University of Khartoum

Faculty of Engineering
Chemical Engineering Department

Advanced Control Strategies

2 References

 D.R. Coughanowr & S.E. LeBlance, Process Systems Analysis and Control, 3rd
Edition McGraw-Hills,2009.
 Parbir K. Sarkar, Process Dynamics and Control, PHI Learning Private Limited,
2014.
3 Feedforward Control
 If a particular load disturbance occurs frequently in a
control process, the quality of control can often be
improved by the addition of feedforward control. Consider
the composition control system shown in figure; in which a
concentrated stream of control reagent containing water
and solute is used to control the concentration of the
stream leaving a three-tank system. The stream to be
processed passes through a preconditioning stirred tank
where composition fluctuations are smoothed out before
the outlet stream is mixed with control reagent. A three-
tank system has been chosen for ease of computation in a
numerical example that follows.
 In the conventional feedback control system shown in
figure, the measurement of composition in the third tank is
sent to a controller, which generates a signal that opens or
closes the control valve, which in turn supplies
concentrated reagent to the first tank.
4  In the block diagram of the process, the symbol for concentration is
denoted by C (capital letter) to denote a deviation variable.
 To obtain some specific control system responses, numerical values of the
time constants of the tanks have been chosen as shown in figure. To study
the response of this control system, the block diagram shown in figure was
simulated on a computer.
5

 The values of 𝐾𝑐 and 𝜏𝑖 were chosen by trial and error to give the response
to a step change in set point shown in curve (II) of figure; this response,
which has a decay ratio of about 1:4, was obtained 𝐾𝑐 = 2.84 and 𝜏𝑖 = 5.
The Ziegler-Nichols settings(𝐾𝑐 = 3.65 and 𝜏𝑖 = 3) give a set point response
shown as curve(I) of figure , which is too oscillatory.
6  Having obtained satisfactory settings for the controller (𝐾𝑐 = 2.84, 𝜏𝑖 = 5), we
obtained the response of the system to a step change in 𝐶𝑖 of 10 units,
shown as curve(I) in figure . Note that the response is oscillatory and has a
long tail. This response illustrates the fact that the feedback control system
does not begin to respond until the load disturbance has worked its way
through the forward loop and reaches the measuring element, with the
result that the composition can move far from the set point during the
transient.
 Responses to a step change in load for
feedforward-feedback control.
Curve I: PI control with 𝐾𝑐 = 2.84, 𝜏𝑖 = 5
−1
Curve II: FF control with 𝐾𝑐 = 2.84, 𝜏𝑖 = 5, 𝐺𝑓 = 5𝑠+1

