SEE TH AMOAHEN HN TENG, MAND EYRERMF ICS, SO MO, ML/H Fuzzy Logic in Control Systems: Fuzzy Logic Controller—Part I CHUEN CHIEN LEE, sropenr Meme, (eer fzaietion and deezicaton strates the eration ofthe dab ‘nd fezy control rae, the dfniion of fury implication, aed ‘asi of fry renasing mechani, LURING the past several years, fuzzy control has smerged as one of the most active and fruitful areas jor research in the application of fuzzy set theory [141] The pionecring research of Mamdani and his colleagues ‘on fuzzy contro! [63]-[66], [50] was motivated by Zadeh's seminal papers on the linguistic approach and. system analysis based on the theory of fuzzy sets [142], [143], [145], [146]. Recent applications of fuzzy control in water quality control [127], [35], automatic train operation sys- tems {135}, [136), (139), automatie container crane opera- tion systems (137]-(159}, elevator control (23), nuclear reactor control [4], [51], automobile transmission control [40), fuzzy logic controller hardware systems (130), 131} fuzzy memory devices (107), {108}, (120), [128), [129] 1133), and fuzzy computers [132] have pointed a way for an effective utilization of fuzzy control in the context of complex ill-defined processes that can be controlled by a skilled human operator without the knowledge of their underlying dynamics. ‘The literature in fuzzy control has been growing rapidly in recent years, making it dificult to present a compre- hensive survey of the wide variety of applications that have been made. Historically, the important milestones i Manuscript eeied May 27 198 eve July 1,199. Tig work was surrey NASA pan NCEE at AFOSR Gran "Th authors withthe Elctronis Research Laboratory, Departs of Elec Enginceing snd Computer Sect, Unie of Car ta Berkeley Ca 9700 EE Log Number #932018, the development of fuzzy control may be summarized as shown in table 1. It should be stressed, however, the choice of the milestones is a subjective mater Fuzzy logic, which is the logic on whieh fuzzy control is based, is much closer in spirit to human thinking and natural language than the traditional logical systems. Ba- sically, it provides an effective means of capturing the approximate, inexact nature of the real world, Viewed in this perspective, the essential part of the fuzzy logic controller (FLC) is set of linguistic control rules related by the dual concepts of fuzzy implication and the compo- sitional rule of inference. In essence, then, the FLC provides an algorithm which can convert the linguistic control strategy based on expert knowledge into an auto- ‘matic control strategy. Experience shows that the FLC yields results superior to those obtained by conventional control algorithms. In particular, the methodology of the FLC appears very useful when the processes are t00 complex for analysis by conventional quantitative tech: niques or when the available sources of information are interpreted qualitatively, inexactly, or uncertainly, Thus fuzzy logic control may be viewed as a step toward a rapprochement between conventional precise mathemati- cal control and human-like decision making, as indicated by Gupta {30} However, at present there is no systematic procedure for the design of an FLC. In this paper we present a suney of the FLC methodology and point tothe problems Which need further research. Our investigation includes. fuzzification and defuzzitication strategies, the derivation of the database and fuzzy control rules, the definition of a fuzzy implication, and an analysis of fuzzy reasoning ‘mechanisms, ‘This paper is divided into two parts. The analysis of structural parameters of the FLC is addressed in Part 1 In addition, Part I contains five more sections. A briet summary of some of the relevant concepts in fuzzy set theory and fuzzy logie is presented in Section Il. The main idea of the FLC is described in Section II, while Section IV describes the fuzztication strategies. In Sec- tion V, we discuss the construction of the data base of an FLC. The rule base in Section VI explains the derivation of fuzzy control rules and rule-modification techniques. Part Il consists of four sections. Section I is devoted to the basic aspects of the FLC decision-making logic. Sev- eral issues including the definitions of a fuzzy implication, (0018-9472 /90 /0300-0404801.00 ©1990 IEEEim gaan i Zadeh in Wilaye oat 18 Komoiw a 190k Mizumoto and Tanaka im Hira and Peder Isms Taka and Sogeno Ions sks, ups a. 18S Tal and Watanabe Toes Yom 188 Duan Pade ‘compositional operators, the interpretations of sentence "and" and “also,” and fuzzy inference mec ‘connectives “an anisms, are investigated. Section II discusses the defu fication strategies. Some of the representative appli tions of the FLC, from laboratory level co industrial process control, afe briefly reported in Section I nally, we describe some unsolved problems and discuss further challenges in this field H. Fuzzy Sets ano Fuzzy Loate For the convenience of the reader, we shall briefly ‘summarize some of the basic concepts of fuzzy set theory and fuzzy logic which will be needed in this paper. A more detailed discussion may be found in (141), (4], (42, {148}, (149] and (21, A, Fuazy Sets and Terminology Let U be a collection of objects denoted generically by (u), which could be discrete oF continuous. U is called the universe of discourse and w represents the generic ele ment of U. Definition I: Fuzzy Set: A fuzzy set F in @ universe of discourse U is characterized by 4 membership function dy whieh takes values in the interval (0,1) namely, sy U's(0,1), A fuzzy set may be viewed as 2 generalization of the concept of an ordinary set whose membership function only takes two values (0,1). Thus a fuzzy set Fin U may be represented as a sct of ordered paits of a gener element w and is grade of membership function: F=((usps(u))u €U). When U is continuous, a fuzzy set F can be written concisely as F = jap(w)/u. When U is ciscrete, a fuzzy set Fis tepresented as PE ald / Definition 2: Support, Crossover Point, and Fuzzy Single- ton: The support of a fuzzy set F is the crisp set of all points w in U such that yp(u)>0. In particular, the element w in U at which 0.5, is called the crossover point and a fuzzy set whose support isa single point in U with jp = L0 is teferted to as fuzzy singlet ‘atone for fury sontal 3) uit spree Seam engine como! ‘Atco shorts SL Be eg nd cee con ‘pia uy cmt eee ray ‘Wantwater eum proses (113 ozzy condo! inference 24) robbie ory se contr 3) Dera fan contol les [15] Predictive zy conte 35 Paring conto of 9 model er 97 aay sytem stay 8) Fuze chip 107) Fuzzy coma hareaze sytem 130) ‘Aprons soning I) B. Set Theoretic Operations Let A and B be wo fuzzy sets in U with membership functions se and pry, respectively. ‘The set theoretic ‘operations of union, ‘intersection and complement for fuzzy sets are defined via their membership functions ‘More specifically, see