See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3621756

Fuzzy logic application to power electronics and drives – an overview

Conference Paper · December 1995

DOI: 10.1109/IECON.1995.483333 · Source: IEEE Xplore

CITATIONS
READS
26
241

2 authors:

G.C.D. Sousa
Bimal Bose
Universidade Federal do Espírito
Santo University of Tennessee

41 PUBLICATIONS 938 CITATIONS 178 PUBLICATIONS 13,481 CITATIONS

Some of the authors of this publication are also working on these related projects:

Design of photovoltaic systems View project

All content following this page was uploaded by G.C.D. Sousa on 11 April 2015.
The user has requested enhancement of the downloaded file.
 Fuzzy Logic Applications to Power Electronics and Drives - An Overview

Gilberto C. D. Sousa Bimal K. Bose

Departamento de Engenharia Elétrica
Electrical and Computer Engineering
Universidade Federal do Espirito Santo The University of Tennessee
C. P. 01-9011
Knoxville, TN 37996 - USA
Vitoria, ES 29060-970 - Brazil

Abstract - Applications of fuzzy logic (FL) to power

outperforming other control techniques, such and MRAC
electronics and drives are on the rise. The paper
and sliding mode controllers.
discusses some representative applications of FL in the
In spite of its practical success, fuzzy logic has been
area, preceded by an interpretative review of fuzzy logic
criticized for its limitations, such as the lack of a formal
controller (FLC) theory. A discussion on design and
design methodology and the difficulty in predicting stability
implementation aspects is presented, that also considers
and robustness of FL controlled systems. In reality, these
the interaction of neural networlcs and fuzzy logic
aspects have evolved considerably in the past few years, as
techniques. Finally, strengths and limitations of FLC are the heuristic approaches, commonly used in FLC design
considered, including possible applications in the area. have gradually been replaced by more formal methods.
In the PE&D field, FL has been applied to various
I - INTRODUCTION problems, such as robust feedback control of a phase-
controlled converter dc drive system [ I], efficiency
Fuzzy logic (FL), the logic of approximate reasoning, optimization of drive systems [3], hierarchical control of
continues to grow in importance, as its application to a circulating-current converters [5], slip gain tuning of vector-
number of practical problems further demonstrates its controlled induction motor drive system [6], servo drives [7]-
usefulness. In fact, fuzzy logic has been used in areas such as [ 10], induction machine direct self control drives [11],
process control, estimation, identification, diagnostics, estimation of power electronic waveforms [12], wind
agriculture, medicine, stock market, etc. However, process generation systems [4], ultrasonic motor control [9], etc.
control is by far its most important and visible application. This paper presents an overview of some relevant work in
Lofti Zadeh, father of fuzzy logic, has classified the area, preceded by an insightful review of FLCs. Salient
computing as hard computing and soft computing. The design and implementation aspects of FLCs are also
computations based on Boolean algebra and other crispy considered, along with an analysis of FLC merits and
numerical computations are defined as hard computing, limitations.
whereas fuzzy logic, neural network and probabilistic
reasoning techniques, such as genetic algorithm, chaos II-FUZZY LOGIC CONTROLLERS REVISITED
theory, belief networks and parts of learning theory are
categorized as soft computing. Several interpretations of fuzzy logic systems have been
Soft computing differs from conventional (hard) comr u ting proposed. One can view them as special type expert systems
in that, unlike hard computing, it is tolerant of imprecision, (ES), where the symbolic processing is not as rigorous as in
uncertainty and partial truth. Soft computing is more the conventional ES. On the other extreme, they might be
analogous to thinking of human mind. seen as pure numeric input/output mapping, quite similar to
FuzZy Logic applications to the control of power that of an artificial neural network (ANN). In most
electronics and drive (PE&D) systems have been increasing • r plications, however, both asr ects are simultaneously
exponentially in the pa st few years. Power converter and present, and this ability of providing a mathematical
drive systems possess inherent characteristics, such as non- interpretation from a qualitative descr 'r tion is its major
linearities, unavailability of a precise model or its excessive strength. Some of these aspects will be reviewed here.
complexity, that make them well suited for FL control.
The fuzzy logic controller (FLC) of a given process is A. Fuzzy logic controllet’s as approximnfc'rs
capable of embedding, in the control strategy, the qualitative
knowledge and experience of an operator or field engineer In many situations FL constitutes an efficient way to
about the process. Therefore, fuzzy logic plays the role of a represent a non-analytical mapping, such that an output
suitable "user interface ", in the ta sk of translating designer's value y oan be inferred from a given input x. In order to
insight about the system into the contro1 I aw. resulting in an construct the fuzzy model, the input/output pattern must be
inherently nonlinear adaptive controller, capat le of known, as illustrated in Fig. 1.

