4976
Bear Mountain Drive Evergreen,
CO 80439 office: 303-670-8392 cell: 303-670-9092 e-mail:
bob@procontrol.net
ProControl,
Inc.
Process Control Education and Technology
Robert V. Bartman, Ph.D., President
 |
Dynamic Reconciliation’s Attributes –
and DR’s Inclusion in both Discover
and the ‘Revelations in Control’ Course
ProControl's Dynamic Reconciliation (DR) Model-Based Control technology offers many advantages. Its inventor, Dr. Bob Bartman, has implemented DR in analyzer, temperature, and difficult level problems requiring feedback, feedforward, constraint, and multivariable control, and plantwide online optimization, on Light-Ends towers, Atmospheric and Vacuum Primary Fractionators and Furnaces, FCC's, Lubes units, Solvents plants, Hydrogen plants, Visbreakers, Blending, MEA units, Hydrofiners, Hydrocrackers, and Chemicals Plants' Reactors. In all such cases, the following DR capabilities and benefits are important:
· Stability in the face of imperfect secondary control (such as analyzers outputting to tray temperature setpoints, where the underlying temperature PV cannot be perfectly held at its setpoint). DR holds secondary setpoints where they should be, and doesn't become confused by variations, or even cycling, arising from imperfection below. Our model-based system allows lower-level loops to solve their own problems, without cycling their setpoints. This contrasts with other available schemes, which can exhibit serious problems if their underlying loops become (for any reason) either mistuned, or impossible to control perfectly due to valve stick.
All the Cascade Control problems identified in ProControl's 'Revelations on Dynamic Process Analysis, Advanced Control, and Online Optimization' Course, for example, completely disappear when DR replaces PID as a cascade control primary. (This assumes a good model between the primary and secondary PV’s.)
· Constraint or Multivariable Control, when signals are not all available at the same frequency. For example, a discontinuous GC signal may be all that's available for one key constraint, or for one dependent variable in a multivariable problem. DR control schemes have no problem combining controller-input signals updated over a wide range of time intervals.
[On the Multivariable topic, note that nearly all Multi-Sidestream Fractionator quality control problems are addressable by DR only, and do not require LP, or any other matrix-based system such as DMC (tm), as a multivariable decoupling aid; this is due to the simple directional manner in which quality interactions propagate in fractionators (one way for partial drawoff configurations such as crude atmospheric towers; the other for total drawoff configurations typified by crude vacuum towers). DR by itself automatically, and properly, handles such quality interactions. You might not want to replace DMC by something simpler on such fractionators – but you could.]
· Precise delineation of Manipulated Variable constraint limits in multivariable applications: a subtle but key need, often neglected, in real-world problems.
· Automated validation and model update from lab data, a capability inherently foreign to many other model-based tools, but simple for DR. Most DR analyzer loops implemented by ProControl have included cross-validation with lab data, and automated fallback to validated lab data usage (via DR model) in the event of analyzer failure.
Many critical quality control loops have relied solely on lab data; a gigantic (and highly profitable) 55-manipulated variable closed-loop blend optimization program implemented by the Instructor in a major European refinery would have been useless if DR techniques had not been used for both lab data validation and control.
· Built-in feedforward control to (multiple) measured load disturbances. A DR controller doesn't have an overreaction problem, as would a PID-based feedforward scheme, if relative deadtimes are in the "unlucky" direction and perfect feedforward cancellation is impossible -- or if cancellation is impractical due to grossly unequal lagtimes, or to pathological feedback dynamics such as inverse response. The DR controller simply holds the manipulated variable setpoint where it needs to be -- in contrast with a PID. DR feedforward is achieved with very little additional overhead, either in code or comprehension, vs. a classical PID-based solution.
DR feedforward implementation times are typically much less than those for feedforward done via other, more complex multivariable products.
Feedforward should be common practice, since it yields the stabilization platform needed for advanced online optimization. DR provides a precise "home" for the relevant process gains, deadtimes and lagtimes in a feedforward problem; there's little structural complexity, and nothing more to tune.
ProControl added DR FeedForward Evaluation Studies, plus key DR FF implementation details, to Discover in 2011.
· Ability to move a manipulated variable quickly to where it should be following either load upsets or setpoint changes, and to leave it there despite long process deadtimes or lagtimes. While this is a classic model-based controller motivation, we believe that DR is more easily implemented and understood than are other predictive controllers commercially available. The better our process understanding, of course, the better will be DR's model-based controller performance - a fact of life for any type of model-based controller.
· Ability to survive model error, i.e. changing process conditions, with robustness similar to that of a PID tuned responsively to combat unmeasured load disturbances. Most model-based control vendors say very little about control scheme vulnerability here, yet it's very crucial.
A real example: An ethane analyzer-to-ethane analyzer-to-temperature-to-reboiler duty cascade in a Light Ends train, where all but the duty controller was a DR model-based controller, and the C2-to-C2 and the C2-to-temperature model knowledge -- especially deadtimes and process gains -- is not perfect. If we choose to accept this uncertainty in process knowledge (see below for a better answer!), DR model-based controllers can adequately survive and still outperform PID's. Model uncertainties in this example arise from variable loading on trays caused by feed changes, from deadtime and lagtime variations due to feed rate changes, and from feed composition shifts (usually unmeasurable).
ProControl's courses, backed by our tools, stress that we can and should acquire as much process understanding as needed; if we choose to understand how feedrate affects gains and deadtimes in the above case, we can utilize DR's next attractive feature:
· Ability to easily update key model parameters as a function of measurable process conditions such as feedrates, feed types, and product specification levels. As another real-world example, at one refinery DR models for both Crude Unit sidestream yields and qualities were updated as a nonlinear function of crude feed quality, with crude quality itself inferred from the mix percentages from six crude tanks of variable (but tracked) composition. This provided dynamic DR feedforward control of key Atmospheric Tower qualities (relevant to plantwide optimization) upon crude composition swings, and allowed a plantwide optimization system to remain in closed-loop control even through quite severe feed composition swings. It also facilitated easy "what if" steady-state case studies by the Refinery Scheduler, who was interested in knowing impacts of crude changes on blend production rates and/or intermediate tankage inventories.
The point: DR models are simple enough to be easily updated dynamically, whether for feedforward control as above on a fast frequency, or simply to (occasionally) improve model adequacy and feedback stability. This ease of drawing on other known relationships to update key control model parameters yields a major advantage for DR over more complex matrix-based model structures.
· Closed-loop Stochastic Adaptive Control, where process parameter validation is essential -- i.e., where "fit-anything" black-box models cannot be safely used.
In the world of true online, statistical, model adaptation -- closed loop, with no human intervention -- DR is an excellent, proven choice. Vendor black-box model structures, which can literally "fit anything", are incompatible with true closed-loop adaptive control; statistical parameter updates in a black-box model should not be accepted without human analysis and approval -- itself often difficult, requiring manual parameter smoothing to achieve DR's inherent lag-based response pattern. ProControl markets unique technology for this updating need -- if you have it. But we suggest you may not need it, despite significant process variations, if you follow the deterministic adaptation strategy above -- as covered in our Revelations class.
· Offline use of online-updated process models; e.g., for optimization , and/or process performance followup.
DR is of great use in helping non-control personnel understand what to expect from control systems -- especially from optimization schemes which require inputs from non-control folks such as Planners, Schedulers, and Coordinators.
Dynamic DR models reduce easily to steady-state models useful for process unit followup -- e.g. on important sensitivities of yields to flows (in our jargon, "gains"). DR facilitates troubleshooting process problems by non-control personnel (e.g., Process Engineers) using dynamic plant data -- a practical capability which some other vendors' model-based tools will never be accused of having. DR also facilitates either online or offline process model updating for use in optimizers -- which most process plants today are considering implementing. Without periodic model update, of course, use of such optimizers can be actually counterproductive.
· Predictive insight into future conditions: sometimes important for Process Operators, as in a real-world tower example covered in class.
· High service factors and maintainability: DR's major tuning entries are those defining the process itself. Associated service factors, Operator acceptance, and applications credits can be confirmed from our many references.
In summary, DR is the simplest solution to a wide variety of real-world control problems. It works well because Control Engineers understand it -- they can clearly relate to deadtimes, lagtimes and gains, and have no hesitation implementing it, or maintaining it afterward. (Both issues can be in some contrast with other model-based solutions!)
Both DR’s technology and tuning are now offered without additional cost in ProControl’s ‘Discover’ software. DR’s motivations are covered by case studies in our Revelations Course.
(c) ProControl, Inc.
Questions? E-mail: bob@procontrol.net