EI6801 Computer Control of Processes

April 2018
Part –A
1. Draw the schematic of reconstruction of continuous signal using zero order hold and
first order hold.

2. Give the state equation of a SISO, second order discrete time system.

3. What are the requirements of a parameter estimation problem?

a) Input-Output data from the process b) A class of models c) A criterion
4. What is the principle of least squares in an estimator? Give the expression for penalty
function J.
The least squares principle states that the SRF (Sample Regression Function) should be
constructed (with the constant and slope values) so that the sum of the squared distance between
the observed values of your dependent variable and the values estimated from your SRF is
minimized (the smallest possible value).

1
 The principle of least squares is that the unknown parameters of a model should be chosen
in such a way that the sum of the squares of the differences between the actually observed
and computed values multiplied by numbers that measure the degree of precision is a
minimum.
1 1 2
J ( )   T   
2 2
where,   y  yˆ
and yˆ  
5. Determine the pulse transfer function of
0.5 0.2  2
x(kh  h)   
0  1  u (kh)
x ( kh )
0  
y(kh)  1 0 x(kh)

6. Using Lyapunov analysis, determine the stability of the equilibrium state of the
system
 0 0.5
X (k  1)  F x(k ) with F   
 0.5 1 
Out of Syllabus
7. What are the implications when the RGA elements.
When ij  1 ;
An input-output paring such that the diagonal elements of the RGA matrix are close to one is
preferable since this paring is related with a diagonal dominant plant.
When ij  0 ; It shows no interaction between the loops.
8. Differentiate between multi loop control and multivariable control.
Multiloop control: Each manipulated variable depends on only a single controlled variable, i.e., a
set of conventional feedback controllers.
Multivariable Control: Each manipulated variable can depend on two or more of the controlled
variables. Examples: decoupling control, model predictive control.
9. List the strengths and weaknesses of Generalized Predictive Controller.
Advantages:
 GPC is robust against model mismatching and unexpected dynamics and superior
to PID control due to the pre-calculation of system behavior.
 It is a competitive control algorithm through its inherent noise rejection without
additional filters.
Disadvantages:
 It experiences higher demand on calculation time.
10. Compare the classical feedback control and internal model control.
A feedback control system maintains a prescriptive relationship between the process output and set
point by comparing them and using the error signal as a means of control. It is the simplest form of
closed loop control scheme. Feedback control system has many daily routine applications; for
instance, consider an automobile speed control or an air conditioner temperature control system
which uses the difference between the actual and the desired speed or temperature to change the
manipulated variable. Since the system output is used to regulate its input, such a device is said to be
a closed-loop control system. The block diagram of a feedback control system is shown below,

2
 The IMC philosophy relies on the internal model principle, which states: Accurate control can be
achieved only if the control systems encapsulates (either implicitly or explicitly) some representation
of the process to be controlled.

Comparison IMC structure and classical feedback structure: both structures are equivalent.

Q( s)
C ( s) 
1  G ( s) Q(s)
C ( s)
Q( s) 
1  G ( s)C (s)

Part –B
11
a

Ans:

3
  2 1 4
F   , g 
 1 1  3

4 5 
U  g Fg    
 3 1

1  p  1  3 4 
p1  3 4 ; P   p F1   10 7 
19  1  19  
zI  F  z  3z  1
2

 1  1 1
zI   F  gk    z  j   z  j   z 2 
 2  2 4
 3  111 18 
k  3 P  
4   76 19 
b)
xˆ (k  1)   F  mc  xˆ (k )   g  md  u (k )  my (k )
c  1 1
 (k  1)  F T  (k )  cT (k )
 (k )  mT  (k )
1 1 
U  cT F T cT    
1 0 
p1  1 1
 p  1 1
P   1T   
 p1 F  3 2 
zI  F T  z 2  3 z  1

 
zI  F T  mT cT  z 2
mT   1 3 P  8 5

c)
x(k  1)  Fx(k )  gu (k )
u (k )  kxˆ (k )
ˆ  1)   F  mc  xˆ (k )  gu (k )  my (k )
x(k
y (k )  cx(k )  du (k )
x1 (k  1)  2 x1 (k )  x2 (k )  5.8 xˆ1  3.79 xˆ2
x2 (k  1)   x1 (k )  x2 (k )  4.38 xˆ1  2.84 xˆ2
and