0-7803-3026-9/95 $4.00 0 1995 IEEE 57

 Fig. 2: Linear PI controller mapping.

not yet zero, but approaches zero at a satisfactory rate. A

typical rule base for this class of FLCs is given in Fig. 3, in
Fig. 1: Fuzzy system representation. which the sliding mode is identified by the zero valued
diagonal. For the linear PI controller, such condition is
Membership functions can then be assigned to ranges of characterized by An(k)=0 or:
values for both input and output variables, such that proper
covering of their universes of discourse (domain and range)
is obtained. A certain amount of overlap among the
••(i) (-i, / k,) e(k› (2)

membership functions is desirable from a robustness

viewpoint. This corresponds to the intersection of the control plane with
A set of fuzzy rules in the form of "IF x is A THEN y is the (e,ce) plane. For the rule base of Fig. 3, when e(k) is not
B" (A and B are fuzzy sets in the universes of discourse of self-correcting, the control increment An(k) depends on the
x and y, respectively) is obtained from expert knowledge, or sign and magnitude of e(k) and ce(k), and its magnitude is
generated by neural or statistical procedures. It has been somewhat proportional to the distance of the (e,ce) cell to the
demonstrated [13] that a fuzzy system can approximate any sliding line [14]. On the other hand, for the conventional PI
continuous function to any degree of accuracy. It can be seen controller, flu(k) is a linear function of e(k) and ce(k) (l) that
from Fig. 1 that the fuzzy system approximates the unknown is directly related to the distance of the (e(k),ce(k)) point to
function with patches. Higher accuracy can be obtained using the sliding line (2). Consequently, the FLC can be viewed as
a larger number of membership functions, at the expenses of a generalization of the linear PI controller, as depicted in Fig.
increased computational cost. 4.

B. The Fuzzy Proportiona 1 Inte gral (PI) Controller NB NM NS ZE PS PM PB

NB NB NB NB NM NS NS ZE
The large majority of applications of FLC lies in this
NM NB NM NM NM NS ZE PS
category. This class of controllers generates the incremental
NS NB NM MS NS ZE PS PM
control signal (Au(k)) from error (e(k)=y’(k)-y(k)) and error
rate (ce(k)=e(k)-e(k-1)). Reasons for such popularity include ZE N8 NM NS ZE PM PM PB
the fact that humans are more sensitive to variations in the Pg NM NS ZE PS PS PM PB
data been sensed or actuated upon, than to their actual PM NS ZE PS PM PM PM PB
values. Furthermore, this type of controller ensures zero PB
M PS PS PM PB PB PB
steady-state error for most applications, what can not be
Fig. 3: Typical rule base for fuzzy PI controllers.
achieved by FLCs that generate the actual control signal (u)
directly from e(k) and ce(k).
Let us consider a discrete time control system. For a linear
(conventional) PI controller, the incremental control signal is
computed from:
A u(k) = k cetk) + ki e{k) (1)

This corresponds to a linear mapping from input space (e,ce)

to output space (Au), as illustrated in Fig. 2. For a PI type
FLC, the rules are designed with a phase plane in mind,
10 4
where the corrective term Au(k) forces system dynamics to
a r roach that of the sliding mode, where the system error is Fig. 4: Sample fuzzy controller wrapping.

