This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Kamyar Mehran Industrial Automation, Robotics and Artiﬁcial Intelligence (EEE8005) School of Electrical, Electronic and Computer Engineering

Newcastle University · 2008 ·

Contents

1 Introduction 2 Takagi-Sugeno fuzzy modeling 2.1 Construction of Fuzzy Models . . . . . . . . . . . . . . 2.1.1 Sector Nonlinearity . . . . . . . . . . . . . . . . 2.2 Basic Fuzzy Mathematics for Modeling . . . . . . . . . 2.2.1 Local Approximation in Fuzzy Partition Spaces 3 Appendix References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 . 3 . 4 . 5 . 10 17 18

ii

Facing complex and nonlinear systems. understandable. However. it is proved that Takagi-Sugeno fuzzy models are universal approximators of any smooth nonlinear system [4. In fact.e. The idea of multi-model approach [1] is not new. The main feature of a Takagi-Sugeno fuzzy model is to express the local dynamics of each fuzzy implication (rule) by a linear system model. Nevertheless. for many nonlinear plans i. the order of the model often becomes very high so that past eﬀects are taken into account. a linear controller is prone to be unstable. and responsible for respective sub-domains. because the nonlinearities in the plant cannot be properly dealt with. that the system model is well achievable through a mathematical model and the parameters of the system model are reasonably well-known. acquired from linearizing the nonlinear system.Chapter 1 Introduction Almost all of the physical dynamical systems in real life cannot be represented by linear diﬀerential equations and have a nonlinear nature. 5]. linear control methods rely on the key assumption of small range of operation for the linear model. even if that is physically unrealistic. the reader will ﬁnd. 1 . but the idea of fuzzy modeling [2] using the concept of the fuzzy sets theory [3] oﬀers a new technique to build multi-models of the process based on the input-output data or the original mathematical model of the system. by some examples. that almost all nonlinear dynamical systems can be represented by TakagiSugeno fuzzy models to high degree of precision. One way to cope with such diﬃculty is to develop a nonlinear model composing of a number of sub-models which are simple. which usually use to design a process control. mechanical and electrical systems does not allow linear approximation. for instance. The overall fuzzy model of the system is achieved by fuzzy ”blending” of the linear system models. In this tutorial. At most a linear combination like a fuzzy model is clearly understandable. building a mathematical model is very diﬃcult and only the input-output data yielded from running the process is accessible for an estimation. in the design process of controllers. Many control problems involve uncertainties in the model parameters. to be valid. At the same time. we have to recognize that modeling is an art and it is important to realize system modeling is generally an act to understand things directly rather than by computer. chemical processes. However due to the high nonlinearity of the process. It is also necessary. the highly nonlinear and discontinuous nature of many. When the required operation range is large. A controller based on inaccurate or obsolete values of the model parameters may show signiﬁcant performance degradation or even instability. Another assumption of linear control is that the system model is indeed linearizable and the linear model is accurate enough for building up the controller. There are some complicated approaches like auto-regressive model based on the input-output data to compensate model uncertainties. The fuzzy model proposed by Takagi and Sugeno [2] is described by fuzzy IF-THEN rules which represents local input-output relations of a nonlinear system.

are diﬀerent. n − 1 and n − 2. however.1) where x1 . (2. M1 . and y (n − 2) are the output of the system to be modeled at sampling time n. x2 . M5 . Thus.. The variables as well as linguistic terms. valve openness). ”IF x1 is M1 AND x2 is M2 AND x3 is M3 ” is called the rule antecedent. Now. u(n) and u(n − 1) are system input at time n and n − 1. and u1 and u2 are output variables (e. In theory. y (n + 1) is system output at the next sampling time. and x3 are input variables (e. let us look at the so-called TS fuzzy rules. Mamdani fuzzy rules and Takagi-Sugeno (TS. which are linguistic if-then statements involving fuzzy sets. its ﬁrst derivative and its second derivative). M4 . respectively. ”High. can be represented by mathematical symbols. An example of a Mamdani fuzzy rule for fuzzy modeling is IF y (n) is M1 AND y (n−1) is M2 AND y (n−2) is M3 AND u(n) is M4 AND u(n−1) is M5 THEN y (n + 1) is M6 (2. M2 . they should be discrete because virtually all fuzzy controllers and models are implemented using digital computers. and M6 are fuzzy sets. fuzzy logic. namely. practically speaking. And.g. and AND are fuzzy logic AND operators. respectively.” ”Small. A simple but representative Mamdani fuzzy rule describing the movement of a car is: IF Speed is High AND Acceleration is Small THEN Braking is (should be) Modest. y (n − 1). and fuzzy inference. where Speed and Acceleration are input variables and Braking is an output variable. n + 1. u2 is M5 . The variables involved. Two major types of fuzzy rules exist. The structure of Mamdani fuzzy rules for fuzzy modeling is the same. M4 .2) where M1 .Chapter 2 Takagi-Sugeno fuzzy modeling A fuzzy controller or model uses fuzzy rules. TS rules use functions of input variables as the rule consequent. M2 . however. such as ”High”. for short) fuzzy rules. y (n). Fuzzy rules play a key role in representing expert control/modeling knowledge and experience and in linking the input variables of fuzzy controllers/models to output variable (or variables). and the ﬁrst two are called input fuzzy sets while the last one is named the output fuzzy set. M3 .” and ”Modest” are fuzzy sets. a Mamdani fuzzy rule for a fuzzy controller involving three input variables and two output variables can be described as follows: IF x1 is M1 AND x2 is M2 AND x3 is M3 THEN u1 is M4 . and M5 are fuzzy sets. Unlike Mamdani fuzzy rules. For fuzzy control.. whereas the remaining part is named the rule consequent. these variables can be either continuous or discrete. a TS rule 2 . error. M3 . Lets ﬁrst start with the familiar Mamdani fuzzy systems.g.

