Вы находитесь на странице: 1из 22

Transactions of the Institute of Measurement and Control 25,4 (2003) pp.

352372
An improved PID-type fuzzy
controller employing individual fuzzy
P, fuzzy I and fuzzy D controllers
S. Bhattacharya, A. Chatterjee and S. Munshi
Electrical Engineering Department, Jadavpur University, Calcutta, West Bengal,
India PIN 700 032
Since the emergence of fuzzy logic controllers (FLCs), control system engineers are in pursuit
of more and more sophisticated versions of these controllers to achieve better performance,
particularly in situations where providing a control action to even a minimal degree of satisfac-
tion is a problem. The present paper is an attempt to contribute in this eld. This work proposes
a fuzzy PID controller comprising fuzzy P, fuzzy I and fuzzy D controllers in parallel. While
fuzzy P and fuzzy I controllers are implemented in incremental form, the fuzzy D controller
is realized in position form. Simulation studies reveal that the new scheme has a signicantly
improved performance compared to the ZieglerNichols tuned PID controller and static fuzzy
PID controllers currently in use.
Key words: decoupled fuzzy rule bases; fuzzy PID; fuzzy P controller; fuzzy I controller; fuzzy
D controller.
1. Introduction
The eld of fuzzy control has been making rapid progress in recent years.
Motivated by the success of fuzzy controllers in controlling nonlinear, complex,
time-varying dynamic processes in the real world, there has been steep increase in
the research work on the theoretical aspects of fuzzy control and their application.
The main reason behind the increasing popularity of fuzzy logic controllers (FLCs)
Address for correspondence: A. Chatterjee, Electrical Engineering Department, Jadavpur University,
Calcutta, West Bengal, India. E-mail: amitavachatterjee@hotmail.com
2003 The Institute of Measurement and Control 10.1191/0142331203tm094oa
Bhattacharya et al. 353
is that FLCs essentially incorporate human expertise in control strategy for con-
trolling processes, exploiting the easier understanding of linguistic interpretation.
Amongst the different types of FLC structures, PI-type FLCs and PD-type FLCs
are very common (Driankov et al., 1993; Passino and Yurkovich, 1998). Various
fuzzy controllers, both in PI- and PD-type of implementations, have been pro-
posed over the years, which employ different adaptation policies to improve one
or more performance indices (Ying et al., 1990; Lee, 1993; Xu et al., 1996, 1998; Oh
and Park, 1998; Mudi and Pal, 1999). In fact, fuzzy versions of commonly
employed more complex conventional control strategies have also been developed
in recent times to improve performance and robustness. Different fuzzy model
reference adaptive controllers are proposed in Fischle and Schroder (1999), Koo
(2001) and Tseng and Chen (2001) as improvements over conventional model
reference adaptive controllers with guaranteed robustness and parameter conver-
gence. Similarly fuzzy model predictive controllers have been successfully
developed for multivariable control in an optimal manner (de Oliviera and Lemos,
2000; Abonyi et al., 2001; Mollov et al., 2002) and fuzzy sliding mode controllers
have been developed to take care of parameter uncertainties and reduce chattering
by providing smoother control action (Yoo and Ham, 1998; Huang and Huang,
2001; Hwang and Kuo, 2001). Hence, logically it should appear as if creating
improved PID-type fuzzy controllers over conventional PID controllers can be
achieved without signicant problems. However, development of PID-type FLCs
was not that popular earlier because they usually needed the construction of a
three-dimensional rule base, which makes the design complicated. Moreover, to
make any PID-type FLC adaptive in nature, the number of free adaptable para-
meters increases and their interaction and interdependence further worsen the
situation (Mann et al., 1999). However, very recent trends show an increasing
interest in development of different architectures and tuning algorithms for PID-
type FLCs with varied degree of success.
A three-input PID-type FLC was proposed by Maeda and Murakami (1992),
which considered error (e), change of error (De) and rate of change of error (D
2
e)
as inputs. This adaptive FLC implemented simultaneous adaptation of input and
output scaling factors (SFs) and control rules to improve control strategy online.
However, their controller failed to provide satisfactory performance for processes
with dominant time delay. Later Chen et al. (1993) implemented another three-
input PID-type FLC with e, De and Se dt as inputs. Different gain scheduling-type
PID controllers with fuzzy-based self-tuning capability have been proposed in
various literatures (He et al., 1993; Zhong et al., 1993). The proposed controller
scheme developed by He et al. (1993) essentially implements a conventional PID
structure, which starts its operation with values of proportional gain (K
p
), integral
time (T
i
) and derivative time (T
d
) obtained from the well known ZieglerNichols
(Z-N) tuning formulae. This scheme implements a supervisory fuzzy controller,
which adaptively changes K
p
, T
i
and T
d
after each sampling instant to improve
the control performance. Although this controller showed superior performance
for linear processes, they could only reduce peak overshoot at the expense of
increasing rise time, and the degradation becomes more and more signicant with
increasing time delay. This restricted their overall acceptance as a good
controller mechanism.
