Professional Documents
Culture Documents
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/314668494
CITATIONS READS
0 14
3 authors:
Ritu Raj
Indian Institute of Technology Kharagpur
6 PUBLICATIONS 2 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Ritu Raj on 25 October 2017.
Abstract—This paper reveals a new mathematical model of observer has been proposed [10] for tuning the scaling factors
the simplest fuzzy PID controller which employs two fuzzy sets of the fuzzy logic controller online.
(negative and positive) on each of the three input variables (error, It has been shown that algebraic product triangular norm
change in error and double change in error) and four fuzzy sets
(−2, −1, +1, +2) on the output variable (incremental control). L - bounded sum triangular co - norm - algebraic product
- type, Γ - type and Π - type membership functions are considered inference method - Centre of Sums (CoS) defuzzification
in fuzzification process of input and output variables. Controller method combination leads to a linear fuzzy PID controller
modeling is done via algebraic product AND operator-maximum [11]. It has been shown that the analytical structures of fuzzy
OR operator-Height (Ht) defuzzification process combination. PID controllers derived via minimum triangular norm are
The new model obtained in this manner turns out to be nonlinear,
and its properties are studied. not suitable for control [12]. Mohan and Arpita [13] have
introduced an analytical structure for fuzzy PID controller by
I. I NTRODUCTION employing algebraic product triangular norm, bounded sum
Conventional (linear) PID controllers have been in extensive triangular co - norm, Mamdani minimum inference method
use in industrial automation and process control. The reason and CoS defuzzification method. They also have derived
behind this is their ease of design, low cost, simplicity of conditions for BIBO stability using the Small Gain theorem.
operation, inexpensive maintenance and effectiveness for most Mohan and Neethu [14] have presented a nonlinear fuzzy PID
of the linear plants. These controllers generally do not work controller model by using algebraic product triangular norm,
well for nonlinear systems, higher order linear systems, delay bounded sum triangular co - norm, Larsen product inference
systems and systems which are complex and vague having method and CoS defuzzification.
no precise mathematical models. To overcome this difficulty, Stabililty analysis of parallel fuzzy P plus fuzzy I plus fuzzy
various kinds of modified linear PID controllers such as auto D control systems was done with the help of mathematical
tuned and adaptive PID controllers have been developed. models of the controllers [15]. A simple design method for
Alternatively, controllers employing fuzzy logic have also been interval type-2 fuzzy PID controllers was presented with
tried upon sometimes. the help of analytical structures of the controllers [16]. A
We now present the historical developments in fuzzy control performance-driven approach has been proposed [17] for gain
technology. A fuzzy PID controller has been constructed [1] tuning of fuzzy PID controller for multi input multi output
by combining a fuzzy PD controller and a fuzzy I controller systems.
in parallel. It has been shown [3] that PID controllers can Till date all researchers have used the CoS method only to
be realized by product-sum-gravity method and simplified find the mathematical models of fuzzy PID controller. The
fuzzy reasoning method. Analytical structure for a fuzzy PID Ht method of defuzzification is computationally attractive and
controller and its Bounded - Input - Bounded - Output (BIBO) satisfies other desirable properties in the context of control.
stability analysis have been studied [4]. Fuzzy PI and fuzzy Mathematical models of fuzzy controller (usually nonlinear)
PD controllers have been combined in parallel to get a fuzzy give better performance than that of conventional PID con-
PID controller [5]. In order to improve the performance in trollers when applied to nonlinear plants. Further, the models
transient and steady states, an adaptive method via function can be easily implemented on general digital platforms like
tuner has been developed [6] to tune the scaling factors of Field Programmable Gate Arrays (FPGA), digital computer
the fuzzy controller online. Tuning of fuzzy PID controller etc. This will avoid the requirement of dedicated hardware
parameters using gain margin and phase margin specifications or software to implement fuzzy controllers. The existing
has been proposed [7]. controller structures in FPGA or digital computer can be easily
Several forms of decomposed PID fuzzy logic controllers replaced with the nonlinear fuzzy controller models. So, in this
have been tested and compared [8]. A function-based evalu- paper an attempt is made to derive a new nonlinear fuzzy PID
ation approach has been proposed [9] for a systematic study controller model using the combination of algebraic product
of fuzzy PID controllers. An adaptive method via relative rate AND - maximum OR - Ht defuzzification.
