CL 329: Process Control

Week 12
Sayantan Dutta

Feedforward controller
 Learning Objectives

• Design of feedforward controller

• Studying the effect of design error in feedforward controller.
• Response of feedforward+feedback controller
 A feedback control block diagram

Disturbance Unmeasured
Fout variable Output
Tout

Fh Fc Process
Th Tc

Manipulated
variable Measured
Output

Controller

control loop
 Response of a feedback control system

P-only controller PI controller PI controller

Smaller τl higher τl

t t

Feedback control is characterized by delay, offset, and overshoot.

 Direct synthesis method
Gd GpGcGf
ȳ(s) = ¯
d(s) + ȳsp(s)
1 + GpGcGf Gm 1 + GpGcGf Gm

Can we design an ideal feedback controller ? (i.e. ȳ(s) = ȳsp(s))

kp
Gp = Gf = Gm = 1
τps + 1
GpGc
=1 Gc = ∞
1 + GpGc

It’s physically (mathematically) impossible to design a feedback controller without any delay.
 Issues with feedback control

• There is always a finite delay period, when the controlled

variable is away from it’s set-point.

• There are offset (at the end), overshoot (on the way), or
oscillatory responses.

• One needs to be careful about stability while designing the

controller.
 A feedforward control block diagram

Fout Fc, Th, Tc

Tout Disturbance
Fh Fc variable
Th Tc Fout Tout
Unmeasured
Controller Process Output
Fh
Manipulated
variable

No delay in achieving control objective.

 Feedback Control Block Diagram

Closed-loop transfer
function
Gd GpGcGf
ȳ(s) = ¯
d(s) + ȳsp(s)
1 + GpGcGf Gm 1 + GpGcGf Gm
Let’s construct feedforward control block diagram

Let’s start with the process block.

What else do we have in common ?
 Let’s construct feedforward control block diagram

We still control the MV with a final control element.

Control action is not based on output any more. It’s based on disturbance.
Let’s construct feedforward control block diagram

Gc = ?
 Let’s construct feedforward control block diagram

We want no deviation. (i.e. ȳ = 0)

Gd Gd
¯
0 = Gdd + Gpm̄ ¯
m̄ = − dGmGcGf m̄ = − d¯ Gc =
Gp GmGpGf
How do we change the set point ?
 Let’s construct feedforward control block diagram

We want the controlled variable to it’s set point (i.e. ȳ = ȳsp)

¯
−Gdd + ȳsp
¯
ȳsp = Gdd + Gpm̄ m̄ = − ¯ m + ȳspGsp)GcGf
m̄ = (− dG
Gp Gm
Gsp =
Gd
Let’s ignore the instrumentation dynamics for the sake of algebra

−Gdd¯ + ȳsp Gd 1
ȳsp = Gdd¯ + Gpm̄ m̄ = − Gc = Gsp =
Gp Gp Gd
 Let’s take an example of first order process

Kd
Gd =
τs + 1
Kp
Gp =
τs + 1

ȳsp = Gdd¯ + Gpm̄

−Gdd¯ + ȳsp
m̄ = −
Gp
Gd 1
Gc = Gsp =
Gp Gd
 Let’s take an example of first order process

Kd
Gd =
τs + 1
Kp
Gp =
τs + 1

Gd kd (τs + 1)
Gc = = Gsp =
Gp kp kd

ȳsp = Gdd¯ + Gpm̄ gain only What does it mean ?

−Gdd¯ + ȳsp
m̄ = −
Gp
Gd 1
Gc = Gsp =
Gp Gd
 Let’s take an example of first order process

kd
Gd =
τs + 1
kp
Gp =
τs + 1

Gd kd (τs + 1)
Gc = = Gsp =
Gp kp kd

ȳsp = Gdd¯ + Gpm̄ gain only What does it mean ?

−Gdd¯ + ȳsp 1 d
[ȳsp + ȳsp]
m̄ = − kd dt
Gp
Gd 1 This is realizable.
Gc = Gsp =
Gp Gd
 Let’s take an example of first order process

kd
Gd =
τs + 1
kp
Gp =
τs + 1

Gd kd (τs + 1)
Gc = = Gsp =
Gp kp kd

ȳsp = Gdd¯ + Gpm̄ gain only What does it mean ?