354 Improved PID-type fuzzy controller
To nullify the basic disadvantages associated with the construction of a three-
dimensional rule base, researchers in the past decade have proposed a few two-
input PID-type FLC structures, as combinations of two-term FLCs. Li and Gatland
(1996) attempted to build an adaptive PID-type FLC structure as a combination
of an adaptive PI-type FLC with an adaptive PD-type FLC. Their proposed scheme
showed better performance compared with conventional controllers in the case
of third-order processes, but failed to give satisfactory results for second-order
processes. Qiao and Mizumoto (1996) proposed another PID-type FLC with simi-
lar architecture but with different tuning philosophy. However, their proposed
scheme could only reduce the settling time of the system without any improve-
ment in peak overshoot and rise time, since their scheme effectively starts oper-
ation only after the rst overshoot of the system is known, which for most practical
systems happens to be the peak overshoot of the system. In both these schemes,
tuning of controller parameters is time consuming because they involve trial and
error methods in setting optimum controller parameters. However, in all these
FLC structures, it is almost impossible to designate any one parameter responsible
for controlling proportional or integral or derivative action solely. Whether in case
of two-input-type or three-input-type existing PID FLCs, change in any one of the
input/output SFs or entries in a fuzzy rule base will involve change in more than
one control action. To solve this particular problem, i.e., to provide the avour of
independent control actions like conventional proportional, integral or derivative
actions, a few hybrid control schemes for PID actions have been proposed in the
past decade. Li (1998) developed an incremental fuzzy proportional controller,
which operated in unison with a conventional integral-derivative controller. This
scheme showed satisfactory performance for both regulation and tracking control
problems. Misir et al. (1996) proposed another fuzzy PID-type controller where a
fuzzy PI controller is implemented in conjunction with fuzzy D controller. How-
ever, none of these structures can individually tune each of proportional, integral
and derivative actions. A systematic classication of possible coupled and deco-
upled PID-type fuzzy structures were discussed in detail by Mann et al. (1999).
However, the theoretical analysis of proposed structures, presented in Mann et al.
(1999), remains to be implemented in practice, as they were not supported by any
practical or simulation study.
The present paper describes the development of a simple, model-free improved
fuzzy PID-type structure as an effective combination of three independent fuzzy
controllers. This controller structure embeds three direct-action-type FLCs in par-
allel form to provide suitable control input. Here an incremental fuzzy P control-
ler, an incremental fuzzy I controller and a fuzzy D controller are connected in
parallel to give the resultant controller structure. With the help of three inde-
pendent fuzzy controllers, independent control actions may be generated which
should necessarily eliminate the problems associated with most practical two-term
or three-term FLCs. Simulation studies have been carried out on second-order
linear processes with and without dead time, and on a marginally stable system
with dead time. The proposed fuzzy control scheme shows a superior performance
for all processes under consideration when compared to a conventional Z-N tuned
PID controller and static PID-type FLC (implemented as a combination of a static
PI-type FLC and a static PD-type FLC).
Bhattacharya et al. 355
2. Proposed controller structure
The proposed fuzzy PID controller is comprized of fuzzy P, fuzzy I and fuzzy D
controllers in parallel form. Figure l gives a block diagram representation of the
proposed controller structure. While the fuzzy P controller (denoted PFLC) and
fuzzy I controller (denoted IFLC) are implemented in incremental form, the fuzzy
D controller (denoted DFLC) is implemented in position form. The incremental
controller output at the nth instant from the fuzzy P controller, i.e., the PFLC in
Figure 1, is given as
Du
P
(n) = K
OP
* f
1
(e
N
(n), De
N
(n))
= K
OP
* f
1
((K
e
* e(n)), (K
De
* De(n))) (1)
where K
e
and K
De
are the input SFs of the FLC inputs for error {e(n)} and change
in error {De(n)} in process output, respectively. f
1
represents the nonlinear fuzzy
function mapping performed by the PFLC on e
N
(n) [i.e., K
e
* e(n)] and De
N
(n) [i.e.,
K
De
* De(n)] to produce Du
P
(n). K
OP
is the output SF of the PFLC, which performs
a function very similar to that of the proportional gain. Hence, the resultant control
action of the PFLC at the nth sampling instant is given by
u
P
(n) = u
P
(n - 1) + Du
P
(n) (2)
The incremental controller output at the nth sampling instant of the fuzzy I
controller, i.e., the IFLC in Figure 1 is given by
Du
I
(n) = K
OI
* f
2
(e
N
(n), De
N
(n))
Du
I
(n) = K
OI
* f
2
((K
e
* e(n)), (K
De
* De(n))) (3)
Here f
2
represents the nonlinear fuzzy function mapping performed by IFLC and
K
OI
is the output SF of the IFLC, which is very similar to the integral gain. The
resultant control force produced by the IFLC is
u
I
(n) = u
I
(n - 1) + Du
I
(n) (4)
Figure 1 Proposed (fuzzy P + fuzzy I + fuzzy D) controller
architecture in schematic representation
356 Improved PID-type fuzzy controller
The control action produced by the fuzzy D controller, i.e., DFLC in Figure 1, at
the nth instant is
u
D
(n) = K
OD
* f
3
((K
e
* e(n)), (K
De
* De(n))) (5)
Here f
3
represents the nonlinear fuzzy function mapping performed by DFLC.