4
 xˆ1 (k  1)  11.84 xˆ1 (k )  5.21xˆ2 (k )  8  x1 (k )  x2 (k ) 
xˆ2 (k  1)  0.38 xˆ1 (k )  8.84 xˆ2 (k )  5  x1 (k )  x2 ( k ) 

OR
11
b

Ans:
The system is completely controllable if and only if none of the elements of the column
matrix g is zero.
The system is completely observable if and only if none of the elements of the row matrix c
is zero.

5
 i) The system is not controllable. The system is observable.
ii) The system controllable. The system is NOT observable.
12
a

Ans:

6
Parameters in discrete-time model can be estimated directly from input-output data based on
linear regression.
y(k )  a1 y(k 1)  a2 y(k  2)  b1u(k 1)  b2u(k  2)
As a specific example, consider the second-order difference equation above. It can be used
to predict y(k) from data available at time (k-1) and (k-2)
In developing a discrete-time model, model parameters a1, a2 and b2 are considered to be
unknown.
Equivalently, S can be expressed as,
S   Y  X   T Y  X  
Where the superscript T denotes the matrix transpose and
 Y1   1 
 
Y   ...  ,    ... 
Yn   p 
 
The least squares solution

ˆ   X T X  X T X
1

 T   a1 a2 a3 a4  , X 1  y (k  1) , X 2  y (k  2)
X 3  u (k  1) , X 4  u (k  2)
t  1 ; for k  0, y(k )  0 and u(k )  0

 0.058   0.058 0 1 0
 0.217  0.217 0 1 1 
  
 0.36   0.36 0 1 1
   
 0.488   0.488 0.058 1 1 
 0.6   0.6 0.217 1 1 
y  y 
 0.692   0.692 0.36 1 1 
 0.772   0.772 0.488 1 1 
   
 0.833   0.833 0.6 1 1
 0.888   0.888 0.692 1 1 
   
 0.925   0.925 0.772 1 1 

Solving
a1  0.984, a2  0.122, b1  0.058, b2  0.101, k  1.159

OR

7
12
b

Ans:

SPECTRAL ANALYSIS
The form of model used in spectral analysis is

Y t h K u t K v t
K 0
(1)
Where,
Y(t) = Output signal
u(t) = Input signal
h (K) = Weighting sequence
v t = Disturbance term.

The system described in (1) is shown in Figure

Taking discrete fourier transforms for the system described in (1) the
following relation for the spectral densities can be derived from cross covariance function
described in (1.1)

ryu ( ) h K ru K (1.1)
K 0

i
Yu H e u

(2)

Yu = Cross spectral density between Input u(t) and Output Y(t)

8
 1 i
Yu rYu ( )e
2
(3)

u = Spectral density of input u(t)

1 i
ru e
2 k 0

(4)

i
H(e ) = Transfer function
iK
h(K)e
k 0

i
Now the transfer function H(e ) can be estimated from (2) as

i
ˆ ( )
Yu
Ĥ( e )
ˆ ( )
u
(5)

Conduct an experiment and collect ‘N’ number of Input u(t) and Output Y(t) data. Then
cross spectral density Yu ( ) and spectral density u ( ) can be estimated from
experimental data as

N
ˆ 1 i
Yu r̂Yu e
2 N

(6)

N
ˆ 1 i
u r̂u e
2 N

(7)

N min ( ,0)
1
r̂Yu Y t u t 0, 1, 2
N t 1 min ( ,0)