etc. u(n − 1)).1. Type 2 is the widely-used Mamdani type fuzzy system which the output function is determined based on overall fuzzy output.2 utilizes a two-rule two-input fuzzy inference system to show diﬀerent types of fuzzy system mentioned above. x3 ). some of them are centroid of area. where F () is an arbitrary function. y (n − 1). a decision-making unit which performs the inference operations on the rules. x2 . Similarly.corresponding to the Mamdani rule (2. a defuzziﬁcation interface which transform the fuzzy results of the inference into a crisp output. Fuzzy inference systems also known as fuzzy rule-based systems or fuzzy models are schematically shown in Figure 2. a TS rule analogous to the Mamdani rule (2. Type 3 is the Takagi-Sugeno type fuzzy system. x3 ). a database which deﬁnes the membership functions of the fuzzy sets used in the fuzzy rules. we need a T-S fuzzy model for a nonlinear system. 7]. Therefore the construction of a fuzzy model represent an important and basic procedure in this approach. Derivation from given nonlinear system equations. To design a T-S fuzzy controller. The identiﬁcation approach to fuzzy modeling is suitable for plants that are unable or too diﬃcult to be represented 3 .1: Fuzzy inference system. maximum of maxima. u2 = g (x1 . where f () and g () are two real functions of any type. Sugeno’s and Kang’s excellent work [6.1 Construction of Fuzzy Models Figure 2. min of maxima. In this tutorial. 2. There has been an extensive literature of fuzzy modeling using input-output data following Takagi’s. a fuzziﬁcation interface which transform the crisp inputs into degrees of match with linguistic values. Identiﬁcation (fuzzy modeling) using input-output data and 2. Figure 2. Figure 2.1) is IF x1 is M1 AND x2 is M2 AND x3 is M3 THEN u1 = f (x1 . In general there are two approaches for constructing fuzzy models: 1. They are composed of 5 conventional block: a rulebase containing a number of fuzzy if-then rules. The procedure mainly consist of two parts: structure identiﬁcation and parameter identiﬁcation. for fuzzy modeling.2) is in the following form: IF y (n) is M1 AND y (n−1) is M2 AND y (n−2) is M3 AND u(n) is M4 AND u(n−1) is M5 THEN y (n + 1) = F (y (n). x2 .3 illustrates the model-based fuzzy control design approach. we focus only on fuzzy models that use the T-S rule consequent. u(n). y (n − 2).

Sector nonlinearity is based on the following idea.4a illustrates the sector nonlinearity approach. it is sometimes diﬃcult to ﬁnd global sector for general nonlinear systems. is more appropriate. the second approach. which derives a fuzzy model from given nonlinear dynamical models. for example.Consequent part Premise part M 11 M 12 Type 1 M4 Type 2 M4 Type 3 u1 = ax1 +bx2 +c u M 21 x1 M 22 x2 x1 x1 x2 x2 Mw u ltip lic a tio n ( o w r m in ) 1 w1 u1 u M5 M5 2 w2 u2 u2 = ax1 +bx2 +r u u W e ig h te d a v e ra g e u = w1 × u1 + w2 × u 2 w1 + w2 m a x u = W e ig h te d a v e ra g e w1 × u1 + w2 × u 2 w1 + w2 u ( centroid of area ) Figure 2. where two lines become the local sectors under −d < x(t) < d. 2.1 Sector Nonlinearity The idea of using sector nonlinearity in fuzzy model construction ﬁrst appeared in [8]. In such cases. This tutorial focuses on second approach. we consider local sector nonlinearity.2: Commonly used fuzzy if-then rules and fuzzy mechanism. ”local approximation. This is reasonable as variables of physical systems are always bounded. Consider a simple nonlinear system x ˙ = f (x(t)).4b shows the local sector nonlinearity. the Lagrange method and the Newton-Euler method. analytical and/or physical models. −d < x(t) < d. However. This approach utilizes the idea of ”sector nonlinearity”. 4 . The following two examples illustrate the concrete steps to construct fuzzy models. This approach guarantees an exact fuzzy model construction. In this case. Figure 2. Figure 2. On the other hand.” or a combination of them to construct fuzzy models. The aim is to ﬁnd the global sector such that x ˙ = f (x) ∈ [a1 a2 ]x(t). nonlinear dynamical models for mechanical systems can be readily obtained by.1. where f (0) = 0. The fuzzy model exactly represents the local region. that is.