K
OD
is the output SF of DFLC, which is very similar to derivative gain. Hence
the resultant control action produced by the proposed PID-type FLC structure is
given by
u
PID
(n) = u
P
(n) + u
I
(n) + u
D
(n)
= u
P
(n - 1) + Du
P
(n) + u
I
(n - 1) + Du
I
(n) + u
D
(n) (6)
The resultant fuzzy PID structure for the implementation of the control actions,
employing decoupled rule bases, as given by (6) and shown in Figure 1, will
henceforth be denoted the DPIDFLC. Since both the IFLC and DFLC employ
essentially the same fuzzy control architecture and the difference only lies in their
SFs and implementations in either velocity or position form, an easier way of
implementing the proposed PID FLC is shown in Figure 2. This form of represen-
tation can be highly appreciated for stand-alone applications, where the system
can be implemented with minimum storage requirements and simplied compu-
tational load. Figure 3 shows the common membership functions (MFs) chosen
for all inputs and outputs of the PFLC, IFLC and DFLC, i.e., e
N
, De
N
, Du
NP
, Du
NI
and u
ND
. Each of these input and output variables are fuzzied using three fuzzy
sets. The shape of the fuzzy MFs are so chosen that the middle MF is a triangle
and at the two extreme ends the two MFs are chosen to be trapezoidal. The linguis-
tic terms associated with the three MFs are negative (N), positive (P) and zero
(Z). Usage of common MFs for all input and output variables should be very
useful for real-world applications, as this will ensure minimum storage space for
the MFs. The universe of discourse for fuzzication of each variable is chosen
as[-1,1]. The choice of as few as three MFs for each fuzzied variable is deliber-
ately done to achieve a structure with reduced complexity. The reduced number
Figure 2 Proposed (fuzzy P + fuzzy I + fuzzy D) controller archi-
tecture in simplied form for practical realization
Bhattacharya et al. 357
Figure 3 Input and output MF (for each of e
N
, De
N
, Du
NP
, (Du
NI
and u
ND
) variations for each of the PFLC, IFLC and DFLC in
Figure 1
of MFs for each input and output variable (compared to the more common choice
of ve or seven MFs for each fuzzied variable), along with the associated reduced
structures of fuzzy rule bases, will ensure less computational burden for the
resultant controller.
3. Construction of fuzzy rule bases
As we have suggested separate FLCs for each of proportional, integral and deriva-
tive actions, three separate fuzzy rule bases are developed. The rule base for gener-
ating the incremental P action is given in Table 1 (Li, 1998). A close investigation
of the rule base for P action will indicate that the PFLC will generate a high control
action for a large magnitude of error and the control input will gradually decrease
with a gradual reduction in error magnitude. This is in close agreement with the
proportional action in a conventional PID controller. In a conventional PID con-
troller, proportional action is high when process output is far away from set-point
and this action gradually diminishes as the error magnitude reduces. In fact, the
magnitude of the incremental output from the PFLC is lower for situations where
Table 1 Rule base for incremental P action (PFLC)
De
N
e
N
N Z P
N N N Z
Z N Z P
P Z P P
358 Improved PID-type fuzzy controller
the e
N
and De
N
are of opposite sign (which implies that process trend is convergent
and the present controller action should not be greatly disturbed) than for
situations where the e
N
and De
N
are of same sign (which implies that process trend
is divergent and the controller action requires immediate correction to improve the
process trend).
It is common knowledge nowadays that many FLCs employ the same fuzzy
rule base to generate both PI-type and PD-type actions (Li and Gatland, 1996; and
Mudi and Pal, 1999). The same controller generating output in incremental or
velocity form can be called a PI-type FLC and when it generates controller output
in position form it is called a PD-type FLC. This also helps in stand-alone
applications, where a common storage space can be used for two types of FLCs
employing identical MFs and fuzzy rule base. The very idea of selecting the same
fuzzy rule base for a PI-type FLC and a PD-type FLC in Mudi and Pal (1999),
and Li and Gatland (1996) inspired us to implement common rule bases for both
the IFLC and DFLC, where the controllers are implemented in velocity and pos-
ition form, respectively, to generate separate integral and derivative action. The
proposed common rule base for producing I action in velocity form and D action
in position form is given in Table 2. The common rule base for IFLC and DFLC
is formed after an in-depth analysis of different situations of the transient and
steady state responses of different processes.
Since each of the PFLC, IFLC and DFLC is implemented independently, each
of them employs a fuzzy rule base where rule activation in any one FLC is not
dependent on the working of the other two FLCs. Hence one can conclude that
each of the fuzzy P, fuzzy I and fuzzy D actions can be controlled separately by
the use of three decoupled fuzzy systems. A proper choice of K
OP
, K
OI
and K
OD
will determine the suitable relative amount of proportional, integral and deriva-
tive actions to be applied to control a process. These parameters for a given process
can be obtained after a few trial runs with different sets of {K
OP
, K
OI
, K
OD
}.
4. Simulation studies
The performance of the proposed DPIDFLC is compared with two other control-
lers, a conventional PID controller whose parameters K
P
, T
i
, T
d
are tuned by the
well known Z-N method (denoted the ZNPIDC) and a static fuzzy PID controller
whose structure is obtained from Li and Gatland (1996) (denoted the PIDSFLC).
