eventually will settle to a different steady state). When the process stabilizes near the new operating mode, a new set of models and tuning constants are used.

What else can I-GMAXC do aside from reformulating the MVPC ?

Aside from MVPC control technology, I-GMAXC also includes sequence control, non-linear control and heuristic control technology capabilities. At every execution, the G2 intelligent system shell is used to generate an integrated control solution in terms of these four technologies for a most optimal solution to the existing process condition (which can vary from start-up to normal to shut-down operating modes).

Do I need a separate G2 license for each I-GMAXC Controller ?

No, multiple I-GMAXC controllers for different process units can run within a single G2 license. Similarly, other G2 based program can also share the same license provided all the applications are programmed in one single knowledge-base.

Reference:

"Intelligent Controller – Delayed Coker Control", Paper presented at the Chemical Engineering Expo, June 1998, Houston, Texas.