Department of Chemical Engineering I.I.T.

Jodhpur, India

CHL4020: Process Control & AI

applications
Department of Chemical Engineering I.I.T. Jodhpur, India

Need of Control Systems?

Tightened product quality specifications

Emphasis on more profitable plant operation
Complex and highly integrated processes in modern plants;
difficult to prevent propagation of disturbances from one unit to
other
Safe and efficient plant operation
Department of Chemical Engineering I.I.T. Jodhpur, India

• Process Dynamics

Process Control
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic Analysis

Cause Process Dynamics Effect

Analysis of the influence of a cause on its effect

How does a cause result in an effect ?

How much ?

How quickly?
Department of Chemical Engineering I.I.T. Jodhpur, India

Control of Process Systems

Design a Process Dynamics Desired
Cause Effect

“Manipulation” of the cause to yield a desired effect.

“Transfer variability from a point where it hurts the most to a point

where it hurts the least” – Jim Downs, Eastman Chemicals.

“Designing a cause” => Structuring and facilitating a

control scheme
Department of Chemical Engineering I.I.T. Jodhpur, India

Transfer of Variability

Fin, Tin

Fout, Tout

Open loop plant Q After Control

Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Example

Cold Stream at Hot Stream at

Flowrate = FC Flowrate = FH
Temperature = TC Temperature = TH

Manipulated Variables enjoy

(We can freely manipulate these)
shower at
Control Objective
Disturbance Variables F = Fsp
T = Tsp
(Beyond our control. Must live with these !!)
Department of Chemical Engineering I.I.T. Jodhpur, India

Process Control: The act of manipulating certain inputs in order to

meet specific operational objectives even in the
presence of disturbances

In the “shower” example:

Operational Objectives: We need to enjoy our shower with water at a

specific flow rate and at a specified (comfortable) temperature.

Disturbances: The temperatures of the cold and hot streams. These

temperatures may change depending on the time of the day, season
etc.

Manipulated Variables: We may operate the cold and hot water valves
as per our fancy and meet the operational objectives even if there are:
changes in the cold and/or hot water temperatures
changes in operational objectives
(“comfortable temperature” may vary from person to person)
Department of Chemical Engineering I.I.T. Jodhpur, India

In Block Diagram Form

Tin hot Tin, cold

Disturbance
dynamics

Fin, cold
Mixing Flow
Dynamics +
Temp.
Fin, hot
Note: Process and Disturbance dynamics are negligible
here, but generally they are not and could be different.
Department of Chemical Engineering I.I.T. Jodhpur, India

Input and Output Variables

Input Variables → Cause Variables that influence the effect

variables
Some of these are manipulable – manipulative variables
Others are not manipulable – disturbance variables

Output Variables → are really the effect variables, also called as

Controlled variables
Department of Chemical Engineering I.I.T. Jodhpur, India

What is the basis for manipulating the cold and hot water valves?
Answer : Look at the deviation from desired value

Step into the shower

“Sense” the output flow rate and temperature → Sensing the variation
(Point of sensing is very important. Disturbance or process variable)

Determine the “deviation” in the measured flow rate and temperature from
what you wanted them to be (target values or setpoints) → Comparison

Adjust the cold and hot water valves suitably → Control Action

Notice that the information on the process outputs (flow rate and temperature)
is used to determine the process inputs (valve positions). This is the basic
concept of Feedback Control.
Department of Chemical Engineering I.I.T. Jodhpur, India

Alternate Scenario

“Sense” the input flow rates and temperatures → Sensing the source of
variation !

Determine if there is any “deviation” in the expected input flow rates and
temperature → Comparison

Adjust the cold and hot water valves suitably → Control Action

Notice that the information on the disturbances (input flow rates and
temperature)is used to determine the process inputs (valve positions).
This is the concept of Feedforward Control.
Department of Chemical Engineering I.I.T. Jodhpur, India

Consider the following scenarios:

What if the disturbances occur frequently? e.g. frequent changes in the

cold and hot water temperatures.

Rather than enjoying the shower, we will end up being busy

changing the valve position all the time

What if my seven year old son wanted to take a shower? – the case of
an inexperienced plant operator.

Can he “sense”, “compare” and take the appropriate “control

action” as well as I (the experienced operator) do? He may
need to be constantly supervised – pressure on my “time”

What if my eighty year old grandfather, who is unable to sense

temperatures properly, steps into the shower? – personnel safety issue

He may end up scalding himself without realizing it !!

Department of Chemical Engineering I.I.T. Jodhpur, India

Wouldn’t it be rather nice if:

1. We have a control panel where we just input the desired flow rate
and the required temperature, step into the shower and just enjoy !

Sensing, comparison, computation of appropriate control action

(valve positions) are all done automatically – by hardware and/or
software.

This leads to “automatic process control” in contrast to the “manual

control” (where a human is very actively involved) discussed earlier

2. To handle the safety issues we must ensure that

The desired temperature cannot exceed a certain value (say 40 oC)

or go below a certain value (say 10 oC). These limits should however
be changeable by only the “expert operator” to handle special
situations
Department of Chemical Engineering I.I.T. Jodhpur, India

Note that the control system cannot deliver “anything that you want”

- you cannot ask for a flow rate that exceeds the sum of
maximum possible cold and hot water flow rates

- you cannot ask for a temperature that is above the incoming

hot water stream temperature or one that is below the
incoming cold water temperature.

- also, only certain “combinations” of the flow rate and the

temperature may be attainable.

for example, if the maximum hot water flow rate is 1 m3 / min

and if the temperature of the cold and hot streams are 30 oC &
40 oC respectively, the system cannot deliver a flow rate of
1.5 m3 / min and a shower tempertaure equal to 40 oC.

No “super-operator” or “advanced control system” can overcome

the limitations posed by a poorly designed process
Department of Chemical Engineering I.I.T. Jodhpur, India

Typical Control Examples

1. Driving your car

Objectives: Go from point A to point B safely (for you and everyone

else) and in a reasonable amount of time

Sense: Position of the automobile, speed of the car (measurable outputs

related to the objective)

Disturbances: curves and bumps on the road, position of other cars,

changing speed limits, …..

Manipulated Variables: Steering wheel, Brake pedal, Accelerator pedal

Must maintain safe & acceptable position, speed of the car by manipulating
the SW, BP and AP despite disturbances such as curves, bumps etc.
Department of Chemical Engineering I.I.T. Jodhpur, India

Typical Control Examples

2. Composition in a Distillation Column
• Objectives: Maintain top and bottom streams at specified values
• Sense: Composition (Temperature) of the top trays
• Disturbances: Feed flowrates, compositions, ….
• Manipulated Variable: Boil-up rates and condenser duty

Qc
D,xd
F,z

Qh
B,xb
Department of Chemical Engineering I.I.T. Jodhpur, India

Bringing in a new concept: “Dynamics”

• I am driving a car on a highway. I
hit the brakes. Does the car
come to an immediate
Q, T1
(instantaneous) halt?
• A certain government introduces
a “financial stimulus package” to
revive a sagging economy. Will
the effects be immediately felt by Q, T2
the common man?
• I open the steam valve of the

Condensate
stirred tank heater system by
10% Will the temperature T2
reach a new steady value
immediately? Steam
Department of Chemical Engineering I.I.T. Jodhpur, India

The answer to the above questions is “No”. Do you know why?

Systems have “Inertia” or “Capacitance”. This means that changes

evolve over time→ Process Dynamics

Final Value of T2

12oC New Steady

Dynamics
Initial Value State
of T2

Final Valve Opening

10%

Initial Valve
Time
Opening
Department of Chemical Engineering I.I.T. Jodhpur, India

Why would knowledge of process dynamics improve in our

ability to control?

Why not work with just steady state information?

Knowledge of Steady-state effects is certainly important (and

perhaps even the most important information about the process)

To improve the sagging economy, a certain government would consider

tax cuts rather than impose new taxes (tax cut is manipulated variable)

If I open the steam valve by 10%, the increase in temperature T2 is

approximately 12oC. So, a change of 1% in the valve opening will affect
the temperature T2 by approximately 1.2oC
However, ……….
Department of Chemical Engineering I.I.T. Jodhpur, India

Temperature Control Problem

B C

Temperature
A

0 time 3

Which is most appropriate ?

Plan A : Sluggish Control Depends on:
Plan B : Aggressive Control Thermal sensitivity
Plan C : Without Control Time for rejection
Department of Chemical Engineering I.I.T. Jodhpur, India

Components of a Feedback Control System

• Remember, the process is the “centerpiece”. A badly designed
process cannot be well controlled even with the most advanced
instrumentation and control system.

• Sensor – to measure the output variable (Controlled Variable). The

value of the controlled output is utilized by the controller to compute
the appropriate value of the input variable (Manipulated Variable)

• Comparator – compares the set point and the measured values of the
controlled variable. Computes the error signal [E = SP – PV]

• Controller – uses the error signal E to compute an appropriate

“controller output” signal – usually an electronic or pressure signal

• Final Control Element – In most process control applications, the FCE

is a valve. The valve responds to the controller output signal and the
valve stem either moves up or down effectively altering a flow
variable (i.e. the manipulated variable)
Department of Chemical Engineering I.I.T. Jodhpur, India

Components of a Feedback Control System

Notes:

1. The controller can either be a human operator or a computing device

2. The comparator is often included within the controller so it is not

often seen separately

3. The control algorithm can either be simple or complex

CO = + K if E > e and – K if E < -e (On/Off Control)

CO = Kc * E (where Kc is the controller gain)
CO = Kc * ( E + KI  E dt) (KI is the integral gain)
etc.

(Later, we will study about control algorithms in detail)

Department of Chemical Engineering I.I.T. Jodhpur, India

Components of a Feedback Control System

Notes:

4. Sensors are one of the most crucial components of the feedback

control system. Sensors are required to measure a variety of
process variables such as flow rate, temperature, pressure, level
and composition.

Even to measure temperature, there are many choices available –

thermocouples, thermistors, bimetallic, resistance temperature
detectors, filled systems and pyrometers.

The choice depends on the required accuracy, range, dynamics

(speed of response) and the cost.
Department of Chemical Engineering I.I.T. Jodhpur, India

Components of a Feedback Control System

Notes:

5. Control Valves (Final Control Element)

The most common method for influencing the behavior of chemical

processes is through the flow rate of process streams.

Usually, a valve with a variable opening in the pipe is manipulated

to influence the flow rate and achieve the desired process behavior.

Again, various types of valves are available. The appropriate valve is

chosen on the basis of the required size, maximum pressure,
temperature of the fluid, required trim characteristic etc.
Department of Chemical Engineering I.I.T. Jodhpur, India

Components of a Feedback Control System

Notes:

6. Besides the components mentioned above, various signal processing

devices such as transmitters, transducers, multiplexers, demultiplexers
etc. are needed.

We may have a controller that outputs a current or voltage signal

but the control valve may be a pneumatic valve and can only be driven
by a pressure signal. A E/P or I/P transducer is needed to convert
the current or voltage signal into an equivalent pressure signal.

Multiplexer is a device that combines multiple inputs into an aggregate

signal to be transported via a single transmission channel.

Remember that “control” involves lots of data processing and communication

Department of Chemical Engineering I.I.T. Jodhpur, India

Why are Chemical & Environmental Systems Controlled?

• Safety and protection of plant personnel, environment

and equipment
• Smooth operation and production
• Keeping consistent product quality
• Minimize the operating cost (utility charges, prolonging
catalyst life, ….)

To ensure safe and profitable plant operation

Department of Chemical Engineering I.I.T. Jodhpur, India

Multivariate (Multivariable) Systems

u1 y1
u2 Process y2
u3 y3

Almost all systems are multivariable in nature

Eg. Driving car: What are the inputs and outputs?

Attending CHL4020 course: What are the inputs and outputs?
Department of Chemical Engineering I.I.T. Jodhpur, India

Issues in Multivariate Control

1. Loop Pairing: To control y1, would you use u1 or u2 or u3?

Similarly for y2 & y3

Steering
Wheel
Speed
Gas Pedal
Process
Brakes Position
Gear

Question: Can we manipulate steering wheel to control the speed?

Department of Chemical Engineering I.I.T. Jodhpur, India

Issues in Multivariate Control

2. Number of independent control objectives vs. number of manipulated variables

u1 y1
Process y2
u2 y3

The process operation team has given us 3 independent control

objectives: desired values specified for y1, y2 and y3
The process design group has provided us two manipulated variables
u1 and u2

It is possible to satisfy the specifications on y1, y2 and y3?

Department of Chemical Engineering I.I.T. Jodhpur, India

3. Centralized or De-centralized control?

De-Centralized (Multi-Loop) Design

+
C1 Y1,SP
-
u1
y1
Process
u2
y2

- +
C2 Y2,SP
Department of Chemical Engineering I.I.T. Jodhpur, India

Centralized (Multivariable) Design

Y1,SP

u1 +
-
y1
Multivariable
Controller
Process
u2 y2 -

+
Y2,SP

The multivariable controller simultaneously determines what u1 and u2 are!