the following, Definition 3: Union: ‘The membership function 4...» Of the union UB is pointwise defined for all w=U' by Bega) = rx (ae 4( 0) s(t) Definition 4 Imersection: The membership function dyn of the intersection 1.0 B is pointwise defined for all weU by Banal) = min (4.464), 4e9()} Definition S: Complement: The membership function 4; of the complement of a fuzzy set A is poinowise defined for all we U by alu) = 1 mau), Definition 6 Cartesian Product: Ayo", Ay are fuzzy sets in U;,--*,U,, respectively, the Cartesian’ prod- uct of Aj:s-s4, is a fuzzy set in the product space U,x <4, with the membership function acelin ay) = i 06) sta) Beggs natty tl) Heats) alta) “Ha (a) Definition 7: Fuzzy Relation: An n-ary fuzzy relation is a fuzzy set in UX "** XU, and sexpresed a8 Rape ata Mayra) pe tte) Mtg) EU =U) Definition 8: Sup-Star Composition: WR and S are fuzzy relations in UxV and V xW, respectively, the composition of R and S is a fuzzy relation denoted by6 seer rman rns sea MK AMMEN Wt 2, A/a, 18 Hepeed 10 os 40 5570 Speed Fp. 1. Diarammatcteresentaton of fury speed. “Speed” i i uc arial with thee fms slow" and hgh Re $ and i defined by osm {[(uw),suplaa(usr) 2.009). wsuverwew) Where + could be any operator in the class of triangular orm, namely, minimum, algebraic product, bounded product, or drastic product (also see Part 11 [1S0D. . Linguistic Variables and Fuzzy Seas Definition 9: Fuzzy Number: A fuzzy number F in a ‘continuous universe U, eg. real line is a fuzzy set F in U which is normal and convex, ie max s(o)=1, (normal) y(Auy += AD) & min (uu) sp) Hit EU, (comes) Ae[0,1] “The use of fuzzy ses provides basis for a sjstematic way’ for the manipulation of vague and imprecise con cepts. In particular, we can employ fuzzy set to represent linguistic variables. A linguistic variable canbe vegarded cither a variable whose value fa fuzzy number or as 8 Variable whose values are-defined in linguistic terms. More special: see the following. Defintion 10: Linguistic Variables: linge variable is characterized by quintuple x, T(),U,G, ) in which 4 Js the name of variable; T(x) i the term St of that i, the set of names of linguini value of with each ‘value being afuzry mimber defined on Ui G is a syntactic role for generating the names of values of sand Mis Semantic rule for asacatng wth each value is meaning. For example, if speed is interpreted ay linguistic var able, then its term set Ti speed) could be T( speed) = (stow, moderate, fast, ‘ery slow, more ores fast, --) where each term in (speed) is characterized by a fuzzy Set in a universe of discourse U =[0,100]. We might inter pret “slow as “a speed below about 40 mph” “mod- erate” as “a speed close to 55 mph," and “fast” as “a speed above about 70) mph.” These terms can be charac: terized as fuzzy sets whose membership functions are shown in Fig, 1 See eae }een ea — od Fig. 2. Bi confiurton of ry kpc conta FLO. D. Fuzzy Logic and Approximate Reasoning In fuzzy logic and approximate reasoning, there are two Important fuzzy implication inference rules named the generalized modus ponens (GMP) and the generalized ‘modus tollens (GMT): premise I: x is 4’, premise 2:if.x is Aten y is B, consequence: y is B (GP) premise I: y is B, premise 2:if xis A then y is B, consequence: x is A wn ‘The fuzzy implication inference is based on the compo- sitional rule of inference for approximate reasoning sug- gested by Zadeh in 1993 [146], Here we introduce fuzzy sets 4,4',B,B" via linguistic variables x,y instead of crisp sets in the traditional logic, The GMP, which re- duces to “modus ponens” when 4’~ 4 and B'= B, is closely related to the forward data-driven inference which Js particularly useful in the FLC. The GMT, which re- duces to “modus tollens” when B'=not B and ’= not A, is closely related to the backward goal-driven inference ‘which is commonly used in expert systems, especially in the realm of medical diagnosis. Definition 11: Sup-Star Compositional Rule of Inference: ICR isa fuzzy relation in U x V, and x isa fuzzy set in U, then the “sup-star compositional rule of inference” as- serts that the fuzzy set y in V induced by x is given by iss] yrxoR where x R is the sup-star composition of x and R. Ifthe star represents the minimum operator, then this defini tion reduces to Zadeh’s compositional rule of inference (1461 HL. Maw Inga or nue FLC In this section, we present the main ideas underlying the FLC. To highlight the issues involved, Fig. 2 shows the basie configuration of an FLC, which comprises fourprincipal components: a ficzificaion interface, a knowl: edge base, decision-making logic, and a defussification interface 1). The fuzification interface involves the following functions: a) measures the values of input variables, ')_ performs a scale mapping that transfers the ange of values of input variables into corre- sponding universes of discourse, ©) performs the function of fuzzfication that converts input data into suitable linguistic values which may be viewed af labels of fuzzy sets 2) The knowiedge base comprises a knowledge of the application domain and the attendant control goals, consists ofa “data base” and a “linguistic (fuzzy) control rule base: a) the data base provides necessary definitions, whieh are used to define linguistic control rules and fuzzy data manipulation in an FLC, ) the rule base characterizes the control goals and control policy of the domain experts by ‘means of a set of linguistic control rules. 3) The decisionmaking logic is the kernel of an FLC; it has the capability of simulating human decision- making based on fuzzy concepts and of inferring fuzzy control actions employing fuzzy implication and the rules of inference in fuzzy logic 4) The defuzzification interface performs the follow- ing functions: 2) scale mapping, which converts the range of values of output variables into corresponding universes of discourse, ) defuzzification, which yields « nontuzzy con- trol action from an inferred fuzzy control We are now ready to deseribe the main ideas underly ing the FLC in terms of fuzzy logic. The structural param- eters involved in the design of an FLC will be discussed at a later point. A. Fuzzy Conditional Statements and Fuzzy Control Rules In an FLC, the dynamic behavior of a fuzzy system is characterized by a set of linguistic description rules based fon expert knowledge. The expert knowledge is usually of| the form 1 (a sot of conditions are satisfied) rw (a set of consequences can be inferred) Since the antecedents and the consequents of these ir=TeN rules are associated with fuzzy concepts (linguistic terms), they ate often called fuzzy conditional statements In our terminology, @ fuzzy control rule is a fuzzy condi- tional statement in which the antecedent is @ condition in its application domain and the consequent is a control action for the system under control. Basically, fuzzy con- trol rules provide a convenient