Table 2 Rule base for incremental I/direct D action (IFLC/DFLC)
De
N
e
N
N Z P
N N Z Z
Z Z Z P
P Z P P
Bhattacharya et al. 359
The PIDSFLC is an effective combination of a PI-type FLC in velocity form and
a PD-type FLC in position form, where each PI-type FLC and PD-type FLC
employs the same rule base as in Li and Gatland (1996). We have considered two
linear second-order processes and one marginally stable process, which were also
considered in Mudi and Pal (1999), and Li and Gatland (1996). To compare the
control actions produced by different types of controllers, several performance
measures such as peak overshoot/undershoot (%OS/US), rise time (t
r
), settling
time (t
s
), integral absolute error (IAE) and integral time multiplied absolute error
(ITAE) have been considered. Each process is tested with a unit step input at
t = 0 s. One possible disadvantage associated with the proposed controller can be
the determination of input and output scale factors needed for each of the three
independent FLCs to produce the improved performance, which may give rise to
signicant associated computational burden. This is because one has to individu-
ally tune the SFs of each FLC. To overcome these difculties, we have developed
PFLC, IFLC and DFLC, where each of them employs the same input SFs (K
e
and
K
De
). Hence the problem of determining six input SFs for three FLCs ultimately
gets reduced to the problem of determining two input SFs. We also propose
empirical rules to determine K
OI
and K
OD
in terms of K
OP
in order to further reduce
the computation load. These empirical rules are proposed after experimentation
on a large class of processes. Once an optimum value of K
OP
is determined, we
can employ K
OI
, which is one-tenth of K
OP
and K
OD
, 30 times K
OP
. Hence, choice
of output SFs is limited to nding an optimum K
OP
, which automatically sets the
values of K
OI
and K
OD
. This indicates that the problem of choosing SFs ultimately
reduces to the choice of two input and one output SFs, a nominal requirement
even for a minimum system employing a single static FLC. We have applied our
proposed empirical rules to a number of processes, which yielded reasonably
satisfactory performance measures. However, if a user is further interested in
improving performance at the cost of increased computational load, he/she can
achieve this by manually tuning any or some or all of the SFs. The proposed
DPIDFLC employs Mamdani-type inferencing and centroid method of defuzz-
ication for each individual FLC. For the numerical integration, a fourth-order
RungeKutta method is used with a sampling interval of 0.01 s for the process
taken from Li and Gatland (1996) and 0.1 s for processes taken from Mudi and
Pal, (1999). The settling band is calculated on the basis of 2% tolerance.
4.1 Second-order linear process with dead time
The chosen transfer function for the second-order linear process considered from
Mudi and Pal (1999) is
G
P
(s) = e
-Ls
/(s
2
+ s + 0.2) (7)
Four different values of dead time (L= 0, 0.1, 0.2 and 0.3 s) are considered, as was
done in Mudi and Pal (1999). The performance of the DPIDFLC for this process
is compared with the performances obtained with Z-N tuned conventional PID
controllers (denoted ZNPIDC) and the static PID-type FLC created as a combi-
nation of a static PI-type FLC and a static PD-type FLC (denoted PIDSFLC). Out-
360 Improved PID-type fuzzy controller
put SFs of the PIDSFLC, i.e., K
UPI
and K
UPD
are chosen 0.01 and 0.3, respectively,
which are obtained after a few trial runs. For the DPIDFLC, K
OP
is chosen as 0.l,
which automatically sets K
OI
= 0.01 and K
OD
= 3 according to the proposed empiri-
cal rule. Input SFs K
e
and K
De
are taken as 0.8 and 10, respectively, for both the
DPIDFLC and PIDSFLC to make a proper comparison of the performances of
different controllers for the process in (7). The performance of the process control
loop with the process in (7) using different types of controllers is given in Table 3.
Table 3 also gives a comparative performance of these controllers for the process
in (7) with a long dead time, L= 1 s. For this sample case, K
OP
is chosen as 0.0257.
Hence K
OI
and K
OD
are automatically set as 0.00257 and 0.771, respectively. In this
case, input SFs K
e
and K
De
are chosen as 0.66 and 12.9, respectively. Figure 4 shows
the variations of process output with time for (7) having different types of control-
lers for ve different values of L. In all the cases of L, it may be noticed that
ZNPIDC fails to produce an acceptable performance because of excessively large
overshoot. The performance improvements employing PIDSFLC are signicant in
improving %OS but the other performance indices are not encouraging because
of highly overdamped response and comparatively high values of IAE and ITAE.