Considers all relationships between the u’s and the y’s simultaneously
Department of Chemical Engineering I.I.T. Jodhpur, India

Control System Development

Information from plants Formulate Control Information from plants
(if available) Objectives (if available)

Computer
Physical & Chemical Develop process Simulation
Principles model
Plant data (if available)

Process Control Devise Control

Theory Strategy Computer
simulation

Select control hardware

& Software Vendor & cost
information

Install Control
Systems

Adjust Controller
Settings Final Control system
Department of Chemical Engineering I.I.T. Jodhpur, India

Summary (1)

Sensors Process
Knowledge
Ingredients needed
Choosing
to reap maximum
Valves the best
benefits from
operating point
APC implementation

Safety Control
Issues Algorithms

A “responsive” plant designed with dynamic considerations

Department of Chemical Engineering I.I.T. Jodhpur, India

Summary (2)
Can APC alone make the plant safe and profitable?

• Safety must be everyone’s concern

• Plant profit determined by
– Business Aspects (Does demand exist? Are profit margins big? ….)
– Plant Design (good chemistry, efficient design “with dynamics in mind”, …)
– Selection of operating conditions (Safety limits, change operating conditions
depending on feed composition to improve profit, ….)
– Process Control (will implement the recommended operating conditions)
– Logistics (Production Schedule, Distribution Network, ….)
– Marketing Strategies – product positioning, customer service etc.
• Process Control does contribute to the overall health of your (current
employer / future employer) business. It helps to learn about this
subject !!
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 3: Analysis of System Dynamics

Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic Analysis

How does a process or plant respond to various signals or stimuli ?

Plant Dynamics
Stimuli / Response
Cause / Effect

Steps in Dynamic Analysis:

•Characterize the input suitably
•Develop or generate a process model
•Mathematically analyze the dynamics
Department of Chemical Engineering I.I.T. Jodhpur, India

Typical kinds of Inputs

Ideal step Ideal ramp Finite pulse

Ideal sine Random noise Noisy sinusoidal

signal
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic Model

Plant Dynamics
Input u Output y

Typical governing Equation:

dy(t)
τ + y(t) = Ku(t)
dt
Remarks: a) If the derivative term =0, plant has no dynamics and
responds instantaneously
b) Higher order relationships (derivatives) can exist – see them later
c) Need to look at a convenient way to represent the inputs and dynamics
to solve for the output behaviour
Department of Chemical Engineering I.I.T. Jodhpur, India

Laplace Transforms

Laplace transforms offer a convenient way of representing and

solving for the system response. They are defined as,

F ( s) = L[ f (t )] =  f (t )e − st dt
0

f(t) must satisfy conditions of piece-wise continuity over 0 < t < inf.
(almost always met in control practice)
Notes: a) Rigorously correct definition is a limit based integral.
b) Inverse Laplace transforms is a contour integral
1
f (t ) = L [ F ( s)] = 
-1 st
e F ( s)ds
2j C
Department of Chemical Engineering I.I.T. Jodhpur, India

Laplace Transforms of typical signals

Constant signals: f(t) =a

 
a − st a
 ae
− st
L[ f (t )] = dt = − e =
0
s 0 s

Step function: S(t) 0 for t  0

S (t ) = {
1 for t  0

1
L[ S (t )] =
s
Department of Chemical Engineering I.I.T. Jodhpur, India

Laplace Transforms of typical signals

Exponential functions: f(t) =e-bt

 

 e dt = 
− bt − st −(b + s ) t
L[ f (t )] = e e dt
0 0

1 −( b + s )t 1
= [ −e ] =
b+s 0 s+b

If b < 0, the laplace transform is unbounded.

Can you guess why ?
Department of Chemical Engineering I.I.T. Jodhpur, India

Laplace Transforms of derivatives


− st
L[df (t ) / dt] = ( df / dt ) e dt
0

Integrating by parts,

− st 

− st
L[df (t ) / dt] = f (t )e dt + fe
0
0

= sL( f ) − f (0) = sF ( s ) − f (0)

Higher order derivatives:

d2 f d
L[ 2
] = L[ ] = s ( s ) −  (0)
dt dt
Department of Chemical Engineering I.I.T. Jodhpur, India

Laplace Transforms of derivatives

Since  ( s) = sF ( s) − f (0)
d2 f
L[ 2
] = s[ sF ( s ) − f (0)] −  (0)
dt
= s 2 F ( s ) − sf (0)] − f ' (0)
General Rule:
dn f n −1 n−2
L[ n
] = s n
F ( s ) − s f ( 0) − s f 1
(0) − ...
dt
sf n − 2 (0) − f n −1 (0)
Department of Chemical Engineering I.I.T. Jodhpur, India

Solution of differential equations by Laplace transforms

Plant Dynamics
Input u Output y
Consider that the plant is perturbed by an ideal step input u
from an initial condition y(0)=0
dy(t)
τ + y(t) = Ku(t)
dt
Taking Laplace transforms of both sides,

K
τsY ( s ) + Y ( s) =
s
Department of Chemical Engineering I.I.T. Jodhpur, India

Solution of differential equations by Laplace transforms

K
Y ( s) =
s (s + 1)
To obtain the response y(t) for the assumed step input

−1 −t /
y(t ) = L [Y ( s)] = K (1 − e )

Steps to obtain dynamic response

1) Take Laplace transforms on both sides of the differential equation
2) Simplify by algebraic transformations
3) Take inverse Laplace transforms to obtain y(t)
Department of Chemical Engineering I.I.T. Jodhpur, India

Advantages of solving via the Laplace transforms route

Consider the coupled system of differential equations,

dy1(t)
τ1 + y1(t) = Ku(t)
dt
dy2(t)
τ2 + y2(t) = y1 (t )
dt
And suppose, we need to solve for y2(t) for a given stimulus u(t)
y1 ( s ) K y2 ( s ) 
= =
u ( s) τ1s + 1 y1 ( s ) τ2s +1
Department of Chemical Engineering I.I.T. Jodhpur, India

Advantages of solving via the Laplace transforms route

y2 ( s ) K
=
u (s) ( τ 2 s + 1)( τ1s + 1)

Easy to invert y2(s) and solve for y2(t)

Notes: 1) Very useful to analyze closed loop behaviour

involving several elements such as sensors, controllers and
actuators.
2) As we shall see later, stability analysis is facilitated by
examining the values of s as poles of the transfer function.
Department of Chemical Engineering I.I.T. Jodhpur, India

Method of partial fraction expansions for higher orders

Given Y(s) as,

N (s) N ( s) n
i
Y ( s) = = n =
D( s )  ( s + bi ) i =1 s + bi
i =1

We can obtain ’’s as ,

N (s)
 i = ( s + bi )
D( s ) s = − bi

Easier to obtain inverses of the Laplace functions and get

the time domain responses.
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative example of partial fraction expansion

Consider the following differential equation,

dy(t)
5 + 4 y(t) = 2, y (0) = 1
dt
5s + 2 1 2
Y (s) = = +
s (5s + 4) s 5s + 4
It can be seen that, 1 = ½ and 2=5/2 and therefore,

y(t ) = 0.5 + 0.5e −0.8t

Caution: Need to use carefully for repeated factors as well as complex

factors.
Department of Chemical Engineering I.I.T. Jodhpur, India

Useful Properties of Laplace Transforms

1. The integral for Laplace transforms need not exist always and so not all
functions can possess Laplace transforms.
2. The transform F(s) may not contain information about f(t) for t <0
3. The function f(t) and its corresponding F(s) occur as pairs and are
unique (Where did we use this property in this class ?)
4. The Laplace operator is linear => L{c1f1(t)+c2f2(t)}=c1L{f1(t)}+c2L{f2(t)}
5. Final Value theorem and Initial value theorems
6. Laplace Transforms of an Integral
7. Shift Properties
Department of Chemical Engineering I.I.T. Jodhpur, India

Initial and Final Value Theorems

Final value theorem: (useful in evaluating the final time value of a
transform)
y () = lim[ sY ( s)]
s − 0

Initial value theorem(useful in evaluating initial slopes of a response)

y (0) = lim[ sY ( s )]
s − 

Transform of an Integral (useful in analyzing integrating systems)

t
1
L{ f (t )dt = F ( s )
0
s
Department of Chemical Engineering I.I.T. Jodhpur, India

Shift Properties of Laplace Transforms

1. Shift in s

L{e at f (t )} = F ( s − a)

Typically useful in inverse transforms, for e.g. L-1{F(s-a)=eat f(t).

Suppose f(t)=1, then F(s)=1/s. So, L-1 (1/s+a)=e-at.1

2. Shift in t

L{ f (t − a)} = e − as F ( s)

Typically useful in modeling time delays in physical process

systems.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 4: Dynamics and Laplace

Transforms
Department of Chemical Engineering I.I.T. Jodhpur, India

First principles Modeling

Step change Step response

F h

time time
During transients, the rate based conservation law should be applied
Department of Chemical Engineering I.I.T. Jodhpur, India

General conservation principle

Basic Equation:

Input + Generation –Output – Depletion = Accumulation

Constitutive Equations
Based on other known physical and chemical principles such
as transport equations, chemical rate expressions and
thermodynamic relations
Notes: a) The above terms are all expressed as rates.
b) They are applicable to each of the individual terms viz. mass,
momentum and energy.
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Examples

Single tank problem

Fin + 0 –  Fout - 0 = d(Ah)

Input rate generation Output Depletion dt

Constitutive relationship

Output flowrate Fout is proportional to the height

Fout = ch
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Examples (contd)

dh
A = Fin − ch
dt
or
A dh 1
+ h = Fin
c dt c
dy A 1
 + y = Ku , with  = , y = h and K =
dt c c
and
y( s) K
=
u ( s ) s + 1
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Examples – Nonlinear case

Assume that the output flowrate Fout is proportional to

the square root of height

Fout = c h
Therefore,
dh
A = Fin − c h
dt

How do you carry out the Laplace transformation ?

Department of Chemical Engineering I.I.T. Jodhpur, India

Linearization of nonlinear processes

Linearize the nonlinear term around an operating point h and Fin,

− c −
Fout =c h =c h + (h − h )
−
2 h

− c − dh − −
Fin − c h + (h − h ) = A and
− dt
F in = c h
2 h

Introduce deviation variables

~ −
Fin = Fin − Fin
~ −
h = h−h
Department of Chemical Engineering I.I.T. Jodhpur, India

Linearization of nonlinear processes (contd)

~
~ ~ dh c
F in − c1 h = A , with c1 =
dt −
2 h

And therefore, ~
h( s ) K'
= '
F in ( s)  s + 1
~

Notes: a) Linearized dynamics dependent on the point of

linearization.
b) Deviation variables enforce h(0) = 0
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Examples – Degrees of Freedom

Stirred tank heater

Ti

T= TR

Q
Department of Chemical Engineering I.I.T. Jodhpur, India

Degrees of Freedom Analysis

dT
mC = wC (Ti − T ) + Q
dt
Degree of freedom analysis can be used to find out how many
variables can be independently manipulated.
NF=NV-NE, where NF , NE and NV are respectively the degrees
of freedom, the number of independent equations, and the
number of variables.
m and C are known parameters related to the design. Thus, w,Ti
and Q must be specified for the model to be completely
determined. Thus, NF = 3.
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Examples – State space model

h1 Fin = c1h1

h2

Fout = c2 h2
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Examples – State space model

dh1(t)
A1 + c1h1(t) = Fin(t)
dt
dh2(t)
A2 + c2 h2(t) = c1h1 (t )
dt
In state space form,

 dh1   − c1 
 dt   A1
0 
 1
h  1 
 dh  =  c    +  A1 Fin (t )
− c2  h2   
 2  1  0 
 dt    A2 A2 
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative Examples – State space model

The state equations completely define the dynamics of the system.

