Dynamic Response Characteristics

of More Complicated Processes

Chapter 6

Dr. Yousef Alsunni

1
Introduction
• We discussed the dynamics of processes modeled by either first- or second-order transfer
functions or an integrator.
• In this chapter, we consider more complex transfer function models that include additional
time constants in the denominator and/or functions of 𝑠 in the numerator.
• We show that the forms of the numerator and denominator of the transfer function model
influence the dynamic behavior of the process.
• We also introduce a very important concept, the time delay, and consider the approximation
of complicated transfer function models by simpler, low-order models.
• Additional topics in this chapter include interacting processes and processes with multiple
inputs and outputs (MIMO).

Dr. Yousef Alsunni

2
Poles and Zeros and Their Effect on Process Response
We have seen that processes response characteristics are determined by the factors of the transfer
function denominator. For example, consider a transfer function:
where 0 ≤ 𝜁 < 1 “underdamped”

we know that the response of this system to any input will contain the following functions of time:
• A constant term resulting from the 𝑠 factor
• An 𝑒 / term resulting from the (τ 𝑠 + 1) factor

• 𝑒 / sin 𝑡

and terms resulting from the (𝜏 𝑠 + 2𝜁𝜏 𝑠 + 1) factor

• 𝑒 / co𝑠 𝑡

Dr. Yousef Alsunni

3
Poles and Zeros and Their Effect on Process Response
where 0 ≤ 𝜁 < 1 “underdamped”

The roots of the factors of the denominator polynomial are:

Control engineers refer to the values of 𝑠 that are roots of the denominator polynomial as the poles of
transfer function 𝐺(𝑠). It is useful to plot the poles and to discuss process response characteristics in
terms of poles locations in the complex 𝑠 plane.
Dr. Yousef Alsunni

4
Poles and Zeros and Their Effect on Process Response
The real pole is closer to the imaginary axis than the complex pair,
indicating a slower response mode (𝑒 / decays slower than 𝑒 / ).
In general, the speed of response for a given mode increases as the pole
location moves farther away from the imaginary axis. “larger 𝜏 ⇒ slower
process and closer pole −1⁄𝜏 to the imaginary axis”.

A pole to the right of the imaginary axis (called a right-half plane pole),
for example, 𝑠 = +1/𝜏, indicates that one of the system response
modes is 𝑒 / . This mode grows without bound as 𝑡 becomes large, a
characteristic of unstable systems.

The complex conjugate poles indicate that the response will contain sine
and cosine terms; that is, it will exhibit oscillatory modes.

Dr. Yousef Alsunni

5
Poles and Zeros and Their Effect on Process Response
𝐼𝑚(𝑠)

𝑅𝑒(𝑠)

Response for impulse input (𝛿(𝑡)) various poles locations in the s plane

Dr. Yousef Alsunni

6
Poles and Zeros and Their Effect on Process Response
The dynamics of a process are affected not only by the poles of 𝐺(𝑠), but also by the values of 𝑠 that
cause the numerator of 𝐺(𝑠) to become zero. These values are called the zeros of 𝐺(𝑠).
The standard form of the transfer function:

The above equation can also be written as:

𝑧 and 𝑝 are zeros and poles, respectively.

It is convenient to express transfer functions in gain/time constant form:
If all factors are real roots:

Dr. Yousef Alsunni

7
Poles and Zeros and Their Effect on Process Response

For practical control systems the number of zeros must be less than or equal to the number of
poles (𝑚 ≤ 𝑛) “as discussed earlier”. When 𝑚 = 𝑛, the output response is discontinuous after a
step input change, as illustrated by the following Example.

Dr. Yousef Alsunni

8
Example
Calculate the response to a step change of magnitude 𝑀 for the process modeled by the transfer
function:
𝐾(𝜏 𝑠 + 1)
𝐺 𝑠 =
𝜏 𝑠+1

Note that 𝒎 = 𝒏 = 𝟏

( )
𝑌 𝑠 =𝐺 𝑠 𝑈 𝑠 =𝐺 𝑠 = ⇒ 𝑌 𝑠 = 𝐾𝑀 +
( )

⁄
⇒ 𝑦 𝑡 = 𝐾𝑀 1 − 1 − 𝑒

Note that 𝑦 0 = 𝐾𝑀(𝜏 ⁄𝜏 ) while from the initial condition 𝑦 0 = 0 (The process initially at steady
state and 𝑦 is a deviation variable). Hence, 𝑦(𝑡) charges suddenly at 𝑡 = 0 i.e. the output 𝑦(𝑡) is
discontinuous at 𝑡 = 0 (when the step change occurs)

Dr. Yousef Alsunni

9
Example
⁄
𝑦 𝑡 = 𝐾𝑀 1 − 1 − 𝑒

Figure (a) shows the response for τ = 4 and five different values of 𝜏 .

Here we notice two distinguishable cases:

(𝜏 > 𝜏 ) → negative slope @ 𝑡 = 0
and (𝜏 < 𝜏 ) → positive slope @ 𝑡 = 0

= 𝐾𝑀

Dr. Yousef Alsunni

10
Example
𝐾(𝜏 𝑠 + 1)
𝐺 𝑠 =
𝜏 𝑠+1

Figure (b) is a pole-zero plot showing the location of the single system zero, 𝑠 = −1/𝜏 , for each of these three
cases.
Note that, if 𝜏 = 𝜏 , the transfer function simplifies to 𝐾 as a result of cancellation of numerator and denominator
terms, which is a pole-zero cancellation.

x is a pole location and □ is a location of single zero

Dr. Yousef Alsunni

11
Second-Order Processes with Numerator Dynamics
The previous example shows that the presence of a zero in the first-order system causes a jump
discontinuity in 𝑦(𝑡) at 𝑡 = 0 when the step input is applied. Such an instantaneous step
response is possible only when the numerator and denominator polynomials have the same
order (𝑛 = 𝑚).

As we mentioned, a physically realizable system should satisfy 𝑛 ≥ 𝑚.

Industrial processes have higher-order dynamics in the denominator (𝑛 > 𝑚), causing them to
exhibit some degree of inertia. This feature prevents them from responding instantaneously to
any input, including an impulse input.

Dr. Yousef Alsunni

12
Example
For the case of a single zero in an overdamped second-order transfer function,

𝐾(𝜏 𝑠 + 1)
𝐺 𝑠 =
(𝜏 𝑠 + 1)(𝜏 𝑠 + 1)
calculate the response to a step input of magnitude 𝑀 and plot the results for 𝜏 = 4, 𝜏 = 1 and several
values of 𝜏 ·

( ) ⁄ ⁄
∵𝑌 𝑠 =𝐺 𝑠 𝑈 𝑠 = ⇒ 𝑦 𝑡 = 𝐾𝑀 1 + 𝑒 + 𝑒 𝜏 = 4, 𝜏 = 1
( )( )

Note that 𝑦 𝑡 → ∞ = 𝐾𝑀. Hence, the effect of including the single zero does not change the final value, nor does
it change the number or locations of the poles. But the zero does affect how the response modes (exponential
terms) are weighted in the solution.

Dr. Yousef Alsunni

13
Example
𝝉𝒂 𝝉𝟏 ⁄ 𝝉𝒂 𝝉𝟐 ⁄
𝑦 𝑡 = 𝐾𝑀 1 + 𝑒 − 𝑒 𝜏 = 4, 𝜏 = 1
𝝉𝟏 𝝉𝟐 𝝉𝟏 𝝉𝟐

Three types of responses are involved here, as illustrated

for eight values of 𝜏 .

Case (i) shows that overshoot can occur if 𝜏 is sufficiently

large. Case (ii) is similar to a first-order process response.
Case (iii), which has a positive zero, also called a right-half
plane zero, exhibits an inverse response which occurs when
the initial response to a step input is in one direction but
the final steady state is in the opposite direction. For
example, for case (iii), the initial response is in the negative
direction while the new steady state 𝑦(∞) is in the positive
direction in the sense that 𝑦(∞) > 𝑦(0). Inverse responses
are associated with right-half plane zeros.
The direction of the initial response can be determined
from the slope of 𝑦 𝑡 | which is (𝑑𝑦 ⁄𝑑𝑡)| .
Dr. Yousef Alsunni

14
Example

⁄ ⁄
∵ =1+ 𝑒 + 𝑒

⇒ = + =

∵ 𝜏 > 0 and 𝜏 > 0

∴ The slope of 𝑦(𝑡)/𝐾𝑀 at 𝑡 = 0 has the sign of 𝜏

Dr. Yousef Alsunni

15
Second-Order Processes with Numerator Dynamics
The phenomenon of overshoot or inverse response results from the zero in the above example
and will not occur for an overdamped second-order transfer function containing two poles but
no zero.
Inverse response or overshoot can be expected whenever two physical effects act on the process
output variable in different ways and with different time scales.
Next, we show that inverse responses can occur for two first-order transfer functions in a parallel
arrangement, as shown in the Figure below. The relationship between 𝑌(𝑠) and 𝑈(𝑠) can be
expressed as

Dr. Yousef Alsunni

16
Second-Order Processes with Numerator Dynamics

After rearranging the numerator into standard gain/time constant form, we have:

The existence of a right half plane zero is a condition to have an inverse response. Thus, 𝜏 < 0:

⇒ For either positive or negative value of 𝑲

This indicates that 𝐾 and 𝐾 have opposite signs, because 𝜏 > 0 and 𝜏 > 0.
The sign of the overall gain 𝐾 is the sing of the slow process (the process with larger 𝜏).
Dr. Yousef Alsunni

17
Processes with Time Delays
Whenever material or energy is physically moved in a process or plant, there is a time delay
associated with the movement. For example, if a fluid is transported through a pipe in plug flow,
as shown in the Figure below, then the transportation time between points 1 and 2 is given by:

a time delay sometimes is referred to as a distance-

velocity lag. Other synonyms are transportation lag,
transport delay, and dead time.

Dr. Yousef Alsunni

18
Processes with Time Delays
Suppose that 𝑥 is some fluid property at point 1, such as concentration, and 𝑦 is the same
property at point 2 and that both 𝑥 and 𝑦 are deviation variables. Then they are related by a time
delay 𝜃

The above equation can be written as:

𝑦 𝑡 = 𝑥 𝑡 − 𝜃 𝑆(𝑡 − 𝜃)
Taking Laplace Transform and rearranging to get
the transfer function:

Dr. Yousef Alsunni

19
Processes with Time Delays

Even when the plug flow assumption is not valid,

transportation processes usually can be modeled
approximately by the transfer function for a time
delay. However, a more general approach is to model
the flow process as a first-order plus time-delay
transfer function:

where 𝜏 is a time constant associated with the

degree of mixing in the pipe or channel.

Dr. Yousef Alsunni

20
Polynomial Approximations to 𝑒

The exponential form 𝑒 is a nonrational transfer function, i.e. it is not a ratio of two
polynomials in 𝑠. Consequently, it cannot be factored into poles and zeros, a convenient form for
analysis. However, it is possible to approximate 𝑒 by polynomials using either a Taylor series
expansion or a Padé approximation.
The Taylor series expansion for 𝑒 (around 0) is:

The Padé approximation for a time delay is a ratio of two polynomials in 𝑠 with coefficients
selected to match the terms of a truncated Taylor series expansion of 𝑒 . The simplest pole-
zero approximation is the 1/1 Padé approximation:

Dr. Yousef Alsunni

21
Polynomial Approximations to 𝑒
There are higher-order Padé approximations, for example, the 2/2 Padé approximation:

The Figure below illustrates the response of the 1/1 and 2/2 Padé approximations to a unit step
input.
Note the discontinuous response when using Padé
approximation because 𝑚 = 𝑛 . The second-order
approximation is somewhat more accurate.

Neither approximation can accurately represent the

discontinuous change in the step input very well.
Step response of 1/1 and 2/2 Padé
approximations of a time delay (𝐺 (𝑠) and
𝐺 (𝑠), respectively).
Dr. Yousef Alsunni

22
Polynomial Approximations to 𝑒
However, if the response of a first-order system with time delay is considered,

The Figure below shows that the approximations are satisfactory for a step response.

