Академический Документы
Профессиональный Документы
Культура Документы
optimal control, -synthesis, and Q-parameterization have also been applied to control
magnetic levitation systems [20]. Control laws based on phase space, linear controller
design, the gain scheduling approach, and neural network techniques [21] have also
been used to control magnetic levitation systems.
In this project design and simulation of a new supervisory control strategy for
magnetic levitation systems that incorporates a fuzzy controller to tune the gains of a
discrete PID controller is studied.
7
2.2 Principle
The basic principle of a simple electromagnetic suspension system is shown in Fig.
2.1. The magnetic force applied by the electromagnet is opposite to gravity and
maintains the suspended steel ball in a levitated position. The magnetic force F
m
depends on the electromagnet current I, electro-magnet characteristics, and the air gap
X between the steel ball and the electromagnet.
Fig. 2.1: Principle of Magnetic Levitation
2.3 System Model Description
Fig. 2.2: Close view of the levitating steel ball
The Magnetic levitation system as shown Fig. 2.2 consists of a magnetic sphere
suspension system. The objective of the system is to control the vertical position of
the ball by adjusting the current in the electromagnet through the input voltage. The
metal sphere is suspended in air by the electromagnetic force generated by an
electromagnet. The Magnetic levitation system consists of an electromagnet, a metal
sphere and an infra-red sphere position sensor. The magnetic ball suspension system
can be categorized into two systems: a mechanical system and an electrical system.
The sphere position in the mechanical system can be controlled by adjusting the
current through the electromagnet where the current through the electromagnet in the
electrical system can be controlled by applying controlled voltage across the
electromagnet terminals.
Magnetic Levitation System
8
Fig. 2.3: Schematic Diagram of the Magnetic Levitation Unit
From Amperes circuit law and faradays inductive law, the magnitude of the force
f(h,i) exerted across an air gap h by an electromagnet through which current i flows
can be described as:
The total inductance L is a function of the distance and given by
Where L
1
is the inductance of the electromagnetic (coil) in the absence of the levitated
object, L
0
is the additional inductance contributed by its presence, and X
0
is the
equilibrium position. The parameters are determined by the geometry and construction
of the electromagnet, and can be determined experimentally. Substituting equation (2)
into (1) yields:
Magnetic Levitation System
)
)
2
2
i dL(h
f h ( , i
dh
= (1)
0 0
1
L h ( ) L = +
L H
h
(2)
2 2
0 0
2
f
L X i
h
h
i
=
=
(3)
9
Fig. 2.4: A magnetic ball bearing system
Eliminating higher order terms give
Evaluating equation (6) using (4) and (5) yields
Where, I
0
is the equilibrium value. At equilibrium, the weight of the object is
suspended by the electromagnet force, f
0
. The force required to maintain equilibrium,
f
1
, is
Combining equations (7) and (8) gives
The voltage equation of the electromagnetic coil is given in equation 1.
Assuming the suspended object remains close to its equilibrium position, h=h
0
, and
therefore
Also assuming that L
1
>> L
0
, equation (10) can be simplified as
Magnetic Levitation System
0 0
2
=
L H
(4)
0 0 0
0 0
f f f f
f = f h
H i h
+ H
+ i +
L
+ L
(5)
0
= f f h
i
h
+
i +
f f
(6)
2
0 0 0
2 3
0 0 0
2
f
i h
H H H
I
= +
I 2 I
(7)
1 0
f = f f
2
0 0
1 2 3
0 0
i f
I
h
H H
I 2 2
(9)
(8)
V i = + R L( ) h
di
dt
(10)
1 0
L( ) h L = + L (11)
10
The principal equation for the suspended object comes by applying Newtons second
law of motion. For this one degree of freedom system, a force balance taken at the
centre of gravity of the object yields
The sensor can be modelled as a gain element,
Where V
s
is the sensor output voltage and K
s
is an experimental gain between the
objects position and the output voltage.
The Laplace transform of above equation obtained as:
The Laplace transform of equation (28) is
The overall transfer function of the Maglev system is obtained as:
Table 1 summarizes the variables and parameters use in this problem. Here the
problem is to maintain the ball at its operating point (position) of 0.03 meters from the
coil.
Magnetic Levitation System
1
V i = + R L
di
dt
(12)
2
1 2
M
d h
= f
dt
(13)
s s
= V K h (14)
2 2
0 0
2 2 3
0 0
M
d h 2 2 I I
h
dt H H
=
i
(15)
2
1 2 2
M
d h
K i = K h
dt
(16)
Where,
0
1 2
0
K =
2I
H
and
2
0
2 3
0
K =
2I
H
(17)
(18)
1
1
2
2
1
G s ( )
s
=
V s ( )
V s ( )
s
K K
ML
K R
L M
s s +
(19)
I s ( ) =
V s ( )
L + s R
(Ms
2
K
2
)H(s) = K
1
I(s)
11
Table1. Parameters of the Magnetic Levitation system
Parameters Description Values
H
0
Equilibrium height of ball (m) 0.03
M Mass of ball bearing (kg) 0.225
R Resistance () 2.48
L
1
Inductor (H) 0.18
Constant related to magnetic force (Nm
2
/A
2
) 7.9310
-5
I
0
Equilibrium current of the coil (A) 5
K
s
Sensor gain factor (V/m) 200
K
1
constant (N/A) 0.882
K
2
constant (N/m) 147
Fig. 2.5: Block diagram of Magnetic levitation system
Magnetic Levitation System
12
Chapter-3
Fuzzy Logic
Fuzzy logic is a logical system providing a mathematical framework to capture the
uncertainties associated with human cognitive systems such as thinking and reasoning.
Simply, it simulates human thinking which operates more likely on symbols than
exact values. In fact, our daily thoughts and communication are full of these symbols
or fuzzy expressions. This chapter gives a brief introduction to the main concepts of
fuzzy logic.
3.1 Fuzzy sets
In conventional set theory, an element either belongs to the set or not. Fuzzy logic is a
generalization of the conventional logic. In fuzzy set theory, the element can belong to
the set partially with a certain degree. The difference between conventional crisp and
fuzzy sets is illustrated in Fig. 3.1. Let us consider tree example sets the poors, the
averages and the richs in the universe of discourse wealth. In conventional logic,
persons are divided into these three groups crisply. The fuzzy sets have no crisp
boundaries, but a person can simultaneously be a member of several groups with
different degrees. For example, a person can be rich with degree of 0.1 and average
with 0.7.
Fig. 3.1: Conventional sets and fuzzy sets.
3.1.1 Linguistic variables
The main advantage of fuzzy logic is that words or sentences can be used as
expressions instead of numeric values. The associative expressions are called
linguistic variables. They are common in our daily life. Let us consider the fuzzy
variable velocity. It can be, for example, divided into three linguistic variables: slow,
medium, and fast which are fuzzy sets, as show in Fig. 3.2. Each linguistic variable is
represented by membership function in the universe of discourse. For example, the
membership function of the linguistic variable slow could be defined by
13
The membership function values can vary between zero and unity and they can have
many shapes, as shown in Fig. 3.3. The selection of the shape for a fuzzy set is
subjective and particular rules do not exist, but the singleton type of membership
function (fuzzy unit set) is usually employed only for the output variables of the fuzzy
reasoning.
Fig. 3.2: Terms of fuzzy logic.
Fig. 3.3: Different shapes of membership functions. (a) Z-shaped, (b) trapezoidal,
(c) bell-shaped, (d) triangular and (e) singleton.
3.2 Fuzzy control
Fuzzy control is useful in some cases where the control processes are too complex to
analyse by conventional quantitative techniques. Fuzzy control design is very
interesting for industrial processes where modelling is not easy to make or conception
of nonlinear controllers for industrial processes with models. The available sources of
information of a process are interpreted qualitatively, inexactly or uncertainly. The
main advantages of fuzzy logic control remains in [1]:
Parallel or distributed multiple fuzzy rules complex nonlinear
Linguistic control, linguistic terms human knowledge
Robust control
3.2.1 Fuzzy control system design
Fig. 3.4 gives the fuzzy controller block diagram, where we show a fuzzy controller
embedded in a closed-loop control system. The plant outputs are denoted by y(t), its
Fuzzy Logic
slow
(v)
1,
1
25
-
v
----
-----
3
---
5
-
,
v [0, 35] ,
v [35, 60] .
=
(20)
14
inputs are denoted by u(t), and the reference input to the fuzzy controller is denoted by
r(t). The design of fuzzy logic controller is based on four main components [1]:
1. The fuzzification interface which transforms input crisp values to fuzzy
values
2. The knowledge base which contains knowledge of the application domain
and the control objectives
3. The decision-making logic which performs inference for fuzzy control
actions
4. The defuzzification interface which provides the control signal to the
process.
Fig. 3.4: Fuzzy controller diagram
3.3 Fuzzy reasoning
3.3.1 Fuzzy rules
Fuzzy reasoning is usually performed using if -then rules. The fuzzy rules define the
connection between input and output fuzzy (linguistic) variables. The rule consists of
two parts: an antecedent and a consequence part. The Inference block is used to link
the input variables to the output variable denoted X
R
and considered as a linguistic
variable given by a set of rules:
X
R
= (I F (condition 1), THEN (consequence 1) OR
I F (condition 2), THEN (consequence 2) OR
OR
I F (condition n), THEN (consequence n).
n corresponds to the product of the number of membership functions of each input
variable of the fuzzy logic controller.
In these rules, the fuzzy operators AND, OR link the input variables in the
condition while the fuzzy operator OR links the different rules. The choice of these
operators for inference depends obviously on the static and dynamic behaviours of the
Fuzzy Logic
15
system to control. The numerical processing of the inference is carried out by three
methods [22]:
1. max-prod inference method
2. max-min inference method
3. sum-prod inference method
3.3.2 Fuzzy inference system
The fuzzy inference system (FIS) performs fuzzy reasoning. The basic FIS is
composed of five functional blocks, as depicted in Fig. 3.5. The knowledge base
consists of the data base and the rule base. The fuzzy sets are defined in the data base
and fuzzy rules in the rule base. The decision-making unit executes fuzzy reasoning
rules taking fuzzified inputs of FIS as inputs and delivering the fuzzy result to the
defuzzifier, which produces the output of the FIS.
Fig. 3.5: The structure of the fuzzy logic inference system.
The operation of the FIS is illustrated in Fig. 3.7. First, crisp inputs x and y are fed into
a FIS. In the second stage, they are fuzzified. After that, the fuzzified inputs are
combined according to the fuzzy rules in the knowledge base. Finally, the results of all
rules are combined and defuzzified. In the following, each stage is described in more
detail.
3.3.3 Fuzzification
In the fuzzification, the crisp input values are transformed to fuzzy values. If the input
has a crisp value, the matching against the membership function of linguistic variable
is shown in Fig. 3.6(a). If the input contains noise, it can be modeled by using a fuzzy
input value. In this case the fuzzy output is the intersection of fuzzy input and the
linguistic variable member- ship functions as shown in Fig. 3.6(b). However, the crisp
input value fuzzification is mostly used because of its simplicity.
The fuzzification block contains generally preliminary data which are obtained from:
Conversion of measured variables with analog/digital converters.
Preprocessing of the measured variables in order to get the state, error,
state error derivation and state error integral of the variables to control
(output variables or other state variables).
Fuzzy Logic
16
Choice of membership functions for the input and output variables namely
the shape, the number and distribution. Usually three to five triangular or
Gaussian membership functions are used with a uniform distribution
presenting 50% of overlapping. More than seven membership functions,
the algorithm processing becomes long and presents a drawback for fast
industrial processes.
Fig. 3.6: Fuzzification of a crisp input (on the left) and a fuzzy input (on the right).
3.3.4 I nference
The decision making unit performs the inference operations on the fuzzy rules. The
fuzzy values within a fuzzy rule are aggregated with connective operators like
intersection (AND), union (OR) and complement (NOT). Due to the use of the multi-
valued logic, the connective operators for fuzzy logic differ from the ones used in the
Boolean logic. The operators can be defined in several ways, but the following are the
best-established ones [23]:
Fig. 3.7: The fuzzy inference using the Min-inference.
Intersection
Fuzzy Logic
, (21)
, (22)
AND (
A
,
B
) = min {
A
,
B
}
AND (
A
,
B
) =
A
B
17
Union
Complement
Where
A
and
B
are membership values which are combined by operators. The firing
strengths of the fuzzy rules are computed by employing above operators. The
operation of the intersection is shown in Fig. 3.7. The final output fuzzy sets are
obtained either scaling (Max-Dot method) or cutting (Max-Min) according to the
firing strength of the fuzzy rules. If the output fuzzy sets are singletons, they are not
handled by the firing strengths in this stage.
3.3.5 Defuzzification
In the defuzzification stage, the outputs of the fuzzy rules are combined to a crisp
output value. Several defuzzification strategies have been suggested [24]. The most
common method is the center of area (COA) defuzzification strategy, illustrated in
Fig. 3.8. Assuming a discrete universe of discount, the crisp output Z is produced by
searching the center of gravity of consequence fuzzy sets according to
where m is the number of quantization levels of the output, z
i
is the amount of output
at the quantization level i, and
i
(z
i
) represents its membership value in C.
If only singletons are used as the consequences of fuzzy rules, the natural
defuzzification method is the weighted average (WA). It can be considered as a
special case of COA defuzzification method is the weighted average (WA). The WA
method combines the consequences of the fuzzy rules to the out- put of the inference
system z according to
where n is number of fuzzy rules,
i
is the firing strength of the rule, and is the output
value of the ith singleton.
Fuzzy Logic
, (23)
,
(24)
OR (
A
,
B
) = max {
A
,
B
}
NOT (
A
) = 1
A
,
(25)
Z
m
i
(z
i
)
i = 0
C
(z
i
) z
i
=
i
----
=
----
0
m
-------------------------
(26)
Z
i
z
i
i = 0
n
i
i = 0
n
= ---------------------
18
Fig. 3.8: Defuzzification methods
On the left, three fuzzy rules which have singleton output fire. The output is computed
by using weighted average strategy. On the right, two fuzzy rules fires. The crisp
output is the centre of the area.
3.4 Different types of Fuzzy Logic controllers
On the basis of the consequence of rules given above, different types of fuzzy
logic controllers are presented.
1. If the consequence is a membership function or a fuzzy set, the fuzzy
controller is Mamdani type. In this case, the processing of inference uses often the
max-min or max- prod inference method while for defuzzification, the center of
gravity method is often used and in some cases we use the maximum value method if
fast control is needed.
2. If the consequence is a linear combination of the input variables of the fuzzy
logic controller. Indeed each rule corresponds to a local linear controller around a
steady state. Consequently, the set of the established rules correspond to a nonlinear
controller. In this case, we use max-min or max-prod inference method and for
defuzzification, we often use the weight average method.
Also, there exist other types of fuzzy logic controllers such as Larsen or Tsukamoto
[25]. Most of time Mamdani and TSK controllers are used in the design of controllers
for nonlinear systems with or without models [22]. The advantages of the design of a
fuzzy logic controller using Mamdani type are an intuitive method, used at a big scale
and well suited for translation of human experience on linguistic rules.
On the other hand, the advantages of a fuzzy logic controller using a Takagi-Sugeno
type are:
Good operation with linear techniques (the consequence of a rule is linear)
Good operation with optimization techniques and parameters adaptation of
a controller
Continuous transfer characteristics very suited for systems with a model
fast processing of information
Fuzzy Logic
19
3.5 Discussion
The different steps followed in the processing of the input variables of the fuzzy logic
controller namely fuzzification, inference and defuzzification, allows obtaining a
nonlinear characteristic. Indeed its an advantage when compared to the classical
control. The nonlinearity of this characteristic depends on some parameters. For
example the number, the type and the distribution of membership functions. Also,
other parameters can be considered such as the number of rules and inference
methods. Finally, the nonlinearity can be more or less pronounced depending on all
these parameters.