However, not all the states are measureable and hence you have
the output equation in terms of the measured variable y,

 h1 
y = 0 1 
h2 
Example: In a distillation column, you cannot sense all the states but
the holdup equations describe the internal dynamics completely.
The outputs could be states themselves or some linear combination
of the states, for e.g. product evolution rates as function of reactants.
Department of Chemical Engineering I.I.T. Jodhpur, India

Q, CAF Chemical Reactor

Consider a CSTR (continuous stirred tank
reactor); feed pure A
V CA
Isothermal Reaction; no heat effects

A→B A → B; − rA = k C A
Q, CA
Constant Volume, V; Constant density 
Variable Description Units

Q Volumetric flow rate m3 / min

CAF , CA Conc. of species A at inlet & outlet resepctively mol / m 3
-rA Rate of Reaction (rate of disappearance of A) mol of A / (m 3 min)
k Reaction Rate Constant *min-1
V Volume of the tank contents m3
*: The units of k are min-1 only for first order reactions
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic Balance Equations

dC A
V = Q C Af − Q C A − kC A V
dt

System comprises of only 2 chemical species A and B

Can write only 2 independent mass balances

- write for species A and species B
- write overall balance & one component balance
(either for species A or B)
Department of Chemical Engineering I.I.T. Jodhpur, India

A Plug Flow Reactor (PFR)

0 L
A

Feed Product

Dz
z Feed is pure A
z z + Dz Isothermal Reaction
No heat effects;
A→B
Control Volume: Element of Volume ADz
Constant density 
Between z & z + Dz
Constant velocity ,v
Department of Chemical Engineering I.I.T. Jodhpur, India

Variable Description Units

v Velocity of Stream m / min

A Cross sectional area of the tube m2
-rA Rate of Reaction (rate of disappearance of A) mol of A / (m3 min)
k Reaction Rate Constant *min-1

Focus on the control volume (A Dz) over the time interval

t to t + Dt

Moles of A Moles of A
Moles of A Moles of A
in CV at time - in CV at time = entering CV
during - leaving CV
during
t + Dt t time interval Dt time interval Dt

Moles of A disappearing
Moles of A accumulated in CV
during the time interval Dt
- in CV due to reaction
during time interval Dt
Department of Chemical Engineering I.I.T. Jodhpur, India

A Dz C A t + Dt
− A DzC A t = v A C A Dt z − v A C A Dt z + Dz
− k C A ADz Dt

Dividing by A Dz Dt throughout, we get

CA − CA v CA − v CA
t + Dt t
= z z + Dz
− k CA
Dt Dz
In the limit both Dz and Dt tending to zero, we get

C A C A A Partial Differential Equation (PDE)

= −v − k CA characteristic of a Distributed
t z Parameter System (DPS)
Department of Chemical Engineering I.I.T. Jodhpur, India

Lectures 5 and 6: Dynamics of low order

systems
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic Order of Systems

Order of the system: is the order of the differential equation that

governs the dynamic behaviour

Working interpretation: number of the dynamic elements /

capacitances or holdup elements between an input variable and a
controlled variable.

Higher order system responses are usually very difficult to

resolve from one another.

The response generally becomes sluggish as the order

increases
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative examples

Qc

F F D,xd
h1 F,z

h
h2
Fout = c2 h2
Q
Order 1 Order 2 B,xb
h
Order n
Department of Chemical Engineering I.I.T. Jodhpur, India

First Order Systems

Driving force
General Equation dy
 + y = Ku
dt
Transfer Function Resisting force
y( s) K
=
u ( s ) s + 1

Special cases: a)  = 0, no dynamics => pure gain systems

b)  is very large and so the dynamics are predominant =>
either one of the driving or resisting force is predominant
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order Systems

Step Response
y(s) K A
= , with u ( s ) =
u ( s) s + 1 s
K A
y( s) =
s + 1 s
=> y(t) = AK(1-e-t/ )

Characteristics: a) Output reaches the value AK at steady state

b) When t=, y attains 63.2 % of the final value,
Department of Chemical Engineering I.I.T. Jodhpur, India

Characteristics of first order step response

c) The initial slope is obtained by differentiating the output equation

dy AK
=
dt t =0 
(Can you think of another method to find initial slopes ?)
d) Measure of reluctance:
Reluctance


y
J =  [ y * − y (t )]dt = AK
0

t
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order System to a Pulse Input

y ( s) K
Plant = G( s) =
u ( s)  s +1

Suppose that the input is a finite width pulse

u(t)
−g
A A A e s
u ( s) = −
s s
t
t=0 t=g
Department of Chemical Engineering I.I.T. Jodhpur, India

Pulse response (continued)

KA KA e −g s
y( s) = −
s ( s + 1) s ( s + 1)
Final value of output
KA e −g s
lim s y ( s) =
KA
− = KA − KA = 0
s→0 ( s + 1) s =0 ( s + 1) s =0

Time profile of output

−g s
 
y (t ) = L  y ( s) = L 
−1 −1 KA KA e
−  = y1 (t ) − y2 (t )
 s ( s + 1) s ( s + 1) 

y1 (t ) = K A S (t ) − K A e −t  S (t )
Department of Chemical Engineering I.I.T. Jodhpur, India

Pulse response (continued)

− (t − g ) 
y2 (t ) = K A S (t − g ) − K A e S (t − g )
y(t)
KA
y1(t)
g Time (t)
0
-KA
- y2(t)
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order System to a Pulse Input

y(t)

KA

Time (t)
0 g

We have implicitly applied the principle of superposition

in this example.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lectures 5 and 6: Dynamics of low order

systems
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic Order of Systems

Order of the system: is the order of the differential equation that

governs the dynamic behaviour

Working interpretation: number of the dynamic elements /

capacitances or holdup elements between an input variable and a
controlled variable.

Higher order system responses are usually very difficult to

resolve from one another.

The response generally becomes sluggish as the order

increases
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative examples

Qc

F F D,xd
h1 F,z

h
h2
Fout = c2 h2
Q
Order 1 Order 2 B,xb
h
Order n
Department of Chemical Engineering I.I.T. Jodhpur, India

First Order Systems

Driving force
General Equation dy
 + y = Ku
dt
Transfer Function Resisting force
y( s) K
=
u ( s ) s + 1

Special cases: a)  = 0, no dynamics => pure gain systems

b)  is very large and so the dynamics are predominant =>
either one of the driving or resisting force is predominant
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order Systems

Step Response
y(s) K A
= , with u ( s ) =
u ( s) s + 1 s
K A
y( s) =
s + 1 s
=> y(t) = AK(1-e-t/ )

Characteristics: a) Output reaches the value AK at steady state

b) When t=, y attains 63.2 % of the final value,
Department of Chemical Engineering I.I.T. Jodhpur, India

Characteristics of first order step response

c) The initial slope is obtained by differentiating the output equation

dy AK
=
dt t =0 
(Can you think of another method to find initial slopes ?)
d) Measure of reluctance:
Reluctance


y
J =  [ y * − y (t )]dt = AK
0

t
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order System to a Pulse Input

y ( s) K
Plant = G( s) =
u ( s)  s +1

Suppose that the input is a finite width pulse

u(t)
−g
A A A e s
u ( s) = −
s s
t
t=0 t=g
Department of Chemical Engineering I.I.T. Jodhpur, India

Pulse response (continued)

KA KA e −g s
y( s) = −
s ( s + 1) s ( s + 1)
Final value of output
KA e −g s
lim s y ( s) =
KA
− = KA − KA = 0
s→0 ( s + 1) s =0 ( s + 1) s =0

Time profile of output

−g s
 
y (t ) = L  y ( s) = L 
−1 −1 KA KA e
−  = y1 (t ) − y2 (t )
 s ( s + 1) s ( s + 1) 

y1 (t ) = K A S (t ) − K A e −t  S (t )
Department of Chemical Engineering I.I.T. Jodhpur, India

Pulse response (continued)

− (t − g ) 
y2 (t ) = K A S (t − g ) − K A e S (t − g )
y(t)
KA
y1(t)
g Time (t)
0
-KA
- y2(t)
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order System to a Pulse Input

y(t)

KA

Time (t)
0 g

We have implicitly applied the principle of superposition

in this example.
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order System to a Sinusoidal Input

y ( s) K
= G( s) =
u ( s)  s +1
Let us introduce a sine input at time t = 0.

Aw
u(t) = A sin (w t) u ( s) = 2
s +w2

KA w
y( s) = 2
( )
s + w ( s + 1)
2
Department of Chemical Engineering I.I.T. Jodhpur, India

Response of First Order System to a Sinusoidal Input

Inverting in the time domain,

K Aw  −t  K A
y (t ) = e + sin (w t +  )
w  +1
2 2
w  +1
2 2

where  = − tan −1 (w  )
Notes: a) The transient term will vanish at large times.
b) The second sinusoidal term will persist
c) Phase lag depends on the frequency of the wave.
Department of Chemical Engineering I.I.T. Jodhpur, India

Frequency responses for First order System

W=0.1 rad/sec

W=1 rad/sec

W=5 rad/sec
Department of Chemical Engineering I.I.T. Jodhpur, India

Pure Capacity Systems

Driving force
General Equation dy
 + y = Ku
dt
Resisting force = 0
Transfer Function
dy
 = Ku
dt
Process Example:
Storage Tanks
F y ( s) K *
=
u (s) s
h Pump
Department of Chemical Engineering I.I.T. Jodhpur, India

Responses of Pure Capacity Systems

* *
Step Response y ( s ) = K A AK
= 2 y(t ) = AK * t
s s s

Sinusoidal Response K* A AK *
y( s) = =
s s +w
2 2
s ( s 2 + w2 )
K* A AK *
y( s) = =
s s +w
2 2
s ( s 2 + w2 )
AK *
y (t ) = [1 + sin(wt − 90)]
w
Department of Chemical Engineering I.I.T. Jodhpur, India

Lead Lag Systems

If a transfer function is of the form,

s + 1
G( s) = K
s + 1

In general, the lead block of the transfer function speeds

up the process.
A pure lead block as in the case of a PD controller is not
realizable.
Approximations of these have to be done to implement
lead or lead/lag blocks (for example in feedforward
control)
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic Response of a Lead Lag System

s + 1
G( s) = K
s + 1

Using partial fraction expansion,

s + 1 A1
G( s) = K = K [ A0 + ]
s + 1 s + 1
where A0=/ and A1=1-A0 = 1-

 K 
y ( s ) = g ( s )u ( s ) =  K + (1 −  )  u ( s)
 s + 1
Department of Chemical Engineering I.I.T. Jodhpur, India

Step Response of a Lead Lag System

Depending on the value of , an effective lead or lag action is seen

Department of Chemical Engineering I.I.T. Jodhpur, India

Lectures 7: Dynamics of higher order

systems
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic behaviour of second order systems

Number of dynamic elements are 2.

F
h1

h2

Second order systems can arise because of two first order

systems in interacting or noninteracting setups
Inherent dynamics in the system is of higher order
Interaction between the
U-tube manometer. holdups could give an
Presence of a controller in a closed loop even if the process oscillatory response.
is first order (such as PI or PID controllers in a closed
loop) can give a second order system.
Department of Chemical Engineering I.I.T. Jodhpur, India

IIlustrative example: Interacting system of tanks in series

F0

h2
h1

First Principles Model dh1

A1 = Fo − c1 (h1 − h2 );
dt
dh2
A2 = c1 (h1 − h2 ) − c2 h2
dt
Department of Chemical Engineering I.I.T. Jodhpur, India

First Principles Model

A1 A2 1 c1
If  1 = ; 2 = ; K1 = ; K 2 =
c1 c2 c1 c2
K1 K 2
y2 ( s ) = u (s)
 1 2 s + ( 1 +  2 + K 2 1 ) s + 1
2

The roots of the denominator (poles of the transfer function)

could be complex. They could give rise to oscillatory
behaviour for different u.
Department of Chemical Engineering I.I.T. Jodhpur, India

Second Illustrative example: Control of a first order

process
+
yd Plant y
controller
- u

 1
t


dy
dt
+ y = Ku and u (t ) = K c ( yd − y ) +
 I 0 ( yd − y )dt
dy  1
t

Therefore, 
dt
+ y = KK c ( yd − y ) +


I 0
( yd − y ) dt 

d2y dy KK c
 1 2 + (1 + KK c ) + y = yd
dt dt I

This is a second order system and will give rise to a second

order transfer function.
Department of Chemical Engineering I.I.T. Jodhpur, India

U-Tube Manometer

L d 2h 4  L dh 1
P1 P2 + + h= DP
2 g dt 2
 g R dt
2
2 g

L = length of fluid in the

h plane of manometer tube
initial rest
h when DP = 0 ,  = density and viscosity
of manometer fluid

R = radius of manometer
tube
DP = P1 - P2
g = gravitation constant
Department of Chemical Engineering I.I.T. Jodhpur, India

General second order transfer function

y(s) K For the manometer

G (s) = = 2 2
(
u ( s )  s + 2  s + 1 ) =
L2
g3
=

R 2
6L
g

The value of  (damping factor) determines the dynamic response of

second order systems. It represents viscous or dissipative forces.

