The Implementation of Analytical and

Experimental Approach for Design a Controller

with Reduced Order Model
Parvendra Kumar Sunil Kumar Chaudhary
Electrical Engineering Department Electrical Engineering Department
SunRise University, Alwar SunRise University, Alwar
Rajsthan, India Rajsthan, India
[email protected] [email protected]

Abstract In this paper, an analytical and experimental Original system or the reduced order model does not match
approach has been proposed for controller design. Scientists and the desired performance of the implementing system, then a
engineers are often confronted with the analysis, design and controller is designed to fulfill the requirement of the industry.
synthesis of real-life problems. The first step in such studies is the The controller may be design by analytical approach or it may
development of a 'mathematical model' which can be considered be with an experimental approach. The implementation of the
as a substitute for the real problem. The analysis and synthesis of controller depends on the plant. If a mathematical model may
the systems which have higher model are difficult and generally get to plant then analytical approach used generally, therefore
not suitable for economic and computational considerations. if a plant is so complex that the mathematical model is not
Thus, it is required to obtain a system with lower order so that, it
possible then experimental approach is used. Here we are
acquires the main characteristics of the original system. Pade
using analytical method [8] Model matching method and
approximation is used to convert the higher order model into a
lower order model. In general, it gives an unstable reduced order experimental approach by the method of Ziegler-Nichols [9].
model, but if it gives a stable reduced order model, then it is the The computational approach may be is used for both,
best among all methods of model order reduction. Sometimes mathematical model and physical plant. Here the comparison
mixed methods may be used to overcome the problem of of the response gives some important suggestion for
instability. Here the first object is to get the stable reduced order implementing the approach.
model. Once we got the stable model, then according to the
II. STATEMENT OF PROBLEM
requirement of industry specific parameters may be designed
with the help of the control. 2.1. Higher order model
Generally two approaches are used for designing the Let the higher order model with unsatisfactory performance
controller. If a mathematical model of the plant can be derived, may be described by the transfer function
then for the transient and steady-state specifications various
design techniques are available. However, for a complicated plant N ( s) A21 A22 s A23 s 2 ...A2n s n 1
the mathematical model cannot be obtained easily, so to G( s) (1)
overcome this problem an experimental approach is used for D( s) A11 A12 s A13 s 2 ...A1n 1 s n
tuning of PID controllers.
2.2. Transfer Function for Lower Order model
Keywords - Reduced order model; Pade approximation; Ziegler-
Nichols Method; Model order reduction; MATLAB; PID For the above continuous system here the requirement is to
controller; Model matching method. find a lower model of rth order where r < n is given as

I. INTRODUCTION a 21 a 22 s a 23 s 2 ...a 2r s r 1
R( s ) (2)
The technique model order reduction is used in all fields of a11 a12 s a13 s 2 ...a1r 1 s r
Electrical, Chemical, Aerospace, Mechanical etc. In the large
process control system and mechanical production houses the Where a2,i and a1,j is scalar constants.
model order reduction plays an important role to take the If the low order system is accurate, then it should retain all
decision for the final product [1]-[3]. Generally the work with characteristics of the original model. For the same type of
large scale system is very complex and time consuming [4]. inputs the response should be as closely as possible.
To check the stability of the system first we make a
mathematical model of the plant, and then reduced the 2.3. PID Controller Transfer Function
obtained model because the analysis of the full order model is The block diagram for a PID controller is shown in Fig.1. The
very complex. For model order reduction a lot of methods are PID controller may be represented in mathematical form as,
existing [5], [6] but here Pade approximation is used. The
reduced order model obtained by Pade approximation is best
among all the methods of model order reduction [7]. If the
 1 t de(t ) For the first (m + n) terms of Eq. (6) and Eq. (7) to be
u (t ) k1 [e(t ) 0 e(t )d Td ] (3) equivalent, it becomes apparent that the following set of
T dt relations must hold:
i
k1 a 0 b0 c 0
a1 b0 c 1 b1 c 0
+
u(s) + e(s) c(s) .
Ti + G(s)
- + .
Td a n 1 b0 c n 1 b1 c n 2 ...bn 1 c 0 (8)
0 b0 c n b1 c n 1 ...bn c 0
GC(s)
.
.
Fig. 1 PID controller block diagram