58
Superior performance is achieved when the higher degree of
freedom in the FLC design is properly harnessed.

III- FLCs APPLICATION TO POWER ELECTRONICS AND DRIVES

A. Control ofa phase -controlled converter dc motor driv•

f*7 While conventional PID controllers can be designed

to
8 ive optimum response for a certain operating condition,
their performance deteriorates when parameter variation,
8'
non-
linearities and load disturbance effects are present.
Fig. 5 shows the proposed FL based control for a phase-
controlled, separately excited, dc drive system. Fuzzy PI
controllers, similar to those discussed in section II, are used
in both speed and current loops. For instance, the speed
controller utilizes speed error e(k) and change in error ce(k)
to infer the change in control Au(k), by sup-min composition. ' Disco *nuou ’’

The actual output (command current I,”) is calculated by

discrete integration of control actions (I,‘ =I,’(k-1)+Au(k)).
A second type of fuzzy controller IS employed in the
linearization of converter characteristics, under
Fig. 6: Linearization of converter characteristics.
discontinuous conduction. Its principle of operation is
illustrated in Fig. 6. For continuous current conduction, the
B. Fuzzy logic control of a wind generation system 5]
converter output voltage Vq varies linearly with cosine of
firing angle o. Linearity is lost when the converter enters the
Fuzzy logic can be used to enhance the performance of a
discontinuous
variable speed cage machine wind generation system, shown
conduction region, where V d becomes a function of both tt
and armature current la, as indicated by the trajectory BC. in Fig. 7. The generated power is a non-linear function of
wind velocity and generator speed, as indicated in Fig. 8. In
Converter linearization can be reestablished by proper
order to extract the maximum power, a on-line search is
addition of a compensating AO, such that for a fixed £t
implemented, by FLC-1. For a given wind velocity of Vq
(=80°), the horizontal trajectory BD is obtained. While a
4,
closed-form equation for Ali does not exist, a fuzzy relation
the output power will initially be at point A, corresponding
can be obtained from the numerical solution, for a limited
to a generator speed tell . The FLC1 will alter the
number of lx and I, values. The inherent interpolation
(command) speed in steps AO„ somewhat proportional to
capability of FLC assures proper computation of AO for any
observed change in output power fiP , until the aotual speed
given lx and I, inputs, thus rendering a linear overall
m,2 is reached, corresponding to maximum output power of
converter characteristics. In this way, faster current loop
point B.
response can be obtained for all modes of operation.

TURBINE z

*0

Fig. 7: Wind generation system with fuzzy control.

Fig. 5: Fuzzy-controlled dc drive system.

59
 Fig. 8: Fuzzy controllers FLC- I and FLC-2 operation.
Fig. 9: Modified slip-recovery drive

The process is restarted if wind velocity changes to Vq , as

indicated by trajectory BD, until the new maximum power
output is reached (point E). A second fuzzy controller F\