way for expressing control policy and domain knowledge. Furthermore, several lin- {uistic variables might be involved in the antecedents and the conclusions of these rules. When this isthe ease, the system will be referred to as a multiinput-multi-output (MIMO) fuzzy system, For example, in the case of two. input-single-output (MISO) fuzzy systems, fuzzy control rules have the form: Reif-x is Ay and y is By then 2 is C,, Reif xis Ay and y is B, then 2 is Cy, x, ‘where sy, and 2 are linguistic variables representing two process site variables and one contol variable; A,B tnd C, are linguistic values ofthe linguistic variables), and 2 in the universes of discourse U, V, and W, respec. tively, with ¢=1.2,-"-.m; and an implicit sentence con- nective als links the rules into a rule set of, equivalently, aerule base ‘A fuzzy contol rule, such as “if (cis A, and y is 6) then (z is €,)" i implemented by a fizzy implication (Guzzy relation) R, and is defined as follows s(uvow) = [ma(u)and e969] + wel) where A, and B, is a fuxzy set A,xB, in UXV: R,8(A, and BC, isa fuzzy implication elation) in UV %Wsand + denotes a fuzzy implication function [As will be seen later there are many ways in which fuzzy implication may be defined. if-xis A, and y is B, then z is C,, Me B. Fuzzifcation Operator A fuzzification operator has the effect of transforming crisp data into fuzzy sets. Symbolically, Fuczifior 9) where xp is a crisp input value from a process; x is a fuzzy set; and fuzifier represents a fuzzifcation operator. © Sentence Connective Operators An FLC consists of a set of fuzzy control rules which are related by the dual concepts of fuzzy implication and the sup-star compositional rule of inference. These fuzzy control rules are combined by using the sentence connec- tives and and aio. Since each fuzzy control rule is represented by a fuzzy relation, the overall behavior of fuzzy system is characterized by these fuzzy relations. In other words, a fuzzy system can be characterized by 2 single fuzzy relation whieh isthe combination of the fuzzy relations in the rule set. The combination in question involves the sentence connective also. Symbolically, Re alsol Ry Ray Rye" Ry) where also represents a sentence connectiveD. Compesitional Operator from the given process states x. y ‘sup-star compositional rule of inference is applied yo(xeR) where + is the sup-star composition E. Defuzzifcation Operator “The output of the inference process so fa is a fuzzy set specifying possibility distribution of control action. In the on-line control, & nonfuzzy (crisp) control action is usually required, Consequently, one must defuzzify the fuzzy control action (output) inferred from the fuzzy control algorithm, namely: 2) = defuazifer( 2) where 2, is the nonfuzzy control output and defi the defuzztication operator. ifr is F. Design Parameters of the FLC ‘The principal design parameters for an FLC are the following: 1) fuzzification stratogies and the fuzzfication operator (fuzzifier), 2) data base: 2) discretization/normalization of universes of discourse, ')_ fuzzy partition of the input and output spaces, ©) completeness, 4) choice of the membership function of a pri= mary fuzzy set; 3) rule base: 4) choice of process state (input) variables and ‘control (output) variables of fuzzy control rules, 'b)_ source and derivation of fuzzy control rules, ©) ypes of fuzzy control rules, «)_ consistency, interactivity, completeness of fuzzy contro rules; 4) decision making logic: 8). definition of @ fuzzy implication, b) interpretation of the sentence connective and, )- interpretation of the sentence connective also, 4) definitions of a compositional operator, ©) inference mechanism; 5). defuzzitication strategies and the interpretation of a defuzzification operator (defuzztier). merpretation of a IV, Fvzziticanon Srratrcies Fuzzification is related to the vagueness and impreci- sion in a natural language. It is a subjective valuation ‘which transforms a measurement into a valuation of @ Subjective value, and hence it could be defined as a mapping {rom an observed input space to fuzzy sets in certain input universes of discourse. Fuczfication plays fan important role in dealing with uncertain information ‘which might be objective or subjective in nature. In fuzzy control applications, the observed data. are usually erisp. Since the data manipulation in an FLC is based on fuzzy st theory, fuzztication is necessary during ‘an eatlier stage. Experience with the design of an FLC ‘suggests the following principal ways of dealing. with fuzztication 1) A fuzzifcation operator “conceptually” comerts a crisp value into a fuzzy singleton within a certain universe of discourse. Basically, a fuzzy singleton is ‘4 precise value and hence 0 fuzziness is intro- ‘duced by fuzification in this case, This strategy has been widely used in fuzzy control applications since it is natural and easy to implement. t interprets an input xy as a fuzzy set A with the membership function y.,(+) equal to zero except at the point sq, at which p(x) equals one. 2) Observed data are disturbed by random noise. In this case, a fuzzification operator should convert the probabilistic data into fuzzy numbers, ie. fuzzy (possibilistc) data. In this way, computational ef ciency is enhanced since fuzzy numbers are much feasier to manipulate than random variables. In 76), an isosceles triangle was chosen to be the fuzzification function, The vertex of this triangle corresponds to the mean value of a data set, while the base is twice the standard deviation of the data set. In this way, we form a triangular fuzzy number Which is convenient to manipulate (42). In this connection, it should be noted that Dubois and Prade [20] defined a bijective transformation which transforms a probability measure into a possibility measure by using the concept of the degree of necessity. Basically, the necessity of an event, E, is the added probability of elementary events in. E ‘over the probabiliy assigned to the most frequent tlementaty event outside of F. Based on the method of Dubois and Prade, the histogram of the measured data may be used (0 estimate the mem- bership function for the transformation of probabi ity into possibility (17. 