Feedforward can make disturbances worse unless intelligence corrects the intensity and timing of the feedforward signal for unknowns. It would be nice if the feedforward signal was the whole story. Unfortunately there are many unknowns in terms of disturbances, dynamics, and errors.
An unknown disturbance can be driving the process variable in the opposite direction of the measured disturbance. The use of feedforward control makes the upset worse because the feedforward correction is in the same direction as the unknown disturbance. The most common feedforward measurement is flow and the least common is concentration. If there is a coincident decrease in feed concentration with an increase in feed flow, the impact of the feedforward control should be decreased. For example, if the light key component concentration or waste acid concentration in the feed to a column or neutralizer decreased when the flow increased, the flow feedforward for temperature or pH control should be reduced, suspended, or even reversed. If the effect of the concentration change is greater than the effect of the feed change, the feedforward correction has the wrong sign. There are many unknown disturbances that may be driving the process variable (PV) away from setpoint when a measured disturbance occurs. The change in the process variable “delta PV” between a new and an old PV created by a deadtime block whose deadtime is set equal to the total loop deadtime can be used to make the feedforward smarter. If the “delta PV” is divided by the process gain and a portion of this correction added to the feedforward signal with dynamic compensation, and the proper feedforward action applied (direct or reverse), the feedforward signal will be automatically adjusted to account for how much the PV is being driven by the unknown upset. The use of a deadtime block in the generation of the “delta PV” is a realization that feedback correction can not occur sooner than one total loop deadtime.
An unknown parameter in the charge, component, or energy balance or an unknown offset in the feedforward measurement can result in a bias error in the feedforward signal. If a feedforward summer is used, the unknown shows up as a sustained feedback controller output correction. An integral only controller whose PV is the process PID feedback correction, whose scale is -100% to +100% correction, and whose setpoint is 0% correction can be used to adapt the feedforward signal. The output of the integral only controller slowly adjusts an input bias to the feedforward signal before being multiplied by the feedforward gain until the feedback correction by the process PID is zero.
In Deminar #11 we saw how important it was to get the feedforward delay correct.
A delay in the feedforward path shorter than the delay in the disturbance path caused inverse response. A simple online identification of the feedforward and disturbance delays can provide an automatic addition of a delay to the feedforward signal to make the total delays equal and the arrival of the feedforward correction coincident with the upset.
Finally, the “near integrator” gain in Deminar #6 identified as the maximum ramp rate of the PV in the first 4 deadtime intervals divided by the change that initiated the change can be used to automatically correct the feedforward gain or the ratio for flow feedforward for all types of processes. The feedforward gain is the ratio of the “near integrator” gain for the disturbance with the feedforward turned off divided by the “near integrator” gain for a setpoint change. The feedforward gain is the change in PID output per change in feedforward input with proper engineering units. Thus, unlike the “near integrator” gain used for the short cut controller tuning method, the primary PID output used in these calculations is in engineering units of the setpoint of the secondary PID rather than in percent.
For more information on smart and adaptive feedward control checkout the September 2007 entries on this website Feedforward Techniques