(FLC2) is used to maximize machine-inverter efficiency, at

light load condition [3). Once FLC1 has driven the generator
speed to its optimum value, such as point E, FLC2 will
reduce the rotor flux by controlling the magnetizing current
command id,’ in adaptive steps. This method guarantees fast
convergence to optimum flux level, and is insensitive to
parameter variation and changes in load condition. The
output power is then further increased to point F, as ’3 ’3
indicated.
A third controller (FLC3) is used to provide robust speed Fig. 10: Drive system phasor diagram.
control against wind vortex and turbine oscillatory torques. It a) Conventional and b) Modified topology;
is a PI type FLC, that infers the change in torque current
reference Aid,’ from speed error and change in error. A hierarchical fuzzy logic control structure, was selected
to implement the control law. The first level, labeled
"instinct", is responsible for Id control. It is a PI-type fuzzy
C. Adaptive fuzzy controller for slip power drives f 6]. controller that generates the corrective duty cycle increment
(Ad) from speed error and change in error. It also
Static Scherbius drives posses inherent low power factor, incorporates I current limiting protection in the rule base.
due to phase-controlled inverter operation, compounded with The second level, that can be likened to the "mechanical
induction machine VAR requirements. A modified scheme, skills", performs two distinct functions. Firstly, it generates
shown in Fig. 9, utilizes a boost converter (GTO1), to the inverter firing angle, based on external speed reference,
directly control the dc rink current Iq, thus freeing the actual speed and duty cycle. Secondly, it generates the actual
inverter to regulate the line fundamental power factor speed reference, required in the first level, by taking into
(displacement factor) to unity value, via proper firing angle account starting conditions, acceleration rate and working
((x) control. parameters.
The principle of operation is illustrated in Fig. 10. In Fig. The third level is responsible for supervisory control. It
10(a) the standard operating mode (with GTO I off) is monitors relevant system variables, including line DF,
shown, for n = 120°, that results in poor line side compare actual state with desired condition, and defines the
displacement factor (DF). In contrast, Fig. 10(b) illustrates corrective actions to be taken by the lower levels.
how a unity DF can be achieved, by simultaneously
controlling inverter firing angle (O=225° ) and GTO1's duty IV-PLC DESIGN AND IMPLEMENTATION ASPECTS
cycle (d), for the same torque and speed condition. Of
course, foroed commutation is required to operate the Typically, the design of a FLC starts with a linguistic
converter in such mode. description of the control strategy, in the form of IF
Proper control strategy had to be learned from extensive (condition) THEN (action) rules, that relate some process
simulation study, that revealed a complex interaction of states (e.g., error and error ohange) to the appropriate control
control variables (lx and d) and output variables (speed, DF, action. The next step consists in deriving a quantitative
Id, etc.) at different system operating conditions.

60
description of the linguistic variables, in the form of fuzzy inference method, that defines how the set of control rules
sets. The design procedure is completed by selecting an (rule base) is evaluated to yield the control action for a
particular process state. Although the general FLC design
procedure is well known, the fine tuning in most cases has
been performed by heuristic methods, and tends to be time
consuming.
Several FLC design and analysis tools have been
introduced in the past few years, such as Togai's TILSHELL.
By providing a graphical interface for the construction of
membership functions and a rule base editor, the design
procedure can be greatly facilitated. However, PE&D
systems are characterized by possessing linear and non-
linear, continuous and discrete subsystems, with time
constants ranging from microseconds to a few seconds, and
consequently are quite demanding on the FL design and
analysis tools. This explains why most researchers still prefer
to use a general purpose simulation tool (such as MATLAB
or SIMNON) in their work.
Recently, a new version of MATLAB was released that
supports a Fuzzy Logic toolbox. Besides greatly increasing
the flexibility of earlier tools, it also helps to consolidate FL
as a mainstream control technique.
A typical FL controlled PE&D system will also contain
conventional control functions, digital filters, plus
monitoring and protection features. Cost considerations, Fig. 11: Neufuz4 block diagram.
combined with the ever increasing capacity of new
microprocessors and DSPs, dictates that all control functions
The multilayered neural net, the Fuzzy Rule Generator
must be implemented in a single processor. Therefore,
(FRG), directly maps weights of different layers into
general purpose processors will be favored, as the solution of
membership functions and fuzzy rules. To simplify the
choice in the majority of applications. Specific fuzzy logic
inference process, the AND operator is interpreted as
ICs will only be competitive at the high end applications. As
product, in the antecedent evaluation (IF portion). The
a consequence, most software tools for FLC design support
consequents (THEN parts) of the rules also consist of
automatic code generation, typically for C or Assembly
singletons, that are accumulated by the output neuron, to
language codes, that can be readily ported into target
yield the crisp control signal.
processors. An example of such tool is discussed next.
The input file should contain enough patters to ensure a
good generalization of the neural net, resulting in the best set
Neuro-fuzzy ! !• f! 5]
of rules and membership functions. Therefore, if the
simulation is not satisfactory, extra iteration are performed,
The assignment of membership functions and fuzzy rules
after supplying more I/O patterns.
generation are normally dependent on the designer's intuitive
The description of the system using fuzzy logic is much
knowledge about the system. As the number of possible rules
cheaper than that using neural nets directly, for real time
increases exponentially with the number of input variables
control. Furthermore, NeuFuz4 will compile the fuzzy
and their fuzzy sets, the difficulty in generating the optimum
rule set increases tremendously with system complexity. control software directly into COP800 microcontroller
A neural net based automated process to generate the fuzzy assembler codes.
rules and membership functions (Neufuz4) has been proposed V- MERITS AND LIMITATIONS OF FLCs
[17], as illustrated in Fig. 11. It requires a numerical
input/output description of the system, that is used to learn While it is true that FL control yields faster, more robust
system behavior, through a modified supervised neural net control systems, the development time tends to be longer
with error baok propagation. Neufuz4 generates bell shaped than that of conventional control systems. The basic reason
membership functions, that at the end are approximated by is the intrinsic higher "degree of freedom" of FLC structure,
trapezoidal ourves, better suited for real time computation. that results in a higher number of parameters to be adjusted.
Another contributing factor is that many application
engineers are not familiar with the new technology. Of
course, the use of design tools can help overcome some of
these limitations, as discussed earlier.

61
FL control has its critics, particularly among the methodologies, including the difficulty in obtaining stability
conventional control theorists. Commonly mentioned and robustness indices. While the criticisms is not unfounded,
weakness include the lack of a formal design and analysis
considerable effort has been devoted to solve these
problems, with relative success. VII - REFERENCES
At the present time, a general design procedure is already
well established. For instance, for the PI type fuzzy [l]G. C. D. Sousa and B. K. Bose, "A fuzzy set theory based control of a
controller, the effects of input and output scaling gains on phase-controlled dc machine drive", IEEE/IAS Annual Meet. Conf. Rec.,
1991, pp. 854-861.
rising time, overshoot and settling time are well understood. (2] L. A Zadeh, "Fuzzy Sets", Information and Control, Vol. 8, 1965, pp.
There is also a olear understanding of the desired structure of 338-353.
the rule base, as discussed in section II. What is left, is the [3]G. C. D. Sousa, B. K. Bose, J. G. Cleland, R. J. Spiegel and P. J.
fine tuning of membership functions and rule base, that is Chappell, "Fuzzy logic based on-line efficiency optimization control of an
normally performed via simulation study or experimentation indirect vector-controlled induction motor drive", IEEE/IECON '93 Conf.
Rec., pp. 1168-1174.
with the actual system. This task can also be automated, as (41M. G. Simoes, B. K. Bose and R. J. Spiegel, "Fuzzy logic based
discussed in [18] for the case of a self-organizing fuzzy intelligent control of a variable speed cage machine wind-generation
controller, and more recently, through the use of genetic system", IEEE/ PESC'95 Conf. Rec., pp. 389-395.
algorithm. [5]L. E. B. da Silva, G. E. April and G. Olivier, "Real time fuzzy adaptive
controller for an asymmetrical four quadrant power converter", IEEE/IAS
It has been also demonstrated [17] that a particular Annual Meet. Conf. Rec, 1987, pp. 872-878.
class of FLC exhibits a behavior similar to that of sliding mode [6]G. C. D. Sousa, B. K. Bose, and Kyung S. Kim, "Fuzzy logic based on-
controllers. Such similarity provides the necessary design line MRAC tuning of slip gain for an indirect vector-controlled induction
gui’del’ines, and at the same time assures the robustness of this motor drive", IEEE/IECON '93 Conf. Rec., pp. 1003-1008.
class of FLC. [7)Y. F. Li and C. C. Lan, "Development of fuzzy algorithms for servo
systems", IEEE Control System Magazine, Vol. 9, no. 3, 1989, pp. 65-72.
The difficulty in deriving a general, non-conservative [8]Y. S. Lu and J. S. Chen, "A self-organizing fuzzy sliding-mode controller
stability criterion for FLCs is also present for other non- design for a class of nonlinear servo systems", IEEE Trans. on Ind.
linear conventional control methods. In fact, most Electronics, Vol. 41, No. 5, Det. 1994, pp 492-502.
approaches to FLC stability analysis are derived from classical [9]Y. Izuno, R Takeda, and M Nakaoka, "New fuzzy reasoning-based high-