3) In large scale systems and other applications, some ‘observations relating to the behavior of such 5ys- tems are precise, while others are measurable only ‘1 statistical sense, and some, referred to as wirids,” require both probabilistic and possibii tic modes of characterization. The strategy of fuzzi- fication in this case is o use the concept of “hybrid numbers” [42] which involve both uncertainty (fuzzy numbers) and randomness (random num bers), The use of hybrid number arithmetic in the design of an FLC suggests a promising direction that is in need of further exploration. V. Dara Base ‘The knowledge base of an FLC is comprised of two ‘components, namely, a data hase and a fuzzy control rule base. We shall address some issues relating to the dataTABLE 1 bo, bo, ba bo Penis vor ere 202 102 ay 404 404i 08 fase ha base in this section and to the rule base in the next section, The concepts associated with a data base are used to characterize fuzzy control rules and fuzzy data manipulation in an FLC. These concepts are subjectively defined and based on experience and engineering judg- ment. In this connection, it should be noted that the correct choice of the membership functions of a term set plays an essential role in the success of an application. In What follows, we shall discuss some of the important aspects relating to the construction of the data base in an FIC. A. Discretization /Nommalization of Universes of Discourse ‘The representation of uncertain information with fuzzy sets brings up the problem of quantifying such informa- tion for digital computer processing, In general, the rep- resentation depends on the nature of the universe of discourse. A universe of discourse in an FLC is either rete or continuous. If the universe is continuous, @ siserete universe may be formed by a discretization of the ‘continuous universe, Furthermore, a continuous universe may be normalized, as wll be seen at a later point in this section, 1) Diseretizaion of a Universe of Discourse: Discretira- tion of a universe of discourse is frequently referred to as quantization. In effect, quantization discretizes a universe into a certain number of segments (quantization levels). Each segment is labeled as a generic element, and forms a discrete universe. A fuzzy set is then defined by assign- ing grade of membership values to each generic element fof the new discrete universe. A look-up table based on aiscrete universes, which defines the output of a con- troller for all possible combinations of the input signals, ccan be implemented by off-line processing in order 10 shorten che running time of the coatroller [90]. In the case of an FLC with continuous universes, the number of ‘quantization levels should be large enough to provide an adequate approximation and yet be small to save memory storage. The choice of quantization levels has an essential influence on how fine a control can be obtained. For ‘example, if a universe is quantized for every five units of measurement instead of ten units, then the conteoller is twice as sensitive to the observed variables, 07 03 no do 0 o0 03 10 03 do 09 oD 83 bo 09 7 00 on 4 00 03 19 03. Oo SD bo no 07 07 OH a For the purpose of discretization, we need scale mapping, which serves to transform measured variables into values in the discretized universe. The mapping can be uniform (linear), nonuniform (nonlinear), or both. The choice of quantization level reflets some a prion’ knowl- edge. For example, coarse resolution could be used for large errors and fine resolution for small errors. Thus, in 4 three-input-one-output fuzzy system, we may have con- ‘ol rules ofthe fo Rj iferror (e) is 4,, sum of errors (ie) is B,. and change of error (de) is C, then output is D, {A simple instance of an FLC can be represented by Kaluk)] = F[Kye(k), Kale), Ksde(k)] where F denotes the fuzzy relation defined by the rule base and K,,i~ 1.2.3.4, represents an appropriate scal- ing mapping. In this relation, we see an analogy to the parameters of @ conventional PID controller [63} (105), in Which as a spectal case F is a linear function of its arguments. An example of discretization is shown in Table 1, where a universe of discourse is discretized into 13 levels with seven terms (primary fuzzy sets) defined 00 it In general, due to discretization, the peformance of an FLC is less sensitive to small deviations in the values of the process state variables, 2) Normalization of a Universe of Discourse: The nor rmalization of a universe requires a discretization of the universe of discourse into-a finite number of segments, with each segment mapped into a suitable segment of the normalized universe, In this setting a fuzzy set is then defined by assigning an explicit function to its member- ship function, The normalization of a continuous universe also involves a priori knowledge of the input/ output space. The scale mapping can be uniform, non- uniform, or both. One example is shown in Table Il, where the universe of discourse, [-60, +45], is trans” formed into the normalized closed interval [~ 1, + 1]. B. Fuzzy Panition of Input and Ougput Spaces A linguistic variable in the antecedent of a fuzzy con: trol rule forms a fuzzy input space with respect to a certain universe of discourse, while that in the consequentTABLE I Nonaatsaen ano Panny Fis Ses Us « Fern Drv (10-05) rT (as 03 ba Nae (03.—ba) [22-00 “92 02 ka (tassa) (200,403) (o0sum oo ng ZE [p02 4hel (run 42s] 02 02S Cog tto 1228348) 9s 02a to at be cose respectively, the maximum rule number is 3%. It should @ o ig Diagrammatic representation of fuzzy parton. (2) Coane Tuzry portion with thee terms: N equines ZE, vera and. foto) Fo zy prion with soon term NB. negate DE Ki negate med, NS, negate smal ZB 2x0 PS. poste Small'P poste meio: and PB, pote of the rule forms a fuzzy output space. In general, a linguistic variable is associated with a term set, with each term in the term set defined on the same universe of discourse. A fuzzy partition, then, determines how many terms should exist in a term set. This is equivalent to Finding the number of primary fuzzy sets. The number of primary fuzzy sets determines the granularity of the con- trol obtainable with an FLC. The primary fuzzy sets inguistic terms) usually have a meaning, such as NB: negative big, NM: negative medium; NS: negative small; ZE: zero; PS: positive small; PM: positive medium; and PB: positive big. A typical example is shown in Fig. 3, depicting two fuzzy partitions in the same normalized universe (1, +1]. Membership functions having the forms of triangle-shaped and trapezoid-shaped functions fare used here. Since a normalized universe implies the Knowledge of the input/output space via appropriate scale mappings, a well-formed term set can be achieved as shown, If this ie not the case, or a nonnormalized universe Js used, the terms could be asymmetrical and unevenly distributed in the universe. Furthermore, the cardinality fof a term set in a fuzzy input space determines the maximum number of fuzzy control ules that we can ‘construct. In the case of two-input-one-output fuzzy sys- tems, if the cardinalities of TC) and 7(y) are 3 and 7, bbe noted that the fuzzy partition of the fuzzy input/ ‘output space is not deterministic and has no unique Solution. A heuristic cut and trial procedure is usually needed to find the optimal fuzzy partition. © Completeness Intuitively, a fuzzy control algorithm should always be able to infer a proper control action for every state of process. This property is called “completeness.” The com- pleteness of an FLC relates to its data base, rule base, or both, 1) Data Base Strategy: The data base strategy is con- cerned with the supports on which primary fuzzy sets are defined, ‘The union of these supports should cover the related universe of discourse in relation to some level set «. This property of an FLC is called «-completeness. In igeneral, we choose the level « at the crossover point as shown in Fig. 3, implying that we have a strong beliet in the positive sense of the fuzzy control rules which a associated with the FLC. In this sense, a dominant rule always exists and is associated with the degree of belief greater than 05. In the extreme case, two dominant rules fare activated with equal belief 0.5. 