Академический Документы
Профессиональный Документы
Культура Документы
680
681
1. Introduction
DC motor drives have occupied a wide spectrum of applications, such as lifts, cranes,
robotic manipulators, and paper and textile industries. Separately excited DC motors find
many applications in industries where precise speed control over a wide range is required.
An intelligent control, based on FL or artificial neural network (ANN), can give
robust performance of a non-linear parameter-varying system with load disturbance, and
design of such a system may not require a mathematical model [1, 2].
General proportional plus integral (PI) and proportional plus integral plus derivative
(PID) controllers are widely used for motor control applications, but they do not give
satisfactory results when control parameters, loading conditions, and the motor itself are
changed. In recent years, FL has gained much popularity in many control applications.
The FL controller can be designed without the exact model of the system. The approach
of the fuzzy logic controller (FLC) design guarantees stable operation, even if there is a
change in the parameters of the motor [35].
The main disadvantage of the fuzzy control system is the heavy computation burden
of translating the linguistic control rules into corresponding control actions. It has been
found that the computation burden in the FLC can be reduced if it can be implemented
in an ANN [6]. Intelligent control techniques based on ANNs has been growing tremendously for industrial applications [7]. Intelligent control techniques involving ANNs are
found to be simpler for implementation and powerful in control applications [810]. It has
been proved that neural controllers are better than fuzzy controllers for microprocessor
implementation [11]. ANNs for DC drive application is becoming popular due to their
high computation rate [12, 13]. Drive behavior under PI, FL, and ANN speed controllers
is thoroughly compared in this article.
Initially, a MATLAB/simulink model of the DC motor with the DCDC converter
is developed and simulated. PI and fuzzy controllers are designed for outer-loop speed
control, and the closed-loop operation is simulated. From the designed PI and fuzzy
controllers, training patterns are generated and used to train the ANN to be used as a
controller. The neuro controller was trained using neural network toolbox with the patterns
obtained from the PI and fuzzy controllers. The closed-loop operation is simulated with
the trained neural network, and performance is studied.
2. Proposed System
Figure 1 shows the block diagram of the DC drive system considered in this work. The
system consists of buck-type DCDC power converter for driving the separately excited
DC motor. The buck-type power converter is used so that the speed can be controlled over
a wide rangefrom zero to the rated speed. A microcontroller is used to generate the
pulse width modulated (PWM) waveform required to switch the DCDC power converter.
The 8051 microcontroller-based system is programmed using embedded system design
tools.
The designed closed-loop control has two loops. One is the outer speed control
loop, and the other is the inner current control loop. In the outer speed control loop, the
speed is fed back and is compared with set speed. After comparison, the error signal
and the change in error are calculated and are given as input to the control algorithm.
The algorithm will attempt to reduce the error to zero by changing the duty cycle of the
switching signal applied to the power converter. The power converter in turn changes
682
the armature voltage applied to the separately excited DC motor whose speed is to be
controlled.
An inner-loop current control of a simple on/off type is used because it has the
advantage of using a motor with any specification. Changing ILREF in Figure 1 can
change the operation and limit the motor current to ILREF . One of the easy methods to
change ILREF is by using a potentiometer.
K Vs
:
.La s C Ra /.Js C B/ C KT2
(3)
683
Table 1
DC motor parameters
DC motor parameters
Value
DC supply voltage
Armature resistance Ra
Armature inductance La
Inertia constant J
Damping constant B
Torque constant parameter Kt
Back emf constant Kb
Speed
110 V
1
46 mH
0.093 Kg m2
0.008 Nm/rad/sec
0.55 Nm/A
0.55 V/(rad/sec)
1500 rpm
5. Fuzzy Controller
The speed is fed back and is compared with the reference speed. After comparison,
the error and the change in error are calculated and are given as input to the fuzzy
controller. In this work, the error is normalized to a per-unit value with respect to the
reference speed, which helps in using the fuzzy controller for any reference speed. The
fuzzy controller will attempt to reduce the error to zero by changing the duty cycle of
switching signal. The general PI-like fuzzy controller structure as shown in Figure 2 is
used in this work.