c o mP a p r a e le n s ld a t io is t n r i ( u P b t De Cd ) Figure 2.3: Model-based fuzzy control design. 3. 2. Here x1 and x2 are nonlinear terms in the equations in the last equations so we make them as our fuzzy variables.5. external disturbances and/or time. we assume that x1 ∈ [0. (b) Local sector nonlinearity. Generally they are denoted as z1 .2 Basic Fuzzy Mathematics for Modeling x ˙ 1 = x2 2 x ˙ 2 = x2 1 + x2 + u Lets consider the nonlinear system below: (2. For simplicity. 4].3) The goal is to derive a T-S fuzzy model from the above given nonlinear system equations by the sector nonlinearity approach as if the response of the T-S fuzzy model in the speciﬁed domain exactly match with the response of the original system with the same input u.4: (a) Global sector nonlinearity. a1 x(t ) f ( x(t )) f ( x(t )) a2 x(t ) x(t ) −d d x(t ) (a) (b) Figure 2. z2 and are known as premise variables that may be functions of state variables. input variables. The following steps should be taken to derive the T-S fuzzy model of (2.3).5] and x2 ∈ [−1. Therefore 5 .

5: Membership functions M1 (z1 (t)). the nonlinear system (2. N1 and N2 as follows: z1 (t) = x1 (t) = M1 (z1 (t)) · 3. We name the membership functions ”Positive”.” respectively. y (t) = Ci x(t). we can generalize that the ith 1 M 2 ( z1 (t )) Negative M 1 ( z1 (t )) Positive 1 N 2 ( z2 (t )) Small N1 ( z2 (t )) Big 0 0. .5] and x2 ∈ [−1. Figure 2.3) is modeled by the following fuzzy rules (we don’t 6 . x(t) is the state vector.5 z1 (t ) 0 3.4) Here. Here. y (t) is the output vector. ..5 −1 z2 (t ) 2. . x2 where x(t) = [x1 (t) x2 (t)]T . they are obviously obtained as follows: max z1 (t) = 3. M2 . . . Therefore.2. min z1 (t) = 0. rule of the continuous T-S fuzzy models are of the following forms: Model Rule i: IF z1 (t) is Mi1 and .5.5. This assumption is needed to avoid a complicated defuzziﬁcation process of the fuzzy controllers [9]. Equation (2.2. z2 (t) = x2 (t) = N1 (z2 (t)) · 4 + N2 (z2 (t)) · (−1). and zp (t) is Mip . It is assumed in this tutorial that the premise variables are just functions of the state variables for the sake of simplicity. M2 (z2 (t)).. it is quite routine. . u(t) is the input vector.5 4 Figure 2. min z2 (t) = −1. To acquire membership functions. Although there is no general procedure for this step and it can be done by various methods predominantly trial and error. .” ”Big. 4]. Therefore x1 and x2 can be represented by for membership functions M1 . . zp (t) are known premise variables as mentioned before.r. The ﬁrst step for any kind of fuzzy modeling is to determine the fuzzy variables and fuzzy sets or so-called membership functions. i=1. . i=1. and because M1 . max z2 (t) = 4. in exact fuzzy modeling using sector nonlinearity.5. ”Negative. (2. .r .3) can be written as x ˙ (t) = 0 x1 1 x(t). N1 and N2 are actually fuzzy sets according to fuzzy mathematics M1 (z1 (t)) + M2 (z1 (t)) = 1. 3. . Mij is the fuzzy set and r is the number of model rules. . Each linear consequent equation represented by Ai x(t) + Bi u(t) is called a subsystem. we should calculate the minimum and maximum values of z1 (t) and z2 (t) which under x1 ∈ [0. Ai is the square matrix with real elements and z1 (t). M2 . THEN x ˙ = Ai x(t) + Bi u(t).z1 = x1 and z2 = x2 .” and ”Small. N1 (z2 (t)) + N2 (z2 (t)) = 1. N1 (z2 (t)) and N2 (z2 (t)).5.5 shows these membership functions.5 + M2 (z1 (t)) · 0.