2) Rule Base Strategy: The rule base strategy has to do with the fuzzy control rules themselves. The property of ‘completeness is incorporated into fuzzy control rules through design experience and enginecring knowledge. ‘An additional rule is added whenever a fuzzy condition is not included in the rule base, or whenever the degree of partial match between some’ inputs and the predefined fuzzy conditions is lower than some level, say 0.5. The former shows that no control action will result, The latter indicates that no dominant rule wil be fire. D. Membership Function of a Primary Fuzzy Set ‘There are two methods used for defining fuzzy sets, depending on whether the universe of discourse is dis: crete of continuous: a) numerical and b) functional 1) Numerical Definition: Yn this case, the grade of mem- bership function of a fuzzy set is represented as a vector ‘of numbers whose dimension depends on the degree of discretization. An illustrative example is shown in Table In this case, the membership function of each primaryFie. 4 Example of fnctionl defn of pinay fizzy et fuzzy set has the form of ae) = Eas /u, where = (03,0.7,1.0,027,03} 2) Functional Definition: A functional definition ex: presses the membership function of a fuzzy set in a functional form, typically a bell-shaped function, triangle- shaped function, trapezoid-shaped function, cte. Such functions are used in FLC because they lead themselves to manipulation through the use of fuzzy arithmetic. The functional definition can readily be adapted to a change in the normalization of a universe. Table III and Fig, 4 show an example of a functional definition expressed as: jl) = emp Note that if the normalized universe is changed, the parameters u,,0; should be changed accordingly. Either @ numerical definition or functional definition may be used 0 assign the grades of membership to the primary fuzzy sets. The choice of grades of membership is based on the subjective criteria of the decision. In particu lar, as we mentioned before, if the measurable data might be disturbed by noise, the membership functions should be sufliciently wide to reduce the sensitivity to noise. This raises the issue of the fuzziness or, more accurately, the specificity of @ membership function, which affects the robustness of an FLC. This issue is discussed in greater detail in (58) VI. Rute Base A fuzzy system is characterized by a set of linguistic statements based on expert knowledge. The expert knowl- edge is usually in the form of “if-then” rules, which are ceasily implemented by fuzzy conditional statements in fuzzy logic. The collection of fuzzy control rules that are expressed as fuzzy conditional statements forms the rule base or the rule set of an FLC. In this section, we shall examine the following topics related to fuzzy control rules: choice of process state (input) variables and control output) variables, source and derivation, justification, types of fuzzy control rufes, and properties of consistency, interactivity, and completeness. A. Choice of Process State Variables and Control Variables of Fuzzy Contol Rules Fuzzy control rules are more conveniently formulated n linguistic rather than numerical terms. The. proper choice of process state variables and control variables is essential (0 the characterization of the operation of fuzzy system, Furthermore, the selection of the linguistic variables has a substantial effect on the performance of an FLLC. As was stated earlier, experience and engincer- ing knowledge play an important role during this selection stage. In particular, the choice of linguistic variables and their membership function have a strong influence on the linguistic structure of an FLC. Typically, the linguistic variables in an FLC are the state, state error, state error derivative, state error integral, et B. Source and Derivation of Fuzzy Control Rules ‘There are four modes of derivation of fuzzy control rules, a8 reported in (103]. These four modes are not tutually exclusive, and it seems likely that a combination of them would be necessary (© construct an effective ‘method for the derivation of fuzzy control rules 1) Expert Esperience and Control Engineering Knowledge Fuzzy control rules have the form of fuzzy conditional Statements that relate the state variables in the an tecedent and process control variables in the consequent. In this connection, it should be noted that in our daily life most of the information on which our decisions are based is linguistic rather than numerial in nature. Seen in this perspective, fuzzy control rules provide a natural frame work for the characterization of human behavior and decisions analysis. Many experts have found that fuzzy control rules provide a convenient way to express their domain knowledge. ‘This explains why most FLCS are based on the knowledge and experience which are ex- pressed in the language of fuzzy if-then rules (64), [47}, {0}, (80), {82}, [59], 1118), (113), (88), (127), (4, The formulation of fuzzy control rules can be achieved by means of two heuristic approaches. The most common fone involves an introspective verbalization of human ex: pertise. A typical example of such verbalization is the operating manual for a cement kiln, Another approach includes an interrogation of experienced experts or opera: {ors using a carefully organized questionnaire. In this manner, we can form a prototype of fuzzy control rules for a practicular application domain, For optimized per- formance, the use of cut and tral procedures is usually & necessity. 