684
5.1. Fuzzification
FL uses linguistic variables instead of numerical variables. The process of converting
a numerical variable (real number or crisp variables) into a linguistic variable (fuzzy
number) is called fuzzification. In the present work, the error and change in error of
speed are fuzzified. Seven linguistic fuzzy sets with triangular membership function are
used as shown in Figure 3. The seven sets used for fuzzy variables error and change
in error are negative big (NB), negative medium (NM), negative small (NS), zero (Z),
positive big (PB), positive medium (PM) and positive small (PS).
5.2. Defuzzification
The reverse of fuzzification is called defuzzification. The FLC produces the required
output in a linguistic variable (fuzzy number). According to real-world requirements, the
linguistic variables have to be transformed to crisp output. Weighted average method is
the best and most well-known defuzzification method for Sugeno type systems, and it is
used in this work. The defuzzified output is the change in duty cycle.
685
The rules are of the format: If error is Ai , and change in error is Bi , then output is C i .
Here, the if part of a rule is called the rule-antecedent and is a description of a
process state in terms of a logical combination of atomic fuzzy propositions; the then
part of the rule is called the rule consequent and is a description of the control output
in terms of logical combinations of fuzzy propositions. This rule table corresponds to
the fuzzy PI controller [15]. From the rule table, the rules are manipulated as follows: If
error is NB, and change in error is NB, then output is NB. As the Sugeno type of fuzzy
control is used, the output is singleton, as shown in Table 2.
5.4.
The Sugeno fuzzy controller uses singleton membership functions for the output variables. This type of fuzzy controller uses the weighted average method of defuzzification
technique. This reduces the calculation required for finding the defuzzified crisp output
after the fuzzy inference mechanism. The controller can be easily implemented in any
embedded system because of the reduction in the calculations. Furthermore, the reduction
in calculation results in the real-time operation. The next change in the duty cycle can
be calculated before a switching cycle. The Sugeno type of controller is selected mainly
for implementation in real-time embedded processors.
NB
NM
NS
PS
PM
PB
NB
NM
NS
Z
PS
PM
PB
1
1
1
1
0.67
0.34
0
1
1
1
0.67
0.34
0
0.334
1
1
0.67
0.34
0
0.34
0.666
1
0.67
0.34
0
0.34
0.67
1
0.67
0.34
0
0.34
0.67
1
1
0.34
0
0.34
0.67
1
1
1
0
0.34
0.67
1
1
1
1
686
Figure 4. Details of the ANN controller trained using fuzzy controller data.
different reference inputs and load conditions. The inaccuracies, non-linearity, and the
limitations of the buck converter and motor are taken care of while creating the training
data from the simulation.
The error (e) and change in error (ce) are the inputs to the controller. The output
corresponds to the change in the duty cycle for the motor control. The data for training
the neural network are obtained from both the conventional PI and fuzzy controllers
designed. The training data are obtained from the simulation of the controller with the
DC motor drive. The details of the designed ANN controller are shown in Figures 4
and 5.
7. Computer Simulation
The simulation of the buck-converter-fed DC motor is done based on the equation
modeling technique in the MATLAB/simulink toolbox. The simulink model developed
is given in Figures 6 and 7.
The fuzzy logic toolbox is used to design the fuzzy controller and the neural network
toolbox is used to design the neuro controller proposed [17]. The simulation was done for
a buck-converter-fed DC motor with the PI, fuzzy, and neuro controllers. The parameters
of the DC motor used are given in Table 1.
687
The computer simulation is run for a step change in motor reference speed, and the
corresponding change in speed is recorded. The step change in load torque is applied,
and the corresponding change in the speed is also recorded.
688
the proper choice of the fuzzy membership functions and inference rules. The control
surface of the designed fuzzy controller is given in Figure 9.
The neuro controller surface can be still made smooth, and the controller can be made
more effective over the given range of the error and change in error. As the training data
are taken during the simulation of the motor, the inaccuracies and the limitations of
the motor are also considered while creating the training data. The controller surface of
the PI-implemented neuro controller is given in Figure 10. The trained neuro controller
689
outperforms the original controllers and can deal with the uncertainties because it has
the different and improved control surface.
The simulated graph of speed variation of the motor for PI, fuzzy, and ANN controllers are given in Figures 1114, respectively. A step change in reference speed from
1000 rpm to 1500 rpm is applied at 3 sec with no load. A step change in load torque from
Figure 11. Graph of speed variation for the step change in reference speed from 1000 rpm to
1500 rpm and for the change in load torque from zero to the rated value at t D 6 sec with the PI
controller.