−Gdd¯ + ȳsp 1 d
[ȳsp + ȳsp]
m̄ = − kd dt
Gp
Gd 1 This is realizable.
Gc = Gsp =
Gp Gd
What happens to a step input in disturbance ?

kd
Gd =
τs + 1 −Gdd¯ + ȳsp
kp m̄ = −
Gp
Gp =
τs + 1

Gd kd (τs + 1)
Gc = = Gsp =
Gp kp kd
 What happens to a step change in disturbance ?

kd
Gd = ȳ = Gdd¯ + Gpm̄
τs + 1
kp −Gdd¯ + ȳsp
Gp = m̄ = −
τs + 1 Gp

Gd kd (τs + 1)
Gc = = Gsp =
Gp kp kd

1 −kd −kd
d¯ = m̄ = −
kps
m(t) = −
kp
H(t) ȳ = 0
s

The manipulated variable instantly get adjusted for the disturbance

No deviation in output.
What happens to a step change in set point ?

kd
Gd =
τs + 1 −Gdd¯ + ȳsp
kp m̄ = −
Gp = Gp
τs + 1

Gd kd (τs + 1)
Gc = = Gsp =
Gp kp kd
 What happens to a step change in set point ?

kd
Gd =
τs + 1
kp −Gdd¯ + ȳsp
Gp = m̄ = −
τs + 1 Gp

Gd kd (τs + 1)
Gc = = Gsp =
Gp kp kd

1 τs + 1 1
ȳsp = m̄ = m(t) = [τ + δ(t)] ȳ = ȳsp
s kp
kps
The manipulated variable hits infinity mathematically
However, the controller is implemented numerically. So in reality it’s not infinity.
 What happens to a step change in set point ?
1 kp
m(t) = [τ + δ(t)] Gp =
kp τs + 1

1 τs + 1
ȳsp = m̄ =
s kps

What does this control action actually mean ?

 What happens to a step change in set point ?
1 kp
m(t) = [τ + δ(t)] Gp =
kp τs + 1

y(t) τ
1
δ(t)
m(t) kp
H(t) + kp

t t

1 τs + 1
ȳsp = m̄ =
s kps
The step input involves a delay for the output to reach steady state.
The transient due to the δ input compensate during the delay and eventually dies.
 There is nothing called a “free lunch”!!
Everything is not so perfect with the feedforward controller.
 What happens to a FOPTD process ?

Kd
Gd = exp(−0.1s)
τs + 1
Kp
Gp = exp(−0.5s)
τs + 1

ȳsp = Gdd¯ + Gpm̄

−Gdd¯ + ȳsp
m̄ = −
Gp
Gd 1
Gc = Gsp =
Gp Gd
 What happens to a FOPTD process ?

Kd
Gd = exp(−0.1s)
τs + 1
Kp
Gp = exp(−0.5s)
τs + 1
Gd kd
Gc = = exp(0.4s)
Gp kp
kd
m(t) = − d(t + 0.4)
kp
ȳsp = Gdd¯ + Gpm̄
The manipulated variable needs information for
−Gdd¯ + ȳsp
m̄ = − distrurbance variable from future.
Gp
Gd 1 Physically not realizable.
Gc = Gsp =
Gp Gd
 Feedforward controller can’t take care of unmeasured disturbances

G′d d′(s)

Unmeasured disturbance remains unattended.



What happens if there is an error in design ?
 Coding assignment

• No submission so far!!! Do you need more time or are you stuck ?

 Coding assignment

• No submission so far!!! Do you just need more time or are you stuck ?

• 3 options!! Let me know the most popular one by 5 pm today through your class
representative.
1. No extension!! We will all submit by tonight.
2. 24 hrs flat extension. It will be due on 22/10 11.59 pm.
3. 72 hrs extension. It will be due on 24/10 11.59 pm. I will give
you an outline of solution tomorrow in class. However, the
groups who submit before class tomorrow (22/10 11.30 am)
will get 2.5 marks bonus point.
 Logistics of course.

• No class on 4/11 Quiz on 7th November, 2024 at LC 102 sharp at 8:35. Question(s)
will be mostly on feedback control (topics from week 7-till last week).

• Problem solving/Doubt session on 5/11 in class at ESE-102 (usual venue) and 8/11
at Room 119. Let me know in advance what you like.

• Please check your ASC attendance reflects the Moodle attendance properly in case
you are in borderline of 23. Send an email with ASC screenshot by 30/10.
Lecture 2
 Recap: issues with feedback control

• There is always a finite delay period, when the controlled

variable is away from it’s set-point.

• There are offset (at the end), overshoot (on the way), or
oscillatory responses.

