System Identification of Cane Carrier Process of

Sugar Mill
Sandeep Kumar Sunori1, Pradeep Kumar Juneja2, Mayank Chaturvedi3, Aayushi Bansal4
1
Graphic Era Hill University, Bhimtal Campus, Uttarakhand, India
2, 3, 4
Graphic Era deemed to be University, Dehradun, Uttarakhand, India
[email protected] , [email protected], [email protected], [email protected]
1

Abstract—System identification refers to finding out an exact In the juice extraction process of sugar factory, the prepared
mathematical model, of a process with unknown characteristics, cane billets are transported to the Donnelly chute from where it is
relating input and output. A control system can only be designed passed to the rolling mill for crushing and extraction of juice. In
for a process when its mathematical model in any form is available. order to yield maximum juice, regulation of the height of
In the present work, the cane carrier system of sugar mill is sugarcane fiber in the chute at a predefined level is very critical
considered and its various mathematical models are obtained from which is in turn controlled by a dc motor running the cane carrier
the measured time domain input-output data using MATLAB. system [3]. The speed of this dc motor depends on the applied
input dc armature voltage [9].
Keywords—System identification; transfer function; polynomial;
cane level; step response In the present paper, black box modeling of this juice
extraction process, with manipulated variable as the armature
I. INTRODUCTION voltage, and the control variable as the cane level in the chute, is
done using two different strategies of linear modeling i.e transfer
Broadly, the system identification techniques can be function modeling and polynomial modeling, from the measured
categorized into white box, black box and the gray box modeling. input-output data. The transfer function modeling is done for a
In case of white box modeling, no input-output data is available given model order .The polynomial modeling can be further
for the process under consideration; the model is obtained only categorized into Auto-regressive exogenous (ARX) and Auto-
on the basis of already known internal properties of that process regressive moving average exogenous (ARMAX) modeling.
using first principles. In case of black box modeling, the internal
properties of the system are not known in advance. The system is II. LINEAR MODELS
identified on the basis of the available input-output data only.
Gray box modeling is done when both, the knowledge about A general linear system is defined by equation (1)
internal properties and the input-output data are partially known.
( ) ( ) ( ) ( ) ( )
Karol Bogdanski and Mathew C Best performed system
identification using Kalman filter [1]. Estimators from both the
Where I(n), O(n) and (n) represent input, output and the
continuous and discrete time and frequency domain models have
added noise respectively. The transfer function P(z) is defined by
been discussed by F.Kozin and H.G.Natke [2].Transfer function
equation (2)
technique has been used by P.Arumugam and V.Anithakumari in
order to anticipate the time series of the production of rubber in
India [4]. ( )
( ) ( )
( )
Maria Silvia de A. Moura et al presented method of least
squares and wavelet expansion for estimation [5]. The
polynomial autoregressive moving average model has been Combining these equations we get equation (3).
discussed and analysis of inevitability and stability is done by
Evelio ( ) ( ) ( )
Hernandez and Yaman Arkun [6], Patryk Chaber and Maciej ( ) ( ) ( )
Lawrynczuk performed the comparison analysis of two different ( ) ( ) ( )
recurrent techniques of process modeling for the neutralization
process [7].The extended Kalman filter technique has been
adopted by Masaru Hoshiya and Osamu Maruyama for
( ) ( ) ( ) ( )
autoregressive process identification [8].
 w(n) represents the zero mean white noise.
( )
( )
( ) ( )
( )
( )
Combining these equations, the general polynomial model is
For model 1,
obtained, as presented in equation (6)

( ) ( ) ( ) ( )
( ) ( )
( ) ( )
For model 2
For an Autoregressive moving average with exogenous
variable (ARMAX) process ( ) ( ) [ ]
So, equation (6) reduces to,
[ ]
( ) ( ) ( ) ( ) ( ) ( ) ( )

An autoregressive with exogenous (ARX) process is a

particular case of ARMAX process with γ (z) =1 i.e.
For model 3
( ) ( ) ( ) ( ) ( ) ( )

III. SIMULATION AND RESULTS [ ]

A. Transfer function(TF) models:

Transfer function simply refers to the s domain output to
input ratio. The transfer function modeling has been done for [ ]
model orders equal to 1, 2 and 3. The estimated TF models are
presented in equations (9), (10), and (11) respectively.

( )

( )

( )

The individual servo responses of these TF models are

presented in Fig 1, 2 and 3 respectively. A comparison of these
step responses is shown in Fig 4. The performance analysis
presented in Table I clearly indicates that TF model with order 3
best matches with the original process with 99.4 % fitness and
least mean square error (MSE) . The corresponding error plots
are showcased in Fig 5.
Now the state space models are determined for these TF
models. The general state space equations are given by equation
(12) and (13), x being the state vector.
Fig 1. Step response of TF model 1
Fig 2. Step response of TF model 2 Fig 5. Comparison of error plots of TF models

TABLE I. PERFORMANCE ANALYSIS OF TF MODELS

Model % Fit MSE

TF model 1 90.25 1.14 x 10-5

TF model 2 97.47 7.73 x 10-7
TF model 3 99.4 4.32 x 10-8

B. Polynomial models
The generated ARX and ARMAX models are as follows:
ARX Model:

( ) ( )

( ) ( )
Fig 3. Step response of TF model 3
No. of poles j=2, (No. of zeros+1) k=2, time delay=1
ARMAX Model:

( ) ( )

( ) ( )

( ) ( )

No. of poles j=3, (No. of zeros+1) k=2, =2, time delay=1

TABLE II. PERFORMANCE ANALYSIS OF POLYNOMIAL MODELS

Model % Fit MSE

ARX Model 71.76 5.96

ARMAX Model 89.67 0.0003723
Fig 4. Comparison of step responses of TF models
 model more closely matches to the original process than the ARX
model.
REFERENCES
[1] K. Bogdanski and M.C. Best, “A New Structure for Non Linear Black-Box
System Identification using the Extended Kaman Filter,” Journal of
Automobile Engineering, Vol. 231(14), 2017, pp.2005-2015.
[2] F.Kozin and H.G. Natke, “System Identification Techniques,” Structural
Safety, Elsevier, Vol.3, Issues 3-4, 1986, pp. 269-316.
[3] Y. Misra and H.R. Kamath, “Simulink Modeling of Fuzzy Controller for
Cane Level Controlling,” International Journal of Industrial Engineering &
Technology (IJIET), TJPRC Pvt. Ltd., ISSN 2277-4769, Vol. 3, Issue 1,
2013, pp.43-50.
[4] P. Arumugam and V. Anithakumari, “Seasonal Time Series and Transfer
Function Modelling for Natural Rubber Forecasting in India,” International
Journal of Computer Trends and Technology (IJCTT), Vol.4, Issue 5,
2013, pp.1366-1370.
[5] M.S.A. Moura, P.A. Morettin, C.M.C. Toloi and C.Chiann, “Transfer
Function Models with Time-Varying Coefficients,” Journal of Probability
Fig 6. Comparison of step responses of polynomial models
and Statistics,Hindawi, Vol.2012, 2012.
[6] E. Hernandez and Y. Arkun, “Control of Nonlinear Systems using
Polynomial ARMA Models,” Process Systems Engineering, AIChE
Journal, Vol.39, Issue 3, 1993, pp.446-460.
[7] P. Chaber and M. Lawrynczuk, “Recurrent Polynomial and Neural
Structures in Modelling of a Neutralisation Process,” Progress in
Automation, Robotics and Measuring Techniques, Advances in Intelligent
Systems and Computing(AISC) Book series, Vol. 350, Springer
International Publishing Switzerland, 2015, pp. 23-32.
[8] M. Hoshiya and O. Maruyama, “Identification of Autoregressive Process
Model by the Extended Kalman Filter,” Reliability and Optimization of
Structural Systems’90, Lecture Notes in Engineering (LNENG) Book
Series, Vol. 61, International Federation for Information Processing,
Geneva, Switzerland, 1991, pp.173-183.
[9] S.K. Sunori, S. Shree and P.K. Juneja, “Control of Sugarcane Crushing
Mill Process: A comparative analysis,” International Conference on Soft
Computing Techniques and Implementations (ICSCTI), 2015, pp.1-5.

Fig 7. Comparison of error plots of polynomial models

The step response and error plot comparisons of the two

polynomial models are displayed in Fig 6 and 7 respectively. The
Table II highlights that ARMAX model more closely matches
with the original process with 89.67 % fitness and negligible
MSE.
IV. CONCLUSION
In this work, the transfer function and polynomial models
have been obtained using system identification toolbox of
MATLAB from the measured input-output data for a cane carrier
system of juice extraction process of sugar factory. The step
response comparison revealed that the third order transfer
function model best matches with the original process with least
mean square error. Comparison is also made between ARX and
ARMAX polynomial models, which exhibited that ARMAX