690
Figure 12. Graph of speed variation for the step change in reference speed from 1000 rpm to 1500
rpm and for the change in load torque from zero to the rated value applied at t D 6 sec with the
fuzzy controller.
zero to the rated value is applied at 6 sec. The results show that the neuro controller
trained using the data from the PI controller has better performance compared with
conventional and fuzzy controllers.
A comparative study of steady-state error and rise time for the controllers under
study is shown in Tables 3 and 4, respectively. The study was done for 60% and 100% of
rated speed without load torque and with full load torque. The artificial neuro controller
trained with data from the PI controller is found to provide low steady-state error and
rise time over the other controllers under comparison.
The robustness of the designed ANN controller was tested by changing the moment of inertia of the motor through simulation. Normally, the conventional controllers
Figure 13. Graph of speed variation for the step change in reference speed from 1000 rpm to 1500
rpm and for change in load torque from zero to the rated value applied at t D 6 sec with the ANN
controller trained with fuzzy controller data.
691
Figure 14. Graph of speed variation for the step change in reference speed from 1000 rpm to 1500
rpm with load torque from zero to the rated value applied at t D 6 sec with the ANN controller
trained with PI controller data.
Table 3
Comparative study of steady-state error (in percent of rated speed)
for controllers under study
Speed
Load
PI
Fuzzy
ANN trained
with fuzzy
controller data
0% of rated torque
100% of rated torque
0% of rated torque
100% of rated torque
1.2
1.0
2.0
8.7
0.55
0.75
1.67
1.86
0.44
0.6
1.0
1.0
ANN trained
with PI
controller data
0.29
0.56
0.05
0.625
Table 4
Comparison of rise time (in sec) for controllers under study
Speed
Load
PI
Fuzzy
ANN trained
with fuzzy
controller
data
0% of rated torque
100% of rated torque
0% of rated torque
100% of rated torque
1.057
11.4
1.9
23.5
0.8313
5.45
1.417
10.5
0.83
4.89
1.42
9.95
ANN trained
with PI
controller
data
0.814
4.6
1.417
9.95
692
Figure 15. Graph of speed variation for the different values of moment of inertia with load torque
from zero to the rated value applied at t D 5 sec with the ANN controller trained with PI controller
data.
performance deviates from the desired parameters, but the ANN controller has given a
better performance that satisfies the design parameters well. The speed regulation curves
for different values of moment of inertia are given in Figure 15. In these curves, a step
change in load torque from zero to the rated value is applied at 5 sec on the motor
running at a speed of 1500 rpm.
9. Experimental Implementation
The designed controllers were implemented by using and programming a Cygnal 8051based processor (C8051F005). A buck converter was built, and the controllers were tested
on the DC motor. A tacho-generator was used to sense the speed and to feed back the
speed signal.
The microcontroller (C8051F005) has a 8051 compatible core with the following
features: 12-bit eight channels ADC, two 12-bit DACs, 2-KB data RAM, and 32-KB
flashin system programmable memory. It also has an in-built PWM waveform generator
that is available as a programmable counter array. The PWM is generated at a frequency
of 14 kHz.
An LEM make current sensor LTS25NP is used to sense the armature current, and
it is compared with the reference current using a comparator. The AND gate is used to
allow the PWM waveform when the actual current is less than the reference current. This
693
Figure 16. Experimental graph of speed variation for the step change in reference speed using the
conventional PI controller.
PWM waveform is then level-amplified and fed to the DCDC power converter through
an isolator and driver chip. The buck converter output is used as a supply to the armature
of the DC motor whose speed is to be controlled. The tacho-generator connected to the
motor shaft gives a DC voltage proportional to the speed, and this DC voltage is fed to
the ADC available in the microcontroller. Figures 1618 show the experimental responses
of a DC motor for a step change in reference speed. Figure 16 shows the response of
the conventional PI controller, while Figures 17 and 18 show the response of the fuzzy
controller and the PI-trained ANN controller, respectively.
Figure 17. Experimental graph of speed variation for the step change in reference speed using the
fuzzy controller.
694
Figure 18. Experimental graph of speed variation for the step change in reference speed using the
ANN controller trained with PI controller data.