5 z1 ∈Negative 0 max z1 z2 ∈Small A1 = A3 = A2 = A4 = 1 .5 0 0.5x1 + 4x2 x ˙ 1 = x2 . 0. Model Rule 4: IF z1 (t) is ”Negative” and z2 (t) is ”Small. if z1 = x1 = 2.5 4 A4 = 0 3. −1 Now. 0 1 . x ˙ 2 = 0. This T-S fuzzy model can exactly represents the nonlinear system in the region [0. Model Rule 2: IF z1 (t) is ”Positive” and z2 (t) is ”Small.5. x ˙ 2 = 3.6). According to the model rules above Model Rule 1: IF z1 (t) is ”Positive” and z2 (t) is ”Big. Model Rule 3: IF z1 (t) is ”Negative” and z2 (t) is ”Big. 1 max z2 .” THEN x ˙ (t) = A1 x(t).5] × [−1. −1 1 . h3 (z (t)) = M2 (z1 (t)) × N1 (z2 (t)). where the subsystems are determined as: A1 = 0 max z1 1 max z2 . A2 = 0 max z1 1 max z2 . x ˙ 2 = 3.” THEN Model Rule 4: IF z1 (t) is ”Negative” and z2 (t) is ”Small.” THEN x ˙ (t) = A2 x(t). 3. we calculate the ﬁnal output of x ˙ for the speciﬁc given values of x1 and x2 . To have a clear picture of the fuzzy modeling procedure above.5x1 + 4x2 x ˙ 1 = x2 . z1 ∈Positive z2 ∈Big z1 ∈Positive z2 ∈Small A3 = which is z1 ∈Negative 0 max z1 z2 ∈Big 1 max z2 . x ˙ 2 = 0. h2 (z (t)) = M1 (z1 (t)) × N2 (z2 (t)). the T-S fuzzy modeling implication can be derived as: 7 . 4] on the x1 − x2 space.” THEN x ˙ (t) = A3 x(t).” THEN Therefore. h4 (z (t)) = M2 (z1 (t)) × N2 (z2 (t)).” THEN x ˙ 1 = x2 . x ˙ can be derived out of defuzziﬁcation process as: x ˙ (t) = h1 (z (t))A1 x(t) + h2 (z (t))A2 x(t) + h3 (z (t))A3 x(t) + h4 (z (t))A4 x(t) where h1 (z (t)) = M1 (z1 (t)) × N1 (z2 (t)).5x1 − x2 x ˙ 1 = x2 .25.consider input u(t) in this stage): Model Rule 1: IF z1 (t) is ”Positive” and z2 (t) is ”Big.” THEN x ˙ (t) = A4 x(t). 3.” THEN Model Rule 3: IF z1 (t) is ”Negative” and z2 (t) is ”Big.5 4 0 1 .5x1 − x2 Model Rule 2: IF z1 (t) is ”Positive” and z2 (t) is ”Small. according to (2.75 and z2 = x2 = 0.

75 3.25 x ˙ 2 = 3.75 = 0.75 and z2 = 0. p wi (z (t)) = j =1 Mij (zj (t)).6225.25 x ˙ 1 = 0.218 + 1.25 + 0.25 0.25 + 0 x 0.25 x ˙ 2 = 0.218 Comparing.218 + 0. in T-S fuzzy defuzziﬁcation process.25 × 0.218 Now.218 = 0. N1 (z2 ) = 0.5 × 2.218 ∧ 0. the values of x ˙ 1 = 0. Generally.25 x ˙ 2 = 0.6).625 × 0.6) = i=1 hi (z (t))Ci x(t).75 0.75 + 0.25 x ˙ 2 = 3. . u(t)) for the model rule (3.375 × 0. N2 (z2 ) = 0.218.25 x ˙ 1 = 0.75 M2 (z1 ) = 0.5 × 2.375 × 0.75 − 0.75 + 2.218 x .25 ˙ 1 = 0.5) = i=1 hi (z (t)){Ai x(t) + Bi u(t)}.5 −1 z2 = 0.25 × 0.25 = 0.125 × 0.75 + 4 × 0.6: Given the value of z1 = 2. we can see that the T-S fuzzy approximation does the good job and small value diﬀerence of x ˙ 2 . 8 .25 and x ˙ 2 = 7.218 0.218 0. the ﬁnal values for x ˙ 1 and x ˙ 2 . can be calculated as: .2775 ˙ 2 = 10.25 0.25 z2 (t ) 4 Figure 2. zp (t)].25 M1 (z1 ) = 0.75 Consequence x ˙ 1 = 0.75 = 0.5 z1 (t ) z1 = 2.75 + 4 × 0.5 × 2.25 Truth value 0.25 to the membership functions Implication Rule 1 Rule 2 Rule 3 Rule 4 Premise M1 (z1 ) = 0.218 + 0.7812.25 × 0.75 + 0.218 + 0.7812 ∧ 0. actually comes from rounding error of the premise fuzzy variables.25 + 0.218 = 7.7812 ∧ 0.25 x ˙ 1 = 0. y (t) = r r i=1 wi (z (t))Ci x(t) r i=1 wi (z (t)) (2.25 M2 (z1 ) = 0.75 − 0.25 × 0.7812 Negative M 1 ( z1 (t )) N 2 ( z2 (t )) 0. N1 (z2 ) = 0.25 = 0.75 Small N1 ( z2 (t )) Big Positive 2. N2 (z2 ) = 0. the ﬁnal outputs of the fuzzy model for the Continuous Fuzzy System (for the general rule of Discrete Fuzzy Systems see the appendix) are inferred as follows: x ˙= r r i=1 wi (z (t)){Ai x(t) + Bi u(t)} r i=1 wi (z (t)) (2. given a pair of (x(t).218. where z (t) = [z1 (t)z2 (t) .M 2 ( z1 (t )) 0.218 ∧ 0.5 × 2.25 + 9 0.7812.75 + 0. .