If 0   < 1 (underdamped system) - oscillatory approach to steady

state

 1 (critically damped, overdamped systems) - non-oscillatory

approach to steady state.
 is also called the natural time constant of the system.
Department of Chemical Engineering I.I.T. Jodhpur, India

Step Response of Overdamped Second Order Systems

A   1 e −t / 1 −  2 e −t /  2 
u ( s) = y (t ) = KA 1 − 
s  1 −  2 

dy  e −t / 1 − e − t /  2 
= KA  
dt   1 −  2 

= 0 at t= 0
Initial Slope = 0. This is
in contrast to that of a
first order system
Department of Chemical Engineering I.I.T. Jodhpur, India

Step Response of Critically Damped Second Order Systems

A   t  −t /  
u ( s) = y (t ) = KA 1 − 1 +  e 
s    

dy  t e − t / 
= KA  2 
dt   

= 0 at t= 0

Again, the initial Slope = 0

Department of Chemical Engineering I.I.T. Jodhpur, India

Step Response of Underdamped Second Order Systems

   1 −  2    1 −  2  
A 
u ( s) = y (t ) = KA 1 − e − t / cos  t + sin  t  
s      1− 2
   

  1 −  2  
− t / 
e sin  t 
dy     
= KA  
dt   1− 2 
 
 

= 0 at t= 0
Again, the initial Slope = 0
Department of Chemical Engineering I.I.T. Jodhpur, India

Step Response of Underdamped Second Order Systems


tr = ( − cos −1  )
1− 2

Rise time tr: Time at which the output

first hits the steady state value

tr
Department of Chemical Engineering I.I.T. Jodhpur, India

Peak time

Time to first peak tp: Time at which the output

hits the first maximum value

tp
Department of Chemical Engineering I.I.T. Jodhpur, India

a = max(y) - D y
a b=Dy

Overshoot (OS) = a / b
b − 
OS = exp( )
1− 2
Department of Chemical Engineering I.I.T. Jodhpur, India

Settling time

Time taken
to reach and
1.05 * b remain within
5% of the total
0.95 * b change in y
(95% response
time)
b=Dy

ts
Department of Chemical Engineering I.I.T. Jodhpur, India

a = value of first peak - D y

c = value of second peak - D y
a
c

Decay ratio (DR) = c / a

−2 
DR = OS  = exp(
2
)
1− 2
Department of Chemical Engineering I.I.T. Jodhpur, India

Period of Oscillation, P
= time between successive peaks
= time between successive valleys

2 
P=
P 1− 2
  1 −  2  
− t / 
e sin  t 
dy     
= KA  
dt   1− 2 
 
 
Department of Chemical Engineering I.I.T. Jodhpur, India

Frequency Response

G (s) =
y(s)
= 2 2
K with u ( s) = A sin(wt )
(
u ( s )  s + 2  s + 1 )
It can be shown that the output y is also a sinusoid with the same frequency
as the input and is given by,
AK
y (t ) t −  = sin(wt +  )
(1 − w  ) + (2w )
2 2 2 2

and
 2w 
 = − tan 
−1
2 2 
 (1 − w  )
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 8: Dynamics of higher order

systems
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic behaviour of higher order systems

Number of dynamic elements =n

K1 K2 K3 Ki Kn
u  1s + 1  2s +1  3s + 1 is +1  ns +1 y

Overall dynamics
n  Ki 
y N ( s) =  u ( s)
i =1  s + 1
 i 

The system would have n poles, each equal to the negative

reciprocal of the time constant.
Department of Chemical Engineering I.I.T. Jodhpur, India

Step Response of the nth order system

n  Ki  1
y N ( s) =  
i =1  s + 1 s
 i 

Expanding by partial fractions,

 A0 N Ai 
y N ( s) = K  +  
 s i =1  i s + 1 
It turns out that A0 =1 and Ai = -i and yN(t) is given by,

 N
Ai −t / i 
y N (t ) = K 1 +  e 
 i =1  i 
Department of Chemical Engineering I.I.T. Jodhpur, India

Limiting case of the nth order system

Consider the special case when :

All the N steady state gains are equal to 1
All the time constants are identical and equal to /N

1
y N ( s) = u (s)

N

 s + 1
N 

In the limiting case as N goes to infinity, G(s) =e-s. This is the

case of a pure time delay.
Department of Chemical Engineering I.I.T. Jodhpur, India

Step responses for the limiting case

Department of Chemical Engineering I.I.T. Jodhpur, India

Higher order systems with zeros

q
( i s + 1)
y(s) = K i =1
p
u ( s)
( i s + 1)
i =1

This would be a (p,q)th order transfer function with p poles

and q zeros.
For realizability, q must be less than or equal to p. The
transfer function would then be called as a proper transfer
function.
As before, lead terms or zeros generally have an effect of
speeding up the response of the system.
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative example of a (2,1) system

K (1s + 1)
g (s) =
( 1s + 1)( 2 s + 1)

The step response expression for this plant can be written as,

   1 − 1  −t /1   2 − 1  −t / 2 
y (t ) = K 1 −  e −  e 
  1 − 2    2 −1  
If 1 =0, the response is that of a pure second order system.
Depending on the value of 1, different responses varying from first
order to second order with lead, can be realized.
Department of Chemical Engineering I.I.T. Jodhpur, India

Effect of the lead for the (2,1) system

Lead time constants: 0.05,1,4 and 8

Lag time constants: 1 and 4
Department of Chemical Engineering I.I.T. Jodhpur, India

Summary of the Low and higher order dynamics

First order : Overall response is a result of an equilibrium

between the driving and the opposing forces.
Response is characterized by a steady stage gain K and the
time constant , which characterizes the speed of response.
Special cases of the response are the pure gain and the pure
integrator processes.
Sinusoidal response shows an output of the same frequency as
the input but differing in amplitude and phase.
Lead-lag dynamics are a result of a pure gain element in paraller
with a pure-lag element.
Department of Chemical Engineering I.I.T. Jodhpur, India

Summary of the Low and higher order dynamics

Second order : Overall response is a result of a two dynamic
elements in an interacting or non-interacting configuration.
Response is characterized by a steady stage gain K, the
damping factor  which characterizes the oscillatory nature and
the natural time constant , which characterizes the speed of
response.
Higher order variants essentially show similar behaviour but
approach the pure time delay in the limiting case.
Higher order variants can also have numerator terms (zeros) and
higher denominator terms (poles) which can exhibit complex
dynamics.
Lead term in generally has a speed-up property over the lag
terms.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 9: Inverse Response and Time

Delay Systems
Department of Chemical Engineering I.I.T. Jodhpur, India

Fed-batch Chemical Reactor

FA -H
A -- > B

FA T
Chemical Reactor

Suppose the objective is to regulate heat generation and temperature

by manipulating FA addition.
With an increase in addition of FA, the concentration would obviously
increase with an increase in the rate and hence temperature.
Department of Chemical Engineering I.I.T. Jodhpur, India

Influence mechanisms

FA affects the temperature T through two different mechanisms :

- Differing enthalpies of mixing causes the reactor to temperature

to initially drop. Mixing dynamics is quick.

- Increase in concentration of reactants results in an increase in

the reaction rate and hence causes the temperature to rise. The
reaction dynamics is relatively slower.

- A balance between the two competing mechanisms causes the

temperature to exhibit “inverse response” dynamics.
Department of Chemical Engineering I.I.T. Jodhpur, India

Inverse Response

Note: Slope @
Department of Chemical Engineering I.I.T. Jodhpur, India

Inverse Response Mechanisms

Reaction dynamics
K1
1 s + 1
+
u(s) y(s)

+
K2
2 s +1
Mixing dynamics
Department of Chemical Engineering I.I.T. Jodhpur, India

Inverse Response Mechanisms

− ( K1 + K 2 )
The zero is located at s=
( K1  2 + K 2 1 )

Note: We consider open loop stable systems. So

1 and 2 are considered to be greater than zero.

1 K2
− (1 + )
1 K1
Equivalently, s=
 2 K2
( + )
1 K1
Department of Chemical Engineering I.I.T. Jodhpur, India

Inverse Response Mechanisms

The zero will be a RHP zero if

K2  2 K2
(1 + ) < 0 and ( + ) >0
K1  1 K1

K2  2 K2
or (1 + ) > 0 and ( + ) <0
K1  1 K1
This implies that the zero will be in the RHP if
 2 − K2  2 − K2
  1 or  1
1 K1 1 K1
Department of Chemical Engineering I.I.T. Jodhpur, India

The zero will be a RHP zero I.e. the system will exhibit
inverse response if:

 2 − K2  2 − K2
 1 or  1
1 K1 1 K1

The gains must have opposite sign

The faster system must have the smaller gain

Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 10: Stability and nonlinear

systems
Department of Chemical Engineering I.I.T. Jodhpur, India

Concept of system stability

If the system is perturbed from its current (steady) state,does it do

either of the following:
Return to its original position
Move to a new position in a “bounded” fashion
Keeps on moving in a “bounded” fashion
Keeps on moving in an “unbounded” fashion ?
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative examples

A closed loop may exhibit any of the following behaviour:

For a small perturbation

in the set-point or a
disturbance, the closed
loop may respond in a
variety of ways.

Could the response be a

function of the input size
as well ?
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative examples (contd.)

Consider the ramp response of a first order plant :

Although the first order plant

is “stable” for all other
inputs, it exhibits
“unbounded” behaviour due
to input “unboundedness”

Input dependence is even

more true for nonlinear
systems.
Department of Chemical Engineering I.I.T. Jodhpur, India

Bounded Input Bounded Output stability

For causal systems therefore, it makes sense to define

stability in terms of bounded-ness to a bounded input.

If in response to a bounded input the dynamic trajectory of a

system remains bounded as t → infinity, then the system is
said to be stable; else it is said to be unstable.

Definition takes care of ramp system responses as well as

sustained oscillations seen in some closed loop systems.
Department of Chemical Engineering I.I.T. Jodhpur, India

Stability of linear systems

Consider the general representation of general high order plants:

bq s q + bq −1s q −1 + bq −2 s q −2 + ... + b0
g (s) =
bp s p + bp −1s p −1 + bp −2 s p −2 + ... + b0

Or, ( s − z1 )( s − z 2 )....( s − z q )
g (s) = K
( s − r1 )( s − r2 )...( s − rp )

The roots determine the behaviour of the system as is evident from

the following:
Department of Chemical Engineering I.I.T. Jodhpur, India

Stability of linear systems

1
y( s) = g ( s)
s
Expanding by partial fractions,
A0 p Ai
y(s) = +
s i =1 ( s − ri )
Inverting in the time domain,
p
y (t ) = A0 +  Ai e rit
i =1
Thus, if ri is positive, the output grows with time else, it decays with
time.
Important Note: The stability is unaffected by the locations of the zeros.
Zeros thus do not affect the system stability.
Department of Chemical Engineering I.I.T. Jodhpur, India

Stability of linear systems

Thus if all the poles of the system lie in the left half plane,then the system
is stable.

For nonlinear systems, one can analyze the stability properties of the
linearized system. If it is stable, the nonlinear system could be stable in a
local neighbourhood. If the linearized system is unstable, the nonlinear
system would also be unstable.

Open Loop unstable systems can be stabilized under feedback.

Open Loop stable systems can exhibit unstable performance under
feedback.
Note: Controller Design is therefore very important.
Department of Chemical Engineering I.I.T. Jodhpur, India

Stability evaluation for state space form

It can be shown that the eigenvalues of the A matrix are negative

reciprocals of the time constants.

The output behaviour can be written in exactly the same manner:

p
y (t ) = A0 +  Ai eit
i =1

where the ’s are eigen values of the matrix A

Thus, if all the eigenvalues of the system A have negative real

parts, the system is said to be stable.
Department of Chemical Engineering I.I.T. Jodhpur, India

Conditional Stability

The transfer function is conditionally stable based on the values

of K

6K
g (s) =
48s 3 + 44s 2 + 12s + (1 + 6 K )

So, K has to be chosen judiciously so that one gets stable

behaviour.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 11: Frequency Response

Department of Chemical Engineering I.I.T. Jodhpur, India

Frequency Response

Response of the process to signals of varying frequencies.

A process can “pass” certain frequencies and block others.

Implicit Assumption: Any signal can be decomposed into

sinusoidal components having different frequencies.

For example, this signal can be

assumed to be consisting of a low
frequency signal superimposed
with a high frequency signal
Noisy sinusoidal
signal
Department of Chemical Engineering I.I.T. Jodhpur, India

Frequency Response

The inherent reluctance of a process and its natural dynamics determines its
response to various frequencies.

The reluctance is mostly a hindrance but sometime it can help !

Qc
F,z D,xd

A surge tank before the feed can

dampen out the disturbances
and stabilize column operation Qh
B,xb
Department of Chemical Engineering I.I.T. Jodhpur, India

Analysis of Dynamics via frequency response

W=0.1 rad/sec

W=1 rad/sec

W=5 rad/sec
Department of Chemical Engineering I.I.T. Jodhpur, India

Frequency Response Generation

Process Transfer function y( s)

= G( s)
u (s)

Input signal
w
u(t) = sin (w t) u ( s) = 2
s +w2
w
y(s) = G (s) 2
(
s +w 2
)
And then invert this in the time domain to get the frequency
response.
Department of Chemical Engineering I.I.T. Jodhpur, India

Basic Complex Algebra

Given a complex number z=a+ j b, it can be represented in the

complex plane as shown below.