In this case we consider the fuzzy logic controller as a nonlinear controller. Another
possibility to get a nonlinear controller is to design and to add a fuzzy supervision to a
PID controller. Industrialists are motivated to keep PID controllers which are well
known and to add a fuzzy supervisor which modifies on-line PID parameters in order
to reach and to maintain high performances whatever the parameters change and
operations conditions maybe. In the design of the fuzzy supervision, the outputs are
the PID parameters to provide on-line to the PID controller.
Fuzzy Logic
20
Chapter-4
Fuzzy Controller
4.1 Fuzzy supervisory control
Fuzzy Supervisory controller is a multilayer (hierarchical) controller with the
supervisor at the highest level, as shown in Figure 4.1. The fuzzy supervisor can use
any available data from the control system to characterize the systems current
behavior so that it knows how to change the controller and ultimately achieve the
desired specications. In addition, the supervisor can be used to integrate other
information into the control decision-making process.
Fig. 4.1: Fuzzy Supervisory controller
Conceptually, the design of the supervisory controller can then proceed in the same
manner as it did for direct fuzzy controllers (fuzzification, inference and
defuzzification): either via the gathering of heuristic control knowledge or via training
data that we gather from an experiment. The form of the knowledge or data is,
however, somewhat different than in the simple fuzzy control problem. For instance,
the type of heuristic knowledge that is used in a supervisor may take one of the
following two forms:
1. Information from a human control system operator who observes the behavior of
an existing control system (often a conventional control system) and knows how
this controller should be tuned under various operating conditions.
2. Information gathered by a control engineer who knows that under dierent
operating conditions controller parameters should be tuned according to certain
rules.
Fuzzy supervisor is characterized by:
1. The outputs which are not control signals to provide to the control system but
they are parameters to provide to the controller in order to compute the
appropriate control.
2. Fuzzy supervision associated to the controller can be considered as an
adaptive controller
21
3. Fuzzy supervisor can integrate different types of information to resolve
problems of control.
4.2 Supervision of conventional controllers
Most controllers in operation today have been developed using conventional control
methods. There are, however, many situations where these controllers are not properly
tuned and there is heuristic knowledge available on how to tune them while they are in
operation. There is then the opportunity to utilize fuzzy control methods as the
supervisor that tunes or coordinates the application of conventional controllers. In this
part, supervision of conventional controllers concerns only PID controllers and how
the supervisor can act as a gain scheduler.
4.3 Fuzzy tuning of PI D controllers
Over 90% of the controllers in operation today are PID controllers. This is because
PID controllers are easy to understand, easy to explain to others, and easy to
implement. Moreover, they are often available at little extra cost since they are often
incorporated into the programmable logic controllers (PLCs) that are used to control
many industrial processes. Unfortunately, many of the PID loops that are in operation
today are in continual need of monitoring and adjustment since they can easily
become improperly tuned.
Because PID controllers are often not properly tuned (e.g., due to plant parameter
variations or operating condition changes), there is a signicant need to develop
methods for the automatic tuning of PID controllers for nonlinear systems where the
model is not well known. In this method, the fuzzy supervisor knows, from a response
time, when the controller is not well tuned and acts by adjusting the controller gains in
order to improve system performances. The principle scheme of the fuzzy PID auto
tuner [1] is given by Fig. 4.2.
Fig. 4.2: Fuzzy PID auto-tuner
Fuzzy Controller
22
The block Behavior Recognition is used to characterize and analyze the current
response of the system and provides information to the PID Designer in order to
determine the new parameters of the PID controllers and to improve performances.
The basic form of a PID controller is given by:
Where u is the control signal provided by the PID controller to the plant.
e is the error deuced from the reference input r and the plant output y.
K
p
is the proportional gain, K
i
is the integral gain, and K
d
is the derivative
gain.
In this case, the adjustment of PID parameters is carried out by some candidate rules
as follows
If steady-state error is large Then increase the proportional gain.
If the response is oscillatory Then increase the derivative gain.
If the response is sluggish Then increase the proportional gain.
If the steady-state error is too big Then adjust the integral gain.
If the overshoot is too big Then decrease the proportional gain.
In these rules conditions are deal with the block Behavior Recognition" and
consequences are evaluated by the block PID Designer of the fuzzy supervisor. In
some applications controller gains are quantified according to different types of
responses a priori identified from experiments on the real process and implemented on
the block Behavior Recognition [1].
4.4 Fuzzy gain scheduling
Conventional gain scheduling involves using extra information from the plant,
environment, or users to tune (via schedules) the gains of a controller. The overall
scheme is shown in Figure 4.3. A gain schedule is simply an interpolator that takes as
inputs the operating condition and provides values of the gains as its outputs. One way
to construct this interpolator is to view the data associations between operating
conditions and controller gains.