x1 (t). 1]. Then. the nonlinear system (2. wi (z (t)) ≥ 0. z2 (t) = (3 + x2 (t))x2 1 (t) = N1 (z2 (t)) · 4 + N2 (z2 (t)) · 0. −1 Next. z2 (t) = 0.r.” ”Negative. They are obtained as follows: x1 (t). .8) i=1. hi (z (t)) ≥ 0. deﬁne z1 (t) ≡ x1 (t)x2 (t) and z2 (t) ≡ (3 + x2 (t))x2 1 (t). r i=1 r i=1 (2.and weighting functions wi should be normalized as hi (z (t)) = wi (z (t)) r i=1 wi (z (t)) (2. .” respectively. . 2 z2 (t) 4 − z2 (t) .9) for all t. we have x ˙ (t) = −1 z2 (t) z1 (t) x(t). we can assume any range for x1 (t) and x2 (t) to construct a fuzzy model. The term Mij (zj (t)) is the grade of membership of zj (t) in Mij .10) For simplicity. z2 (t) = 4. N2 (z2 (t)) = ..7) for all t. where M1 (z1 (t)) + M2 (z1 (t)) = 1. (2. we should calculate the minimum and maximum values of z1 (t) and z2 (t) under x1 (t) ∈ [−1.10) is represented by the following fuzzy model. we have hi (z (t)) = 1. we assume that x1 ∈ [−1.x2 (t) min From the maximum and minimum values. Then.2. N1 (z2 (t)) = 9 . z1 (t) and z2 (t) can be represented by z1 (t) = x1 (t)x2 2 (t) = M1 (z1 (t)) · 1 + M2 (z1 (t)) · (−1). N1 (z2 (t)) + N2 (z2 (t)) = 1. 1] and x2 ∈ [−1. Example 1 Consider the following nonlinear system: x ˙ 1 (t) x ˙ 2 (t) = −x1 (t) + x1 (t)x3 2 (t) −x2 (t) + (3 + x2 (t))x3 1 (t) (2. . 2 M2 (z1 (t)) = 1 − z1 (1) . Since wi (z (t)) > 0. Equation (2.x2 (t) max x1 (t). i=1.. 1]. Therefore the membership functions can be calculated as M1 (z1 (t)) = z1 (t) + 1 .x2 (t) max z1 (t) = 1. For 2 the nonlinear terms.x2 (t) min z1 (t) = −1. . .” ”Big.” and ”Small. 1] and x2 (t) ∈ [−1.r.2. 4 4 We name the membership functions ”Positive. x ˙ (t) = 2 −1 (3 + x2 (t))x1 (t) 2 where x(t) = [x1 (t) x2 (t)]T and x1 (t)x2 2 (t) and (3 + x2 (t))x1 (t) are nonlinear terms. Of course.10) can be written as −1 x1 (t)x2 2 (t) x(t). x1 (t).