Im. axis


Real axis

Magnitude of z denoted by |z|= (a2+b2)0.5 and angle is tan-1 (b/a)

and z can be more compactly represented as z=|z| ej
Department of Chemical Engineering I.I.T. Jodhpur, India

Short cut Method for Frequency Response Generation

Step 1: s=jw in G(s) to obtain G(jw)

Step 2: Rationalize G(jw) to obtain Real and imaginary parts
Step 3: The output sine wave will have an amplitude Aout=A(Real2 +
Imag2) and the phase angle is =tan-1( Imag/ Real).

Illustrative example
1
1 G ( jw) =
G( s) = jw + 1
s + 1
Department of Chemical Engineering I.I.T. Jodhpur, India

Short cut Method for Frequency Response Generation

− jw + 1 1 − jw
G( jw) = 2 2 = 2 2 + 2 2
 w +1  w +1  w +1
1 − w
Re( jw) = 2 2 and Im( jw) = 2 2
 w +1  w +1
Output Amplitude is therefore

 1   − w 
2 2 1
 2 2  + 2 2 
  w +1   w +1  2 w2 + 1

Phase is therefore − tan −1 (w )

Department of Chemical Engineering I.I.T. Jodhpur, India

Representation of Frequency Response Information

Bode Plots: Plot the amplitude ratio (AR) and the phase lag on
two separate plots called the Magnitude and the phase plots.
For the illustrative example,

Magnitude plots are

typically logarithmic
but phase plots are
semi-logarithmic.
Department of Chemical Engineering I.I.T. Jodhpur, India

Representation of Frequency Response Information

Nyquist Plots: Plot the amplitude ratio (AR) and the phase lag
on the same plot in the imaginary plane.Treat G(jw) as a
complex number and plot the Real(G) and Imag(G) on the
diagram.

Note: Unlike the bode plot, the frequency is implicit in the plot here.
Department of Chemical Engineering I.I.T. Jodhpur, India

Frequency Response of several systems in series

If g(s) = g1(s)g2(s)g3(s) … gN(s), then they can all be expressed in the

frequency domain as g(jw)=g1(jw)g2(jw)g3(jw) … gN(jw)

Using the polar form of representation g(jw)=|g1||g2||g3||g4| …. |gN|

ej, where , is the summation of the individual phase angles.

Thus, the bode plots of higher order systems can be constructed

from the local building blocks.
Department of Chemical Engineering I.I.T. Jodhpur, India

General behaviour of the frequency response plots

The amplitude ratio plots generally decay at higher frequencies,

more so for chemical processes. The rate of decay is a function
of the order of the process.

The phase plot asymptotically approaches (p-q) times –90o at

high frequencies, where p is the number of poles and q is the
number of zeros. Can you guess why ?

Also, what is the frequency response of a pure delay process ?

Department of Chemical Engineering I.I.T. Jodhpur, India

Interpretations from the frequency response

K
First Order Processes G ( jw) =
jw + 1
K
AR and Phase are: − tan −1 (w )
 w +1
2 2

It is convenient to work in terms of a magnitude ratio:

AR 1
log( ) = − log( 2 w2 + 1)
K 2
As w → 0, AR/K → 1 and log (AR/K) → log(1); gives the low
frequency asymptote
As w → , and log (AR/K) → -log(w) ; gives the high frequency
asymptote
(The two asymptotes meet at w =1), I.e. at w=1/. Thus the natural
time constant has a role in the frequency response.)
Department of Chemical Engineering I.I.T. Jodhpur, India

Pure Capacity System: Frequency Response

K
G (s) = ,
s
K
G ( jw) =
jw
K
AR = ,
w
 = − tan −1 () = −90o
AR
log( ) = − log( w)
K
Department of Chemical Engineering I.I.T. Jodhpur, India

Pure Lead System: Frequency Response

G ( s ) = K (s + 1) ,

G ( jw) = K (jw + 1)

AR = K 1 + w2 2 ,

 = tan −1 (w)
Department of Chemical Engineering I.I.T. Jodhpur, India

Lead - Lag System: Frequency Response

Effective Lag

(s + 1)
G ( s) = K ,
s + 1

(jw + 1)
G ( jw) = K
jw + 1

1 + w2 2 Effective Lead

AR = K ,
1+ w  2 2

 = tan −1 (w) − tan −1 (w)

Department of Chemical Engineering I.I.T. Jodhpur, India

Second-order Systems: Frequency Response

Amplitude
(1)
G (s) = K ,
( 1s + 1)( 2 s + 1)
(1) Phase
G ( jw) = K
( 1 jw + 1)( 2 jw + 1)
1
AR = K ,  → −180o at higher frequencies
1 + w2 1 1 + w2 2
2 2
 → −0o at lower frequencies
AR 1 1
) = − (1 + w  1 ) − (1 + w2 2 )
2 2 2
log(
K 2 2
AR
log( ) → log(1), at low frequencies
K
and to - 2log( w ) at higher frequencies
Department of Chemical Engineering I.I.T. Jodhpur, India

General Second-order Systems: Frequency Response

Effect of damping factor
Transfer Function

(K )
G (s) = ,
( s + 2 s + 1)
2 2

AR 1
= ,
K (1 − w  ) + (2 w )
2 2 2 2

 −2 w 
 = tan  −1
2 2 
 1 − w  

It can be shown that the AR curve shows a hump for all values of  less
than 0.707 and the maxima occurs at
1
wr = 1 − 2 2

Department of Chemical Engineering I.I.T. Jodhpur, India

Time Delay Systems

G ( s ) = e −s , G ( jw) = e −jw

AR
= 1,
K
 = −w
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 13: Introduction to Feedback

Control
Department of Chemical Engineering I.I.T. Jodhpur, India

Feedback Control

F,Ti
Steam T
TC Valve Plant

TS TS
+ -

Feedback Loop
TC

Steam
Department of Chemical Engineering I.I.T. Jodhpur, India

Typical Elements of the Feedback Loop

The plant
Steam T
Controller Signal
(4-20 mA/ 1-5V/ 3-15 Valve Plant
psi)
Sensor
T TS Tm, mV signal

Comparator
+

Setpoint Error signal, mV

-
Tm
Department of Chemical Engineering I.I.T. Jodhpur, India

Typical Elements of the Feedback Loop

Controller

Error signal TC Signal to valve

Valve / Actuator

% valve opening /
Controller output TS
steam flowrate

Typically valve and sensor dynamics are of simple first order type
and are lumped into the plant dynamics.
Department of Chemical Engineering I.I.T. Jodhpur, India

Controller types for feedback control

Proportional Controller

c(t) = p(t) -ps = Kc (t) , where c is the controller output, (t) is the
error and ps is the output at zero error.

In the laplace domain, c(s) = Kc (s) => gc(s) = c(s)/ (s) =Kc

Kc is called the proportional gain of the controller and is sometimes

represented as proportional band, PB.
100.(maximum range of controller output)
PB =
K c (maximum range of measured variable)
Department of Chemical Engineering I.I.T. Jodhpur, India

Controller types for feedback control

Proportional + Integral Controller

 1
t

c(t ) = p (t ) − ps = K c  (t ) +   (t )dt 
 I 0 

where c is the controller output, (t) is the error and ps is the output
at zero error.
c( s )  1 
g c ( s) = = K c 1 + 
 ( s)   I 
s

The term I is called the integral time, reset time and the reciprocal
is called the reset rate.
Department of Chemical Engineering I.I.T. Jodhpur, India

Integral action

Integral action is usually used in conjunction with proportional mode.

Interpreted as repetition of the proportional action after every
integral time.
With valve saturation, reset windup occurs
Department of Chemical Engineering I.I.T. Jodhpur, India

Controller types for feedback control

Proportional + Integral + Derivative Controller

 1
t
d (t ) 
c(t ) = p (t ) − ps = K c  (t ) +   (t )dt +  D 
  I 0 dt 

where c is the controller output, (t) is the error and ps is the output
at zero error.
c( s )  1 
g c ( s) = = K c 1 + +  D s
 ( s)  Is 

The term D is called the derivative time constant.

Department of Chemical Engineering I.I.T. Jodhpur, India

Closed Loop Transfer Functions

Unlike in the open loop case, the closed loop has two independent
inputs, viz. the set point and the disturbance.

Assuming that all other elements such as valve and sensors are lumped
with the process dynamics, the block diagram can be drawn as,
d

disturbance
+
+
yd Plant y
controller
- u +
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed Loop Transfer Functions

From the previous block diagram, the following are true:

u(s) = g c (s) (s)

y ( s) = g p ( s)u ( s) + g d ( s)d ( s)
= g p ( s) g c ( s) ( s) + g d ( s)d ( s)

 ( s ) = yd ( s ) − y ( s )
Therefore,
y ( s ) + g p ( s ) g c ( s ) y ( s ) = g p ( s ) g c ( s ) yd ( s ) + g d ( s ) d ( s )

g p ( s) g c ( s) g d ( s)
or , y ( s) = yd ( s ) + d ( s)
1 + g p ( s) g c ( s) 1 + g p ( s) g c ( s)
Servo response Regulatory response
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed loop response of first order + proportional (P) control

K
For gp(s)= and gc(s)=Kc,the response to a unit step at the
s + 1
set point is given by,
KK c
y ( s ) = s + 1
1 KK c 1
=
KK c s 1 + KK c 
1+ s +1
s + 1 1 + KK c

The system responds like a first order process with a smaller time
constant and a gain close to unity. The steady state error between the
set point (unit step) and the process output y(t) is called the offset.
Most processes under proportional control exhibit offset.
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed loop response of first order + proportional (P) control

Existence of off-set which decreases with increasing

gain.
Large gain can also result in instability and amplification
of noise.
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed loop response of first order + (PI) control

 1
t


dy
dt
+ y = Ku and u (t ) = K c ( yd − y ) + 
 I
0 ( yd − y )dt
dy  1
t

Therefore, 
dt
+ y = KK c ( yd − y ) +


I 0
( yd − y ) dt 

2
d y dy KK c
 1 2 + (1 + KK c ) + y = yd
dt dt I
Therefore the closed loop transfer function is second order with lead,
KK c ( I s + 1)
y(s) = yd ( s )
 1 I s +  I (1 + KK c ) s + KK c
2

This could yield oscillatory response with possible overshoot. For a unit
step change, it will reach set-point and exhibit zero offset. Integral action
therefore increases the effective order of the system but gives zero
offset.
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed loop response of first order + (PI) control

Integral action causes off-set to be zero

High proportional gain causes oscillatory response
High integral action leads to sluggish response
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed loop response of first order + (PD) control

dy  d 
 + y = Ku and u (t ) = K c  +  D
dt  dt 

Therefore the closed loop transfer function is first order with lead term,
KK c ( D s + 1)
y( s) = yd ( s )
( + KK c D ) s + 1 + KK c

The overall order is therefore actually reduced. Transient response will

be that of a lead lag system. It will exhibit steady state offset.

Likewise for PID control, the closed loop transfer will have 2 poles and 2
zeros and due to integral mode, it will not exhibit offset.
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed loop response of first order + (PID) control

Derivative action leads to stability due the lead term in the

transfer function.
Department of Chemical Engineering I.I.T. Jodhpur, India

Derivative mode illustration : on-off control of a heater.

Department of Chemical Engineering I.I.T. Jodhpur, India

Anticipatory action of the derivative mode

Period Sign of Type of More or Corrective sign of Effect of

error current less action de/dt derivative
action necessary action
(0,t1) positive heating more decrease negative decrease
heat heat
(t1,t2) negative cooling less increase negative increase
cooling cooling
(t2,t3) negative cooling more decrease positive decrease
cooling cooling
Department of Chemical Engineering I.I.T. Jodhpur, India

Integral action and offset removal

Consider that at steady state, the steam flow rate is 100

kg/hr and the temperature is 200o C.

Suppose that a setpoint change to 205 oC is introduced. Let us now

consider how a P and PI controller would behave at steady state in
closed loop.
du(t ) d (t )
The equation is u (t ) = K c (t ) and therefore
= K c
dt dt
du(t ) d (t )
At steady state, rate of change is zero and therefore = =0
dt dt
This could happen only if  is a zero or a non-zero constant. If  is a
zero, the deviation in steam flowrate u will be zero and also error
between setpoint and CV is zero. This is not possible physically as you
cannot get temperature to rise without supplying additional steam.
Therefore,  ≠0
Department of Chemical Engineering I.I.T. Jodhpur, India

Integral action and offset removal

t
Kc
For PI control, the equation is u (t ) = K c (t ) +
I   (t )dt
0

du(t ) d (t ) 1
and therefore = Kc +  (t )
dt dt I
du(t ) d (t )
At steady state, rate of change is zero and therefore = =0
dt dt

This means that the error must go to zero at steady state and the
extra control effort comes from the integral term. This ensures that a
nonzero deviation in the control effort is available through integral
action.
Department of Chemical Engineering I.I.T. Jodhpur, India

Summary of P+I+D control

Mode Symbol Increase of Advantages Disadvantages Remarks
mode
Proportional Kc increase 1. Higher speed 1. Nonzero
of response offset
2. Smaller offset 2. Stability
problems
Integral I decrease 1. Improves 1. Reset Windup Order of system
speed of 2.May become increases
response unstable
2. Zero offset
Derivative D increase 1. Reduction in Not good in
overshoots noisy systems
2. Stabilising
effects

Proportional action speeds up the response but gives an offset at steady state.
Integral action gives zero offset at steady state but destabilizes the loop. Also problem of
reset windup exists.
Derivative action gives anticipatory control action as well as a stabilizing effect to the closed
loop.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lectures 14 : Stability with Feedback

and Controller Design
Department of Chemical Engineering I.I.T. Jodhpur, India

Closed Loop Stability

Characteristic Equation 1 + g c ( s) g p ( s) = 0

Analysis of the roots of this equation tells us about the stability of the
closed loop system.
But this is really a function of the controller parameters through the
term gc(s) and therefore design methods need to include stability as a
first criterion.
The stable region is first determined before the controller design task
is taken up.
Department of Chemical Engineering I.I.T. Jodhpur, India

Methods for determining stability

Routh’s Stability Test: tells us how many roots are in the RHP which
would thus lead to instability. The test is applicable to polynomials only.

Steps 1. Write the characteristic equation in standard form

a0sn+ a1sn-1 + a2sn-2 + ….. + an-1s + an = 0
Leading coefficient a0 must be positive.
If even one of the coefficient is negative, then at least one root
lies in the RHP and the system is unstable. Else,
2. Generate the Routh array: This array will have n+1 rows for an nth order
polynomial. The first two rows can be generated from the characteristic
equation. The next (n+1-2) rows are generated algebraically.
Department of Chemical Engineering I.I.T. Jodhpur, India

Routh’s stability test

Routh array

Row 1 a0 a2 a4 … …..

Row 2 a1 a3 a5 … …..
a1a2 − a0 a3 a a −a a
Row 3 b1 b2 b3 … ….. b1 = and b2 = 1 4 0 5
a1 a1
Row 4 c1 c2 c3 … ….. b1a3 − a1b2 ba −ab
c1 = and c2 = 1 5 1 3
b1 b1

Row n+1 z1

The first column must contain all elements that are positive, for stability.
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative examples
Consider the polynomial s4+ 5s3 + 3s2 +1 = 0. What can you say
about its stability ?

Consider the polynomial 10s3 +17s2 + 8s +1 + Kc =0. What can

you say about its stability as a function of Kc?

Construct the routh array

Row 1 10 8
System will be stable for –1 <
Row 2 17 1+Kc Kc < 12.6
Row 3 7.41-0.588Kc 0
Row 4 1+Kc 0
Department of Chemical Engineering I.I.T. Jodhpur, India

Routh Stability limits

If one of the rows (say n) becomes zero, this means that a pair of
purely imaginary roots exist. These can be evaluated from the n-1th
row.
Limits on the controller gain can therefore be calculated by requiring
all the first column elements to be zero.

An alternate way of calculating the stability limits is to simply set s=jw

and check at what values of K, the roots have positive real parts.

For, the polynomial 10s3 +17s2 + 8s +1 + Kc =0, it can be shown

that the same limits are obtained.
Department of Chemical Engineering I.I.T. Jodhpur, India

Stability limits by direct substitution

For, the polynomial 10s3 +17s2 + 8s +1 + Kc =0, set s=jw and solve
for the real and imaginary parts.

So, -10jw3 +17w2 + 8jw +1 + Kc =0, which yields

(1+Kc-17w2) + j(8w-10w3) =0.

Setting the real and imaginary components to zero gives

1+Kcm-17w2 =0 and 8w-10w3 = 0 so w2=0.8 or w = ±0.894 and Kc =

12.6

Thus a sustained oscillation of frequency w=0.894 rad/min occurs for

Kc=12.6.
Department of Chemical Engineering I.I.T. Jodhpur, India

Stability Limits by Root Locus

The root locus is a plot of the roots of the characteristic equation as

the controller gain varies from 0 to infinity.

The root locus begins at the open loop poles and ends either at the open
loop zeros or at infinity. Thus it has as many branches as the number of
open loop poles. Also, it is symmetric about the real axis (why ?!).

On the root locus, the magnitude of the gp(s)gc(s) is always unity.

The closed loop system is non-oscillatory, when the root locus lies on the
real axis and becomes oscillatory as it departs from the real axis. Also,
as it becomes unstable, when the locus crosses the imaginary axis.
Department of Chemical Engineering I.I.T. Jodhpur, India

Root Locus: Illustration

Consider that the process is described by

2
G(s) =
( s + 1)( s + 2)( s + 3)
and is under proportional control. Therefore the characteristic equation
would be
(s + 1)(s + 2)(s + 3) + 2Kc = 0

The root locus plot shows that any value of Kc beyond 30 takes the locus
into the RHP and therefore results in closed loop instability. For values of
Kc=0.2, the closed loop shows non-oscillatory behaviour beyond which the
closed loop would becomes oscillatory.
Department of Chemical Engineering I.I.T. Jodhpur, India

Root Locus: Illustration

Department of Chemical Engineering I.I.T. Jodhpur, India

Lectures 15 : Feedback Controller

Design
Department of Chemical Engineering I.I.T. Jodhpur, India

Issues in Feedback Controller Design

Choice of Sensors and Transmitters

These are the dynamic elements of the loop. Typically, they give out a
current (4-20mA signal) or a voltage (1-5V DC) signal. {Concept of a
live zero !}
Need a balance between the cost and accuracy of sensors. Precise
sensors are often used in cascade control schemes.

Smart transmitters – configure themselves to generate accurate

measurements. Intelligent sensors also manage the data
communication between themselves and the control room.
Department of Chemical Engineering I.I.T. Jodhpur, India

Control Valves / Actuators

Valve Action: Fail-open (Air-to-close) or Fail closed (Air-to-open) based

on safety considerations.
Valve Characteristics : relates to how the flow changes with respect to
valve stem. Could be linear or nonlinear. Inherent characteristics may
be linear but installed characteristics may be nonlinear.
Quick opening

Linear valve
% valve

Equal percentage

% stem position
Department of Chemical Engineering I.I.T. Jodhpur, India

Control Valves / Actuators

Application 1: Valve regulates steam flow in a reactor

heating coil

Reactor overheating is catastrophic. Hence a fail-close (FC)

valve is suitable

Application 2: Valve regulates flow of effluents from a

wastewater treatment holding tank into a
river

Flow of untreated wastewater is not acceptable, hence we

must select a valve that guards against this. Fail-close valve
is suitable.
Department of Chemical Engineering I.I.T. Jodhpur, India

Control Valves / Actuators

Application 3: Valve regulates cooling water flow to a

distillation condenser

The vapor coming out from the top of the column must be
condensed before it goes to the receiver.

If cooling water flow shuts off completely (due to

transmitter failure), the vapors would not be condensed.

Fail-open valve would be desirable.

Department of Chemical Engineering I.I.T. Jodhpur, India

Direct and Reverse Acting Controllers

Consider the use of a proportional controller with proportional gain Kc.

CO(t) = Kc e(t) = Kc { ysp(t) - y(t) } = Kc ysp(t) - Kc y(t)

If Kc is positive : CO(t)  as y(t)   CO(t)  as y(t) 

This is termed Reverse Acting Controller

If Kc is negative : CO(t)  as y(t)   CO(t)  as y(t) 

This is known as Direct Acting Controller

What kind of controller must you use for a given situation? - this
depends on the process and the valve!!
Department of Chemical Engineering I.I.T. Jodhpur, India

Direct and Reverse Acting Controllers

Consider the liquid level control system shown in the following
figure. The output from the level transmitter increases if the liquid
level increases.
(Direct Acting Transmitter). The valve is a AC(air-to-close) valve.
Should the proportional controller have direct or reverse action?

LT LC The transfer function

between the MV (exit
h flow) and the CV (h)
has negative gain

X
Department of Chemical Engineering I.I.T. Jodhpur, India

Direct and Reverse Acting Controllers

If the level in tank increases from its set point value, then the exit flow rate
(MV) should be increased.

This means the valve has to open more

We have a AC valve. Therefore, to open it more, the pressure signal to the

valve must be decreased

The controller output CO must decrease in order to reduce the pressure

signal sent to the valve.

If y  , we want the CO to . This means we would require a reverse

acting controller i.e. Kc must be positive.

What if the inlet flow rate was used as the MV?

Department of Chemical Engineering I.I.T. Jodhpur, India

Commercial Forms of PID Controllers

PD Controllers
 1+ Ds 
g PD ( s) = K c  
 1 +  D s 

PID Controllers

 1  1 +  D s 

g PID ( s) = K c 1 +  
  I s  1 +  D s 
Department of Chemical Engineering I.I.T. Jodhpur, India

Controller Design Principles

Choose the controller type and parameters such that some objective
criteria of closed loop performance is satisfied.
Performance criteria would generally involve:
Stability criteria – closed loop response must be stable
Steady state criteria – offset must be zero
Dynamic response criteria – closed loop system must reject
disturbances and track setpoints in an acceptable manner.
Department of Chemical Engineering I.I.T. Jodhpur, India

Dynamic response criteria

Need smaller rise times,

settling times,overshoots etc.
Department of Chemical Engineering I.I.T. Jodhpur, India

Quantitative measures of response


• Integral Absolute error IAE =  |  | dt
(IAE) 0

• Integral squared error
(ISE) ISE =   2 (t )dt
0
• Integral time weighted 
absolute error (ITAE) ITAE =  t | | dt
0
• Integral time weighted 
squared error (ITSE) ITSE =  2
t | | dt
0
Department of Chemical Engineering I.I.T. Jodhpur, India

Lectures 16: Design, Stability and

Tuning
Department of Chemical Engineering I.I.T. Jodhpur, India

Getting Controller Parameters: Integral Criteria

Aims to minimize area measure
based on the set point response
or load rejection response.
Based on a FOPDT model
including sensor, valve,
transmitter etc)
Write out the error
analytically in terms of the
controller parameters or use
an optimizer

Methods based on Integrals

tends to be conservative
Department of Chemical Engineering I.I.T. Jodhpur, India

Getting Controller Parameters: Integral Criteria

Start with a model of the plant
and guess controller parameters
New values of
parameters
Simulate the time profile of the
error

Optimizer
Get the Integral value and check
for constraints

Are the controller parameters Design

optimal ? complete
No Yes
Department of Chemical Engineering I.I.T. Jodhpur, India

Getting Controller Parameters: Integral Criteria

Empirical parameters for a FOPDT model

From: Seborg et al. (1989)

Department of Chemical Engineering I.I.T. Jodhpur, India

Relative comparison of the three criteria

ISE and ITSE result in

short rise times but
larger overshoots
Department of Chemical Engineering I.I.T. Jodhpur, India

Direct Synthesis tuning

A desired closed loop trajectory that reflects the performance

requirement is specified.
Such a trajectory must be chosen with care (realizability).
If there is a delay in the process, the trajectory must also have a
delay term.
The controller is directly obtained by equating the closed loop
response equation (in terms of the controller) to the desired
closed loop trajectory.

Synthesis could result in a non-PID type controller and therefore

simplifying assumptions need to be made.
Department of Chemical Engineering I.I.T. Jodhpur, India

Direct Synthesis Procedure and example

gg c Actual closed loop

y(s) = y (s)
1 + gg c d transfer function
y ( s) = q( s) y ( s)
d Desired closed loop
gc = (
1 q 1
)  gc = [
1
] transfer function
g 1− q g 1 −1
q
Example 1 :
K 1
g ( s) = ;q =
s + 1 cs +1
 1
gc = (1 + )
cK s
Department of Chemical Engineering I.I.T. Jodhpur, India

Direct Synthesis Procedure and example

Example 2: Presence of time delay in the loop

Ke −s e − c s
G( s) = ;q =
s + 1  cs +1
 1
gc ( s) = (1 + )
K ( +  c ) s

Notes: q(s), the desired closed loop transfer function must be

specified carefully. A rough rule of thumb is to start with a time
constant that is half of the process time constant.
Department of Chemical Engineering I.I.T. Jodhpur, India

Direct Synthesis Procedure: Another example

Designed Achieved
Let the process g =
K
r
(1s + 1)( 2 s + 1)( 3s + 1) r = ;  = 2 K r  r
gg c 2 r
y(s) = y (s)
1 + gg c d ( s + 1)( s + 1)
y ( s) = q( s) y ( s)
d gc ( s ) = 1 2
1 q 1 1 s
gc = ( )  gc = [ ]
g 1− q g 1
−1 ( +  )
q K c = 1 2 ;  = ( +  )
1  I 1 2
q=
 r 2 s 2 + 2 r r s + 1  
(1s + 1)( 2 s + 1)( 3s + 1)  = 1 2
 gc =
( s )(s + 1)
D  +
1 2
This is a PID controller
Department of Chemical Engineering I.I.T. Jodhpur, India

Methods based on approximate models

Generation of approximate models

Ke −s
G ( s) =
s + 1

Method is a little approximate

but is adequate for feedback
controller design purposes.

For more advanced control

strategies, model accuracy is
important.
Department of Chemical Engineering I.I.T. Jodhpur, India

Approximate model tuning rules

(valid for 0.1 < / < 1.0)
Cohen-Coon
Controller Kc I D
P 1/K(/) - -
PI 0.9/K(/) 3.3 -
PID 1.2/K(/) 2.0 0.5

Direct Synthesis
Controller Kc I D
PI / K(r+)  -
PID ((2+)/ +(/2) r/(2K(r+)
(2K(r+))
Department of Chemical Engineering I.I.T. Jodhpur, India

Approximate model tuning rules (Cohen-

Coon) (valid for 0.1 < / < 1.0)
Department of Chemical Engineering I.I.T. Jodhpur, India

Approximate model tuning rules (ITAE)

(valid for 0.1 < / < 1.0)
Department of Chemical Engineering I.I.T. Jodhpur, India

Methods based on Stability Margins

Controller parameters calculated as a back-off from stability limits

The stability limit is defined based on the value of the proportional

gain that sets sustained oscillations of constant amplitude in the
system.

Any of the methods for example, root locus or routh criteria could be
used. Alternately, back off from the stability limits on the bode and
nyquist diagrams could also be used.

Stability limits can indicate the proportional gain and the frequency at
which sustained oscillations are observed in the system.
Department of Chemical Engineering I.I.T. Jodhpur, India

Example of stability limits and stability margin

Plant
2
G(s) =
( s + 1)( s + 2)( s + 3)

Controller

(s + 1)(s + 2)(s + 3) + 2Kc = 0

Sustained oscillation are observed at Kc=30 (call this Kcu) and the
frequency of oscillations can be read from the root locus plot(or
calculated by direct substitution method). This gives w=3.32 rad/s
and the period is Pu=2/w.
Department of Chemical Engineering I.I.T. Jodhpur, India

Zeigler Nichols Stability margin based tuning

Controller Kc I D
P 0.5 Kcu - -
PI 0.45Kcu Pu/1.2 -
PID 0.6Kcu Pu/2 Pu/8

So, for the previous example, if a P controller were used, Kc = 15; For
a PI controller, Kc=13.5 and I = 1.577; For a PID controller, Kc=18, I
= 0.946, D=0.24
Department of Chemical Engineering I.I.T. Jodhpur, India

Lectures 17 : Stability& Design in the

Frequency Domain
Department of Chemical Engineering I.I.T. Jodhpur, India

Stability limit based on sustained oscillations

+
yd Plant y
controller
-  u

b
Consider the following simple experiments:
1) Let the setpoint be perturbed in a sinusoidal fashion with the loop open at b
If at that frequency, the controller and plant effectively add a phase lag of –180o,
then b will be signal that is out of phase with the set-point.
2) When the oscillations become steady, suppose the loop is closed at b and
the setpoint is set to zero; the –ve sign will introduce another phase change of –
180o and the wave at b will pass through the loop over and over. If the controller
and plant gain is less (greater, equal) than unity, these oscillations will die
(grow,stay constant) as the signal traverses repeatedly through the loop.
Department of Chemical Engineering I.I.T. Jodhpur, India

Bode Stability Criteria

If the AR of the open loop transfer function is greater than 1 when the
phase lag is –180o, the closed loop system is unstable. The frequency at
which phase lag = -180o is called the cross-over frequency.

Cross over frequency

Implications
1) A first order process can never be made unstable with proportional control (why ?)
2) As the order of the process gets higher, the cross-over occurs at lower frequencies
and therefore you cannot expect aggressive (rapid) control with feedback.
3) Time delays also limit the range of acceptable controller gains for stability.
Department of Chemical Engineering I.I.T. Jodhpur, India

Controller Design based on Bode Stability Criteria

Establish the stability limits using the bode plot of the open loop system:
Steps:
a) Obtain the open loop bode diagram and locate the crossover
frequency and the AR at this frequency. Denote this by (AR)c.
b) If (AR)c is less than 1, find that value of Kc, that will push Arc =1. This
is given by Kcu (AR)c=1 which means Kcu= 1/Arc.
c) Knowing Kcu and wc, one can use stability margins based tuning
methods.

Note: The Bode stability criteria is only valid for phase plots that monotonically
decrease. Generally the nyquist stability criteria is more powerful than the
Bode criteria
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative example: Bode Stability Criteria

Let G( s) =
5 What is the ultimate gain for a P
( s + 1)(0.5s + 1) controller ?
Case 2: (dashed line): Integral
time=0.2
Case 3: (dash-dot line) Integral
time =1

If the crossover frequency is

not crossed, then the system
will be stable for any gain.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 18 : Cascade and Feedforward

Control
Department of Chemical Engineering I.I.T. Jodhpur, India

Effect of disturbances in feedback control

d1 d2
yd +
C P y
-

Traditional feedback control strategies cannot handle processes

with significant disturbances.
Even if they could, corrective action is taken only after the
disturbance affects the process
Perfect Control is generally not possible using feedback only.
Department of Chemical Engineering I.I.T. Jodhpur, India

Transients in the Steam header affects steam flow rate

(From text)
Department of Chemical Engineering I.I.T. Jodhpur, India

Cascade scheme for the reboiler

(From text)
Department of Chemical Engineering I.I.T. Jodhpur, India

Analysis of the cascade scheme

Department of Chemical Engineering I.I.T. Jodhpur, India

Reduced block diagram

The new transfer functions g1* and g2* represent the inner closed loop and
Therefore, if shaped properly, could yield good overall closed loop behaviour
Department of Chemical Engineering I.I.T. Jodhpur, India

Possible Remedies

If the disturbances can be measured and a good model of the

disturbance dynamics is available, it may be possible to employ a
more complex control structure for better regulation.
If the disturbance is closer to the input side, better regulation is
possible through the use of cascade control structure.
If the disturbance is closer to the output side, better regulation may
be possible through the use of feedforward control structure.
The schemes may require additional sensors and detailed
characterization of the disturbances, but better regulation may offset
these costs.
Department of Chemical Engineering I.I.T. Jodhpur, India

Requirements for Cascade Control

Cascade Control needs to be implemented only if traditional single loop

feedback fails.
Presence of a measured secondary variable is a must.
It must indicate the occurrence of a disturbance.
Its dynamics must be significantly faster than the primary dynamics (at least
three times as as the primary).
Measurements need only be precise not necessarily accurate.
Disturbances that affect the secondary (slave) variable are detected and
eliminated early. Controllability of the outer loop is improved because the
inner loop speeds up the response.
Department of Chemical Engineering I.I.T. Jodhpur, India

Tuning considerations in Cascade Control

Master Controller is like the traditional controller in a feedback loop.
Slave controller has to ensure that changes in its setpoint are adhered
to.
Typically tuning is doing from the inner loop to the outer loop
- the inner loop must be tuned very “tightly” ( typically a high gain
controller).
- the outer loop may be tuned using traditional methods.
In general, integral mode is not used in the slave controller, especially if
gain is limited by stability.
However, it may be used if large disturbances affect the process and the
offset will affect the primary process variable.
Derivative mode compensates for sensor lag or loop deadtime and allow
for a higher slave controller gain.
Department of Chemical Engineering I.I.T. Jodhpur, India

Typical slave loops in the process industry-- Flow Loop

Flow loops are quite commonly seen as slave loops in cascade

schemes. They have rapid dynamics and are amenable to be driven by
remote set-point changes.
Example: Reflux flow in a distillation column
Flow transmitter as a sensor is quick and also cancels
nonlinearities.
Flow controller is typically tuned tight with a PI scheme with the
integral time set equal to the valve time constant.
A high gain will take care of the hysterisis or deadband in the valve
(valve stiction etc).
Department of Chemical Engineering I.I.T. Jodhpur, India

Typical slave loops in the process industry-- Flow Loop

Tsp

TI FI
C C
L
FT C

TT
Department of Chemical Engineering I.I.T. Jodhpur, India

Other choices for the slave loop

Temperature can be a slave variable if

sensor lag is taken care of by adding derivative action
the derivative action must act on the measured variable only and not
on the error to prevent two derivative units in series.
In a reactor control problem, temperature has the advantage over coolant
flow as it compensates for disturbances in coolant header pressure and
temperature.
Pressure is a good slave variable to use as it can be measured easily, fast
and reliably.
Example : Distillation column reboiler where pressure determines the
steam condensing temperature and hence the heat transfer rate.
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative examples

Single loop control Cascade Control

[From Marlin (1995)]

Department of Chemical Engineering I.I.T. Jodhpur, India

Feed-forward control -- a simple illustration

The inlet temperature is sensed and the steam flow is adjusted

before it affects the outlet temperature T
Department of Chemical Engineering I.I.T. Jodhpur, India

Requirements for Feedforward Control

Feedforward control is desirable when :

Feedback control does not provide good performance
A measured feedforward variable is available
A feedforward variable must satisfy the following:
the variable must indicate the occurrence of a important
disturbance
disturbance dynamics must be significantly faster than the
process dynamics
Department of Chemical Engineering I.I.T. Jodhpur, India

Aspects of Feedforward Control

Feedforward control compensates for disturbance before it affects the

process
Presence of FF controller does not affect the stability of the system
FF control cannot eliminate steady state offsets
FF control requires a sensor and a model for each disturbance
The scheme needs a perfect process and disturbance model for good
disturbance.
It is not useful if the disturbance cannot be measured.
Department of Chemical Engineering I.I.T. Jodhpur, India

Block Diagram of Feedforward Control

Gff

Gd(s)

+
+ +
Gst Process Gp(s)
+

For perfect disturbance rejection, Gff= -Gd/Gp

Department of Chemical Engineering I.I.T. Jodhpur, India

Derivation of the feedforward controller

-s
Ke
Let the process be modelled as G(s) = and
s + 1
- s
Ke
Let the disturbance be modelled as G d ( s ) =
ds +1
The feedforward controller will have the form,
K ff (s + 1)
−s
G ff ( s ) = − e
ds +1
Department of Chemical Engineering I.I.T. Jodhpur, India

Implementation aspects

If  and  are approximately equal, the controller is a simple lead-lag

block and can be implemented.
If  <  , the controller has a positive delay exponent => that a
prediction is required. This is physically unrealizable. Implement only
as a steady state gain block.
Gain adjustments can be done with the feedback controller on manual
or automatic:
If manual, mismatch between the setpoint and controlled variable is used to tune
the gain of the feedforward block
If automatic, mismatch between the values of the manipulated variables before
and after occurrence of the disturbance is used to tune the gain.
In either case, correction is done only after the system reaches steady state.
Department of Chemical Engineering I.I.T. Jodhpur, India

FF controller is a lead lag block

Lead lag block determines the appropriate lead or lag action

necessary to give perfect control
Department of Chemical Engineering I.I.T. Jodhpur, India

Simple feedforward control of a heater

Department of Chemical Engineering I.I.T. Jodhpur, India

Feedforward augmented with Feedback

Feedback stability is unaffected by

the presence of feedforward control
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 19 : Model Based Control

Department of Chemical Engineering I.I.T. Jodhpur, India

Some examples of processes with difficult dynamics

Set Plant 1 Plant 2

I K Ke −s
s + 1 s + 1
K (1 + s) K (1 − s)
II
s + 1 s + 1
K K
III s − 1
s + 1

Within each set, the transfer functions exhibit similar AR behavior but different
phase behavior. This results in a relative difficulty in controlling such plants.
Department of Chemical Engineering I.I.T. Jodhpur, India

Basic Idea of Model based Control

Benefits of feedforward control are not normally achieved if the

disturbances are not measureable.
Question: Can we estimate the disturbances somehow and implement
feedforward control ?
Yes ! Use a model to predict the effect of the input. If the plant model is
perfect, the residual variation in the output is due to the disturbance.
This is the basic idea behind internal model control; Other highlights:
Explicit use of a model to design the controller.
Directly addresses the issue of model plant mismatch and
robustness
Department of Chemical Engineering I.I.T. Jodhpur, India

Internal Model Control

If the plant behaviour is y( s) = g ( s)u ( s) + d ( s) ,we want y=yd , therefore

u (s) =
1
 yd ( s ) − d ( s ) This is like a feedforward controller !
g (s)

However, this requires the value of d(s). Hence we estimate it using a

model as,
^
d ( s) = y − g m ( s)u ( s)

If we denote 1/g(s) by c(s), then we can get a controller law as,

 ^

u ( s ) = c( s )  yd ( s ) − d ( s )
 

This controller can be realized by the following block diagram.