In each case, DPIDFLC gives the best overall performance with signicantly
reduced IAE and ITAE. The controller can be further tuned to improve upon any
of the performance indices, as desired by the user, by varying the relevant free
parameter. For example, reduction in %OS (may be at the cost of increased t
r
)
may be achieved by reducing K
OI
and/or increasing K
OD
. Table 4 demonstrates
Table 3 Performance comparison of the DPIDFLC vis-a`-vis other controllers for the
process in (7)
L (s) FLC %OS t
r
(s) t
s
(s) IAE ITAE
0 ZNPIDC 57.04 5.5 23.6 6.33 64.78
PIDSFLC 1.08 6.0 9.9 3.77 10.62
DPIDFLC 7.46 2.6 9.9 1.75 4.31
0.1 ZNPIDC 75.66 0.8 8.2 1.82 3.55
PIDSFLC 1.05 6.1 10.1 3.79 10.48
DPIDFLC 10.46 2.6 12.8 1.81 5.09
0.2 ZNPIDC 72.06 1.2 9.5 2.38 5.78
PIDSFLC 1.03 6.2 10.2 3.80 10.36
DPIDFLC 15.33 2.4 10.8 1.82 5.25
0.3 ZNPIDC 67.2 1.5 10.4 2.58 6.44
PIDSFLC 1.01 6.3 10.2 3.82 10.25
DPIDFLC 14.94 2.20 11.2 1.74 5.24
1 ZNPIDC 48.47 3.3 10.5 3.92 11.46
PIDSFLC 0.17 5.8 12.4 3.94 12.96
DPIDFLC 14.30 4.5 15.3 3.51 11.5
Bhattacharya et al. 361
Figure 4 Unit step responses of second-order process in (7)
employing ZNPIDC (- - -), PIDSFLC ( ) and DPIDFLC ()
with (a) L= 0 s, (b) L= 0.1 s, (c) L= 0.2 s, (d) L= 0.3 s and (e)
L= 1 s.
362 Improved PID-type fuzzy controller
Table 4 Performance analysis of the DPIDFLC, when activated to produce different
control actions, for the process in (7)
L (s) FLC %OS t
r
(s) t
s
(s) IAE ITAE
0 PFLC 27.3 3.2 25.7 3.97 24.34
PFLC+IFLC 29.90 3.0 24.9 3.94 24.34
PFLC+DFLC 6.80 2.6 12.2 1.80 4.89
DPIDFLC 7.46 2.6 9.9 1.75 4.31
1 PFLC 49.6 6.0 9.64 94.77
PFLC+IFLC 55.26 5.7 9.90 98.38
PFLC+DFLC 11.42 4.7 18.1 3.65 13.82
DPIDFLC 14.30 4.5 15.3 3.51 11.5
the individual responses obtained from the resultant DPIDFLC structure when
it is selectively activated to produce different control actions like conventional
P/PI/PD/PID control actions. These results are reported with L= 0 and 1 s. The
values of %OS, t
r
and t
s
are signicantly changed when we add/delete integral
and/or derivative action to the DPIDFLC, when used only as a P controller. These
performance changes clearly demonstrate the effectiveness of PFLC, IFLC and
DFLC individually in producing the desired proportional, integral and derivative
actions. Introduction of integral action through the IFLC along with PFLC achieves
reduction in both rise time and settling time at the cost of increased %OS, when
compared to the performance of the PFLC alone, as expected. Application of
derivative action through the DFLC introduces increased damping and reduces
the peak overshoot signicantly. Introductions of all three actions simultaneously,
in an optimum manner, produce signicant improvements in all performance
measures. Figure 5 gives these variations in process output with time when the
DPIDFLC is activated to produce different controller actions like conventional
P/PI/PD/PID control actions.
To give a pictorial representation of variation of gain of the DPIDFLC with
time, we have chosen the sample case with dead time L= 0 s for the process in
(7). For this process the ZNPIDC is employed with constant proportional gain,
(K
P
= 25.11), integral gain, (K
i
= K
p
/T
i
= 55.8) and derivate gain,
(K
d
= K
p
T
d
= 2.825). However, each component FLC within the DPIDFLC gives a
nonlinear variation of gain as shown in Figure 6. The nonlinear gain of the PFLC,
i.e., (Du
NP
/e
N
) is plotted against time. For both the IDFLC and DFLC the gain
variation plots are identical. This is because they employ identical input and out-
put MFs and a fuzzy rule base, as shown in Figure 2. The difference lies in the
fact that the IFLC is employed in velocity form and the DFLC is employed in
position form. Hence, the gain plots of IFLC and DFLC are shown by a single
curve in Figure 6, where the nonlinear gain is [(Du
NI
/e
N
) = (u
ND
/e
N
)].
Bhattacharya et al. 363
Figure 5 Unit step responses of second-order process in (7) with
(a) L= 0 s and (b) L= 1 s employing only PFLC ( ), PFLC
and DFLC (- - -), PFLC and IFLC ( ) and DPIDFLC ()
Figure 6 Nonlinear gain variations with time for PFLC ()
and IFLC/DFLC ( ) for the process in (7) with L= 0 s
364 Improved PID-type fuzzy controller
4.2 Marginally stable system with dead time
The chosen transfer function for the marginally stable system, also considered in
Mudi and Pal (1999) is
G
P
(s) = e
-Ls
/(s(s+1)) (8)
Control of a marginally stable system is considered a difcult problem in the
process industry. The presence of dead time usually further worsens the situation.
Three different values of dead time (L= 0.1, 0.2 and 0.3 s) are considered. The
output SFs of the PIDSFLC i.e., K
UPI
and K
UPD
, are chosen as 0.01 and 0.3, respect-
ively. For the DPIDFLC, K
OP
= 0.1 is chosen, which sets K
OI
= 0.01 and K
OD
= 3.
Input SFs K
e
and K
De
are taken as 0.8 and 10, respectively, in both the DPIDFLC
and PIDSFLC. The performances of different types of controllers for the above
process are shown in Table 5. Figure 7 shows the variations of process output with
time for different types of controllers and for various L. As in the previous case
study, it may also be noticed here that excessive overshoot associated with the
ZNPIDC restricts its successful implementation. The performance of the PIDSFLC
is also not satisfactory for all cases because it often produces a high overshoot.
