2016 4th International Conference on Control, Instrumentation, and Automation (ICCIA)

27-28 January 2016, Qazvin Islamic Azad University, Qazvin, Iran

Dynamics Modelling and Control of a Fluidized Bed Dryer

Elias Razzaghi1, Payam Zarafshan2, S. Razi Karimi Akandi3

Department of Agro-Technology, College of Aburaihan, University of Tehran, Tehran, Iran

Corresponding Author: p.zarafshan@ut.ac.ir

Abstract - Husk rice must be dry before any treatment in dependent variables) behaviour of a dynamical system with
industrial processes. For food industries, to preserve the quality respect to changes in the input variables (i.e. independent
of materials, the temperature must remain in a specific domain. variables), [9].
So, in this paper, control of the moisture content and
temperature of the grains are studied. To this end, a
mathematical model is used to shape the control algorithm. In
addition, the linear model is prepared for the control process. Outletof
Outlet ofexhauster
exhauster
High-passed Bessel filter, low-passed Butterworth filter, and
Linear Quadratic Regulator (LQR) controllers are designed Cyclone for solid Exhauster for
collection solid
and compared to find the best control gain matrix for the transportation
dryer. Using a comprehensive simulation routine, the stated
control methods are used to control the final moisture content
and the temperature of husk rice grains. It is shown that the
controller is able to control the process in about 50 minutes. Fresh solid feeder to dryer
Index Terms - Fluidized Bed Dryer, Optimal Control, Dynamics Steam generator
Modelling, Thermal System.

I. INTRODUCTION Pressure sensor

D rying is considered as a complex phenomenon of mass

and energy transfer which is affected with many
different parameters. Between different models of fluidized
dryers, a well-mixed continuous fluidized bed dryer (Figure-
(1)) is one of the most common dryers which is used in the
industry, [1]. In all of the designing methods such as
simulation, optimization and process control, a model of the
plant is needed as the core of our computational problem.
Electronic heater
The model is a set of equations connecting all process
parameters and a set of constraints in the form of inequalities
OPTO
which adequately describes the behaviour of system. On the system Fan
other hand, it is difficult to control fluidized bed drying temperature
registration
processes because of the lack of online tools to measure the
solids moisture, which is the most important parameter for
monitoring and control of fluidized bed dryers, [2]. Pressure
Although the temperature and relative humidity of the outlet sensor
air can be measured, it is difficult to use them directly to
estimate the solids moisture for the feedback control. It is Recycle
well known that a drying process has two stages: first is the
constant-rate drying and second is the falling-rate drying,
[3]. In addition, several methods have been used for solids Dry solid
moisture measurement, e.g., microwave resonance, [4], discharg
capacitance probes, [5], electrical conductive and resistance
measurement, [6], and microwaves, [7]. However, most of Figure 1. Flow diagram of a well-mixed fluidized bed dryer
them are not suitable for fluidized bed applications because
it is difficult to obtain the real time control, [8]. Also, a Air flow regime in the fluidized drying technology has a
model predictive control relies on dynamic models of the crucial role in the drying process. The terminal velocity of
process, which are used to predict the output variables (i.e. different grains is different. So, for the case of dry particles
(or partially dry, no surface moisture), if the fluidizing gas is
2- Assistant Professor
1, 3- PhD Student
978-1-4673-8704-0/16/$31.00 ©2016 IEEE 46
further increased, the bed of particles goes through different where all of parameters are stated in Table (1). Starting from
types of fluidization regimes depending on the types of the energy balance equations in the differential volume
particles with reference to the Geldart classification of element, the equation that predicts the temperature profile of
powders, [10-11]. In this regard, many software and control the particles and gas in the downer bed dryer is obtained as:
systems have been produced to optimize industrial dryer
systems. Also, 17 commercial and semi commercial dTs 6S Vs 1 - ε ª h Tg - Ts - N λ 0 - Cal Ts º
programs for drying calculations were presented that they = « »
dz dpWs Vg ε « Css - Cal X »¼
were able to identify on the market, [12]. Only a few of them ¬
perform process calculations of dryers including dryer (2)
Cal Ts dX
dimensioning, usually for fluid bed dryers. Typically, a -
program for dryer calculations performs balancing of heat Css + Cal X dz
and mass and, if dimensioning is possible, the program
requires empirical coefficients, which has supplied by the Table 1 – Parameters and theirs experimental data, [16]
user. Similarly, the drying process is designed in commercial Variable Definition Value
process simulators using the chemical and process Ca Specific heat (water steam) 1.642KJ/Kg°C
engineering. However, with present-day computer Cal Water specific heat 4.144 KJ/Kg°C
technology, construction of such software is possible;
Cb Gas specific heat 1.022KJ/Kg°C
whereas dryPAK is a program that evolves in this direction,
[13-14]. The main concept of dryPAK is that all models Cd Drag coefficient 0.7
share the same database of humid gas, moist material Css Specific heat of dry solid 1.61KJ/Kg°C
properties, methods for calculation of drying rate, etc. The D Duct diameter 0.2 m
results are also visualized in the same way, [15]. dp Particle diameter 0.03 m
In this paper, the objective is to control the difference of
moisture and temperature of grains between the entrance and fp Gas particle friction factor 0.01
output of the dryer. To this end, a mathematical model is fs Particle duct wall friction factor 0.45
used to shape the control algorithm. Next, High-passed filter, fw Blasius factor 0.005kg/s^2
low-passed filter, and Linear Quadratic Regulator (LQR)
G Gravity acceleration 10m/s^2
controllers are designed and compared to find the best
control gain matrix for the dryer. In fact, the controller must Gs Flux density Kg/s
keep the difference of moisture and temperature of grains. h Convective heat transfer coeff. 30W/m^2
Finally, using a comprehensive simulation routine, the N Drying rate 0.08Kg/m^2s
merits of the designed controllers is fully studied. P Pressure 10^5pa
R Duct ratio 0.05m
II. DYNAMICS MODELLING Ta Ambient temperature 20°C