(7.1)
Using (7.1) equation (6) can be written as

9
 N N max( ,0)
ˆ 1 i (8)
Yu
2 N
Y t u(t) e
N
t 1 min( ,0)

Next make the substitution s t . Figure (1) illustrates how to derive the limits for the
new summation index.
Since (e i ) e 1(s t)

is
e . eit
We get,

Figure 1. Change of summation variables

(Summation is over the shaded area)
N N
ˆ 1 is
Yu Y s u(t)e .eit
2 N s 1 t 1

1
YN UN
2 N
Where,
N
is
YN y(s)e = Discrete fourier transform of y(t)
s 1

N
is
UN u(s)e = Discrete fourier transform of u(t)
s 1

2 4
For 0, , ....., , YN and U N can be computed efficiently using Fast
N N
Fourier Transform (FFT) algorithms. Then using (4.30) we can estimate cross spectral density
ˆ .
Yu
Similarly ,

10
 N N
ˆ 1 is it
u
2 N
u(s) u(t)e .e
s 1 t 1

1 1 2
UN UN UN ( )
2 N 2 N
(9)

By computing U N by using FFT algorithm, we can estimate spectral density ˆ u .

Then we can estimate transfer function

i
ˆ ( )
Yu
Ĥ( e )
ˆ ( )
u

1
YN UN
2 N
1
UN UN
2 N

YN
Ĥ ei
UN
(10)

This transfer function is called empirical transfer function estimated.

CORRELATION ANALYSIS
The form of model used in correlation analysis is

Y t h K u t K v t (2)
K 0
Where,
Y(t) = Output signal
u(t) = Input signal
h (K) = Weighting sequence
v t = Disturbance term.

11
 Figure 1. System used in correlation analysis

Assume that input u(t) is a stationary stochastic process (white noise) which is
independent of the disturbance v t . Then the following relation holds for the cross
covariance function.

ryu h K ru -K
k 0

(2)
Where,
ryu = Cross covariance function between output Y(t) and input
u(t)
= E Y(t ) u(t) rˆu rˆu
ru = Cross covariance function of input u(t)
= E u(t ) u(t)
Note: E is called expected value = mean value
Conduct an experiment and collect Input u(t) and Output Y(t) data.
The covariance functions in (2) can be estimated from the Input and Output data as

N min ( ,0)
1
r̂Yu Y t u t 0, 1, 2
N t 1 min ( ,0)

(3)

N
1
r̂u u t u t
N t 1

rˆu rˆu 0, 1, 2

Where,
N = Number of experimental data
r̂Yu = Estimated covariance function between Y(t) and
u t
12
 from Input and Output data.
r̂u = Estimated covariance function of u t from Input data.

Then an estimate ĥ(K) of the weighting function h(K) can be determined by solving
following equation

rˆyu hˆ K rˆ u -K
k 0

(4)

rˆu 0 ... rˆu

r̂Yu 0 hˆ 0
. r̂u 1 ... . .

rˆYu rˆu ... rˆu 0 ĥ

Solving above infinite dimensional equation is very difficult. This problem can be
2
simplified by using white noise (mean value = 0, variamce 1 as input
For white noise,
r̂u = 0, for t = 1, 2…

r̂u = Constant value for = 0 (i.e non zero value)

r̂u 0 = Constant value

Since input is a white noise, the equation (4) can be written as

rYu k
h (K ) K = 0, 1… (5)
ru k

rYu 0
ru 0
h 0
rYu 1
h1
. ru 0
h .

rYu
ru 0

13
 From the above equation the weighting function h(K) i.e, h 0 , h 1 ,.....h
can be easily estimated.

13
a

Ans:

14
OR

15
13
B

Ans:

16
14. Consider the following process with three inputs and three outputs.
a) 1 2 0.1
y1 ( s )  m1 ( s )  m2 ( s )  m3 ( s )
i) s 1 ( s  2)( s  3) 3s  1
5 1
y2 ( s )  2 m2 ( s )  m3 ( s )
s  2s  1 s 1
10 7 15
y3 ( s )  m1 ( s )  m2 ( s )  m3 ( s )
5s  1 (4 s  1)(0.1s  1) ( s  1)( s  5)
Using Bristol’s relative gain array, select the control loops with minimum steady state
interaction.
Ans:

17
18
14. Select the control loops with minimum steady state interaction for the following
a) system with two outputs and three inputs.
ii) 3 0.5 1
y1 ( s )  m1 ( s )  m2 ( s )  2 m3 ( s )
2s  1 ( s  1)( s  3) s  3s  2
2 4
y2 ( s )  10m1 ( s)  m2 ( s)  m3 ( s)
s 1 ( s  1)(3s  1)
Ans:
19
20
 OR

14. Explain the significant difference between partial decoupling and static decoupling.
b) Ans:
i) In partial decoupling, the decoupling matrix for a 2x2 example is restricted to the following
 1 d12 ( s ) 
form, D( s )  
1 
. Here we specify a decoupled response and the decoupler
 d 21 ( s )
with the structure above.
 * 
 g11 ( s ) 0 
G p ( s) D( s) 
 * 
 0 g 22 ( s) 
 * 
 g11 ( s ) g12 ( s )   1 d12 ( s )   g11 ( s ) 0 
 g ( s) 
 21 g 22 ( s )   d 21 ( s ) 1   * 
 0 g 22 ( s ) 

21
And we can solve four equations in four unknowns to find,
g ( s)
d12 ( s )   12
g11 ( s )
g 21 ( s )
d 21 ( s )  
g 22 ( s )
* g12 ( s) g 21 ( s)
g11 ( s)  g11 ( s) 
g 22 ( s )
* g 21 ( s ) g12 ( s )
g 22 ( s )  g 22 ( s) 
g11 ( s )
The effect of the synthetic inputs are being fed forward to minimize control loop
interaction. Some factorization of the decoupling elements must be performed to make
certain that they are stable and physically realizable. The feedback control diagram is as
shown below. The result is a true multivariable strategy.
In practice, static decoupling based on steady state gains is used much more often than
dynamic decoupling, since the dynamic version may not be physically realizable or the
dynamic parameters may be too uncertain. In this case simply set s=0 in the decoupling
equations.

22
14. Explain the strategies for reducing control loop interactions.
b) Ans:
ii) When the designer is confronted with two strongly interacting loops, he introduces in the
control system special new elements called decouplers. The purpose of decouplers is to cancel
the interaction effects between the two loops and thus render two non-interacting control
loops. Consider the process whose input-output relationships are given the following
equations,
y1 ( s )  H11 ( s )m1 ( s )  H12 ( s )m2 ( s )
y2 ( s )  H 21 ( s )m1 ( s )  H 22 ( s )m2 ( s )
Form two loops by coupling m1 with y1 and m2 with y2. Assume that initially noth outputs are
at the desired set point values. Suppose that a disturbance or set point change causes the
controller of loop 2 to vary the value of m2. This will create an undesired disturbance for loop

23
1 and will cause y1 to deviate from its desired value. From the above equations, we find that
in order to keep y1 constant m1 should change by the following amount.
H (s)
m1   12 m2 This implies that we can introduce a dynamic element with transfer
H11 ( s )
H (s)
function D1 ( s )  12 which uses the value of m2 as input and provides as output the
H11 ( s )
amount by which we should change m1 in order to cancel the effect of m2 on y1. This dynamic
element is called decoupler and when it is installed in the control system it cancels any effect
that loop 2 might have on loop 1, but not vice versa. To eliminate the interaction from loop 1
to loop 2, we can follow the same reasoning as above and we find that the transfer function
H (s)
of the second decoupler is given by D2 ( s )  21 . The following two closed loop input-
H 22 ( s )
output relationships is as follows;
Gc1[ H11  H12 H 21 / H 22 ]
y1  y1sp
1  Gc1[ H11  H12 H 21 / H 22 ]
Gc 2 [ H 22  H12 H 21 / H11 ]
y2  y2 sp
1  Gc 2 [ H 22  H12 H 21 / H11 ]
The last two equations demonstrate complete decoupling of the two loops since the controlled
variable of each loop depends only on its own set point and not on the set point of the other
loop.