The proposed DPIDFLC, for all cases of L, produces signicant improvement in
all performance measures. The values of IAE and ITAE are signicantly less com-
pared to other types of controllers. Table 6 lists the performance indices obtained
using the DPIDFLC when it is activated to produce different controller actions
like conventional P/PI/PD/PID control, for the process without dead time.
Figure 8 shows plots of these responses. The ordered nature of variation of the
system response, as different components of the proposed PID-type FLC are
employed either individually or in different combinations, reiterates the fact that
different controllers can be separately employed and tuned for individual P, I and
D actions as per users wish. Hence, the exibility and usefulness of the proposed
controller can be easily appreciated. Here, also, we have chosen the delay-free
Table 5 Performance comparison of the DPIDFLC vis-a`-vis other controllers for the
process in (8)
L (s) FLC %OS t
r
(s) t
s
(s) IAE ITAE
0.1 ZNPIDC 79.5 0.8 7.3 1.73 3.05
PIDSFLC 39.60 5.6 28.5 6.69 57.01
DPIDFLC 15.04 2.5 11.4 1.85 5.15
0.2 ZNPIDC 78.06 1.2 10.6 2.52 6.27
PIDSFLC 41.34 5.6 28.5 6.90 59.30
DPIDFLC 19.60 2.3 11.9 1.87 5.38
0.3 ZNPIDC 75.02 1.4 11.2 2.94 8.25
PIDSFLC 43.16 5.6 30.1 7.12 62.40
DPIDFLC 26.70 2.0 15.3 2.02 6.47
Bhattacharya et al. 365
Figure 7 Unit step responses of the second-order process in (8)
employing ZNPIDC (. . . .), PIDSFLC () and DPIDFLC ()
with (a) L= 0.1 s, (b) L= 0.2 s and (c) L= 0.3 s
366 Improved PID-type fuzzy controller
Table 6 Performance analysis of the DPIDFLC, when activated to produce different
control actions, for the process in (8)
L (s) FLC %OS t
r
(s) t
s
(s) IAE ITAE
0 PFLC 48.74 3.0 53.6 7.83 105.98
PFLC+IFLC 52.35 2.9 51.4 7.53 96.78
PFLC+DFLC 11.05 2.5 10.9 1.77 4.656
DPIDFLC 11.61 2.4 10.9 1.73 4.250
Figure 8 Unit step responses of the second-order process in (8)
with L= 0 s employing only PFLC ( ), PFLC and DFLC (- - -),
PFLC and IFLC ( ) and DPIDFLC ()
case, to give graphical representation of the nonlinear variations of gain for the
PFLC. IFLC and DFLC. Figure 9 shows these gain variations. It is worth men-
tioning that, in this case, ZNPIDC was employed with constant K
P
= 25.1, K
i
= 55.8
and K
d
= 2.825.
4.3 Second-order linear process without dead time
A second-order linear process, which was considered in Li and Gatland (1996) is
given as
G
P
(s) = 2.2/(s
2
+ 0.5s + 1) (9)
In Li and Gatland (1996), the proposed PID-type fuzzy controller (denoted as
Bhattacharya et al. 367
Figure 9 Temporal variation of nonlinear gain for PFLC ()
and IFLC/DFLC ( ) for the process in (8) with L= 0 s
FZAUKPID) failed to provide a better result than the chosen conventional PID
structure (denoted as CONAUKPID) for this process. Now, performance of
DPIDFLC for this process is compared with the performances obtained by
implementing CONAUKPID and FZAUKPID in the process control loop. For both
CONAUKPID and FZAUKPID, identical values for free controller parameters are
chosen, as was used in Li and Gatland (1996). For successful implementation of
the DPIDFLC, we have chosen K
e
= 0.5 and K
De
= 25. The value of K
OP
is chosen
as 1 after a few trial runs. The other output scale factors are automatically determ-
ined using our proposed empirical formulae, i.e., K
OI
= 0.1 and K
OD
= 30. Perform-
ance measures obtained with different types of controllers for this process are
shown in Table 7. Figure 10 shows corresponding process responses with time.
Table 7 shows signicant improvements in performance obtained with the
DPIDFLC compared to other controllers. While the proposed PID-type FLC in Li
and Gatland (1996) could not improve upon the performances for the process in
(9) compared to the conventional PID controller, our proposed scheme shows
improvements in t
s
, IAE and ITAE with a marginal increase in rise time. In fact,
this marginal increase in rise time is accompanied by a negligible peak percentage
Table 7 Performance comparison of the DPIDFLC vis-a`-vis other controllers in Li and
Gatland (1996) for the process in (9)
Controller %OS t
r
(s) t
s
(s) IAE ITAE
CONAUKPID 36.07 0.99 5.62 1.15 1.47
FZAUKPID 3.00 5.11 1.25 1.99
DPIDFLC 1.05 3.86 0.582 0.645
368 Improved PID-type fuzzy controller
Figure 10 Unit step responses of the second-order process in
(9) employing CONAUKPID (- - -), FZAUKPID ( ) and
DPIDFLC ()
overshoot compared to CONAUKPID, which shows more than 36% OS. The sig-
nicantly less values of IAE and ITAE in the DPIDFLC indicate the improvements
in both transient and steady state performance of the process. Table 8 shows the
performance measures when the DPIDFLC is implemented to produce control
actions like P/PI/PD/PID controllers employing one or combinations of PFLC,
IFLC and DFLC separately. When the DPIDFLC is activated to produce only P
or only PI action, the DPIDFLC fails to give reasonable performance because of
excessive overshoot. When the DPIDFLC is activated to produce PD actions by
activating both the PFLC and DFLC and deactivating the IFLC, it provides an
improved damping effect, which necessarily makes the resultant response an over-
damped one with high settling time. Addition of integral action by activating the
IFLC along with PFLC and DFLC, i.e, activating the DPIDFLC in full strength,
we could reduce the settling time by almost 14% with improvements in both IAE
and ITAE without any associated degradation in t
r
. The system can still maintain
negligible peak overshoot. Figure 11 shows graphical representations correspond-
ing to Table 8.