The paper is organized as follows: The next section deals with 0
if −He ≤ es (k) ≤ −he
fundamental components of a typical fuzzy PID controller. es (k)+he
µPe (es (k)) = if −he ≤ es (k) ≤ he (4)
Section III presents a mathematical model of nonlinear fuzzy 2he
1 if he ≤ es (k) ≤ He
PID controller. Properties of the model are described in
Section IV. The computational aspects of the fuzzy controller Similarly, the mathematical descriptions of the other member-
are discussed in Section V. Section VI concludes the paper. ship functions on two scaled inputs ∆es (k) and ∆2 es (k) can
II. P RINCIPAL C OMPONENTS OF A TYPICAL F UZZY PID be defined. Notice that µNe (es (k)) + µPe (es (k)) = 1, ∀es (k).
C ONTROLLER This is true for the other two inputs. The output variable is
fuzzified by L - type, Π - type and Γ membership functions
The incremental control effort generated by a discrete-time
and are shown in Fig. 4, where ∆us (k) = S∆u · ∆u(k).
PID controller is given by:
These membership functions were used by the researchers
extensively for deriving the mathematical models of fuzzy
∆u(k) = KPd ∆e(k) + KId e(k) + KD
d
∆2 e(k) (1)
controllers. It is also possible to use nonlinear member-
∆u(k) = ∆es (k) + es (k) + ∆2 es (k) (2) ship function like Gaussian type for obtaining mathematical
models. However, such a choice will restrict the usage of
where ∆u(k) = u(k) − u(k − 1), ∆e(k) = e(k) − e(k − 1),
other components of fuzzy controller while deriving their
∆2 e(k) = ∆e(k) − ∆e(k − 1) and
mathematical models. Moreover, the models obtained by using
KPd , KId and KD d
are respectively the proportional, integral L - type, Π - type and Γ - type membership functions in the
and derivative constants of discrete-time PID controller. The literature are nonlinear and were applied to control linear as
inputs to fuzzy PID controller are shown in Fig. 1. The block well as nonlinear plants.
diagram of a typical fuzzy PID controller is shown in Fig. 2
−1
in which Se , S∆e , S∆2 e and S∆u represent the scaling factors
of the fuzzy controller. The scaled inputs are given by es (k)
= Se · e(k), ∆es (k) = S∆e · ∆e(k) and ∆2 es (k) = S∆2 e ·
∆2 e(k). The principal components of fuzzy PID controller
are fuzzification and defuzzification modules, and control rule
base which are described in the following.
A. Fuzzification Module
The inputs are fuzzified by L - type and Γ - type mem- B. Control Rule Base
bership functions [2], shown in Fig. 3, whose mathematical Total eight (23 ) rules are required as there are two fuzzy
description is given by sets defined on each of the three input variables. In this paper
rules having the same consequent part are merged into one
1
if −He ≤ es (k) ≤ −he rule. As there are four fuzzy sets defined on output variable
µNe (es (k)) = −es2h (k)+he
e
if −he ≤ es (k) ≤ he (3) we have the following four control rules :
0 if he ≤ es (k) ≤ He R1 : IF es (k) is Ne AND ∆es (k) is N∆e AND ∆2 es (k) is
N∆2 e THEN ∆us (k) is ∆U−2 (B − 0.5A) · (µ+2 − µ−2 ) + B3 · (µ+1 − µ−1 )
= (5)
R2 : IF (es (k) is Ne AND ∆es (k) is N∆e AND ∆2 es (k) is µ−2 + µ−1 + µ+1 + µ+2
P∆2 e ) OR (es (k) is Ne AND ∆es (k) is P∆e AND ∆2 es (k) where p(j) represents the peak value of the j th fuzzy set of the
is N∆2 e ) OR (es (k) is Pe AND ∆es (k) is N∆e AND output. Peak values of the fuzzy sets ∆U−2 , ∆U−1 , ∆U+1 and
∆2 es (k) is N∆2 e ) THEN ∆us (k) is ∆U−1 ∆U+2 are −(B − 0.5A), −B B
3 , 3 and (B − 0.5A) respectively.
R3 : IF (es (k) is Ne AND ∆es (k) is P∆e AND ∆2 es (k) is
P∆2 e ) OR (es (k) is Pe AND ∆es (k) is N∆e AND ∆2 es (k) III. M ATHEMATICAL M ODELS OF THE S IMPLEST F UZZY
is P∆2 e ) OR (es (k) is Pe AND ∆es (k) is P∆e AND ∆2 es (k) PID CONTROLLER
is N∆2 e ) THEN ∆us (k) is ∆U+1
For simplicity let h1 = he , h2 = h∆e , h3 = h∆2 e , x1
R4 : IF es (k) is Pe AND ∆es (k) is P∆e AND ∆2 es (k) is
= es (k), x2 = ∆es (k), and x3 = ∆2 es (k). Then upon
P∆2 e THEN ∆us (k) is ∆U+2
substituting the values of µ−2 , µ−1 , µ+1 and µ+2 defined in
Algebraic product t - norm is considered to perform the AND Table I in Eq. (5), we have the following expressions:
operation in the rule base and is defined as N
∆us (k) = D
Algebraic product: t(µA (x), µB (y)) = µA (x) · µB (y) where
Maximum t - co - norm is considered to perform the OR Case 1: In cells (1, 2, 3), (1, 3, 4), (2, 3, 1), (3, 1, 2), (3, 4, 1),
operation in the rule base and is defined as (4, 1, 3)
Maximum: s(µA (x), µB (y)) = max{µA (x), µB (y)} N = S1 B
D=3
We consider all possible combinations of three input vari-
ables in a 3D space. A point, say (xl , yl , zl ), in a 3D and S1 defined in Table I. Here S1 is the sign term either + or
space can always be distinctly shown by taking its projections −. Such a term is introduced here to present the mathematical
on the xy, yz and zx planes. So, as shown in Fig. 5, models in all the possible cells in a concise manner.