Thus, although neither approximation can accurately

represent the step input, they are satisfactory
accurate for a step response for a first-order system
with time delay.

Step response of a first-order plus time delay

process (𝜃 = 0.25𝜏) using 1/1 and 2/2 Padé
approximations of 𝑒 .
Dr. Yousef Alsunni

23
Approximation of Higher-Order Transfer Functions
In this section, we present a general approach for approximating higher-order transfer function
models with lower-order models that have similar dynamic and steady-state characteristics.
For small values of 𝑠, Taylor series expansion truncated after the first-order term provides a
suitable approximation for the time delay transfer function:

An alternative first-order approximation consists of the transfer function:

These expressions can be reversed to approximate the pole or zero on the right-hand side of the
equation by the time-delay term on the left side.

Dr. Yousef Alsunni

24
Skogestad's "Half Rule"

Skogestad (2003) has proposed a related approximation method for higher-order models that
contain multiple time constants. He approximates the largest neglected time constant in the
denominator in the following manner. One-half of its value is added to the existing time delay (if
any), and the other half is added to the smallest retained time constant. Time constants that are
smaller than the largest neglected time constant are approximated as time delays using (𝑒 =
1⁄(1 + 𝜃𝑠)). A right-half plane zero is approximated by (𝑒 = 1 − 𝜃𝑠).
𝐾 1 − 𝜏 𝑠 1 − 𝜏 𝑠 …𝑒
𝐺 𝑠 = 𝜏 >𝜏 >𝜏 >⋯
𝜏 𝑠+1 𝜏 𝑠+1 𝜏 𝑠+1 …
Approximate as first order → larges neglected 𝜏 is 𝜏
𝝉𝟐
⋯
𝐾𝑒 𝟐
⇒𝐺 𝑠 ≈
𝝉𝟐
𝜏 + 𝑠+1
𝟐
Dr. Yousef Alsunni

25
Example
Consider a transfer function:
𝐾(−0.1𝑠 + 1)
𝐺 𝑠 =
(5𝑠 + 1)(3𝑠 + 1)(0.5𝑠 + 1)
Derive an approximate first-order-plus-time-delay model,
𝐾𝑒
𝐺 𝑠 =
𝜏𝑠 + 1
using two methods:
(a) The Taylor series expansions.
(b) Skogestad's half rule.
Compare the normalized responses of 𝐺(𝑠) and the approximate models for a unit step input.
(a) The dominant time constant (5) is retained. Apply the approximations (𝑒 = 1⁄(1 + 𝜃𝑠)) and (𝑒 = 1 − 𝜃𝑠) :

and

Substitution into the transfer function gives the Taylor series approximation, 𝐺 𝑠 :

Dr. Yousef Alsunni

26
Example

𝐾(−0.1𝑠 + 1)
𝐺 𝑠 =
(5𝑠 + 1)(3𝑠 + 1)(0.5𝑠 + 1)

(b) The largest neglected time constant in has a value of 3. Thus, the new time constant and time delay are:

𝐾𝑒 .
𝜏 =5+ = 6.5 and 𝜃 = 0.1 + + 0.5 = 2.1 ⇒𝐺 𝑠 =
6.5 𝑠 + 1

Skogestad's method provides better agreement

with the actual response.

Dr. Yousef Alsunni

27
Example
Consider the following transfer function:

𝐾 1−𝑠 𝑒
𝐺 𝑠 =
(12𝑠 + 1)(3𝑠 + 1)(0.2𝑠 + 1)(0.05𝑠 + 1)
Use Skogestad's method to derive two approximate models:
(a) A first-order-plus-time-delay model
(b) A second-order-plus-time-delay model
Compare the normalized output responses for 𝐺(𝑠) and the approximate models to a unit step input.
(a) The largest neglected time constant in has a value of 3. Thus, the new time constant and time delay are:

𝐾𝑒 .
𝜏 = 12 + = 13.5 and 𝜃 =1+1+ + 0.2 + 0.05 = 3.75 ⇒𝐺 𝑠 =
13.5 𝑠 + 1

(b) The largest neglected time constant in has a value of 0.2. Thus, the new time constants and time delay are:

. . 𝐾𝑒 .
𝜏 = 12 , 𝜏 = 3 + = 3.1 and 𝜃 =1+1+ + 0.05 = 2.15 ⇒𝐺 𝑠 =
(12 𝑠 + 1)(3.1 𝑠 + 1)

Dr. Yousef Alsunni

28
Example

The actual and second-order model responses are

almost indistinguishable.