h2 (z (t)) = M1 (z1 (t)) × N2 (z2 (t)).7a and 2. It is also clear from Figure 2. Model Rule 4: IF z1 (t) is ”Negative” and z2 (t) is ”Small.” THEN x ˙ (t) = A4 x(t).9a. A1 = A3 = −1 4 −1 4 1 .7b illustrates the above membership functions. 1] on the x1 − x2 space. The spirit of the approach is to approximate nonlinear terms by judiciously chosen linear terms. h3 (z (t)) = M2 (z1 (t)) × N1 (z2 (t)). For instance. if we try to exactly represents the inverted pendulum by T-S fuzzy model as the way in the previous section. 2.” THEN x ˙ (t) = A1 x(t).8 shows the implementation of the above fuzzy model in Matlab/Simulink. carried out as x ˙ (t) = i=1 4 hi (z (t))Ai x(t). h4 (z (t)) = M2 (z1 (t)) × N2 (z2 (t)). In comparison.” THEN x ˙ (t) = A3 x(t). This fuzzy model exactly represents the nonlinear system in the region [−1.9b that even outside of the boundaries of x1 and x2 . M2 (z1 (t)). −1 A2 = A4 = −1 0 −1 0 1 .2. −1 −1 . Model Rule 3: IF z1 (t) is ”Negative” and z2 (t) is ”Big. −1 −1 . which means the fuzzy model can exactly represents the original system in the pre-speciﬁed domains.Model Rule 1: IF z1 (t) is ”Positive” and z2 (t) is ”Big. The defuzziﬁcation is 1 M 2 ( z1 (t )) Negative M 1 ( z1 (t )) Positive 1 N 2 ( z2 (t )) Small N1 ( z2 (t )) Big 0 −1 z1 (t ) 0 0 (a) z2 (t ) 1 0 2 4 (b) Figure 2.” THEN x ˙ (t) = A2 x(t). where h1 (z (t)) = M1 (z1 (t)) × N1 (z2 (t)). −1 Figures 2. Here. This procedure leads to reduction of the number of model rules. the above approach can accurately represents the original system (2.7: Membership functions M1 (z1 (t)).1 Local Approximation in Fuzzy Partition Spaces Another approach to obtain T-S fuzzy models is the so-called local approximation in Fuzzy Partition Spaces. N1 (z2 (t)) and N2 (z2 (t)). Figure 2.10). it ends up with 16 rules. the time responses of the fuzzy model can exactly follow the responses of the original diﬀerential equations. using 10 . Model Rule 2: IF z1 (t) is ”Positive” and z2 (t) is ”Small. As it is evident in Figure 2. 1] × [−1.

(b) fuzzy operations block. real and exact fuzzy modeling. 1. 11 .10) modeled with T-S fuzzy. 11]). (b) Function surface of x ˙ 2 . local approximation.6 0.2 1 0. a T-S model with 4 or 2 rules can be constructed.8 0.2 0 0 2 4 Time [s] 6 8 10 Fuzzy model of x1 Fuzzy model of x2 original x1 x1 & x2 original x2 original DE surface Fuzzy model surface 20 10 xdot2 0 -10 -20 40 40 x2 20 0 0 20 x1 (a) (b) Figure 2. This is because the number of model rules for the overall T-S fuzzy control system is basically the combination of the model rules and control rules (for further reading on Model-based T-S fuzzy Controllers based on LMI design see [10.9: (a) Time response of the fuzzy model and the original system.x1 weights x2 x1 & x2 Fuzzy Operations 1 s x OA1* u Product OA2* u Product1 OA3* u Product2 OA4* u Product3 Add dx Integrator1 (a) 1 x1 2 x2 Triangular MF (M1) 1 Triangular MF (M2) u Weights Normalization Add 1 weights T riangular MF (N1) 3 Constant Add1 T riangular MF (N2) (b) Figure 2. The number of model rules is directly related to the complexity of analysis and design LMI control laws for the T-S fuzzy controller.8: (a) Simulink implementation of the diﬀerential equations (2.4 0.

and u is the force applied to the cart (in newtons).13) When x1 (t) is near ±π/2. When x1 (t) is near zero. 0 B1 = B2 = 0 .15) where β = cos(88◦ ) Just remind that (2. Membership functions for Rule 1 and Rule 2 can be simply deﬁned as shown in ﬁgure 2. Rule 2 1 Rule 1 0 −90 0 x1 90 (deg) Figure 2.14) (2. 4l/3 − amlβ 2 (2. x ˙ 2 (t) = 2gx1 (t)/π − aβu(t) . the nonlinear equations can be simpliﬁed as x ˙ 1 (t) = x2 (t). g sin(x1 (t))−amlx2 2 (t)sin(2x1 (t))/2−a cos(x1 (t))u(t) x ˙ 2 (t) = . aβ − 4l/3− amlβ 2 0 2g π (4l/3−amlβ 2 ) and β = cos(88◦ ). the nonlinear equations can be simpliﬁed as x ˙ 1 (t) = x2 (t). g = 9.10: Membership functions of two-rule model.15) are now linear systems. 4l/3−aml cos2 (x1 (t)) (2. 0 1 .8 m/s2 is the gravity constant.10. m is the mass of the pendulum. Here. We arrive at the following fuzzy model based on linear subsystems: Model Rule 1: IF x1 (t) is about 0 THEN x ˙ (t) = A1 x(t) + B1 u(t). M is the mass of the cart.a = 1/(m + M ). A1 = A2 = g 4l/3−aml 0 1 .12) (2. a − 4l/3− aml 0 . Model Rule 2: IF x1 (t) is about ±π/2(|x1 | < π/2) THEN x ˙ (t) = A2 x(t) + B2 u(t).11) where x1 (t) denotes the angle (in radians) of the pendulum from the vertical and x2 (t) is the angular velocity.Example 2 The equations of motion for the inverted pendulum [12] are x ˙ 1 (t) = x2 (t). An important and natural question arises in the construction using local approximation in fuzzy partition spaces or simpliﬁcation before sector nonlinearity is that ”Is it possible to 12 . 2l is the length of the pendulum.12)-(2. x ˙ 2 (t) = gx1 (t) − au(t) 4l/3 − aml (2.