Fig. 4.3: Conventional fuzzy gain scheduler
Fuzzy Controller
t
D
0
d
u(t)=K e(t)+K e() d+K
e(t)
dt
(27)
P I
23
The controller gains are established on the basis of information collected from the
plant to control, the operator or the environment. Three approaches are proposed for
the construction of the fuzzy gain scheduling [1]:
Heuristic Gain Schedule Construction
Construction of gain schedule by fuzzy identification
Construction of gain schedule using the PDC method (Parallel
Distributed Compensation method)
4.4.1 Construction of a heuristic schedule gain
This method is applied for plants with specific particularities not involved in the
design of classical controllers. The PID parameters are deduced intuitively and the
rules used for the adjustment of parameters are heuristic. This is for example the case
of a tank with an oval shape (Fig. 4.4). In the heuristic rules, the condition
corresponds to the water levels and the consequence corresponds to the values of the
controller gain [1]. Each rule covers a set of water levels taking into account the tank
section. For low levels, the gain is higher in order to get high flow rates and for high
water levels, the gain is small in order to get small flow rates. This approach is very
useful for systems without models.
Fig. 4.4: Tank
4.4.2 Construction of a schedule gain by fuzzy identification
This approach is useful for plants where we know a priori how to adjust the controller
gains under different operation conditions [1]. For example if a control engineer
knows how to adjust gain controller according to certain rules, he can represent this
data by a fuzzy model of Mamdani or TSK type. Indeed its the equivalent of a set of
controllers which are active in terms of the operation points. Also, the gain controllers
are deduced on-line by the inference mechanism between controllers for any operation
point. Indeed its a soft transition from controller to another one.
4.4.3 Construction of a gain schedule using the PDC method
This approach is applied particularly for processes that can be modelled. Most of time,
the established models are nonlinear. In this case, the nonlinear model is replaced by a
sum of linearized models around different operation points [1]&[26]. For each
Fuzzy Controller
24
linearized model, a linear controller is designed (Fig. 4.5). These linear controllers
could be PI, PD or PID state controllers. The set of the designed controllers is finally a
nonlinear controller which is a fuzzy controller.
In this approach the n linearized models and the n corresponding controllers are rules
which are active simultaneously two by two since the condition is similar for both,
thus the name of the method "Parallel Distributed Compensation.
In all the approaches presented above, performances are not used directly when
designing controllers. Also non linearity, disturbances and variation parameters of the
plant are not taken into account in the systems with models.
In some cases stability is not ensured particularly when a change set point occurs or
when a disturbance is present. In the case of the PDC approach, local and global
stabilities are checked using Lyapunov theory [1]. For the other approaches, stability
is checked when implementing fuzzy supervision for classical controllers.
Fig. 4.5: PDC concept
Usually specifications and performances in closed loop are a priori defined. Therefore,
its more interesting to use them and to design controllers ensuring stability and same
performances in closed loop whatever the operation conditions maybe.
Fuzzy Controller
25
Chapter-5
Tuned PID Controller
5.1 Overview
Tuning of PID controllers has always been an area of active interest in the process
control industry. Ziegler Nichols Method (ZN) is one of the best conventional
methods of tuning. Though ZN tunes systems very optimally, a better performance is
needed for very fine response and this is obtained by using Fuzzy Logic (FL)
methodology which is highly effective.
5.1.1 Tuning and its Purpose
A PID may have to be tuned when:
Careful consideration was not given to the units of gains and other parameters.
The process dynamics were not well-understood when the gains were first set,
or the dynamics have (for any reason) changed.
Some characteristics of the control system are direction-dependent (e.g.
actuator piston area, heat-up/cool-down of powerful heaters).
You (as designer or operator) think the controllers can perform better.
Always remember to check the hardware first because there are many conditions
under which the PID may not have to be tuned. These conditions are when:
A control valve sticks. Valves must be able to respond to commands.
A control valve is stripped out from high-pressure flow where the valves
response to a command must have some effect on the system.
Measurement taps are plugged, or sensors are disconnected. Bad
measurements may have you correcting for errors that dont exist. Once fix
these hardware problems then depending on the responses we obtain an
appropriate decision can be taken whether or not to tune a PID controller. [27]
5.2 Trial and error method
This process is a very time consuming process as a lot of permutations and
combinations are involved. Though much iteration is performed the final result is not
satisfactory. A balance is not obtained between the rise time and % overshoot even
though a lot of possible combinations of the gains are incorporated. Continuous
cycling may be objectionable because the process is pushed to the stability limit.
Consequently, if external disturbances or a change in the process occurs during
controller tuning, it results in unstable operation. The tuning process is not applicable
to processes that are open loop unstable because such processes typically are unstable
at high and low values of K
c
but are stable at an intermediate range of values. It can be
observed in Fig. 5.1 that large overshoot is obtained as the program is written for
faster rise time hence compromising with overshoot. All the time response
specifications cannot be balanced using trial and error method.
26
Fig. 5.1: Unit step response of the system G(s) tuned with trial and error method.
5.3 Pole placement method
If the process is described by a low-order transfer function, a complete pole placement
design can be performed. Consider for example the process described by the second-
order model.
This model has three parameters. By using a PID controller, which also has three
parameters, it is possible to arbitrarily place the three poles of the closed loop system.
The transfer function of the PID controller in parallel form can be written as
The characteristic equation of the closed loop system becomes
A suitable closed-loop characteristic equation of a third-order system is
which contains two dominant poles with relative damping () and frequency (), and a
real pole at . Identifying the coefficients in these two characteristic equations
determines the PID parameters K, T
I
and T
D.