To demonstrate the effectiveness of the proposed DPIDFLC, we have also
simulated its performance for input variations, which are different from unit step.
We have subjected each process in (7) and (8) to a combination of input variations.
For each process, two values of dead time, L= 0 and L= 0.3 s, are considered. The
input initially varies in a parabolic manner from zero and reaches 1 p.u. at t = 5 s.
A parabolic variation of input is chosen to subject the system to small set- point changes
initially and large set-point changes later as t = 5 s is approached. From t = 5 s to 50 s,
the input stays at 1 p.u. At t = 50 s, the input takes a large negative step
change of 0.8 p.u. and remains at 0.2 p.u. until t = 100 s. Figure 12 and Table
9 demonstrate the performance comparisons of the various controllers
Bhattacharya et al. 369
Figure 11 Unit step responses of the second-order process in (9)
employing PFLC and DFLC (- - -) and DPIDFLC ().
Table 8 Performance analysis of the DPIDFLC, when activated to produce different con-
trol actions, for the process in (9)
FLC %OS t
r
(s) t
s
(s) IAE ITAE
PFLC+DFLC 1.05 4.43 0.597 0.67
DPIDFLC 1.05 3.86 0.582 0.645
Table 9 Dynamic performance comparison of the DPIDFLC vis-a`-vis other controllers for
varying input combination
Process L (s) Controller IAE ITAE
Process in (7) 0 ZNPIDC 11.525 450.864
PIDSFLC 5.146 175.917
DPIDFLC 3.255 140.260
Process in (7) 0.3 ZNPIDC 44.230 301.130
PIDSFLC 5.139 174.856
DPIDFLC 4.370 175.930
Process in (8) 0 ZNPIDC 43.526 301.260
PIDSFLC 4.445 152.841
DPIDFLC 3.479 142.780
Process in (8) 0.3 ZNPIDC 44.487 301.940
PIDSFLC 4.688 158.265
DPIDFLC 3.491 132.421
370 Improved PID-type fuzzy controller
Figure 12 Dynamic performance comparison for the varying
input combination with ZNPIDC (- - -), PIDSFLC ( ) and
DPIDFLC () with (a) process in (7), L= 0, (b) process in (7),
L= 0.3 s, (c) process in (8), L= 0 s and (d) process in (8), L= 0.3 s
implemented. Results show that the DPIDFLC exhibits the best dynamic
performance among the competing controllers.
5. Conclusion
From the preceding discussions, followed by the simulation studies, the salient
features of the proposed control scheme can be summarized as follows:
1) The improved control action obtained by replacing Z-N-tuned PID controllers
with the static fuzzy PID controller (PIDSFLC), proposed by earlier
investigators, can be further augmented by using the proposed fuzzy PID con-
troller (DPIDFLC).
2) For stable second-order systems with dead time, where the PIDSFLC results
in an overdamped closed-loop step response, the proposed controller yields a
fast underdamped response, but with a not-too-large overshoot, and as a
Bhattacharya et al. 371
consequence there is a remarkable enhancement in performance with regard
to all other performance indices.
3) In particular, for marginally stable processes with dead time, for which the Z-
N-tuned three-term PID controllers fail to provide good control action, and the
PIDSFLC also does not work satisfactorily, the vastly improved performance
of the new control scheme is praiseworthy.
4) No degradation in the performance of the DPIDFLC could be observed with
increase in the process dead time. This fact is amply demonstrated by the case
study conducted for the delayed process in (7) with long dead time of L= 1 s.
5) For each of the three sample processes considered, we have shown the effec-
tiveness of our proposed DPIDFLC with each input and output variable of
PFLC, IFLC and DFLC fuzzied using three MFs. However, DPIDFLC is
model-independent and it is expected to be implemented for a wide class of
plant models. Hence, there may be situations where, for a given plant model,
3 3 rule base for each component FLC may not provide satisfactory results.
In those situations, a user may have to choose a more common 5 5 or 7 7
rule base to provide smoother control actions.
6) One advantage of employing our proposed DPIDFLC can be that one can
switch on/off any of the three principal actions, i.e., proportional, integral and
derivative, depending on the process. The basic decoupled nature of the PFLC,
IFLC and DFLC makes this possible. This is in sharp contrast with many of
the present day PI/PD/PID FLCs available. A probable area of further enhanc-
ing the performance of the system in future can be by making each of these
P, I and D controllers adaptive in nature. Future research work will face the
challenging problem of making an improved system with self-adaptation of
each/some component unit and at the same time the structure should be kept
relatively simple to implement as far as practicable.