performance speed/position servo control schemes incorporating ultrasonic
non-linear dynamic systems theory. It is normally assumed motor, IEEE Trans. on Ind. Applications, Vol. 28, No. 3, May-June 1992.
that a formal model of the plant to be controlled is known, at [10]C. Y. Won, S. C. Kim and B. K. Bose, "Robust position control of
least approximately. induction motor using fuzzy logic control", IEEE/IAS Annual Meet. Conf.
Rec., 1992, pp. 472-481.
VI - CONCLUSION [l1]S. A Mtr, D. S. Zinger, and M. E. Elbuluk, "Fuzzy controller for
inverter fed induction machines" , IEEE Trans. on Ind. Applications, Vol.
30, No. 1, Jan./Feb. 1994, pp. 78-84.
Application of FL to power electronics and drives will [12]M. G. Simoes and B. K. Bose, "Application of fuzzy logic in the
continue to flourish in the near future. Most promising estimation of power electronic waveforms", IEEE/IAS Annual Meet. Conf.
applications are those where classical control solutions can Rec., 1993, pp 853-861.
[l3]B. Kosko, "Fuzzy systems are universal approximators", Proc. of Fuzzy
not provide satisfactory performance, typically when the
IEEE, 1992, pp. 1153-1162.
model of the plant is not available or highly non-linear, or [14]K. L. Tang and R J. Mulholland, "Comparing fuzzy logic with classical
when the system is subject to significant parameter variation. controller designs", IEEE Trans. on Systems, Man, and cybernetics,
FL constitutes a powerful tool to translate human expert Vol. SH4G-17, No. 6, Nov./Dec. 1987, pp. 1085-1087.
knowledge into automatic control strategy, a feature that can [15]S. Thaler, "Fuzzy rule generation based on a neural network approach",
also be exploited in the design of diagnosis and supervisory Electronic Engineering, July 1993, pp 43-30.
control systems. [16]L. E. B. da Silva, B. L. Torres, V. F. da Silva, K. Nakashima, G. E.
Apn 1, G. Olivier, "Adaptive fuzzy techniques for slip-recovery drive
To harness the potential benefits of FL, PE&D engineers control", Proc. of Fuzzy IEEE, 1992, pp. 381-388.
need to become more familiar with the advancements in FLC [17]D. Driankov, H. Hellendoorn, M. Reinfrank, "An introduction to fuzzy
theory, the capabilities of the new design and analysis tools, control", Springer-Verlag, 1993.
in particular those that take advantage of the growing I l . Kosc, G. Griva, "Self-tuning position/speed fuzzy logic controllers for
torque controlled drives", Proc. IPEC-Yokohama, 1995, pp. 1758-65.
interaction of FL with other emerging technologies, [19)T. Takagi and M. Sugeno, "Fuzzy identification of systems and its
particularly neural networks. A challenge worth taking! applications to modeling and control", IEEE Trans. on Syst., Man and
Cyber.,Vol. IS, Jan./Feb., 1985, pp. 116-132.
[20]B. K. Bose, "Expert systems, fuzzy logic, and neural network
applications in power electronics and motion control", Proceedings of UEE, 1994,
pp. 1303-1323.
[21]B. K. Bose, "Fuzzy logic and neural network applications in power
electronics", Proc. IPEC-Yokohama, 1995, pp. 41-45.

62