With the given block diagram u(s) denote control signal and 0 b0 c 2 n 1 b1 c 2 n 2 ...bn 2 c n c n 1
e(s) denotes the error signals of the system. Here k1 represent Once the coefficients ci, i = 0, 1, 2... are find out [7] using Eq.
the proportion gain and Ti , Td used for the integral and (9) and cj = (-1)j aj+2,1
derivative time constants respectively. The transfer function For the full model,
Gc (s) of the corresponding PID controller is given as
d 0 d1 s d 2 s 2 ...d m1 s m 1
1 G( s) (9)
Gc ( s) k1 [1 Td s ] (4) e0 e1 s e 2 s 2 ...e m1 s m1 e m s m
Ti s
Equation (8) can be written in matrix form as
Equation (4) can be rewritten as
k2
G c ( s ) k1 k3 s (5) c n c n 1 .... c1 b c
s 0 0
c n 1 c n .... c 2 b1 c1
Here k2 and k3 used for integral gain and derivative gain values c n 2 c n 1 .... c3 .
of the controller respectively. . .
. bn 2 c n 2
The objective is to derive a controller such that the .

performance of the augmented process matches with that of c 2 n 1 c 2 n 2 .... c n bn 1 c n 1
the reference model. In computational system the desired
performance should be satisfied by the closed loop control b0 a 0
system. A PID controller is derived for both, reduced order c0 0 0 0
system and original plant model. c c0 0 0
b1 a1
1 . .
III. MODEL ORDER REDUCTION BY PADE . . . ..
APPROXIMATION bn 2 a n 2
c n 2 c n 1 c1 c0
Here the used model order reduction approach generate from bn 1 a n 1
the theory of Pade [7] and was later used by Shamash [10]. (10)
Before a formal presentation of the method is done, consider IV. CONTROLLER DESIGN USING MODEL MATCHING
the following definition. METHOD
Consider a function, Step 1 Construction of a specified model whose closed
loop system must approximate to that of the original
f (s) c0 c1 s c 2 s ...
2
(6)
closed loop response [8]. Let it be specified as:
and a rational function Um(s)/ Vn(s) where Um(s) and Vn(s)
are mth and nth order polynomials in s respectively, and m <= a 0* a1* s a 2* s 2 ...a m
* m
s
T ( s) (11)
n. The rational function Um(s)/ Vn(s) is said to be a Pade b0* b1* s b2* s 2 ...bn* s n
approximate of f(s) if and only if the first (m+n) terms of the
power series expansions of f(s) and Um(s)/ Vn(s) are identical. c0 c1 s c 2 s 2 ... (12)
For the function f(s) in Eq. (6) to be approximated, let the Where Eq. (12) is the power series expansion of Eq. (11)
following Pade approximate be defined. about s=0.
U n (s) a 0 a1 s a 2 s 2 ...a n 1 s n 1 Step 2 Specify the structure of the controller and express
(7) that in the form of the transfer function as given
Vn ( s) b0 b1 s b2 s 2 ...bn 1 s n 1 s n
below.
 k2 In this method, we first set T = and Td = 0. Using the
G c ( s ) k1 k3 s (13)
s proportional control action only increase K, from 0 to a critical
value K cr at which the output first exhibits sustained
Step 3 Determination of the transfer function for closed loop
of unknown controller parameters. oscillations.

k2 h h1 s h2 s 2 ...
( k1 k 3 s) 0
g 0 g 1 s g 2 s ...
s 2
GCL ( s) (14)
k h h1 s h2 s 2 ...
1 ( k1 2 k 3 s ) 0
g 0 g 1 s g 2 s ...
s 2
Fig-2 Closed loop system for proportional controller
This can be written as
Thus, the critical gain K c and the corresponding period Pcr are
a 0 a1 s a 2 s 2 ...a n 1 s n 1
GCL ( s) (15) determined by experiment. According to Ziegler-Nichols
b0 b1 s b2 s 2 ...bn 1 s n 1 s n method the values of the parameters K p , Ti and Td can
Step 4 Comparing the coefficients of both transfer function obtained by the formulas shown in Table
given in Eq. (14) and Eq. (15) we get TABLE 1: FOR CRITICAL GAIN AND CRITICAL PERIOD

a 0 h0 k 0 Type of Kp Ti Td
controller
a1 h0 k1 h1 k 2
P 0.5 K cr 0
a 2 h1 k1 h2 k 2 h0 k 3
a 3 h2 k1 h3 k 2 h1 k 3 (16) PI 0.45 K cr 1/1.2 Pcr 0
a 4 h3 k1 h4 k 2 h2 k 3
PID 0.6 K cr 0.5 Pcr 0.125 Pcr
a 5 h4 k1 h5 k 2 h3 k 3
a 6 h5 k1 h4 k 3
a 7 h5 k 3 The PID controllers tuned by the second method of Ziegler-
And Nichols rules give [11].
1
b0 h0 k 0 Gc ( s) K p (1 Td s ) (19)
Ti s
b1 g 0 h0 k1 h1 k 2
b2 g 1 h1 k1 h2 k 2 h0 k 3 1
0.6 K cr (1 0.125Pcr s)
b3 g 2 h2 k1 h3 k 2 h1 k 3 0.5Pcr s
(17)
b4 g 3 h3 k1 h4 k 2 h2 k 3 4 2
b5 g 4 h4 k1 h5 k 2 h3 k 3 (s )
Pcr
0.075K cr Pcr (20)
b6 g 5 h5 k1 h4 k 3 s
b7 g 6 h5 k 3 Thus, the PID controller has a pole at the origin and double
Step 5 Now comparing the transfer function of the closed zeros at s=0.4/ Pcr
loop plant and controller with model transfer
function, we get: VI. EXAMPLES
The control problem is taken from [8] is
a 0 b0 c 0
a1 b0 c1 b1 c 0
a 2 b0 c 2 b1c1 b2 c 0 (18) 0 1 0 0 0 0
0 0
a 3 b0 c 3 b1c 2 b2c1 b3 c 0 0 1 0 0
a 4 b0 c 4 b1c 3 b2 c 2 b3 c1 b4 c 0 0 0 0 1 0 0
A
Step 6 Now substituting the values of as, bs and cs and 0 0 0 0 1 0
solving linear simultaneous equations. We can find 0 0 0 0 0 1
out the values of k1, k2 and k3.
2 33.6 155.94 209.46 102.42 18.30
V. CONTROLLER DESIGN USING ZIEGLER-NICHOLS B [0 0 0 0 0 1]T
SECOND METHOD
C [2 3 16 20 8 1] A: CONTROLLER DESIGN FOR FULL MODEL BY
MODEL MATCHING METHOD
D0
Choosing r = 1 and By using plant transfer function and PID controller transfer
function we get the closed loop transfer function of the plant
Q = dig ([1, 1, 1, 1, 1, 1]) as
Optimal transfer function is k2
k 3 s)G ( s) ( k1
T * (s) [0.8944 12.5772s 139.8524s 2 1365.2s 3 12536s 4 ] GCL ( s) s
k
c0= 0.8944 1 (k1 2 k 3 s)G ( s)
s
c1= -12.5772
Using Eq. (16), (17) and (18) and solving linear simultaneous
c2=139.8524 equations we get the values of k1, k2, and k3 as
c3= -1365.2 k1=0.5405
c4=12536 k2=0.0380
The matrix A, B and C gives the plant transfer function as k3=1.2156
1
G(s) [SI A] B On substituting the value of k1, k2 and k3 in Eq. (16) and (17)
we get values of as and bs closed loop transfer function as
2 3s 16s 2 20s 3 8s 4 s 5
G( s)
2 33.600s 155.9400s 2 209.4600s 3 102.4200s 4 18.300s 5 s 6

0.0760 1.1950s 4.6607s 2 13.0548s 3 30.5636s 4 28.6740s 5 10.2653s 6 1.2156s 7

GCL ( s)
0.0760 3.1950s 38.2607s 2 168.9948s 3 240.0236s 4 131.0940s 5 28.5653s 6 2.2156s 7

0.0026864 0.03236s 0.05991s 2 0.008384s 3

Step Response for the close loop of full model GCL ( s)
1 0.0026864 0.04502s 0.28011s 2 1.008384s 3
0.8 Step Response for the closed loop of reduced order model
Amplitude

0.6 1.2

1
0.4
Amplitude

0.8
0.2
0.6

0 0.4
0 50 100 150 200
0.2
Time (sec)
Fig -3: step response of close loop plant of original model
0
0 50 100 150 200
Time (sec)
B: CONTROLLER DESIGN FOR REDUCED ORDER Fig -4: step response of close loop plant of reduced order model
MODEL BY MODEL MATCHING METHOD
First the original plant is reduced by Pade approximation [12], C: CONTROLLER DESIGN FOR REDUCED ORDER
for which the algorithm is given in III. MODEL BY ZIEGER-NICHILS METHOD
According to Pade approximation the second order reduced Here first requirement is to find out the starting point for K p
order model is
and double zeros.
0.02645s 0.01266
R2 ( s ) Let start the tuning with considering the K p only. Here the
s 2 0.2202s 0.01266
closed loop response is
Now, after getting the reduced order model, let us design the
controller for it by the mode matching method. The algorithm C (S ) (0.02645s 0.01266)
2
is given in IV. R( S ) s (0.2202 0.02645K p ) s (0.01266 0.01266 K p )
Close loop response of Pade approximation By the help of Routh-Hurwitz criteria the value of K cr for
sustained oscillations is K cr =-8.325.
So after putting this value in characteristic equation the value Gain (K) = 10
of critical frequency is w=0.1456
Zeros (a) = 0.9270
2
Pcr = 43.15 Maximum overshoot (m) =1.0719
0.1456
Now with the help of table other values are
Unit-Step Response for the close loop of full model
K p = -4.995
1
Ti=21.575
Td= 5.3937
0.8

Amplitude
So here the initial values are obtained. As the requirement of
industry, we can set the value of maximum overshoot in
0.6
programming. Generally, according to the better establishment
of the system the overshoot should be between10% to 40%.
Using the MATLAB program, we vary the gain 50 to -50 with 0.4
step size -0.2 and zeros as 0.927 to -0.0092 with step size -
0.02. Fine tuning gives the following results 0 50 100 150 200

Gain (K) = 50 Fig -6: step response of tclose

Sec (sec)
loop plant of original model

Zeros (a) = 0.9270 TABLE 2: COMPARISON FOR PERFORMANCE SPECIFICATION

Maximum overshoot (m) =1.0719 OF DESIGNED CONTROLLER

Controller design Model Matching Method Ziegler-Nichols Method

Unit-Step Response for the close loop of reduced order model Full Reduced Full Reduced
1.2 Models Model Order Model Order

Specifications G (s) R (s) G (s) R (s)

1 Rise time (sec) 54.9 9.06 1.65 0.984

101
Amplitude

Settling time (sec) 32.9 31 9.11

Peak amplitude 1 1.13 1.07 1.1
0.8
Overshoot (%) 0 12.9 7.19 9.75
At time (sec) 200 20.5 17.9 2.62
0.6

0 50 100 150 200 VII. CONCLUSIONS

t Sec (sec)
FigK
50
-5:= step response of close loop plant of reduced order model On behalf of the result shown in table some important point
a0.927
= may be described for tuning of the controller. If the desired
D: CONTROLLER DESIGN FOR ORIGINAL PLANT BY specification is not given and a reference model is available,
ZIEGER-NICHILS METHOD then use of model matching method is preferred. But if desired
Here first requirement is to find out the starting point specification is given then Ziegler-Nichols method gives better
for K p and double zeros for the plant. result. In case of Model Matching Method, for the given
mathematical model the designing of the controller is very
2 3s 16s 2 20s 3 8s 4 s 5 much complex. On the other hand, if the controller is designed
G( s) for the reduced order model, then it gives the best performance
2 33.600s 155.9400s 2 209.4600s 3 102.4200s 4 18.300s 5 s 6
in the sense of all specifications. Here the maximum overshoot
Let start the tuning with considering the K p only. Here all is 12.9 % more as the full model, but its not a problem
steps are the same as above. The consideration of starting because due to this, the response is fast and till 40 % the
point for gain and double zeros are same because original overshoot is accepted in the industry.
plant and the reduced order model save the same origin. If the plant is so big that the mathematical model is not
So possible, and then Ziegler-Nichols method is preferred on the
other hand it may be used for the given mathematical model.
K p = -4.995 and double zeros are a= 0.0927. Here the obtained initial value is used in MATLAB program.
In the reduced order model the performance specification is
Using the MATLAB program, we vary the gain 10 to -2.5 better as compared to a full model.
with step size -0.2 and zeros as 0.927 to -0.0092 with step size
In industry for design consideration, two parameters have big
-0.02. Fine tuning gives the following results
importance. One is, the settling time should be minimum and
other is the maximum overshoot should be in the given range.
So for this the Ziegler-Nichols method is preferred.
For fast developing nations, manufacturing units play an
important role and here the maximum work is done for stable
product. So to save the time, effort, energy and to reduce the
complexity the computational approach is preferable.
VIII. REFERENCES
[1] Per-Olof, G. Carl Fredrik, Mannerfel and P. Molander, Contributions
to the model reduction problem, IEEE Transactions on Automatic
Control, Vol. AC-27, No. 2, April 1982, pp. 454-455.
[2] B. Bandyopadhyay, A. Rao and H. Singh, On Pade approximation for
multivariable systems, IEEE Trans. Circuits and Systems, Vol. 36, No.
4, pp. 638-639, 1989.
[3] Y. Smamash Truncation method of reduction, A viable alternative
electronic letters 22nd January, 1981, Vol. 17, No. 2, pp. 97-98.
[4] M. Jamshidi, An overview on the aggregation of large scale systems,
1981, pp. 9-92.
[5] J. Pal, Stable reduced-order Pad approximants using the Routh-
Hurwitz array, Electronic Letters, Vol. 15, No. 8, April 1979.
[6] T. C. Chen, C. Y. Chang, and K.W. Han, Stable reduced order Pade
approximants using stability eq. method, Electron Letter, Vol. 16, pp.
345-346, 1980.
[7] S. Janardhan, Model Order Reduction and Controller Design
Techniques, 2005.
[8] P. Kumar and R. Prasad, Controller design using Pade approximation
and mixed methods, International Journal of Computer Technology and
Applications, vol. 3, No. 4, 2012, pp. 1504-1509.
[9] K. Ogata, Modern Control Engineering, 4th ed., Englewood Cliffs, NJ:
Prentice Hall, 2001.
[10] Y. Shamash, Model reduction using the Routh stability criterion and
the Pade approximation technique, Int. J. Control, Vol. 21, No. 3, pp.
475-484, 1975.
[11] C.C. Hang, K. J. Astrom and W. K. Ho,Refinement of the Ziegler-
Nichols tuning formula, IEEE Proceeding a Control Theory and
Application, Vol. 138, No. 2, pp. 111-118, 1991.
[12] P. Kumar and S. K. Chaudhary, A novel approach in classic Pade
approximation, IEEE 7th International Conference on Cloud
Computing, Data Science & Engineering, Noida, 12-13 Jan 2017.