14 shows the construction of fuzzy models with 2 and 4 rules respectively Matlab/Simulink.11: Time responses of original Inverted Pendulum system and their fuzzy approximations. aβ − 4l/3− amlβ 2 A4 = 0 0 1 . 0 B4 = 0 . π ]. the original vector ﬁeld plus its ﬁrst-order derivative can be accurately approximated(for further reading please see [10]).5 Time [s] 1 1. 2. Figure 2. π ] except for a thin strip near ±π/2. The answer is fortunately Yes if we consider the problem in C 0 or C 1 context.5 Figure 2. 0 B3 = 0 . Model Rule 3: IF x1 (t) is about ±π/2(|x1 | > π/2) THEN x ˙ (t) = A3 x(t) + B3 u(t). Model Rule 1: IF x1 (t) is about 0 THEN x ˙ (t) = A1 x(t) + B1 u(t). We add two more rules (Rules 3 and 4) to the fuzzy model. That is. Here A1 .13 and 2. In this subsection. Now suppose the pendulum on the cart system is built in such a way that the work space of the pendulum is the full circle [−π.12. Balancing the pendulum for the angle range of π/2 < |x1 | ≤ π is referred to as a swing-up control of the pendulum. Model Rule 4: IF x1 (t) is about π THEN x ˙ (t) = A4 x(t) + B4 u(t). we extends the result to the range of x1 ∈ [−π. A2 and B2 are the same as above and A3 = 0 2g π (4l/3−amlβ 2 ) 1 .approximate any smooth nonlinear systems with Takagi-Sugeno fuzzy models (3. Model Rule 2: IF x1 (t) is about ±π/2(|x1 | < π/2) THEN x ˙ (t) = A2 x(t) + B2 u(t). B1 . a − 4l/3− aml The membership functions of this four-rule fuzzy model are shown in Figure 2.1) having no consequent constant term?”. Figures 2 0 -2 -4 x : Fuzzy model with 4 rules -6 -8 -10 -12 0 1 x1: Angle (rad) & x2 : Angular velocity x2: Fuzzy model with 4 rules x : Original model 1 x2: Original model x : Fuzzy model with 2 rules 1 2 x : Fuzzy model with 2 rules 0. Recall that for x1 = ±π/2 the system is uncontrollable.11 compares the time response of the original system with 13 .

fuzzy models.13: (a) Simulink implementation of Inverted Pendulum modeled with T-S Fuzzy with 2 rules (b) Fuzzy operation block. One can see that the 4-rule fuzzy model is a better approximation of real system comparing to 2-rule fuzzy model.12: Membership functions of four-rule model. Figure 2. 14 .1 Rule 4 Rule 3 Rule 2 Rule 4 Rule 2 Rule 3 Rule 1 0 −180 −90 x1 0 90 180 (deg) Figure 2. x1 w1 x1 & x2 & xdot_2 x1 w2 Terminator Fuzzy Operati ons Terminator1 x2 x2 1 s Integrator1 A1* u B1* u Pulse Generator Add1 A2* u Product Product1 B2* u Add2 Add3 (a) 1 x1 MF Min & Max Triangular MF (Rule 1) 1 w1 MF Min & Max First Part Triangular MF (Rule 2) 2 w2 MF Min & Max Add Second Part Triangular MF (Rule 2) (b) Figure 2.15 depicts the error between the real and fuzzy approximation.

15 .x1 & x2 & xdot_2 w1 Terminator x1 w2 w3 Terminator1 w4 Fuzzy Operations 1 s Integrator1 Pulse Generator A1* u Product Add1 A2* u B2* u Add2 A3* u B3* u Add3 A4* u B4* u Add4 Product3 Add8 Product2 Product1 B1* u (a) 1 x1 Membership Min & Max Rule 1 1 w1 Rule 2 2 w2 Add5 Rule 2 Rule 3 3 w3 Add6 Rule 3 4 Rule 4 Add7 Rule 4 w4 (b) Figure 2. (b) Fuzzy operation block.14: (a) Simulink implementation of Inverted Pendulum modeled with T-S Fuzzy with 4 rules.

