Documente Academic
Documente Profesional
Documente Cultură
I. INTRODUCTION
Iron and steel production process consumes a lot of active
lime as a supplementary raw material. Many steel companies
have their own lime kilns to produce active lime. The
implementation of reasonable control of lime kiln will
effectively reduce energy consumption in the lime production
process, improve the passing rate of final product, and thus
improve the quality of steel, iron and steel products. It can also
improve the economic performance and reduce pollutant
emissions [1].
According to different structural characteristics, Lime kiln
can be divided into rotary lime kiln and vertical gas firing kiln
[2]
. The control systems of different kilns vary a lot. However
the overall of the control systems can be divided into feeding
materials, discharging lime, combusting, cooling, cleaning
dust, hydraulic station control sections. The object researched
in this paper is the combusting control section of the rotary
lime production process.
The rotary lime kilns in iron& steel plants are fuelled with
blast-furnace gas (BFG), coke oven gas (COG) or the mixture
of different gases [2]. The kilns are heated to high temperature
*
This work is supported by a grant from the National High Technology Research and Development Program of China (863 Program)No.2008AA042901
5814
(2)
T f = T _ fr T _ fr 0
(3)
Tb = T _ bk T _ bk 0
(4)
In this way, input and output variables are defined as the
increment to the physical quantity of the balance point.
B. System Identification Method
If we assume that the relationship of the four variables of
the system to be identified can be approximated as a linear
time-invariant systems near the equilibrium point, we can
describe the system with the transfer function matrix D (S).
T ( s ) = D ( s )U ( s )
(5)
In the equation:
T ( s ) = (T f ( s ), Tb ( s )) T ;U ( s ) = (U 1 ( s ),U 2 ( s )) T
They denote the Laplace Transform results of the input
and output variables, respectively.
D (s)
D ( s ) = 11
D 21 ( s )
D12 ( s )
D 22 ( s )
(6)
In order to determine the transfer function matrix D (s),
we have designed two experiments.
The first experiment is to keep U2 = 0 (i.e. u2 = 0.5), U1 =
0.5u (t) (i.e. u1 step 0 to 1 at the time of 0 second). The
experimental results (simulation) are shown in Figure 4.
From the figure we can see that the input variables of the
temperature control system of rotary kiln are the fuel flow rate
and the damper position. Here, u1 denotes the real fuel flow
rate relative to the percentage of full flow rate, and
u2denotes the percentage of the damper opening.
The outputs of the system are the front temperature T_fr
and the back temperature T_bk.
According to the information of production operating, the
kiln has a balance operating point, namely: When [u10, u20]
Rotary kiln
=[0.5, 0.5], [T_fr0, T_bk0] =[2225, 425].
operation is performed in the vicinity of the point. Therefore,
the linear object recognition may wish to use this point as the
reference point for the establishment of an incremental
mathematical model. Each variable is defined as:
U 1 = u1 u10
U 2 = u 2 u 20
After the simulation, the precise data of T_fr (t) and T_bk
(t) corresponding to time t can be seen in the workspace.
These data can be transformed to increment data to the
equilibrium point in accordance with equation (1) - (4) So, we
get:
Tf () = 165 (i.e. T_fr () = 2390); To reach stable 65%
of amplitude, the corresponding time is 35 seconds;
Tb () = 34 (i.e. T_bk () = 459); To reach a stable 65%
of stable amplitude, the corresponding time is 5 seconds.
The trends of the curve can be approximated with firstorder elements. In accordance with the characteristics of firstorder elements, we can derive:
D11 ( s ) =
(1)
5815
330
35s + 1
(7)
D21 ( s ) =
68
5s + 1
(8)
The second experiment: to maintain the input U1 (t) = 0,
U2 (t) = 0.5u (t), we can get the temperature curves shown in
Figure 5.
Fig. 5 The temperature curves when U1 (t) = 0U2 (t) = 0.5u (t)
348
4.2 s + 1
280
D22 ( s ) =
3s + 1
D12 ( s ) =
(9)
(10)
C.
Model Validation
In order to verify whether the model identified out of the
original system can describe the working process, we have
designed a verification model diagram shown in Figure 6. The
results of the simulation to the model with the S function and
the identified model can be compared when they are inputted
with the same signals.
T ( s) = ( s ) R( s)
i.e.
T f ( s ) = 1 ( s ) R1 ( s )
Tb ( s ) = 2 ( s ) R2 ( s )
(11)
(12)
(13)
By this way, the system changes into two SISO systems.
So the controllers can be designed.
By equation (5) and (11) we can see;
D( s )U ( s ) = ( s ) R ( s )
(14)
U ( s ) = D( s ) 1 ( s ) R( s ) = P( s ) R( s ) (15)
5816
P( s ) = D( s ) 1 ( s )
(16)
It should be pointed out that: a reasonable choice of(s) is
very important, because the matrix P(s) must be physically
achievable, or the degree of numerator polynomial of each
element in the matrix should not higher than that of the
numerator polynomial. Therefore, the design should take into
account of the compromise between system simplification and
realization.
Now we select proper(s)according to the actual situation
as follows:
1
35s + 1
( s) = diag
3s + 1
(17)
This just can eliminate a factor section from each
numerator of the elements in the inverse matrix of D (s).
After the final solution, we get:
P(s) =
7
(21s+ 5)(5s+1)
36
-17 (3s+1)(21s+ 5)
360
29
(5s+1)(35s+1)
24
11
(21s+ 5)(5s+1)
48
(18)
T f ( s ) = 35s + 1 R1 ( s )
1
Tb ( s ) =
R2 ( s )
3s + 1
(19)
After decoupling, the two sections of the system both are
first-order objects. Regulators are easy to design, and we can
predict to gain a good performance system.
If we design the system into typical I-type system, all we
need are two integrators, just as shown in figure10. That
makes the system fast adjusting with no error. The parameters
k1,k2 of the integrators can either be preset or tune according
to the production demands.
5817
REFERENCES
[1] , ,
,2009.7(Vol.26.3),pages:225-228.
[2]
2008.8(vol.34.4) pages:21-25
[3] 2004
[4] Ribeirp, B.; Dourado, A.; Costa, E. A neural network based control of a
simulated industrial lime kiln. Proceedings of 1993 International Joint
Conference(Vol 2)25-29 Oct. 1993Page(s):2037 - 2040
[5] Brown, Kevin; Rastogi, Laxmi, Mathematical Modeling and Computer
Control of Lime Kilns, American Control Conference, 22-24 June 1983
Page(s):1 6
[6] Hassell, G., Statistical modelling for lime kiln control. Dynamic
Modelling Control Applications for Industry Workshop, IEEE Industry
Applications, 30 April-1 May, 1998 Page(s):20 28
[7] Barreto G. A.,Lime kiln hybrid control system. Dynamic Modeling
Control Applications for Industry Workshop, IEEE Industry Applications
Society,26-27 May, 1997 ,Page(s):44 50
[8] Hagemoen S.W., An expert system application for lime kiln automation
Pulp and Paper, Industry Technical Conference, Annual, 1993
Page(s):91 97
[9] Chao-Ying Liu; Kai Li; Zhe-Ying Song; Xue-Ling Song; Hui-Fang
Wang, Decoupling Control and Simulation for Variable Flow Heating
Systems, 2007 International Conference on Machine Learning and
Cybernetics Volume 1,19-22 Aug. 2007 Page(s):419 - 424
V CONCLUSION
In this paper, rotary lime kiln and the kiln temperature
control problem has been studied. A model described by the
S-function in Matlab is treated as a black box on behalf of
real lime-kiln, and the method of step response is applied to
identify the system with two inputs and two outputs. The
mathematical model with a transfer function matrix is derived
to describe system. Based on this model, we design a
decoupling controller in accordance with decoupling control
method.
Through the previous discussion we get the following
conclusions:
5818