Dr. Yousef Alsunni

29
Interacting and Noninteracting Processes
For a noninteracting process, changes in a downstream unit have no effect on upstream units. By
contrast, for an interacting process, downstream units affect upstream units, and vice versa.
An example of a system that does not exhibit interaction is storage tanks connected in series in
such a way that liquid level in the following tank did not influence the level in the preceding, see
the Figure below.
For two tanks connected in series. The following transfer
functions relating tank levels and flows were derived as:

A series configuration of n noninteracting tanks.

where 𝐾 = 𝑅 , 𝐾 = 𝑅 , 𝜏 = 𝐴 𝑅 , 𝜏 = 𝐴 𝑅
Dr. Yousef Alsunni

30
Interacting and Noninteracting Processes
Consider an example of an interacting process that is similar to the two-tank process discussed
earlier. The process shown in the Figure below is called an interacting system because ℎ
depends on ℎ (and vice versa) as a result of the interconnecting stream with flow rate 𝑞 .
𝒉𝟏
Not 𝒒𝟏 =
𝑹𝟏

Thus, a more complicated expression results:

It is of the form:

Two tanks in series whose liquid levels interact.

Dr. Yousef Alsunni

31
Interacting and Noninteracting Processes
or

It can be shown that 𝜁 > 1 by analyzing the denominator of the two equation; hence, the
transfer function is overdamped and second-order, and has a negative zero.
The transfer function relating ℎ and ℎ ,

Consequently, the overall transfer function between 𝐻 and 𝑄 is:

The denominator polynomial can no longer be factored into two first-order terms, each
associated with a single tank. Hence, the interacting two-tank system is more complicated than
that for the noninteracting system.
Dr. Yousef Alsunni

32
Multiple-Input, Multiple-Output (MIMO) Processes
Most industrial process control applications involve a number of input (manipulated) variables
and output (controlled) variables. These applications are referred to as multiple-input/multiple-
output (MIMO) systems to distinguish them from the single-input/single-output (SISO) systems
that have been emphasized so far.
For example, consider the thermal mixing process
shown in the Figure.
Controlled variables: ℎ and 𝑇
Manipulated variables: 𝑤 and 𝑤
Disturbance variables: 𝑇 and 𝑇
Constant parameters: 𝑤 and all liquid properties
The energy and mass balances for this process are:

A multi-input, multi-output thermal mixing process.

Dr. Yousef Alsunni

33
Multiple-Input, Multiple-Output (MIMO) Processes
After linearizing the differential equations, putting them in deviation form, and taking Laplace
transforms, we obtain a set of eight transfer functions that describe the effect of each input
variable (𝑤 , 𝑤 , 𝑇 , 𝑇 ) on each output variable (𝑇′ and ℎ′):

where 𝜏 = 𝜌𝐴ℎ/𝑤 is the average residence time in the tank.

• all four inputs affect the tank temperature through first-order

transfer functions and there is a single time constant that is
the nominal residence time of the tank 𝜏.

• The inlet flow rates affect level through integrating transfer

functions that result from the pump on the exit line.

• It is clear that inlet temperature changes have no effect on

liquid level.
Dr. Yousef Alsunni

34
Multiple-Input, Multiple-Output (MIMO) Processes
A very compact way of expressing these equations is by means of a
transfer function matrix in which two transfer function matrices can
be used to separate the manipulated variables, 𝑤 and 𝑤 , from the
disturbance variables, 𝑇 and 𝑇 :

Dr. Yousef Alsunni

35
Multiple-Input, Multiple-Output (MIMO) Processes

The block diagram illustrates how the four input

variables affect the two output variables.

Final remark:
The development of physically-based MIMO models
can require a significant effort. Thus, empirical models
rather than theoretical models often must be used for
complicated processes.
Dr. Yousef Alsunni

36