2) Based on Operator’ Conrol Actions: In many indus- trial man-machine control systems, the input-output re- lations are not known with sufficient precision to make it Possible to employ classical control theory for modeling nd simulation. And yet skilled human operators can control such systems quite successfully without having any {quantitative models in mind, In effect, human operator employs-consciously or subconsciously—a set of Tuzzyoe i « Fe 5 if-then rules 10 control the process. As was pointed out by Sugeno, to automate such processes, itis expedient to express the operator's control rules as fuzzy if-then rules temploying linguistic variable, In practice, such rules can be deduced from the observation of human controller's actions in terms of the input-output operating data (97}-(99} '3) Based on the Fuzzy Model of a Process: a the liguis tic approach, the linguistic description of the dynamic characteristics of a controlled process may be viewed as a fuzzy model of the process. Based on the fuzzy model, we ccan generate a set of fuzzy control rules for attaining ‘optimal performance of a dynamic system. The set of fuzzy control rules forms the rule base of an FLC. Al- though this approach is somewhat more complicated, it yields better performance and reliability, and provides a more tractable structure for dealing theoretically with the FLC. However, this approach to the design of an FLC has not as yet been fully developed. +4) Based on Leaming: Many FLCs have been built to emulate human decision-making behavior, but few are focused on buman learning, namely, the ability to create fuzzy control rules and to modify them based on experi- ence, Procjk and Mamdani [87] described the first self- ‘organizing controller (SOC). The SOC has a hierarchical structure which consists of two rule bases. The first one is the general rule base of an FLC. The second one is constructed by “meta-rules” which exhibit human-like learning ability to ereate and modify the general rule base based on the desired overall performance of the system. Recently, further studies relating to the SOC have been carried out at Queen Mary College and elsewhere (60), {94}, [102], [95], 1106]. A very interesting example of fuzzy rule-based system which has a learning capability is Sugeno's fuzzy cat [97], [99]. Sugeno’s fuzzy car can be trained to park by itselt ©. Jusification of Fuzzy Control Rules There are two principal approaches to the derivation of fuzzy control rules. The first is a heuristic method in ‘which a collection of fuzzy control rules is formed by Analyzing the behavior of a controlled process. The con- trol rules are derived in such a way that the deviation ute ustication by wing phase pla. a) Pase-plan test. Sytem tp respon from a desited state can be corrected and the control ‘objective can be achieved. The derivation is purely heutis- tie in nature and relies on the qualitative knowledge of process behavior, Several methods of adjustment of rule Selection have been studied (1), (49), (7h (6). A brief review of these results is given in the following. The second approach is basically a deterministic method which can systematically determine the linguistic structure and or parameters of the fuzzy control rules that satisfy the control objectives and constraints (111], (103), (108), on. ‘Mamdani (1] proposed a prescriptive algorithm for de- riving the “best” control rules by restricting system re- sponses to a “prescriptive fuzzy band” which is specified by fuzzy control rules. However, the convergence of the prescriptive method requires a careful analysis. King and Mamdani [49] introduced another useful ‘method for rule justification. So-called “scale mappings” should be adjusted first so that the system trajectory can terminate on a desired state. The rule jusifiation is done by referring to a closed system trajectory ina phase plane, A knowledge of parameter-adjusting based on phase plane analysis (eg, overshoot, rise time) and an intuitive feel for the behavior of the closed loop system are required. “The principle of global rule modification in symmetry and monotonicity is also employed. For example, Fig, 5 shows the system response of a process to be controlled, where the input variables of the FLC are the error (E) and error derivative (DE). The output is the change of the process input (C1). We ‘assume that the term sets of input/output variables have the same cardinality, 3, with a common term negative, zero, positive). The prototype of fuzzy control rules is tabulated in Table IV and a jusifieation of fuzzy control rules is added in Table V. The corresponding rule of region { can be formulated as rule R, and has the effect fof shortening the rise time. Rule R,, for region ii de- freases the overshoot of the system's response, More specifically, R, if (E is positive and DE is negative) then CI is postive, Ry: iC CE is negative and DE is negative) then CY is negativeTABLE 1v Paororyreor Fozzy Covert Rots win Tra Sor ‘Nrcarir Zeno, Por) Se eee tetepy Rove Jusnnicamio wre Tena Sere (Nrcarivn Zeno, Pose) Cr Reference Range ‘ PN Pitre med 5 NN tenemos see eee en 5 oR OP no 0 peewee T " NP a Paororvre of Fuzzy Cowan Rutas win Tea Ses (is NS. EPSP Bue EE )__-Reteence Fant Ze Pe * ceo : as i Rw Nir ' NS OM i OND é ie Mw : NS t ote 4 mB s os 7 ne setpoint ‘eho Ss NB. NMS, ZE, PS, PMP) @ PONS PM Arse ume) Is PS NBM towers) NDP ‘i ONS Po i i® PMs Ze bo OS s Better control performance can be obtained by using finer fuzzy partitioned subspaces, for example, with the term set (NB, NM, NS, ZE,PS, PM, PB). The prototype and the justification of fuzzy control fules are also given in Table Vi and Table VII. ‘A slightly modified method was suggested in (7), It tracked the linguistic trajectory ofa closed loop system in a “linguistic phase plane.” The main idea is that scale ‘mappings should be adjusted frst to yield approximately a Xesited trajectory behavior. This can be inferred from the linguistic trajectories. Then rule modification can be ac- complished by using the Linguistic trajectory behavior to ‘optimize the system response in the linguistic phase plane. ra Fg. 6 Rate jutiseation by using ingutic phase plane (0) Ling i etary wth nia ates 6) Lig rakes. (Hom Bra se Rutherford Pn me AAn additional advantage of this approach is that the measurement noise appearing in the linguist phase plane isles ofa problem than that in the noalingustc phase plane. An example is shown in Fig. 6 ‘An approach to generating the rule base of an FFLC, which is analogous to the conventional controller design by pole placement, is described in [6], Braae and Rutherford assumed that the fuzzy control rules of an ‘open system (process) and a desited closed-loop sistem ‘were intially given, The purpose isto synthesize a linguis tie control element (FLC) based on the fuzzy models , ys Rycit ris Ayes and y is B, then 2 = f(xs09) izzy control rules of this type, which are referred 10 as “state evaluation fuzzy control rules,” evaluate the process state (estate, state error, state integral at time 1 and compute a fuzzy control action at time tas a function of (++, y) and the control rules in the ruleset 2) Object Evaluation Fuzzy Control Rules: Yasunodu, Miyamoto, and Ihara [135] proposed another algorithm Which predicls present and future control actions and ‘evaluates control objectives. I is called “object evaluation fuzzy control,” oF “predictive fuzzy control.” The rules in question, which are derived from skilled operator's expe- rience, are referred to as “object evaluation fuzzy control rules." A typical rule is described as Rif (us C,-+(xis A, and y is B)) then wis C; A control command is inferred from an objective evalu. ation of a fuzzy control result that satisfies the desired states and objectives. A control command w takes a crisp Set as a value, and x,y are performance indices for the evaluation of the ith rule, taking values such as “good” or “had.” The most likely control rule is selected through predicting the results (x, y) corresponding to every con: trol command C, In linguistic terms, the rule is interpreted as: “if the performance index x is 4, and index y is B, when a control command w is chasen to be C,, then this rule is selected and the control command C, is taken to be the ‘output of the controller.” In automatic train operation, a typical control rule is if the control notch isnot changed and if the train stops in the predetermined allowance zone, then the conarol notch is not changed. tis well known that systems control encounters dif cults in satisfying multiple performance indices simulta neously and in achieving accurate control in the presence of disturbances. In such circumstances, zy control pro- vides an effective framework for solution. However, the state evaluation fuzzy control does not evaluate the com- puted control actions as human operators do. By contrast, the predictive fuzzy control provides a mechanism for evaluation so that the desired states and control objec- tives can be achieved more easily. I should be noted that predictive control has been successfully applied to auto- ‘matic train operation {135}, (136), [139] as well as 10 ‘automatic container erane operation systems [1371-1139]. ‘Tests have shown that this type of control is capable of‘operating trains and cranes as skillfully as an experienced ‘operator. E, Properties of Consistency, Interactivity, and Completeness 1) Completeness: Please refer to Section V of this pa: per. 2) Number of Fuzzy Control Rules: There is no general procedure for deciding on the optimal number of fuzzy ‘control rules since a number of factors are involved in the decision, performance of the controller, efficiency of ‘computation, human operator behavior, and the choice of linguistic variables. 3) Consistency of Fuzzy Control Rules: If the derivation ‘of fuzzy control rules is based on the human operator experience, the rules may be subjected to different per- formance criteria, In practice, iti important to check on the consistency of fuzzy control rules in order to minimize the possibility of contradiction. (64) (121 ‘4 Interactivity of Fuzzy Control Rules: Assuming that @ collection of fuzzy control rules has the form Ryif-x is A, then 2 is C), If an input xp is Aj, we would expect that the control action 2 is C,. In fact, the control action z may be @ subset or a superset of C, [12], [26 [85], [18], [19) depending on the definition of fuzzy implication and sup-star composition, This may happen as a consequence of interaction between the rules. ‘The problem of interaction is complex and not as yet well understood. The reported research in [12], (26, (85) [18], (19) indicates that interactivity of rules can be con” trolled by the choice of fuzzy implication and sup-star ‘composition. The consistency of rules may be improved through the use of the concept of a fuzzy clustering of fuzzy control rules. In this connection, it should be noted that Sugeno’s reasoning and identification algorithm pro- vides an alternative solution to these problems {104}, fio1) AcxsowLeDaMenT 1 am greatly indebted to Professor Lotfi A. Zadch of the University of California, Berkeley, for his encourage- ment of this research, The assistance of Professor Zadeh is gratefully acknowledged. The author would like to thank Professor M, Tomizuka of the University of Califor- nia, Berkeley, and the reviewers for their helpful com- ments and suggestions Rererences UU). Sabin and, H. Madani, “Precritive methods for ering cond ply in fase contol,” cron Lt sel hp 68-620 197 tay J'balann and N. Gull, "Modeling comtolersasing tay teats" Aves vl pp 233-20, (0) JF" Bhgner tnd BW. Pwo, “Avomatic approach to implication Yor spprocinaie reaming wh Ry logs Fuse Soe Serva pa la} 1°" Bernard "Uae of rule based tem for proces contol TRE Gory Maps Bo. 5p SHS ro or 6 1 uo) uu ay us) a) us) is) uw us) i) bo kn ea) by ba es os 7 es) 9) 0) on) i ba M. Brose and D. A. Rutherford, “Faz relations in control hing Kieren sr np 68-188 197 SIN Theoretical and igus aspects of the fay lope Sater ata, vl Sno Se pp 35-57, 179 satan of parameter for faz lope conga.” Fay Sea Spe a2 np 8-198, 197, KGa, Use, and ink, “Pralaspests ofthe Komledge ised contol fs mobile abet motion e.Prae Sih Midcest Sop on Cet and Sens Space, NY, 1H E'Conpln, Probe Ser in Decor Mating and. Cel. ‘Cologne: Verag TUV Rheinland, 8 ‘Probaisic fzzy controler asa generalization ofthe Beep of tony contact Fas Sor Spt wo 28, Dp 215- EE Gaogla and K. Hirota, Polite So: Fas and Seca Approach 10" Decision,” Conta and Reception Prost (Googe: Verlag TOV Rheinand, 198, onic f oem of deckion taking in fanzy tens hel Mam Mach Stade, ol 3, pp. 201-22, 18 "On Mention in fuzzy yma is pian in Sail problems" Fy Ses St 6m pp. 73-6, 181 ‘Control problems in fury ystems” Fst) Ss Ss, vl Trae h pp 8 "FBaay ele generation fr tury contro” Cem. St, GOES oo. pp 20, 1a, E-Caopl ah Raw, "Modeling of zy conte with pplation tthe convo of Nolopesl processes" Pury Ses Sh vol 3, pp. 13-22, 108, B'Bharah Bee! and'V. V. S. arma, “Estimation of fay ‘memberhipe fom histograms” inform Se, vo 38 pp. 48-39, is D. Dubois and H. Pade, “Fury topes and the generalized Iosus pores ested,” Cer. Sa. 18, 9p. 34 1984 "The geeralized ds pone unde sp-min comps ion" theoreti! age” im_apyrumate Resonng Sls MOM, Gop anal Bode Sod Kin ‘Uni eis and necely meanies: Toward» pos ‘Site iterpettion af Mtge” acy Ses St 1090. pp 15-20, 8S, 7. Posy Ther: An Apron 0 Comper Processing Uncen. New Yor: Plemam Pres, 8h 5 Etathiu,“Rulebsed proces conl using fz log" ‘ppraimate’ Reson oI Sem, Deion ant Con TE Sancher and L.A. Zadeh Eds New York Pergamon, {ost pp. 148-158 F Fuge, “FLEX-880 series elewtor group contol stem, Fj Cn, 1, Ona, pan, 18k 'S"Futams, ML Nirumot, and Tanaka, "Some considerations tan conditional inference,” Fasy Se Sty 00 pp. 2-273 LR. Gaines and LJ, Katou, “The ary decade: A bibliog iy of fuzzy stems and last felted topics” it. 7. Man Mak Sais v 9p 181977 S-Goutald and Wi Perc, "Problems of the design of tun) eottolers” i AppesimateResoning i Exp Swems, MM Gopi A Kandel W- Bander, and 3B. Kok Ed” Amster ‘ham: No-Hollan 8p. 388-405, 1p. Graham and RB. Newell "Fuzzy identification and sont of agi eet i Fy So Sah 3p. 385-2, azz adaptive cont of a fistoder proses.” Fuzy Sar Sr. vol 3s pp. aT-#8 1589 Graham and Eons, Ee Sstene Kowier, Uncen ‘ant Becton New York: Charan and Val 1986 MM Gupta and". Tsukamoto, "Fuzzy igi controller — 1X perigee” tm Prot doi una Cntr. Conf San France Avg. 1980. pp. FAID.C MM Gupta 0. M. Tian and JB. Kick, “Control of fen contd stem” TEE Trans Sy" ManCitem Ya SMC-I6, 904, pp. S82 1986, KC Mita. Ari abd'S.Hachisy, “Moving mark recognition tu moan objet manpulstion a fz ont ste Con Thy ab Techn wo 9.3 pp. Weal 18 KHkols and W. Pedncs, "Andis and sates of fizzy systems ty the we of fury sets” Fu) Sas Sv. 10, no ppinis, 83ba. bs be wn bs we) to) on (21 fe) tl ks es (a ves) ko 0) is (2 is) 9 iss) ts) i i381 9) \ lo we , tsk deniation and uz ues cont of mie foot motion. in Pos IEEE Ie Syonaon Ie Co. Paden PAL 196 (Otto Goa T- Nakayama, and 8. Takamizaa, “Appi ‘ion of zz contol tata sludge presen Pr Da [SA Congress, olyo. apa, Jay. 198). pp. 282-28, ‘A tones, &Kauimnans aod H. J. Zinmemann, Fey Se Thea and Appeatns. Boson: D. Reidel Polishing Com pany, 186, IT Racpem and. A. Ono, Optimization Made: wing Puc Sra Pty Penson DR Pasig Co FP Rbcprak and M, Fei, Combining Fussy Imreion with Probataliste“Cnceray im Decision Making.” New York SpringeVerle 988. 