A. Non-Linear Model Tg u1 Air temperature u 01 80 C

Ts x1 Solid temperature x 01 20 C
The resultant mass, energy, and momentum balances,
applied to a differential volume, by assuming a system U Overall heat transfer coefficient 10W/m^2
formed only for spherical particles of a single size could be S Duct diameter 0.2m
expressed as, [16]: Vg u2 Air velocity u 02 7 m s
Vs x3 Solid velocity x 03 0.01m s
dVg ª fs º
Vg
vt
= «g - » Terminal velocity 9m/s
dz εW Vg - VS ¬ ρs ¼ Wg Air flow 0.129Kg/s
Ws u3 Solid flow u 03 0.1Kg s
dVs Vs ª fs º
= «g - » Xs x2 Moisture of solid (%) x 02 16Kg Kg
dz 1 - ε W Vg - Vs ¬ ρs ¼ Xg Humidity of air(dry base) 0.4Kg/Kg
dε Vg ª fs º ε x4 Porosity (0< ε<1) x 04 0.6
= «g - » fi Spehrcity of particulate material 0.37
dz W Vg - Vs ¬ ρs ¼ (1)
ρs Density solid 700 Kg/m^3
V Vg ρg Density gas 1.225Kg/m^3
W=- - s
ε 1- ε λ0 Latent heat 2432.4KJ/Kg
2 z Dryer length 6m
3Cd ρ g ε1 7 Vg -Vs 3-2ε dV
fs = + ρg Vs s Considering Table (1) and rearranging equations Eq. (1)
4dp 2ε dz and Eq. (2), a set of useful equations are yields as:

47
 2 while the term dz dt in Eq. (6) is defined as the average
3Cd ρ g x 41 7 u 2 - x 3
α= speed of the solid particles inside the column and it is
4dp equivalent with the average process speed of the solid
particles. Also, the average process speed is set to be 30
3 - 2x 4 ρ g x 3
β= minutes constant. For a dryer with 6 meters length, the
2x 4 average speed of the particle can be calculated as:
1
γ= dz 6
u 2 - x3 u 2 1 - x 4 - x3x 4 (3) v= = | 0.003m s = cte (7)
dt 1800
u 2 1 - x 4 + x3x 4
W= Then, Eq. (6) can be rewritten as:
x4 1- x4
dΔx dz dz
αρs - βx 3 x 4 γρs g = Ac Δx + Bc Δu (8)
fs = dt dt dt
ρs - βx 3 x 4 γ
where state matrices A, B can be obtained as:
where x i for i = 1,2,3,4 are defined as state variables,
dz dz
ui for i = 1,2,3 are inputs and so, nonlinear equations are: A = Ac , B = Bc (9)
dt dt
6Sx 3 1 - x 4 h u1 - x1 - N λ 0 - Cal x1
x1 = and so, time dependent state equations are yields as:
d p u 3u 2 x 4 Css + Cal x 2
§ Cal x1 · § -6x 3 1 - x 6 NS · Δx t = Δx = AΔx + BΔu
(10)
-¨ ¸ ¨¨ ¸¸ Δy = CΔx + DΔu
© Css + Cal x 2 ¹ © d p u 3u 2 x 4 ¹
-6x 3 1 - x 4 NS Also, it should be noted that in Eq. (10), Δx, Δu, Δy are
x2 = (4)
d p u 3u 2 x 4 the differences of state variables, inputs, and outputs
between the entrance and output of dryer respectively
ª f º (Figure (2)). Grains inter the dryer column with the initial
x 3 = - x 3 x 4 γ «g - s » temperature and moisture of 25°c and 16% respectively.
¬ ρs ¼
x 4 1 - x 4 γ αρs - βx 3 x 4 γρs g
x 4 = - x 4 1 - x 4 γg +
ρs ρs - βx 3 x 4 γ
­ x 01 State (0); u 03
B. Linear Model °x
° ­u
As stated before, constant parameters in Eq. (1) to Eq. (4) State (0); ® 02 State (1); ® 12
are obtained from different experimental data as studied in ° x 03 ¯ u13
°̄ x 04
Table (1). So, after linearization, state equations can be
attained as:

Δx z = AcΔx + BcΔu (5)

in which the solid velocity, gas velocity, porosity, and solid

temperature derivatives are obtained with respect to the
length of dryer. In another word, these equations are length
dependent. The performance of controller is always
evaluated with respect to time. So, the controller must be
able to regulate the process parameters in a reliable time.
Changing in solid properties are also a time consuming
phenomena. To explain the controller performance, Eq. (5) ­ x11
can be rewritten as below: °x
° State (1); u13
State (1); ® 12
dΔx dΔx 1 ° x13 ­u
= . = AcΔx + BcΔu °̄ x14 State (0); ® 02
¯ u 03
dz dt dz (6)
dt
Figure 2. Changing in particle properties during the dryer column

48
 Since the complexity of dynamic system is existed, III. CONTROLLER DESIGN
instead of trying to regulate the system output variables in a In this section, three control methods are compared to
specific point, it is preferred to control the difference of choose a better control performance for the dryer system.
initial and final variables of the solid particles. In fact, the Also, the block diagram of these designed controllers is
initial values of input and states variables are considered as: shown in Figure (3). The high-passed filter is used to
measure the control gain G High . In this case, the poles are
x 01 = 25, x 02 = 16, x 03 = 0.01, x 04 = 0.6 placed in [-0.0023+0.0010i, -0.0023-0.0010i, -0.0010+0.0023i, -0.0010-
(11)
u 01 = 80, u 02 = 7, u 03 = 0.1 Figure (4) shows the response to the initial values of
0.0023i].
system (when Δu=0). Moreover, the low-passed filter is used
Substituting these parameters in Eq. (4), state matrices A, to measure the control gain G . Also, Figure (5) shows the
Low

B, C and D can be calculated as: response to the initial values of system with high-passed
filter (when Δu=0). The poles of Bessel filter were placed in
ª-0.005 -0.0149 24.4633 -1.0193 º [-0.0025+0.0008i, -0.0025-0.0008i, -0.0015+0.0019i, -0.0015- 0.0019i]. In
« 0 0 -0.0914 0.0038 »» comparison with Figure (4), it is shown that the state
A=« variables receive their equilibrium values in a shorter time.
« 0 0 7.9021u10-4 2.9043 u10-5 » The low-passed filter poles provide a more speed regulation
« »
¬ 0 0 -8.0657 u10-5 3.69 u10-4 ¼ for the states variables than the high-passed filter. In fact,
0.0349 2.4463º this is predictable because the poles in this case are far from
ª0.005
« 0 the poles of high-passed filter poles. Moreover, e1, e2, e3, e4
1.3061 u104 0 »» (12) in Figures (4) and (5) represent the error which is previously
B=«
« 0 2.5835 u106 0 » defined in Eq. (17).
« »
¬ 0 1.0334 u106 0 ¼
Δx Δx Δy
ª1 0 0 0 º
C=« »
¬0 1 0 0 ¼
D = [0]
Δu
The differences of input variables in the steady state can
be assumed as:
Figure3. Block diagram of the fluidized dryer controller

Δu 0 = u out,0 -u in,0 = >Δu 01 Δu 03 @

20
Δu 02 e1
(13) 15

= > 20 1 0@
state variables(error)

10

5
and state variables can be calculated as:
0
e3

Δx 0 = x out,0 -x in,0 = > Δx 01 Δx 02 Δx 03 Δx 04 @ -5 e4

(14)
= > 2.2592 6 0.0103 0.2819@ -10
e2

-15
0 500 1000 1500 2000 2500 3000 3500
Considering the desired changes in input variables, new time(s)
sets of differences of the input and state variables can be Figure 4. Response to initial values for high-passed filter
obtained as: 20

15
Δu1 = u out,1 -u in,1 (15)
state variables(error)

e1
10

Δx1 = x out,1 -x in,1 (16) 5

e3
0
and finally, the error can be defined as: e4
-5

e = Δx1 -Δx 0 (17) e2

-10
0 500 1000 1500 2000 2500 3000 3500
time(s)
This means that the controller must be able to make the Figure 5. Response to initial values for a low-passed filter
error into zero.

49
 The Linear Quadratic Regulator (LQR) controller is also bucket to bucket. This fact increases the complexity of the
used to measure the control gain. The LQR method place the dryer system dynamics. For this reason, the control system
poles of the closed loop system in [-0.1124, -0.0023+0.0002i, - may never have an ideal regulation activity. Furthermore,
0.0023-0.0002i, -0.0018]. The response of the system with the using the difference values of states or Δx instead of state
designed LQR controller to initial values is shown in Figure variables or x, the control system is designed more
(6) whereas the assumed characteristics for the LQR method facilitated. On the other hand, instead of exact variables as
are: system outputs, an acceptable domain is more achievable.
20
ª50 0 0 0º
« 0 100 0 0 » -------butterworth
Q=« » 15

solid tempreture (error-0c)

-------bessel
«0 0 100 0 » -------LQR
« »
¬0 0 0 1¼ 10
(18)
ª1 0 0 º
5
R = «0 0.1 0 »»
«
«¬0 0 1000 »¼ 0

1
-5
e1 0 500 1000 1500 2000 2500 3000 3500
0 e3 time(s)
e4 Figure 7. Performance comparison of the three mentioned controller for
state variables (error)

-1
solid temperature
-2

-3
As shown in Figures (7) and (8), the regulation time of
grain temperature is more than the time which is needed to
-4 regulate the output moisture content. So, this time of
-5 regulation depends on the time of moisture content control.
e2
Since mechanical limitations are existed in the fluidized bed
-6 dryer actuators, the values of input variables must be set at
-7 an allowed range. Considering this fact, the regulation time
0 500 1000 1500 2000 2500 3000 3500 is increased for the fluidized bed dryer. Also, Figure (9)
time(s)
shows the performance of the designed LQR controller to
Figure 6. Response to initial values for LQR controller control the states/inputs/outputs variables considering the
saturation point of inputs.
4
IV. OBTAINED RESULTS AND DISCUSSIONS
changes of solid moisture (error%)

The most important state variables for grains are the 2

moisture content, and the solid temperature. The temperature
has a crucial effect on the quality of dry grains. On the other 0

hand, the goal of a control system for a dryer is to reduce the

-2
moisture content to a desirable value, and to avoid the
temperature to reach a destructive temperature. A -4
comparison of the three control methods for the solid
temperature and the solid moisture control is shown in -6
Figures (7) and (8) respectively. In fact, the LQR controller ------- Butterworth
is found to be more flexible in the control gain design. It is -8 ------- Bessel
shown in Figure (7) that the speed of regulation for ------- LQR

temperature using the LQR controller is more than other -10

0 500 1000 1500 2000 2500 3000 3500
methods. Also, it is also shown that the LQR controller keep time (s)
the differences of initial and final grain temperature very Figure 8. Performance comparison of the three mentioned controller for
small. This characteristic is an advantage since the control solid moisture
system decreases the moisture content to the desirable value
and avoids the temperature to increase so high. In addition, In addition, Figure (10) shows the performance of the
Figure (8) shows the performance of the LQR controller controller in regulation of outputs variables. The regulation
which controls the difference of initial and final grain time is about 3000 seconds which is needed for the dryer
moisture content. For a dryer plant, the properties of input with this length. After 3000 seconds, when the error is
grains may differ very quickly. So, the moisture content and converge to zero, the dryer would return to its equilibrium
the temperature may differ from time to time and from and the difference between the temperature/moisture in the

50
input and output of the dryer would be set to the initial was designed as the best method to place the closed loop
desired values. poles in the optimum position. As shown in the results, this
controller was able to control the process in about 50
air temperatureΔ (0c)

1 minutes. Since a dryer system has the complex dynamics

0 comparing with other mechatronics systems, this time is
-1 found to be natural. Also, the controller needed more time to
-2
regulate the moisture content than the grain temperature
whereas regulation of moisture content is an area of work to
-3
0 500 1000 1500 2000 2500 3000 3500 reduce the regulation time and optimal performance of the
controller.
air velocityΔ (m/s)