twenty input combinations are considered in each input plane Case 2: In cells (2, 2, 2), (4, 4, 4)
(es (k)∆es (k), ∆es (k)∆2 es (k) and ∆2 es (k)es (k)) so that the
N = S1 (2B − A)
input point (es (k), ∆es (k), ∆2 es (k)) can be uniquely located
in the 3D cell (subspace) represented by the triplet (na , nb , nc ) D=2
where na , nb , nc = 1, 2, 3, ..., 20. For example, the triplet (9, Case 3: In cells (1, 7, 9), (1, 8, 10), (3, 11, 5), (3, 12, 6)
11, 20) represents the 3D cell with 9 from (a), 11 from (b),
and 20 from (c) of Fig. 5. A cell is said to be valid if and N = S1 Bx3
only if the relations between es (k) and ∆es (k), and ∆es (k) D = 3h3
and ∆2 es (k) produce the relation between ∆2 es (k) and es (k).
For example, the cell (13, 14, 16) is a valid cell because by Case 4: In cells (2, 7, 6), (2, 8, 5), (4, 11, 10), (4, 12, 9)
combining the relations |∆es (k)| ≥ es (k) (valid in 13 of Fig. N = (4B − 3A)x3 + S1 (8B − 3A)h3
5(a)) and |∆2 es (k)| ≥ |∆es (k)| (valid in 14 of Fig. 5(b)) will
D = 12h3
produce the relation |∆2 es (k)| ≥ es (k) which is valid in 16
of Fig. 5(c). The control rules R1 to R4 are used to evaluate Case 5: In cells (5, 2, 8), (6, 2, 7), (9, 4, 12), (10, 4, 11)
appropriate control law in each valid cell (na , nb , nc ). Let the N = (4B − 3A)x1 + S1 (8B − 3A)h1
outcomes of premise parts of rules R1 , R2 , R3 and R4 be µ−2 ,
µ−1 , µ+1 and µ+2 respectively. By using the algebraic product D = 12h1
triangular norm and the maximum triangular co - norm, the Case 6: In cells (5, 3, 11), (6, 3, 12), (9, 1, 7), (10, 1, 8)
values of µ−2 , µ−1 , µ+1 and µ+2 are found in each valid cell
N = S1 Bx1
and shown in Table I.
D = 3h1
C. Defuzzification Module Case 7: In cells (7, 6, 2), (8, 5, 2), (11, 10, 4), (12, 9, 4)
Defuzzification is done here using the Ht method. Ht N = (4B − 3A)x2 + S1 (8B − 3A)h2
method is used instead of the popular CoS defuzzification
D = 12h2
because it gives simpler yet nonlinear expressions for the
control law. According to this method [2], the defuzzified value Case 8: In cells (7, 9, 1), (8, 10, 1), (11, 5, 3), (12, 6, 3)
of scaled control output is given by N = S1 Bx2
P4 (j)
∗ j=1 p · µj D = 3h2
∆us (k) = P4
j=1 µj Case 9: In cells (5, 7, 16), (5, 7, 17), (5, 8, 18), (6, 7, 15),
Fig. 5. (a) es (k) − ∆es (k) plane, (b) ∆es (k) − ∆2 es (k) plane, and (c) ∆2 es (k) − es (k) plane of 3 - dimensional scaled input space.
TABLE I
O UTCOMES OF ALGEBRAIC PRODUCT AND AND MAXIMUM OR AND ATTRIBUTES OF S1 IN DIFFERENT CELLS
Cells µ−2 µ−1 µ+1 µ+2 S1
Case a: All three inputs lie in the outer cuboid
(1,2,3),(2,3,1),(3,1,2) 0 1 0 0 −
(1,3,4),(3,4,1),(4,1,3) 0 0 1 0 +
(2,2,2) 1 0 0 0 −
(4,4,4) 0 0 0 1 +
Case b: One input lies in the inner cuboid and two inputs lie in the outer cuboid
(1,7,9),(1,8,10),(3,11,5),(3,12,6) 0 N∆2 e P∆2 e 0 +
(2,7,6),(2,8,5) N∆2 e P∆2 e 0 0 −
(4,11,10),(4,12,9) 0 0 N∆2 e P∆2 e +
(5,2,8),(6,2,7) Ne Pe 0 0 −
(5,3,11),(6,3,12),(9,1,7),(10,1,8) 0 Ne Pe 0 +
(7,6,2),(8,5,2) N∆e P∆e 0 0 −
(7,9,1),(8,10,1),(11,5,3),(12,6,3) 0 N∆e P∆e 0 +
(9,4,12),(10,4,11) 0 0 Ne Pe +
(11,10,4),(12,9,4) 0 0 N∆e P∆e +
Case c: Two inputs lie in the inner cuboid and one input lies in the outer cuboid
(5,7,16),(5,7,17),(5,8,18),(6,7,15) Ne N∆2 e Pe N∆2 e Pe P∆2 e 0 −
(5,8,19),(6,7,14),(6,8,13),(6,8,20) Ne N∆2 e Ne P∆2 e Pe P∆2 e 0 −
(7,14,6),(8,13,6),(8,19,5),(8,20,6) N∆e N∆2 e P∆e N∆2 e P∆e P∆2 e 0 −
(7,15,6),(7,16,5),(7,17,5),(8,18,5) N∆e N∆2 e N∆e P∆2 e P∆e P∆2 e 0 −
(9,11,13),(9,11,20),(9,12,14),(10,11,19) 0 Ne N∆2 e Ne P∆2 e Pe P∆2 e +
(9,12,15),(10,11,18),(10,12,16),(10,12,17) 0 Ne N∆2 e Pe N∆2 e Pe P∆2 e +
(11,13,9),(11,19,10),(11,20,9),(12,14,9) 0 N∆e N∆2 e P∆e N∆2 e P∆e P∆2 e +
(11,18,10),(12,15,9),(12,16,10),(12,17,10) 0 N∆e N∆2 e N∆e P∆2 e P∆e P∆2 e +
(13,6,8),(14,6,7),(19,5,8),(20,6,8) Ne N∆e Pe N∆e Pe P∆e 0 −
(13,9,11),(14,9,12),(19,10,11),(20,9,11) 0 Ne N∆e Pe N∆e Pe P∆e +
(15,6,7),(16,5,7),(17,5,7),(18,5,8) Ne N∆e Ne P∆e Pe P∆e 0 −
(15,9,12),(16,10,12),(17,10,12),(18,10,11) 0 Ne N∆e Ne P∆e Pe P∆e +
Case d: All three inputs lie in the inner cuboid
−he ≤ es (k) ≤ he , −h∆e ≤ ∆es (k) ≤ h∆e , −h∆2 e ≤ ∆2 es (k) ≤ h∆2 e
(13,14,16), (14,14,15), (19,13,16),(19,13,17)
Ne N∆e N∆2 e Pe N∆e N∆2 e Pe P∆e N∆2 e Pe P∆e P∆2 e +
(19,19,18),(19,20,17),(20,14,16), (20,14,17)
(13,15,16),(13,15,17),(13,16,18),(14,15,15)
Ne N∆e N∆2 e Pe N∆e N∆2 e Pe N∆e P∆2 e Pe P∆e P∆2 e +
(19,18,18),(20,15,17),(20,16,18)(20,17,18)
(13,16,19),(13,17,19), (14,15,14),(14,16,13)
Ne N∆e N∆2 e Ne N∆e P∆2 e Pe N∆e P∆2 e Pe P∆e P∆2 e +
(14,16,20), (14,17,20),(19,18,19),(20,17,19)
(15,14,14), (16,13,14), (16,19,13),(16,20,14)
Ne N∆e N∆2 e Ne P∆e N∆2 e Ne P∆e P∆2 e Pe P∆e P∆2 e −
(17,19,13),(17,19,20),(17,20,14),(18,19,19)
(15,14,15), (16,13,15), (17,13,15),(17,20,15)
Ne N∆e N∆2 e Ne P∆e N∆2 e Pe P∆e N∆2 e Pe P∆e P∆2 e −
(18,13,16),(18,19,18),(18,20,16),(18,20,17)
(15,15,14), (15,16,13), (15,17,20),(15,17,13)
Ne N∆e N∆2 e Ne N∆e P∆2 e Ne P∆e P∆2 e Pe P∆e P∆2 e −
(16,18,13),(16,18,20),(17,18,20),(18,18,19)
(9, 11, 13), (9, 11, 20), (9, 12, 14), (10, 11, 19) (16, 13, 15), (17, 13, 15), (17, 20, 15), (18, 13, 16),
N = (10B − 3A)h1 x3 + S1 (2B − 3A)x1 x3 + (6B − (18, 19, 18), (18, 20, 16), (18, 20, 17), (19, 18, 19),
3A)(h3 x1 + S1 h1 h3 ) (20, 17, 19)