The solution is
Tuned PID Controller
) =
( )
2 1
1+ sT 1+ )( sT
G(s
K
P
(28)
)
I
I D I
C
sT
G (s) =
K s T T
2
(1+ sT +
(29)
1 1 1
1 2 1 2 1 2 1 2 2 1
2 3
= +
+ +
+ +
TTT
KK
TT
KK
TT
s
TT
KKT
T T
s +s
I
P P P D
0
(30)
(s + )(s
2
+ 2s +
2
) = 0
(31)
( )
P
K
1
K =
T T
2
1+ 2
2 1
(32)
2
27
Fig. 5.2: Response of a system tuned with Pole Placement Method.
5.4 Ziegler Nichols method
Ziegler Nichols formula ensures good load disturbance attenuation, but it generally
provides a poor phase margin and therefore it produces a large overshoot and settling
time in the step-response. The overall control scheme for Ziegler Nichols Method is
shown in Fig 5.3.
Fig. 5.3: Control Scheme for Ziegler Nichols Method.
Tuned PID Controller
( )
3
2 1
2 1
1 2 1
T T
T =
T T
I
+
(33)
( )
( ) 1 2 1
2
2
2 1
2 1 2 1
+
T
+
T T
T
T =
T T
D
(34)
(35)
(t)
+
1
e()d
t
i
d p
dt T
u(t) =K e(t) +T
de
0
28
Fig. 5.4: Unit step response of the system G(s) tuned with ZL method.
Tuned PID Controller
29
Chapter-6
Fuzzy Supervised PID-Controller
The fuzzy logic controller for the given Maglev system is designed in the following
sections. The various blocks related to this controller and rules governing the
operation of the controller are also detailed in the subsequent sections.
6.1 Fuzzification
The success of this work, and the like, depends on how good this stage is conducted.
In this stage, the crisp variables e and de are converted in to fuzzy variables e and
de respectively. The membership functions associated to the control variables have
been chosen with triangular shapes as shown in Fig. 6.1 - Fig. 6.5. The universe of
discourse of all the input and output Variables are established as the suitable scaling
factors are chosen to brought the input and output variables to this universe of
discourse. Each universe of discourse is divided into seven overlapping fuzzy sets: NL
(Negative Large), NM (Negative Medium), NS (Negative Small), ZE (Zero), PS
(Positive Small), PM (positive Medium), and PL (Positive Large). Each fuzzy variable
is a member of the subsets with a degree of membership varying between 0 (non-
member) and 1 (full-member). All the membership functions have asymmetrical shape
with more crowding near the origin (steady state). This permits higher precision at
Steady state [27].
Fig. 6.1: Triangular Membership functions of input variable error.
Fig. 6.2: Triangular Membership functions of input variable error-rate.
30
6.2 I nference engine
Knowledge base involves defining the rules represented as IF-THEN statements
governing the relationship between input and output variables in terms of membership
functions. In this stage, the variables e and de are processed by an inference engine
that executes 25 rules (5x5) as shown in Table 2. These rules are established using the
knowledge of the system behavior and the experience of the control engineers. Each
rule is expressed in the form as in the following example: IF {e is Negative Large}
AND {de is Positive Large} THEN {*fuz is Zero}. Different inference engines can be
used to produce the fuzzy set values for the output fuzzy variable *fuz. In this work,
the Max-product inference method [27] has been used.
Table 2: Rule base Parameters for K
P
.
de NL NS ZE PS PL
e
NL PVL PVL PVL PVL PVL
NS PML PML PML PL PVL
ZE PMS PS PVS PMS PMS
PS PML PML PML PL PVL
PL PVL PVL PVL PVL PVL
Fig. 6.3: Triangular Membership functions of output variable K
P
.
Table 3: Rule base Parameters for K
I
.
de NL NS ZE PS PL
e
NL PM PM PM PM PM
NS PMS PMS PMS PMS PMS
ZE PS PS PVS PS PS
PS PMS PMS PMS PMS PMS
PL PM PM PM PM PM
Fuzzy Supervised PID-Controller
31
Fig. 6.4: Triangular Membership functions of output variable K
I
.
Table 4: Rule base Parameters for K
D
De NL NS ZE PS PL
e
NL PVS PMS PM PL PVL
NS PMS PML PL PVL PVL
ZE PS PMS PVS PL PVL
PS PML PVL PVL PVL PVL
PL PVL PVL PVL PVL PVL
Fig. 6.5: Triangular Membership functions of output variable K
D
.
6.3 Rule Base
A decision making logic which is stimulating a human decision process, infers fuzzy
control action from the knowledge of control rules and linguistic variable definitions.
The rules are If-Then format and formally the If side is called the condition and the
Then side is called conclusion. The computer is able to execute the rules and compute
the control signals depending on the measured input error (e) and change in error (de).
In a rule based controller the control strategy is stored in more or less natural
language. A rule base controller is easy to understand and easy to maintain for a non-
specialist end user and equivalent controller could be implemented using control
techniques.
Fuzzy Supervised PID-Controller
32
Fig. 6.6: Fuzzy Logic Rule-Base in SIMULINK.
Fig. 6.7: Surface plot of Fuzzy Logic Rule-Base for variable K
P
.
Fuzzy Supervised PID-Controller
33
Fig. 6.8: Surface plot of Fuzzy Logic Rule-Base for variable K
I
.
Fig. 6.9: Surface plot of Fuzzy Logic Rule-Base for variable K
D
.
6.4 Defuzzification
The reverse of fuzzification is called as defuzzification. The use of Fuzzy Logic
Controller (FLC) produces output in linguistic variables (Fuzzy number). According
to real world requirements, the linguistic variables have to be transformed to crisp
output.
In order to define fuzzy membership function, designers choose many different shapes
based on their preference and experience. There are generally four types of
membership functions used:
Fuzzy Supervised PID-Controller
COGS
=
C
(x
i
).x
i
G
(x
i
)
(36)
34
Trapezoidal MF
Triangular MF
Gaussian MF
Generalized bell MF
Fig. 6.10: Structure of fuzzy logic controller.
Implementation of an FLC requires the choice of four key factors:
I. Number of fuzzy sets that constitute linguistic variables.
II. Mapping of the measurements onto the support sets.
III. Control protocol that determines the controller behavior.
IV. Shape of membership functions.
PID parameters fuzzy self-tuning is to find the fuzzy relationship between the three
parameters of PID and "e" and "de", and according to the principle of fuzzy control, to
modify the three parameters in order to meet different requirements for control
parameters when "e" and "de" are different, and to make the control object a good
dynamic and static performance [29].
6.5 Adjusting fuzzy membership functions and rules
In order to improve the performance of FLC, the rules and membership functions are
adjusted. The membership functions are adjusted by making the area of membership
functions near ZE region narrower to produce finer control resolution. On the other
hand, making the area far from ZE region wider gives faster control response. Also the
performance can be improved by changing the severity of rules. An experiment to
study the effect of rise time (T
r
), maximum overshoot (M
p
) and steady-state error
(SSE) when varying KP, KI and KD was conducted. The results of the experiment
were used to develop 25-rules for the FLC of KP, KI and KD are the output variables
and from error and change of error are the input variables. Triangular membership
functions are selected.
Fuzzy Supervised PID-Controller
35
Fig. 6.11: Block diagram of self-tuning Fuzzy Supervised PID controller.
6.6 Results and Discussion
Fig. 6.12: SIMULINK model of self-tuning Fuzzy Supervised PID controller.
Fig.6.12 shows the SIMULINK model of the Maglev system in addition to the
Fuzzy tuned PID controller. The results shows that the controlled electromagnet
current can stabilize the disturbances that otherwise, would cause the ball to either fall
or attach itself to the electromagnet. From the simulation results shown in Fig. 6.14, it
can be observed that the fuzzy controller has better transient response than the
classical controller. The overshoot of the FLC controller is 6% compared to 18% in
the classical case. Furthermore, FLC has a faster transient response; it reaches to
steady state in 0.9 sec to that of 2 sec in classical PID. In comparison to the steady
state value, both controllers satisfactorily attain the steady state value.
According to the different responses, its clear that performances are better with the
fuzzy control (Fuzzy supervised PID controller) in comparison to the classical control.
Indeed, the designed Fuzzy PID supervisor provides on-line the PID parameters K
P
,
K
I
, and K
D
allowing to reach the desired position with good performances and we
consider the performances better because the PID parameters vary in order to control
the ball position very quickly and without overshoot.
Fuzzy Supervised PID-Controller
36
Fig. 6.13: Error rate (de) v/s time response of Fuzzy tuned PID controlled Maglev
System model.
Fig. 6.14: Output v/s time response of Fuzzy tuned PID controlled Maglev System
model.
Fuzzy Supervised PID-Controller
37
Fig. 6.15: Error (e) v/s time response of Fuzzy tuned PID controlled Maglev System
model.
Fuzzy Supervised PID-Controller
38
Chapter-7
Conclusion
In the control of a nonlinear process, classical control is robust but not optimal for the
complete range of operation conditions. Indeed, the design of one controller is not
sufficient to ensure good performances and stability for all the operation set points.
Also, the variation of physical parameters of a process over time affects the
performances. Therefore a continuous adjustment of controller gains is necessary to
improve and eventually to maintain performances. On the basis of the proposed
approach, performances are used a priori in the design of the fuzzy PID supervision
taking into account the variation of parameters and operation conditions. Indeed in
terms of both, the designed fuzzy PID supervision provides on-line the appropriate
gains to the PID controllers ensuring the same performances whatever the operation
conditions maybe. From the simulation results, it has been shown that the fuzzy
controller can stabilize the system efficiently. Also the performance during the
transient period of the fuzzy system is better in the sense that less overshoot was
obtained. Moreover, the fuzzy controller provides a zero steady state error. Based on
the simulation results it is concluded that the Fuzzy Logic supervision based controller
can stabilize the system efficiently and accurately compared to a classical PID
controller.
Further work in this direction could be the analysis of the Maglev system and design
of a controller using more advanced analysis techniques like GA, Ant Colony
algorithm, Neuro-Fuzzy algorithm and ultimately practical implementation of the
designed controller onto the magnetic levitation system.
39
References
[1] Passino, K.M. & Yurkovich, S. (1998). Fuzzy control, Addison-Wesley-
Longman, Menlo Park, CA.
[2] PASSINO, K.M., STEPHEN Y., Fuzzy Control, Book; Addison Wesley
Longman, Inc. 1997 ISBN 0-201-18074-X.
[3] MIZUMOTO, M., Realization of PID controls by fuzzy control methods.
ELSEVIER Fuzzy Sets and Systems 1995; 70:171-182.
[4] COPELAND, R.P., RATTAN K.S., A Fuzzy Logic Supervisor for PID Control
of Unknown Systems. In: Proceedings of IEEE international Symposium of
Intelligent Control, Ohio, USA. 1994; pp. 22-26.
[5] GARCIA-BENITEZ, E., YURKOVICH S, PASSINO K.M., Rule-based
supervisory control of a two-link flexible manipulator. Journal of Intelligent and
Robotic Systems 1993; 7(2):195213.
[6] JANTZEN, J., Fuzzy Supervisory Control, Technical University of Denmark,
Denmark. Tech. report no 98-H-875 (proc), Nov. 1998.
[7] JIAN, L., YONG K, JIAN C., Fuzzy-Tuning PID Control of an Inverter with
Rectifier-Type Nonlinear Loads. In: Proceedings of the Third International
Power Electronics and Motion Control Conference, Beijing, China. 2000;
v1:381-384.
[8] LOU, K.N., KUO CY, SHEU L.T., A Novel Method for Fuzzy Self-Tuning PID
Controllers. In: Proceedings of the Asian Fuzzy Systems Symposium 'Soft
Computing in Intelligent Systems and Information Processing. Kenting, Taiwan,
1996; pp. 194-199.
[9] SONG, S., LIU W., Fuzzy Parameters Self-Tuning PID Control of Switched
Reluctance Motor based on Simulink/NCD. In: Proceedings of International
Conference on Computational Intelligence for Modelling, Control and
Automation, 2006.
[10] DE QUEIROZ, M.S., DAWSON D.M., Nonlinear control of active magnetic
bearings: a back stepping approach, IEEE Trans Control Sys Tech, 1996.
[11] HUNG, J.Y., Magnetic bearing control using fuzzy logic. IEEE Trans Indus
Appl 1995; 31: 1492-1497.
[12] DHAR, D, BARRETT L.E., KNOSPE C. R., Optimum design of decentralized
magnetic bearings for rotor systems. In: Proceedings of the third international
symposium on magnetic bearings, Alexandria Virginia. 1992; pp. 47-59.
[13] HERZOG, R., A comparison between passively and actively controlled
magnetic bearings. In: Proceedings of the third international symposium on
magnetic bearings, Alexandria Virginia. 1992; pp. 223-231.
[14] CHARARA, A, CARON B., Magnetic bearing: comparison between linear and
nonlinear functioning. In: Proceedings of the third international symposium on
magnetic bearings, Alexandria Virginia. 1992; pp. 451-460.
40
[15] ABDEL-HADY, F., EL-MOGAHZY Y., ABUELENIN S, ABDEL-KADER,
R., Innovative approach to high-speed spinning using magnetically-elevated
spinning ring AUTEX Res J, 2006; 6:113-121.
[16] D. L. Trumper, M. Olson, and P. K. Subrahmanyan, Linearizing control of
magnetic suspension systems, IEEE Transactions on Control Systems
Technology 5 (1997), no. 4, 427438.
[17] Charara, J. DeMiras, and B.Caron, Nonlinear control of a magnetic levitation
system without premagnetization, IEEE Transactions on Control Systems
Technology 4 (1996), no. 5, 513523.
[18] M. R. Filho and C. J.Munaro, A designmethodology of tracking controllers for
magnetic levitation systems, Proc. 2001 IEEE International Conference on
Control Applications, (Mexico City), 2001, pp. 4751.
[19] S. A. Green and K. C. Craig, Robust, design, nonlinear control of magnetic-
levitation systems, Journal of Dynamics, Measurement, and Control 120 (1998),
no. 4, 488495.
[20] M. Fujita, F. Matsumura, and K. Uchida, Experiments on the H disturbance
attenuation control of a magnetic suspension system, Proc. 29th IEEE
Conference on Decision and Control, (Hawaii), vol. 5, December 1990, pp.
27732778.
[21] M. Lairi and G. Bloch, A neural network with minimal structure for maglev
system modelling and control, Proc. 1999 IEEE International Symposium on
Intelligent Control/Intelligent Systems and Semiotics, (Massachusetts), 1999,
pp. 4045.
[22] Bhler, H. (1994). Rglage par Logique Floue. Presses Polytechniques et
Universitaires Romandes.
[23] C. C. Lee, Fuzzy logic in control systems: fuzzy logic controller part I, IEEE
Trans. Syst., Man, Cybern., vol. 20, no. 2, pp. 404418, 1990.
[24] C. C. Lee, Fuzzy logic in control systems: fuzzy logic controller part II, IEEE
Trans. Syst., Man, Cybern., vol. 20, no. 2, pp. 419435, 1990.
[25] Driankov, D., Hellendoorn, H. & Reinfrank, M. (1993) An Introduction to
Fuzzy Control, Springer Verlag, Berlin Heidelberg.
[26] Vermeiren, L. (1998), Proposition de lois de commande pour la stabilisation de
Modles Flous, Thse de Doctorat, Universit de Valenciennes et du Hainaut
Cambrsis, France.
[27] I.J.Nagrath and M.Gopal. 1999 Control systems engineering.
[28] Zhang, N. Wang and S. Wang, A developed method of tuning PID controllers
with fuzzy rules for integrating process, Proceedings of the American Control
Conference, Boston, 2004, pp. 1109-1114.
[29] Wang Xiao-kan, Sun Zhong-liang, Wanglei, Feng Dong-qing, "Design and
Research Based on Fuzzy PID-Parameters Self-Tuning Controller with
MATLAB," Advanced Computer Theory and Engineering, International
Conference on, pp. 996-999, 2008 International Conference on Advanced
Computer Theory and Engineering, 2008.
References