20

10 REFERENCES
[1] Alvarez, P. I., Vega, R., and Blasco, R., “Downer Fluidized
0
Bed Dryer: Modelling and Experimental Results”, 2nd
-10 Mercosur Congress on Chemical Engineering and 4th
0 500 1000 1500 2000 2500 3000 3500 Mercosur Congress on Process Systems Engineering, Costa
1 Verde-Rd- Brazil, 2005.
solid flow Δ (kg/s)

[2] Buschmuller, C., Wiedey, W., Doscher, C., Dressler, J., and
Breitkreutz, J., “Inline monitoring of granule moisture in
0 fluidized-bed dryers using microwave resonance technology”,
European Journal of Pharmaceutics and Bio pharmaceutics,
69, pp. 380–387, 2008.
-1
0 500 1000 1500 2000 2500 3000 3500 [3] Cristea, V. M., Baldea, M., and Agachi, S. P., “Model
Time(s) Predictive Control of an Industrial Dryer”, European Sym.
Figure 9. Input variables regulation for LQR controller with saturation Computer Aided Process Engineering, pp. 271-276, 2000.
[4] Davies, C. and Talion, S., “In-line density-compensated
moisture measurement in free-flowing bulk solids with an off-
This is crucial to point out that as a natural limitation, the the-shelf capacitance level probe”, in proceedings of AIChE
dryer could not increase the moisture content of the grains Annual Meeting, pp. 3197–3200, 2000.
and also could not decrease the temperature and the porosity [5] Gawande, N. A., Reinhart, D. R., Thomas, P. A., Townsend,
of solids. Consequently, positive and negative values in P. T., and Municipal, T.G., “Solid waste in situ moisture
Figure (4) to (10) are referred to the differences between content measurement using an electrical resistance sensor”,
changes in the initial time and changes in the final time (as Waste Management, 23, pp. 667–674, 2004.
seen from Eq. (17)). [6] King, R. J., King, K. V., and Woo, K., “Microwave moisture
measurement of grains”, IEEE Transactions on Instrument
solid moistureΔ (%) solid tempretureΔ ( )

and Measurement, 41, pp. 111–115, 1992.

0c

15
10
[7] Majundar, A. S., “Handbook of industrial drying”, Taylor &
Francis Group LLC, 2006.
5
[8] Nava, J., Palencia, M. A., Salgado, G. C. R., and Rodriguez,
0
M. A., “Robustness of a proportional-integral with
-5 feedforward action control in a plant pilot spray dryer”
-10 Chemical Engineering. Journal, 86, pp. 47–51, 2002.
0 500 1000 1500 2000 2500 3000 3500
[9] Vojtech, V. C., Markvart, M., and Drbohlav, R., “Fluidised
Bed Drying”, Leonard Hill, London, 1966.
0 [10] Wang, H. G., Senior, P. R., Mann, R., and Yang, W. Q.,
-2 “Online measurement and control of solids moisture in
fluidized bed dryers”, Chemical Engineering Science, 64, pp.
-4
2893–2902, 2009.
-6 [11] Ginzburg, A. S. and Savina, I. M., “Mass transfer
0 500 1000 1500 2000 2500 3000 3500 characteristics of food products”, Handbook, LiPP, Moscow.
Time (s) 1982.
Figure 10. Regulation of state variables for LQR controller [12] Menshutina, N. V. and Kudra, T., “Computer aided drying
technologies”, Drying Technol., 19(8), pp. 1825–1849, 2001.
[13] Silva, M. A. and Correa, J. L. G., “Using dryPAK to simulate
V. CONCLUSION drying processes”, IDS98, pp.303–310, 1998.
In this paper, a mathematical model of the fluidized bed [14] Pakowski Z., “Simulation of the process of convective drying:
dryer was studied. Then, a linear model was generated from identification of generic computation routines and their
the original nonlinear model. Next, to reduce the complexity implementation in computer code dryPAK”, Comput.
of system, difference values of states or Δx were used Chemical Eng., 23(1), pp. 719–722, 1999.
[15] Geldart, D., “Types of gas fluidization”, Powder Technology,
instead of state variables x. Also, to control the moisture
7, pp. 285–292. 1973.
content and temperature of the grains, three control methods
[16] Geldart, D., “Characterization of fluidized powders, in Gas
were used and compared for the fluidized bed dryer. Finally, Fluidization Technology”, John Wiley & Sons, New York,
using a comprehensive simulation routine, a LQR method chap. 3. 1986.

51