5 (a) 0 -5 -10 -15 xdot2 -20 -25 -30 -35 4 2 0 1 2 Error range of x 1 -2 -0.15: Error surface (a) with respect to time (b) of 2-rule fuzzy approximation with respect to x ˙ 2 (c) of 4-rule fuzzy approximation with respect to x ˙ 2.1.1 0.5 2 2.5 0.3 0.5 1.5 Error range of x (b) 0 -5 -10 -15 xdot2 -20 -25 -30 -35 -40 -45 4 2 0 -2 Error range of x 1 -0.5 1 Time [s] 0.6 Error range of x 2 (c) Figure 2.5 0 0.2 0.1 0 0. 16 .5 2 2.5 1 1.4 0.5 0 4 2 0 x -2 0 0.5 x 2 1 -0.

Since wi (z (t)) > 0. p wi (z (t)) = j =1 Mij (zj (t)). THEN x(t + 1) = Ai x(t) + Bi u(t). .. Ai is the square matrix with real elements and z1 (t). y (t) is the output vector.. The term Mij (zj (t)) is the grade of membership of zj (t) in Mij . y (t) = r r i=1 wi (z (t))Ci x(t) r i=1 wi (z (t)) (3.3) hi (z (t)) = for all t.2. x(t) is the state vector. u(t)). Here. and zp (t) is Mip . . . r i=1 (3. where z (t) = [z1 (t)z2 (t) .4) 17 . Given a pair of (x(t). y (t) = Ci x(t). . . the ﬁnal outputs of the fuzzy model are inferred as follows: x(t + 1) = r r i=1 wi (z (t)){Ai x(t) + Bi u(t)} r i=1 wi (z (t)) (3. . u(t) is the input vector. zp (t) are known premise variables as mentioned before.1) = i=1 hi (z (t)){Ai x(t) + Bi u(t)}.2) = i=1 hi (z (t))Ci x(t). Mij is the fuzzy set and r is the number of model rules. . .2. wi (z (t)) ≥ 0. i=1. . . . .r. Each linear consequent equation represented by Ai x(t) + Bi u(t) is called a subsystem. .Chapter 3 Appendix Discrete Fuzzy System: DFS Model Rule i: IF z1 (t) is Mi1 and . i=1. wi (z (t)) r i=1 wi (z (t)) (3. zp (t)].r.. . i=1. .r. .2. .

.6) where x(t) = [x(t) x(t − 1) · · · x(t − n + 1)]T ..2. .2.r. z1 (t) = x(t − 1). ..we have hi (z (t)) = 1. . THEN x(t + 1) = Ai x(t) + Bi u(t). r i=1 (3.r. . y (t) = Ci x(t). Example Assume in the DFS that p = n.2. the model rule can be represented as follows: Model Rule i: IF x(t) is Mi1 and . . . . i=1. hi (z (t)) ≥ 0.. z1 (t) = x(t). and x(t − n + 1) is Min . . i=1. 18 . . i=1. . .r. zn (t) = x(t − n + 1). . (3. . Then.5) for all t. .

Tanaka. TAKAGI and M. Tanaka. 15–33. North-Holland. 2000. Wang. Zadeh. “Structure identiﬁcation of fuzzy model. 19 . [6] M. “About a multimodel control methodology. 1992. [7] M. IEEE International Conference on.” 1994.” Automatica (Journal of IFAC).” Information Sciences.Bibliography [1] Z. algorithm. and K. H. vol. Dynamics of Physical Systems. Rovatti. Fuzzy Control and Modeling: Analytical Foundations and Applications. 1992. vol. Ying. 981– 986. 1996. and Fuzzy Systems: Selected Papers. Sugeno and G..” IEEE transactions on systems. “Fuzzy identiﬁcation of systems and its applications to modeling and control. pp. [10] H. SUGENO. pp. 1988. Wiley-IEEE Press. “Dynamic parallel distributed compensation for Takagi–Sugeno fuzzy systems: An LMI approach. 1. Buckley. 28. 3-4. vol. 2. implementation and application. multi-processors.” in Fuzzy Systems. Sugeno. Proceedings of the Fifth IEEE International Conference on. vol. [12] R. no. pp. Taniguchi. Dover Publications. Tada. 2003. “An approach to stability analysis of second order fuzzy systems.” Fuzzy Sets. Ishigame. “A Theory of Advanced Fuzzy Control. 1427–1434. Fuzzy Logic.” in Fuzzy Systems. Li. K. 28. 2000. 123. [3] L. “Universal fuzzy controllers. no. 1245–1248. 6. [8] S. 1.. [4] C. vol. pp.” Fuzzy Sets and Systems. 116–132. [9] K. Binder. and T. Fantuzzi and R. Kang. 1996. 1992. Fuzzy Control. 1991. 1988. A. “On the approximation capabilities of the homogeneous Takagi-Sugenomodel. pp. (Kyoto). pp. D.” in 8th IFAC World Congress. “Fuzzy sets. [5] J. 1985. 201–221. Cannon. Kawamoto. man. 1996. no. [11] J. and cybernetics. [2] T. 15. Niemann. no..

TAKAGI SUGENO DESIGN

TAKAGI SUGENO DESIGN

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd