Feeding on Feedforward

This Month’s Topic—Feedforward Control

By Greg Mcmillan and Stan Weiner, PE

Stan: This month’s topic—feedforward control.

Greg: I have so much feedforward I eat before I am hungry.

Stan: Maybe engineers are into preemptive action—always the first to the cafeteria. But seriously, timing is often overlooked. A feedforward action that arrives too soon causes an inverse response and an initial reaction of the control loop in the opposite direction of desired compensation.  Consider the example of a feed-rate compensation feedforward for the temperature control of a distillation column, where the controlled temperature is on a tray near the bottom of the column. An increase in the feed flow should increase the steam flow to the reboiler. The corrective action of an increased boil-up may reach the temperature sensor before the feed flow disturbance and cause the bottom temperature to initially increase. This causes the controller to back off the steam, taking out the desired correction.

Greg: Fortunately, the slow and gradual response of a distillation column from the large number of interacting lags offers a big time interval where feedforward signals arriving sooner or later still offer significant benefits. For faster processes, the timing gets more critical to the point where feedforward can do more harm than good. For liquid pressure control, most feedforward signals are old news. On the other hand, the large process time constant for gas pressure control of large volumes provides a nice timing window for feedforward control. The number and size of process lags and delays for the manipulated and disturbance variables are often different. Dynamic compensation of the feedforward is used to get the timing right. Delays and lags are added to feedforward signals to prevent the feedforward from arriving too soon. A lead-lag is added to compensate for a lag in the manipulated variable’s (MV) path that doesn’t exist in disturbance’s path to the control point in the process. In this case, the lead of the lead-lag is set equal to the additional MV lag. The lag of the lead-lag is set to reduce the amplification of feedforward measurement noise by the lead. If there is an additional delay in the manipulated variable path, there is no compensation for this deficiency in feedforward action.

Stan: Feedforward is predicated on the availability of a representative measurement of a prevalent disturbance. Consequently, flow feedforward is the most common type. In this case, what we really have is flow ratio control where the manipulated flow is ratioed to a feed flow to a unit operation. The Coriolis flow meter provides the most accurate flow measurements for ratio control and may provide a valuable correction based on an inference of stream composition from the meter’s density measurement.

Greg: Flow feedforward works much better if there is a cascade loop where the primary loop (e.g. temperature, composition, pH or overhead pressure) corrects a remote cascade (RCAS) setpoint to a secondary flow controller for the manipulated flow. The setpoint is the process feed flow multiplied by a ratio factor. The ratio factor needs to be readily visible and adjustable by the operator via a faceplate displaying desired ratio (SP) and the actual ratio (PV) as corrected by the primary loop. Ideally, the feedforward correction should be a multiplier, since it is the slope (ratio) on a plot of the manipulated flow versus the process feed flow. However, due to scaling issues and bias errors, the feedback correction is usually added to the computed manipulated flow (a feedforward summer instead of a feedforward multiplier is used).   

Stan: Without the secondary flow loop, the feedforward gain would have to deal with the nonlinearity of the installed characteristic of the control valve—not a pretty sight.

Greg: The improvement in the loop performance of the feedforward control ultimately hinges on how accurately the feedforward measurement represents the disturbance. For flow disturbances, the accuracy of the feed rate measurement is obviously important, but so is the accuracy of the computed ratio factor. In most cases, the ratio factor depends upon other process variables and parameters. For a heat exchanger, the ratio factor depends upon the temperatures and specific heats of the coolant and process streams. (See equations 2-45 and 2-46 in Advanced Control Unleashed, ISA, 2003.)

For flow feedforward control of column and reactor temperature, the ratio also depends upon the composition of the feed. For reactors, a secondary loop of jacket temperature is used to remove the effect of coolant unknowns (e.g., disturbances). For neutralizers, the acid and base composition of the feed is critical. When the influent is on the flat part, and the set point is on the steep part of the titration curve, small errors in the influent pH measurement cause gross errors in the feedforward. Also, the response and reliability of the pH electrode for the higher acid and base concentrations of the influent is often unacceptable. Consequently pH feedforward is rarely used for an influent pH on the extremes of the titration curve (see page 25 of Advanced pH Measurement and Control, 3rd Edition, ISA, 2005).

Stan: A valuable technique is to plot the process variable for the primary loop versus the ratio of the manipulated flow to the feed flow for a given set of process variables and parameters. The slope at the operating point (primary loop set point) is the process gain. The family of curves shows how the process gain and flow ratio change as the process changes.

Greg: An adapted-velocity, limited feedforward has been used to provide a smoother batch-to-continuous flow transition for level control in a surge tank. The velocity limit is computed based on the time-to-violation of a level constraint and the imbalance between a filtered batch inlet flow and manipulated outlet flow for the surge tank that is the feed to continuous operations downstream. The filter time for the batch flow used in the adaptation is set equal to the batch cycle time. The filter time for the flow feedforward signal itself is large enough to reduce flow measurement noise (See Appendix B in Advanced Control Unleashed, ISA, 2003). 

Stan: Feedforward control has been used effectively even when a direct measurement of flow was not possible. For example, feeder speed and amps have been used for calciner, extruder, and dryer temperature control. For more info on feedforward control, checkout Greg’s Sept. 1-24, 2007, entries at  www.modelingandcontrol.com/continuous_control.

Greg: Retirement has its process control opportunities too. Top Five Opportunities for Retired Automation Professionals

(5) Automating lawn care
(4) Regulating swimming pool temperature and pH
(3) Avoiding weather disturbances
(2) Optimizing golf ball trajectories
(1) Scheduling grandchildren visits.

This month’s puzzler

When Feeding the Forwards Is a Problem

Puzzler: When does a steam rate feedforward for three-element boiler drum level control cause problems?

Send an e-mail with your answer to the Puzzler, CONTROL questions, or comments to controltalk@putman.net.

Show Comments
Hide Comments

Join the discussion

We welcome your thoughtful comments.
All comments will display your user name.

Want to participate in the discussion?

Register for free

Log in for complete access.


No one has commented on this page yet.

RSS feed for comments on this page | RSS feed for all comments