References
Abonyi, J., Nagy, L. and Szeifert, F. 2001:
Fuzzy model based predictive control by
instantaneous linearization. Fuzzy Sets and
Systems 120, 10922.
Chen, C.L., Chen, P.C. and Chen, C.K. 1993:
Analysis and design of fuzzy control
systems. Fuzzy Sets and Systems 57, 12540.
Driankov, D., Hellendorn, H. and Reinfrank,
M. 1993: An introduction to fuzzy control. New
York: Springer-Verlag.
Fischle, K. and Schroder, D. 1999: An
improved stable adaptive fuzzy control
method. IEEE Transactions on Fuzzy Systems
7, 2740.
He, S.Z., Tan, S. and Xu, F.L. 1993: Fuzzy self-
tuning of PID controllers. Fuzzy Sets and
Systems 56, 3746.
Huang, S.J. and Huang, K.S. 2001: An adaptive
fuzzy sliding-mode controller for servo-
mechanism disturbance rejection. IEEE
Transactions on Industrial Electronics 48, 845
52.
Hwang, C.-L. and Kuo, C.-Y. 2001: A stable
adaptive fuzzy sliding-mode control for
afne nonlinear systems with application to
four-bar linkage systems. IEEE Transactions
on Fuzzy Systems 9, 23852.
Koo, T.J. 2001: Stable model reference adaptive
fuzzy control of a class of nonlinear systems.
IEEE Transactions on Fuzzy Systems 9, 62436.
Lee, J. 1993: On methods for improving
performance of PI-type fuzzy logic con-
trollers. IEEE Transaction on Fuzzy Systems 1,
298301.
Li, H.X. and Gatland, H.B. 1996: Conventional
fuzzy control and its enhancement. IEEE
372 Improved PID-type fuzzy controller
Transactions on Systems, Man and Cybernetics
Part B: Cybernetics 26, 79197.
Li, W. 1998: Design of a hybrid fuzzy logic
proportional plus conventional integral-
derivative controller. IEEE Transactions on
Fuzzy Systems 6, 44963.
Maeda, M. and Murakami, S. 1992: A self-
tuning fuzzy controller. Fuzzy Sets and Sys-
tems 51, 2940.
Mann, G.K.I., Hu, B.G. and Gasine, R.G. 1999:
Analysis of direct action fuzzy PID controller
structures. IEEE Transactions on Systems Man
and Cybernetics Part B: Cybernetics 29,
37188.
Misir, D., Malki, H.A. and Chen, G. 1996:
Design and analysis of a fuzzy proportional-
integral-derivative controller. Fuzzy Sets and
Systems 79, 297314.
Mollov, S., van den Boom, T., Cuesta, F.,
Oilero, A. and Babuska, R. 2002: Robust
stability constraints for fuzzy model
predictive control. IEEE Transactions on Fuzzy
Systems 10, 5064.
Mudi, R.K. and Pal, N.R. 1999: A robust self-
tuning scheme for P1- and PD-type fuzzy
contro1lers. IEEE Transaction on Fuzzy System
7, 26.
Oh, S.-Y. and Park, D.-J. 1998: Design of new
adaptive fuzzy logic controller for nonlinear
plants with unknown or time-varying dead
zones. IEEE Transactions on Fuzzy Systems 6,
48291.
de Oliveira, J.V. and Lemos, J.M. 2000: A com-
parison of some adaptive-predictive fuzzy
control strategies. IEEE Transactions on
Systems, Man and Cybernetics Part C: Appli-
cations and Reviews 30, 13845.
Passino, K.M. and Yurkovich, S. 1998: Fuzzy
control. Menlo Park, California: Addison
Wesley.
Qiao, W.Z. and Mizumoto, M. 1996: PID type
fuzzy controller and parameters adaptive
methods. Fuzzy Sets and Systems 78, 2335.
Tseng, C.S. and Chen, B.S. 2001: H
`
decentral-
ized fuzzy model reference tracking control
design for nonlinear interconnected systems.
IEEE Transactions on Fuzzy Systems 9,
795809.
Xu, J-X., Lin, C. and Hang, C.C. 1996: Tuning
of fuzzy PI controller based on gain/phase
margin specication and ITAE index. ISA
Transactions 35, 919.
Xu, J-X., Lin, C. and Hang, C.C. 1998: Tuning
and analysis of a fuzzy PI controller based
on gain and phase margins. IEEE Trans-
actions on Systems, Man and Cybenetics Part
A: Systems and Humans 28, 68594.
Ying, H., Siler, W. and Buckley, J.J. 1990:
Fuzzy control theory: a nonlinear case. Auto-
matica 26, 51320.
Yoo, B. and Ham, W. 1998: Adaptive fuzzy
sliding model control of nonlinear system.
IEEE Transactions on Fuzzy Systems 6, 31521.
Zhong, H.S., Shaohua, T. and Zhuang, W.P.
1993: Fuzzy self-tuning of PID controllers.
Fuzzy Sets and Systems 56, 3646.

Вам также может понравиться