24
25
 Explain the design procedure for an IMC.
15. Ans:
a) The internal model principle and structure
i) The IMC philosophy relies on the internal model principle, which states: Accurate control can be
achieved only if the control systems encapsulates (either implicitly or explicitly) some
representation of the process to be controlled.

Suppose ˜ G(s)is a model of G(s).

• By setting C(s)as the inverse of the model:
C(s)= ˜ G(s)−1
And assuming that
˜ G(s)= G(s)
Then the output y(t)will track the reference input yd(t) perfectly.
• However in this example no feedback.

15. Explain the IMC tuning procedure.

a) Ans:
ii) The IMC design procedure for SISO systems is identical to the design procedure that we
developed for open-loop controller design. Model uncertainty is handled by adjusting the
filter design filter factor for robustness and speed of response. The IMC design procedure
consists of the following four steps;
~ ~ ~
1. Factor the process into invertible and non-invertible elements. g p ( s)  g p  ( s) g p  ( s)
This factorization is performed so that the resulting controller will be stable.
2. From the idealized IMC controller, the ideal internal model controller is the inverse of the
~ ~ 1
invertible portion of the process model. q ( s )  g p  ( s )
3. Add a filter to make the controller proper. A transfer function is proper if the order of the
denominator polynomial is at least as high as the numerator polynomial.
~ ~ 1
q( s)  q( s) f ( s)  g p  ( s) f ( s )

26
 If it is most desirable to track step set point changes, the filter transfer function usually
1
has the form f ( s )  and n is chosen to make the controller proper. If it is most
(s  1) n
n s  1
desirable to track ramp set point changes then f ( s ) 
(s  1) n
4. Adjust the filter tuning parameter to vary the speed of the closed loop system. If the λ is
small, the closed loop system is fast, if λ is large, the closed loop system is more robust.

OR
15. Explain the concept of MPC with a block diagram.
b. Ans:
i The general design objective of model predictive control is to compute a trajectory of a future
manipulated variable u to optimize the future behaviour of the plant output y. The optimization is
performed within a limited time window by giving plant information at the start of the time window.
To help understand the basic ideas that have been used in the design of predictive control, we examine
a typical planning activity of our day-to-day work. The day begins at 9 o’clock in the morning. We
are, as a team, going to complete the tasks of design and implementation of a model predictive control
system for a liquid vessel. The rule of the game is that we always plan our activities for the next 8
hours work, however, we only implement the plan for the first hour. This planning activity is repeated
for every fresh hour until the tasks are completed.
There are three key elements required in the planning. The first is the way of predicting what might
happen (model); the second is the instrument of assessing our current activities (measurement);and
the third is the instrument of implementing the planned activities (realization of control). The key
issues in the planning exercise are: 1. the time window for the planning is fixed at 8 hours; 2. we
need to know our current status before the planning; 3. we take the best approach for the 8
hours work by taking the constraints into consideration, and the optimization is performed in
real-time with a moving horizon time window and with the latest information available.
The planning activity described here involves the principle of MPC. In this example, it is described
by the terms that are to be used frequently in the Introduction to the
following: the moving horizon window, prediction horizon, receding horizon control, and
control objective. They are introduced as below.
1. Moving horizon window: the time-dependent window from an arbitrary time ti to ti +Tp. The length
of the window Tp remains constant. In this example, the planning activity is performed within an 8-
hour window, thus Tp = 8, with the measurement taken every hour. However, Ti, which defines the
beginning of the optimization window, increases on an hourly basis, starting with Ti = 9.
2. Prediction horizon: dictates how ‘far’ we wish the future to be predicted for. This parameter equals
the length of the moving horizon window, Tp.
3.Receding horizon control: although the optimal trajectory of future control signal is completely
described within the moving horizon window, the actual control input to the plant only takes the first
sample of the control signal, while neglecting the rest of the trajectory.
4. In the planning process, we need the information at time Ti in order to predict the future. This
information is denoted as x(Ti) which is a vector containing many relevant factors, and is either
directly measured or estimated.
5. A given model that will describe the dynamics of the system is paramount in predictive control. A
good dynamic model will give a consistent and accurate prediction of the future. 6. In order to make
the best decision, a criterion is needed to reflect the objective. The objective is related to an error
function based on the difference between the desired and the actual responses. This objective function
is often called the cost function J, and the optimal control action is found by minimizing this cost
function within the optimization window.

27
 Principles of MPC
Model Predictive Control (MPC) is a multivariable control algorithm that uses:
 an internal dynamic model of the process
 a history of past control moves and
 an optimization cost function J over the receding prediction horizon,
to calculate the optimum control moves.

15. Explain the implementation of MPC.

b) Ans:
ii) The steps involved in the implementation of MPC can be summarized as follows;
1. Initial controller design
2. Pretest activity
3. Plan tests
4. Model development
5. Control system design and simulation
6. Operator interface design and operator training
7. Installation and commissioning
8. Measuring results and monitoring performance
Initial controller design:
The first step is to select the controlled, manipulated and measured disturbance variables.
These choices determine the structure of the MPC control system and should be based on
process knowledge and control objectives.
Pretest activity:
During the pretest activity, the plant instrumentation is checked to ensure that it is working
properly. In the pretest experiments, each manipulated variable is bumped atleast once by
making a small step change steady state gains and settling ties are estimated from the step
response data. Each measured disturbance variable should also be bumped if possible.
Plant tests:
The dynamic model for the MPC calculations is developed from the data collected during
special plant tests. The plant testing can be very time consuming. The required test duration
28
depends on the settling times of the outputs and the numbers of MVs and DVs. The excitation
for the plant tests usually consists of changing the input variable or a disturbance variable
from value to another, using either a series of step changes with different durations or the
pseudorandom binary sequence. The magnitude of the maximum allowable input changes can
be estimated from knowledge of output constraints and the estimated steady state gains from
the pretest.
Model development:
The dynamic model is developed from the plant test, data by selecting a model form and then
estimating the model parameters. Parameter estimation is usually based on least squares
estimation. As part of the model development step, the model accuracy should be
characterized because this information is useful for subsequent system design and tuning.
Control system design and simulation:
The MPC design is based on the control and optimization objectives, process constraints and
the dynamic model of the process. The preliminary control system design from step 1 is
critically evaluated and modified if necessary. Then the MPC design parameters are selected
including the sampling periods weighing factors and control prediction horizons. Next the
closed loop system is simulated using the identified process model and a wide variety of
process conditions to evaluate the control system performance.
Operator interface design and operator training:
Because plant operators play a key role in manufacturing plants, it is important that the MPC
operator interface meet their demands. Operator training is also important because MPC
concepts such a predictive control, multivariable interactions and constraint handling are very
different from conventional regulatory control concepts.
Installation and commissioning:
After an MPC system is installed it is first installed in prediction mode, model predictions are
compared with measurements, but the process continues to be controlled by the existing
control system. After the output predictions are judged to be satisfactory, the calculated MPC
control moves are evaluated to see if they are reasonable. The commissioning period requires
some troubleshooting and can take as long as or even longer than the plant tests.
Measuring results and monitoring performance:
The evaluation of MPC system performance is not easy and widely accepted metrics and
monitoring strategies are not available. However usual diagnostic information is provided by
basic statistics, such as the means and standard deviation for both measured variables and
calculated quantities such as control errors and model residuals.

29