• One needs to be careful about stability while designing the

controller.
 Recap: a feedforward control block diagram

Fout Fc, Th, Tc

Tout Disturbance
Fh Fc variable
Th Tc Fout Tout
Unmeasured
Controller Process Output
Fh
Manipulated
variable

No delay in achieving control objective.

 Recap: Feedforward controller design

−Gdd¯ + ȳsp Gd 1
ȳ = ȳsp = Gdd¯ + Gpm̄ m̄ = − Gc = Gsp =
Gp Gp Gd
 Effect of design error

design real
Gd ≠ Gd
−Gdd¯ + ȳsp
ȳsp = Gdd¯ + Gpm̄ m̄ = − design
Gp ≠ real
Gp
Gp
 Effect of design error

design real
Gd ≠ Gd
design real
Gp ≠ Gp

design ¯
−Gd d + ȳsp
real ¯ real m̄ = −
ȳ = Gd d + Gp m̄ design
Gp
 Effect of design error

design real
Gd ≠ Gd
design real
Gp ≠ Gp

design ¯
−Gd d + ȳsp
real ¯ real m̄ = −
ȳ = Gd d + Gp m̄ design
Gp
 Effect of design error

design real
Gd ≠ Gd
Results in
permanent offset design
Gp ≠ real
Gp

real ¯
design ¯
−Gd d + ȳsp ȳ ≠ ȳsp
ȳ = Gd d + real
Gp m̄ m̄ = − design
Gp
design real However, it is difficult to
G Gp
[ design ]
ȳ = real
Gd − real d
Gp d¯+ ȳ estimate the process
design sp
Gp Gp parameters exactly.
 Issues with feedforward control

• It can’t take care of unmeasured disturbance.

• It’s physically impossible to design feedforward controller for
processes with finite time delay.

• Feedforward controller can’t account for design errors resulting

in offset.
Both feedforward and feedback controller have their own issues
Why don’t we have both ?
Feedforward+Feedback Controller
Feedforward+Feedback Controller

Let’s do our algebra for Gf = Gm = 1.

 Feedforward+Feedback Controller

m̄(s) = c̄(s) = c̄ (s) + c̄fb(s)

Let’s do our algebra for Gf = Gm = 1.

ff
 Feedforward+Feedback Controller

m̄(s) = c̄(s) = c̄ (s) + c̄fb(s)

design ¯
−Gd d + ȳsp
c̄ (s) = − design
Gp

c̄fb(s) = Gc[ȳsp(s) − ȳ(s)]

real ¯ real
ȳ = Gd d + Gp m̄
ff
ff
 Let’s take an example of first order process and feedback control

real ¯ real
m̄(s) = c̄(s) = c̄ (s) + c̄fb(s) ȳ = Gd d + Gp m̄

design ¯
−Gd d + ȳsp
c̄ (s) = − c̄fb(s) = Gc[ȳsp(s) − ȳ(s)]
design
Gp
real real design design
real
kd kp design
kd kp
Gd = Gp = Gd = Gp =
τ reals + 1 τ reals + 1 τ designs + 1 τ designs + 1

Gc = Kc
ff
ff
 ¯
d(s)
 Practice problem
Consider a distillation column, where the feed flow rate (F) is a source of
disturbance and you want to control the distillate mole fraction (xd) by controlling
the reflux rate (R). Your process design says that
x̄d 10 exp(−10s) x̄d 5 exp(−tds)
= =
F̄ 10s + 1 R̄ τs + 1
Gf = Gm = 1
(a) Show the feedforward contol block diagram.
(b) Design the feedforward controller that rejects any disturbance in F. What’s the
limit of td ? Can you design a controller with set point tracking capability ?

(c) Suppose td = 0.5 min and τ = 1 min. How will R vary as a function of time for
an unit step change in F ?
 Practice problem
Consider a distillation column, where the feed flow rate (F) is a source of
disturbance and you want to control the distillate mole fraction (xd) by controlling
the reflux rate (R). Your process design says that
x̄d 10 exp(−10s) x̄d 5 exp(−tds)
= =
F̄ 10s + 1 R̄ τs + 1

x̄d 7 exp(−12s)
(d) Now Imagine the actual transfer function = . Compute and sketch
F̄ 8s + 1
the response of xd(t) with the designed controller. Still td = 0.5 min and τ = 1 min.
More practice problems

V.24, V.25, V.16

From Stephanopoulos textbook