10. Conclusion
In this article, the comparison of PI, fuzzy, and ANN controllers for the speed control of
a DC motor is presented. The studies show that the neuro controller is found to be more
optimal than the PI and fuzzy controllers. This advantage arises from the fact that the
neural network has the property of generalization and the neural controller can have a
smooth control surface. Also, the architecture of the designed neural network controller
is simple and uses only two layers with three biased neurons. The neuro controller is
easy to implement and reduces the computation burden in embedded systems. Moreover,
the neuro controller trained using training data from the PI controller has a slightly better
performance compared to neuro controller trained using fuzzy data.
References
1. Bose, B. K., Power electronics and motor drivesrecent technology advances, Proceedings
of the IEEE International Symposium on Industrial Electronics (ISIE), pp. 2225, LAquila,
Italy, 811 July 2002.
2. Bose, B. K., Power electronics and motion controltechnology status and recent trends,
IEEE Trans. Industry Appl., Vol. 29, No. 5, pp. 902909, September/October 1993.
3. Gupta, T., and Boudreax, R., Implementation of a fuzzy controller for DC-DC converters
using an inexpensive 8-bit micro controller, IEEE Trans. Indust. Electron., Vol. 44, No. 5,
pp. 661667, October 1997,
4. Guillemin, P., Fuzzy logic applied to motor control, IEEE Trans. Industry Appl., Vol. 32,
No. 1, pp. 5156, January/February 1996.
5. Senthil Kumar, N., Sadasivam, V., and Prema, K., Design and simulation of fuzzy controller
for closed loop control of chopper fed embedded DC drives, IEEE International Conference
(POWERCON), Vol. 1, pp. 613617, Singapore, 2024 November 2004.
6. Buja, G. S., and Todesco, F., Neural network implementation of a fuzzy logic controller,
IEEE Trans. Indust. Electron., Vol. 41, No. 6, pp. 663665, December 1994.
7. Meireles, M. R. G., Almeida, P. E. M., and Simes, M. G., A comprehensive review for
industrial applicability of artificial neural networks, IEEE Trans. Indust. Electron., Vol. 50,
No. 3, pp. 585601, June 2003.
695
8. Linkens, D. A., and Nyongesa, H. O., Learning systems in intelligent control: An appraisal
of fuzzy, neural and genetic algorithm control applications, IEE Proc. Control Theory Appl.,
Vol. 143, No. 4, pp. 367386, July 1996.
9. El-Sharkawi, M. A., Huang, T. C., and El-Samahi, A., Intelligent control for high performance drives, IEEE International Conference on Electric Machines and Drives Conference
(IEMDC97), Invited Paper, Milwaukee, WI, 1821 May 1997.
10. Lin, B. R., Analysis of neural and fuzzy-power electronic control, IEE Proc. Sci. Measur.
Technol., Vol. 144, No. 1, pp. 2533, January 1997.
11. Wilamowski, B., and Binfet, J., Do fuzzy controllers have advantages over neural controllers
in microprocessor implementation, International Conference on Recent Advances in Mechatronics, pp. 342347, Istanbul, Turkey, 1999.
12. Lazzerini, B., Reyneri, L. M., and Chiaberge, M., A neuro-fuzzy approach to hybrid intelligent
control, IEEE Trans. Industry Appl., Vol. 35, No. 2, pp. 413425, March/April 1999.
13. Barsoum, N., Artificial neuron controller for DC drive, IEEE Power Engineering Society
Winter Meeting, pp. 398402, Singapore, 2327 January 2000.
14. Ericson, R. W., Fundamentals of Power Electronics, Chapman & Hall, New York: International
Thomson Publishing, 1998.
15. Drainkov, D., Hellendoorn, H., and Reinfrank, M., An Introduction to Fuzzy Control, New
Delhi: Narosa Publishing House, 1996.
16. Zurada, J. M., Introduction to Artificial Neural Systems, Mumbai: Jaico Publishing House,
1992.
17. MATLAB, Neural Network Tool Box Users Guide, Version 3, Massachusetts: The Mathworks
Inc.
18. Gomez-Ramirez, E., and Chavez-Placencia, A., How to tune fuzzy controllers, IEEE International Conference on Fuzzy Systems, Vol. 3, pp. 12871292, Budapest, Hungary, 2529 July
2004.