Curve III: FF control with 𝐾𝑐 = 2.84, 𝜏𝑖 = 5, 𝐺𝑓 = −1

Curve IV: FF control with 𝐾𝑐 = 2.84, 𝜏𝑖 = 5, 𝐺𝑓 = −0.5
  If the change in load disturbance 𝐶𝑖 can be detected as soon as it occurs
7 in the inlet stream, this information can be fed forward to a second
controller that adjusts the control valve in such a way as to prevent any
change in the outlet composition from the set point. A controller that uses
information fed forward from the source of the load disturbance is called a
feedforward controller. The block diagram that includes the feedforward
controller 𝐺𝑓 as well as the feedback controller 𝐺𝑐 is shown in figure .
8 Analysis of Feedforward Control
 The response of C to changes in 𝐶𝑖 and R can be
written from figure as follows:
𝐶 𝑠 = 𝐺1 𝑠 𝐺𝑝 𝑠 𝐶𝑖 𝑠 + 𝐺𝑓 𝑠 𝐺𝑝 𝑠 𝐶𝑖 𝑠 + 𝐺𝑐 𝑠 𝐺𝑝 𝑠 𝐸 𝑠
Where 𝐸 𝑠 = 𝑅 𝑠 − 𝐶(𝑠). To determine the transfer
function of 𝐺𝑓 𝑠 that will prevent any change in the
control variable C from its set point R, which is 0, we
solve equation above for 𝐺𝑓 𝑠 with 𝐶 = 0 and 𝑅=0.
The result is:
−1
𝐺𝑓 𝑠 = −𝐺1 𝑠 =
5𝑠 + 1
This transfer function can be implemented easily with
existing control hardware.
 If the load response of the control system, with
−1
𝐺𝑓 𝑠 = 5𝑠+1, were obtained for a step change in 𝐶𝑖 ,
there would be no deviation of C from the set
point (i.e., perfect control). This response is shown
as curve (II), which of course is a horizontal line at 𝐶
= 0.
9 −1
 Rather than use the 𝐺𝑓 = 5𝑠+1 in the feedforward
controller, one can try using only the constant
term of 𝐺𝑓 ; that is, 𝐺𝑓 = −1.
 The response for (𝐺𝑓 = −1) gives curve (III); this
response has a very large undershoot before
the feedback controller returns C to the set
point. If we try using 𝐺𝑓 = −0.5, we obtain curve
(IV); the undershoot is less in this case, but the
response is still unsatisfactory. As shown by
curves (III) and (IV), omitting the dynamic part
of 𝐺𝑓 (𝑠) can give very poor results. The success
of using a feedforward controller depends on
accurate knowledge of the process model, a
luxury that may not be available in many
applications.
10 Tuning Rules for Feedforward
Feedback Control
 In the practical application of feedforward control, one does not have a block
diagram with transfer functions; by introducing a step change in the disturbance
that enters the feedforward controller (𝐶𝑖 in figure ) and then applying some
tuning rules. The rules to be discussed here are from a training film on
feedforward control produced by Foxboro Co. (1978).
11 Feedforward Rules
 In describing these rules, reference will be made to the general block diagram
for a feedforward-feedback system shown in figure . It is assumed that 𝐺𝑓 𝑠 will
be a lead-lag transfer function of the form:
𝐾𝑓 (𝑇1 𝑠 + 1)
𝐺𝑓 𝑠 = −𝐺1 𝑠 =
𝑇2 𝑠 + 1
Where 𝐾𝑓 steady-state gain of feedforward controller
𝑇1, 𝑇2 time constants of dynamic part of feedforward controller.
 Commercial microprocessor-based controllers provide this lead-lag transfer
function.
  The tuning rules listed below are explained with
the help of figure. In which, a unit step is
selected for the disturbance 𝐶𝑖 , and 𝐾𝑓 has been
12 taken as 1. In practice, 𝐾𝑓 will, of course,
depend on the particular process being
controlled.
1. Remove the control action in 𝐺𝑐 (𝑠) by setting the
controller to manual.
2. Set the feedforward controller to the computed
steady-state gain (𝐾𝑓 ) necessary to compensate
ultimately for a step change in 𝐶𝑖 . This means
that the dynamic portion of 𝐺𝑓 (𝑠) will be
removed, and only the constant term (𝐾𝑓 ) will
remain.
3. Make a step change in 𝐶𝑖 and observe the
open-loop transient of 𝐶. The general shapes of
the response to be expected are shown in
figure.
4. If the response shown in figure(a) occurs, lead
must predominate in 𝐺𝑓 (𝑠) of equation above
(i.e., 𝑇1 > 𝑇2). If the response of figure(b) occurs,
lag must predominate in 𝐺𝑓 (𝑠) (i.e., 𝑇1 < 𝑇2). The
values of 𝑇1 and 𝑇2 are found by use of the
information in Table below . The value of 𝐾𝑓 has
been obtained in step 2.
13
14 Example
 Apply the feedforward tuning rules to the system in figure. Since this
example is concerned with the application of the tuning rules to a system for
which a mathematical model is not generally available, assume that the
transfer functions for 𝐺1 𝑠 and 𝐺𝑝 𝑠 are unknown. The determination of
𝐺𝑓 𝑠 is to be obtained solely by information from open-loop transients. We
must first determine the steady-state gain 𝐾𝑓 for the system.
 If a step change in 𝐶𝑖 is made, 𝐶 will
undergo a transient and eventually level
out at a steady-state value. If the
controller parameters are properly
selected, the value of 𝐶 at the end of the
transient will be the same as it was before
the transient occurred. By computation or
experiment, one can determine the value
of 𝐾𝑓 needed to obtain no change in 𝐶.
For the system in figure, one can see that
𝐾𝑓 must be equal to -1.
15
 We must now apply the feedforward tuning rules to obtain 𝑇1 and 𝑇2 . After
removing the feedback controller action [𝐺𝑐 𝑠 ], we have the equivalent
diagram shown in figure. A unit-step change in 𝐶𝑖 produces the transient for
𝐶 shown as curve (I) in figure below . Comparing the 2 shape of the
transient with those of figure above, we see that lead must predominate in
𝐺𝑓 𝑠 . The peak value occurs at 𝑡𝑝 = 2. Applying the rules in gives:
𝑇1 = 1.5 𝑡𝑝 = 3
𝑇2 = 0.7 𝑡𝑝 = 1.4
The feedforward controller transfer function is therefore:
− (3𝑠 + 1)
𝐺𝑓 𝑠 =
1.4𝑠 + 1
16

 It is of interest to show the response of C for feedforward only when the

− (3𝑠+1)
feedforward transfer function [𝐺𝑓 𝑠 = 1.4𝑠+1 ] is used. MATLAB and Simulink
can be used to simulate the result for a unit-step change in 𝐶𝑖 . It is shown as
curve (II).
17

 Comparison of conventional feedback control with feedforward-feedback

control for Example.
Curve I: PI control with 𝐾𝑐 = 2.84, 𝜏𝑖 = 5
− (3𝑠+1)
Curve II: Feedforward-feedback control with 𝐾𝑐 = 2.84, 𝜏𝑖 = 5 and 𝐺𝑓 𝑠 = 1.4𝑠+1