Department of Chemical Engineering I.I.T. Jodhpur, India

Internal Model Control

d1

disturbance
d +
+
yd Plant y
Controller
- c(s) u + +
Model
-

𝑑መ

gc 1 − g mc
y( s) = yd + d
1 + c( g − g m ) 1 + c( g − g m )

servo regulatory
Department of Chemical Engineering I.I.T. Jodhpur, India

Properties of Internal Model Control

u ( s ) = c ( yd − d )
Nominal model stability
y = gc( yd − d ) + d

So long as g(s) is open loop stable and c(s) is stable, the closed
loop is stable.

Perfect Control: If gm=g and c=gm-1, then we get perfect control.

However, c= gm-1 is difficult to satisfy in practice.

Offset-free Control: So long as c(s) is chosen such that,

1
lim c( s ) =
s − − 0 g m ( 0)

IMC guarantees offset free control, regardless of model mismatch

Department of Chemical Engineering I.I.T. Jodhpur, India

Implementation issues

1) For c(s) to be stable, g(s) should not contain RHP zeros or

time delays. So, if they exist, factor them out (gm+)
+ −
gm = gm gm
2) Choose c(s) to be inverse of the gm-. This will inevitably not
be realizable, so multiply by a filter f(s)
1
c( s ) = − f ( s )
gm
1
f (s) =
(s + 1) n

3) Convert into regular feedback form and implement the

controller.
Department of Chemical Engineering I.I.T. Jodhpur, India

Interpretation of the IMC filter

f(s) is used for causality but can also be interpreted as a filter for
robustness in the presence of model plant mismatch.
Also, assuming gm=g, it can be shown that the servo transfer function
reduces to f(s). Thus, IMC is also equivalent to the direct synthesis
controller with f(s) as the desired closed loop transfer function.
Most advanced control schemes can be shown to be based on the
IMC principle, i.e. explicit use of the model in controller design. For
example, smith predictor scheme uses a model to negate the effects
of the delay.
Department of Chemical Engineering I.I.T. Jodhpur, India

Illustrative examples

K
Consider g (s) = s + 1 The IMC filter c(s)=1/g-m(s)f(s) and therefore,
1 s + 1
c( s ) =
K s + 1
Converting this back to feedback form,
  1
g c ( s) = 1 +  Which is a PI controller.
K  s 

If g(s) had a time delay, c(s) would still have the same form but gc(s)
would have a different form than the above.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 20 : Multivariable Systems

Department of Chemical Engineering I.I.T. Jodhpur, India

Multivariable Process Characteristics

Multivariable processes have more than one controlled variable and

manipulative variables.
Unlike single input-single output processes, multivariable, such
processes are characterised by interaction i.e. every
manipulative variable affects every other controlled variable.
Department of Chemical Engineering I.I.T. Jodhpur, India

Forms of representation

Transfer function
 y1 ( s)   g11 ( s) g12 ( s)   u1 ( s) 
 y ( s) =  g ( s) g ( s) u ( s)
 2   21 22  2 

Or, in general y(s)=G u(s)

State Space .
X = AX + Bu

y = CX + Du

The output y and inout u are vectors unlike in the SISO case
Department of Chemical Engineering I.I.T. Jodhpur, India

Zeros and Poles of Multivariable Systems

The poles of a multivariable system are the collection of all poles

of the individual transfer functions. They are also the eigenvalues
of the equivalent state space matrix A.

The zeros of the multivariable system are the solution of the

determinant of G(s) =0
Closed loop expressions
y( s) = ( I + GGc ) −1 GGc yd ( s) + ( I + GGc ) −1 Gd d ( s)
The equivalent denominator is actually a matrix and is called the
return difference matrix.
The determinant of (I+GGc)=0 gives the closed loop poles and
must be located in the LHP for stability.
Department of Chemical Engineering I.I.T. Jodhpur, India

Block diagram representation

+
u1 g11 y1
+
g12

g21
+
u2 g22 y2
+
Two options for control: a) Think “single loop” and use multiloop control
b) Look at decoupling and implement multivariable/ centralized control.
Choice depends on the analysis of Interaction
Department of Chemical Engineering I.I.T. Jodhpur, India

Interaction Analysis

1) Assume a loop pairing say y1-u1 and perform the following experiments
2) With all loops open, make a step change in u1 to u1+ u and measure
the change in output y1. We will term this as a direct effect.
3) With all loops except the u1-y1 loop closed, repeat the change in u1.
There will be change in y1 because of the direct effect but also there
will be a retaliatory effect because u2 changes to keep y2 constant. We
will term this change as  y1+ yr.
4) The ration of these two terms can be defined as λ11 (for the y1-u1
pairing) as 11=  y1/( y1+  yr.)
5) Depending on the values of this index for various assumed loop
pairings (step 1), decision will be taken on the final loop pairing.
6) This decision making is based on steady state analysis only.
Department of Chemical Engineering I.I.T. Jodhpur, India

Loop Interactions for a 2 x 2 system

Department of Chemical Engineering I.I.T. Jodhpur, India

Analysis of the Indices

If11 =1 retaliatory action is not present. So assumed loop pairing is
correct because there is no interaction from the other loop.
If 0 <  11 < 1 Þ retaliatory action is comparable to the direct action but is
in the same direction. The assumed loop pairing may be chosen only
if the index is closer to 1 (say 0.8).
If  11 =0, Retaliatory action is much greater than the direct action. The
assumed loop pairing is incorrect. The loop pairing u1-y2 is preferable.
If  11 > 1, Retaliatory action is in opposite direction to the direct action
but is smaller in magnitude than the direct. The assumed loop pairing
may be chosen only if the index is close to 1.
If  11 < 0, Retaliatory effect is larger and opposite in direction to the main
effect. Do not chose this loop pairing.
Department of Chemical Engineering I.I.T. Jodhpur, India

Relative Gain Array (RGA)

The collection together of all the ij in a matrix, it is called the RGA.
(y j / ui )loops _ open
i , j =
(y j / ui )loops _ except ( i , j ) closed
The RGA elements sum to 1 on a row and on a column. They are
a measure of steady state interaction.
The RGA elements are easily calculated from the steady state gain
matrix K. Denote by R, the inverse transpose of K. Then, ij =rij*kij

The loop pairing may be based on which ij are close to 1. If they
are generally closer to 0.5, this is an indication that the loops are
strongly interacting and so need to go centralized control. Else,
pair the ith input to the jth output for which ij is closer to 1.
Department of Chemical Engineering I.I.T. Jodhpur, India

RGA for non-square systems

For under-defined systems, leave out the variables that are least
likely to cause stability problems and then pair the loops.

For over-defined systems, look at various combinations of the inputs

that have favorable RGA (i.e. lesser interaction) and then pair the
loops.

For over-defined systems, look at various combinations of the inputs

that have favorable RGA (i.e. lesser interaction) and then pair the
loops.
Controller design will typically require detuning the loops to account
for interaction.
Department of Chemical Engineering I.I.T. Jodhpur, India

Lecture 12: AI in Process Control:

Process Identification
Department of Chemical Engineering I.I.T. Jodhpur, India

First principles Modeling

Model generated has high fidelity and is fairly accurate under various
scenarios.
Gives greater fundamental understanding of the process dynamics.
Need fairly accurate and complete knowledge of the process
parameters.
Not particularly useful for poorly understood process.
Very time consuming to generate.
Involves parameter estimation
Department of Chemical Engineering I.I.T. Jodhpur, India

Alternate Approaches to Model Building

Empirical model forms

Data + apriori Purely data

Knowledge based Heuristics based
based
Department of Chemical Engineering I.I.T. Jodhpur, India

Data + Apriori Knowledge based models

Examples: First order plus time delay models etc..

Models are generated using output/input data to estimate parameters of an
assumed model form.
Useful when partial knowledge of the process is available. For e.g., linear or
nonlinear, information regarding order of the process, magnitude of time
delay
Gives models that are relatively more fidel than purely data based methods.
Department of Chemical Engineering I.I.T. Jodhpur, India

Purely data based models

Such models are useful when no a priori knowledge of the process

is known in any form.
Typical examples are neural network (NN) models, wavelets or
Statistical models based on PLS (Projection to Latent Structures)
and/or NN.
Such models are based on explaining correlation in the data
collected. Hence, such models have poor fidelity if used for
extrapolation.
Department of Chemical Engineering I.I.T. Jodhpur, India

Qualitative information based models

Useful when process outputs/inputs cannot be exactly quantified.

Examples -- Models based on Fuzzy Logic.
Models are essentially databases that use qualifiers such as ‘more’ or
‘less’, ‘high’ or ‘low’, ‘increase’ or ‘decrease’, etc.
Typical applications are in Cement Kiln Control etc.
Department of Chemical Engineering I.I.T. Jodhpur, India

Data+ Apriori knowledge based methods

Unknown Plant
Dynamics Measured
Perturbation
signal Output

Steps
Assume a model form based on apriori knowledge
Choose a perturbation signal
Regress and estimate the model parameters
Cross validate the model.
Department of Chemical Engineering I.I.T. Jodhpur, India

Graphical methods

Restricted to simple step type perturbations

Perturb the input by a

unit step and generate
the response.
Graphically fit an
assumed model form
to the data
Department of Chemical Engineering I.I.T. Jodhpur, India

Graphical methods

Express in deviation
variable form

Estimate the gain

from the final value of
the normalized step
response
Department of Chemical Engineering I.I.T. Jodhpur, India

Estimation of the time delay

Get an estimate of the time

delay by considering the
time when the output
actually began to change. Output does
not respond
Input was until here
Higher order system response stepped up
can be approximated by a here
first order plus time delay q
q = 3 minutes
model.
Department of Chemical Engineering I.I.T. Jodhpur, India

Alternative graphical method

We know that the response to a step is y(t) = K (1-e-(t-)/)

with y=K.
y − y and therefore,
Consider that = e −( t − ) /
y
 y − y   t
ln  = −
 y   
time


Slope and
delay are  y − y 
ln 
estimated  y 
simultaneously Slope =-1/
Department of Chemical Engineering I.I.T. Jodhpur, India

Regression methods

Propose more complex candidate model structures and solve

the model parameters through regression.
A big advantage is that special perturbation tests other than the
step and more complex model forms can be used.

time

time
Department of Chemical Engineering I.I.T. Jodhpur, India

Regression methods

Candidate model forms could be of first order plus time delay or

second order and the model parameters could be estimated through
regression.

Regression could involve solving a linear or non-linear optimization

problem and is typically done by least squares error minimization.

Models developed using these methods cross-validate well, i.e give

accurate predictions for inputs that have not been “seen” during the
model building step.
Department of Chemical Engineering I.I.T. Jodhpur, India

Effect of model order

Suppose that the following plant

15
G plant ( s) =
( s + 1)(3s + 1)(5s + 1)

is approximated by
15e −1.5 s
Gml ( s ) =
8s + 1

Apparent mismatch in the dynamics could be more serious than it

appears above
Department of Chemical Engineering I.I.T. Jodhpur, India

Effect of model order

Mismatch is considerable at high frequencies, that is important from

control viewpoint.
Department of Chemical Engineering I.I.T. Jodhpur, India

Pulse Testing
For a model to be fidel and useful in closed loop control, it must
match the plant at important frequencies.
A step change is a long time perturbation that is relatively problematic
during plant operation. Also, a step change give fairly accurate
identification of the steady state gain but not of dynamics.
It can be shown that an ideal impulse (infinite height and
infinitesmal width can contain all frequencies. Thus, it is important
to analyze variants of the

A finite pulse of variable width and waveform could be used to bring

in the required dynamics. As well it is acceptable in closed loop
plant operation as the “change” is not long term.
Department of Chemical Engineering I.I.T. Jodhpur, India

Pulse Testing

Typical waveforms

h
h 2tw
2tw
-h 0
-h 0
Department of Chemical Engineering I.I.T. Jodhpur, India

Summary and study guide

1. Overview of various controller design schemes: Feedforward, Feedback, how to

construct block diagrams from schematics, component of control loops , multivariable
systems.
2. Analysis of system dynamics, Linearization, Laplace transforms, state space model
representations, how to convert from one form to the other. Analysis of first order, second
order, higher orders, time delays, inverse response and lead-lag behaviour.
3. Frequency response, Bode plots.
4. Process Identification methods – graphical and pulse testing based methods.
5. Feedback control, controller modes (P, PI, PID) and their relative merits and demerits,
anti-reset windup schemes, direct and reverse acting controllers, types of valves air-to-
open and air-to-close.
6. Stability aspects of feedback control, Routh array, Root locus, Bode and Nyquist
stability criteria, tuning PID controllers, Direct synthesis, Internal Model control.
Department of Chemical Engineering I.I.T. Jodhpur, India

Summary and study guide

7. Cascade and feed-forward control, when to use, design and tuning aspects

8. Multivariable systems, Analysis and concepts of zeros and poles and closed loop
transfer functions, RGA analysis, how to evaluate and interpret RGA, loop pairing,
decoupling control.
9. Model Predictive Control – preliminary concepts of prediction, correction and
optimization.