1A Kandel, Fay Techniqes in Pater Recogiton ‘ile ntencene, 1K Rul and Y. Morioto, “Eletonielycontoled coating ‘ish variable raemeson in Pac It, Conpet om rap tow Eero, Dear, MI 1988, ‘3 Kaui Pectin Yo Tho of Finy Subs ork Academic 197. [A Koulman and MM. Gap, Induction to Fay Andee. ‘New Work Was Nostrand. 1086 a rhea Mahemarcal Modes in Engen and Monae ‘rat Sconce “Amscrdam: North-Holland 88 IHC Ketemen, Fzanes and stacarl sbi.” Ayers a 10, pp sta W. JPM Rikert Towards an anaysof gute modeling Pas See Sts vl 2 pp 35-07 9. re Parhcr anal and ppl of faz lope contol.” imterial Rep. F/WK/78, Qseen Mary College, Lindon. 195 (WLM Rskert and HR’ Van Natta Lemke Appin of 3 {hazy controler in 3 warm water plan” Automat 12 m0. pp 30-3089 BEAUM. Riden and E, HL Madani Anas of ny li nile,” Fuzy Sty Suton. pp. 23-48, 1978 King and EH. Maman ~The appcson of zy sotrot ‘ystems to indus processes” In EAC Word Cones. MIT. Boson, 1973, Bese application of fzey control gyms to indestit Proeses” tomar vo 13 no pp. 289-20, 1977 AeSKinota T Fukuzat, Satoh aed ML Make, “An ‘tomate operation msthod or conta ads BWR pli a ‘Pres Speci” Metig nt D-Cor Insomeraon and Resor (Coe Atenert,Cabrache, Fras, 8 SB Kiata, ME, Kocbann and’ D. S. Swink, “The invence of same parameters on the acuracy fy mode" Indsina Apeaton of Fay Conol M.Sapen, Ed. Ambier. hi Nov ond, ssp. 87-250, “The nance of ome fz implication operators onthe uray of fay ml art Psy Soe Shr a0 1. pp Tie ois! ee Th influence of some fry implication operator onthe Tick MM, Gupta, a P. 8. Nikoruk, “Energetic Stbiity ot fazzydymamie stems” IEEE. Trns. St Man {hiv sok SMC m8, pp 9792 19S Goi klinand TA. Foiger Fas Son Unc, and Informa fon. Englewood Clit i: rence Hal 1968, SV. Koln, 8. P Makes, and IF Shaknow, “Opti! tol e's Hint seman wth fz constr id aes {lige hems 1 mo 6p. 88-80, 199. LE Larkin," fy ge cooler fr art fight conto” i nds Apcations of Puy Con, ML Se, Ed A= Aendam Non aid: 18S pp 7 1G PIM Larsen “Industral appli of fzz lei com 4 Mon ch, Stade, Bw ep. Th E, Lembessh "Dynami Lering tchator of raetased feleonpning contri,” Phi, tess, Queen Mary Cols, Une of Landon 8t- YE Ui lind CC Las, “Development of fz agoritns for sero ystems” FEEE Cm. Ss Mag 9. Bo 3: 43-72, ioe 1, Maies and ¥. . Shen, “Appleton of fuzzy st then, IEEE Tran” St Mon Cyber SMC 15, 90.1 pp 75-18, (os) (os ts ts) tw (os) (1 po. ru m bs, ba, vs 081 vm os) om wo oy ws 3 tsa) bss ro) wr (os) wo (so) EH. Maran, “Appiations of fuze alrite for simple ‘hour pan Pow TEE, wl 12m 12 11S EH Minin and S-Asan. “AN experiment inHg Syne with fy hgh come lt! Mon Mach Sal Irv To tpt 1s 1s TM Mima Agence in the ingatie ste of faz Cotes J. bon Mack Sta so ea fp 09-47% ie cava of fz log 1 approxima reasoning aig hngo test” ERE Pam Col 38 m8 pp ssn te plan fey st they t cote stems 2 Saveyin'Basy ants and ct Pres, MM Gap GPR side bod" R. Gains, Ede Amend Non Saad 97, pp. 77-8 POH" Mamdinl and BR. Grinen Fir Resmi ond Apptctons don: Academe. 1 ‘Stacia na Sova,“ Homan opsratrs zy made it muremacine qua mht nonin cota eet Pst Ine Conon Fas normaton Provan. 188 MM Mitta "Note ot ei Ta by Za or uy Stent nirece," Cite Sy mh 3. py. 27~ 198 MMi nd Hamman, “Compan 9 ey aime meth gy Sand Some 8 pp SS ime Mt Mwnot."Compason o vars zr eaoning meta 2 Pow ira Conger Tigo Dpan Je OW pp. 2 Panny cones wider mrs apposmat, soning ‘Ba Be bd 184 Cone, Te, apa J 1 poe Patani, “Apptiation of foxy convaler 10 somobie Speed cord gntim Dc ofthe TEAC Sp on Fy IBmmon Kroes erent nd Deon “nays. MancieFnce D8 pa 3 Mfuratamt and M, Mar, "Apples ff cotter vo Butte speed cotrl em i tnd Apna of ag Cou M. Sugeno Ed” Amserdam’ Non Holand ma and Terao, “Opin como of diesel aint nl pps of Fay En MSs EE Rtwerdom Noneffand 198. pp. eh“ ENV Negot "On testy of zy tem in Pre TE tia Conon Ceres and Sooty. 1 pp anor ‘eo Sononr und Pacem Mis Pat: Bex ja /Catming,Potishing Cope 0 IR Na mal AG. St Vanier The Maserati of Fre Shum Gesamiescing Verlag TUV Rhian. 1988 Png Pury lage conto of ea uhange proces. in Pua dunwnan and Decon Press: MM Gap GN Sui and BR. Cains Bak,” Amstidamt Nonlin 19Ph pp 285-0. 37, Bi and BK. Data Majer, Fie Maratea rows Patem Recgtn. Nw Delhi Whey Ear Li C,Papis and. H. Man.“ top soto for aloo IEEE Tun Sra Cb 0) SMC no. {tgp ori 197 19, Flanex, "A approach to the anf zy ster” cot Sp ah Siar vl pp 3-16. 184 Pe Taian of fry flat equations or meta oF ‘Sng a presents fy data” Fs Se 0 pp ieserst i HiFi and CV. Newt Fg: Lage in Knee Ei $F, Rook and En Mari hgh cevguiing proc cota” Aaa 1c pp 18-30, 1. ESP Ras and B.A" Ruteron. “Apposnae eration ‘lapping non fom Hpustie Sevptan in problems ot fay a apps to em Som” Creme, 120.3 R'S"Ray ana D.D. Majumder. ~Appication of orl criteria for Sthiy Seabee flincar SISO a8 MIMO 9steme swore Sh fry ie cote: EE Try. tor Ce SMC! pp, st Din. Raterots and ©. C. Bloc. “The implementation of fey agri fo contain Ps. TERE, Wm pp Fass Seon 2 a) wa wos) bo} on 8) 3) ico) tuo) tue uns) 0 {105 tio tion tos) uo) i) uu any ts a tus) bin) BoA. Rutherford and G. A, Cater, “A Heute adupine Ceairer for sinter plants in Poe. nd TEAC. Simp on Sinem Stong hve! and Ata) Procsong, Homes tig RSA, Sep 9% AY Satay A facy controler a trnng process automation” ia Instr Appleone Cora MSapena Ed AWste. dhe Nov land. se pp. 159-152 E'Soncner and LA Zadeh, Apron Reasoning in eigen ‘Syems, Devon and Conrat New Yorks Pergamon, 1988 BPM Sehort and NZ Mande "The sppcaton of fay fontrle to the control oa mutidegeeeedom robot am in dust Apcatons of Fay Col M. Sugeno, Ed. AMS {enum Novi lund 18S pp 41-0. Si, "Fuzzy selon controler and its aplication fr Shmamic proces.” Frey Se Ss wo, 26 pp IST To4 18K. WMS" Stathowier and ME Kochara Fay modelag of the Proce" im Pc 2nd IFSC. Toyo, Japan July 198, tran RE Staeno and K Murakami, “Fuzzy parking conto of model acini IEEE Conon Deion and Con. Las Vee, rr IM. Sugeno ond M, Nishida, “Fuze contol of model ear” Fy Sas Sev 16, pp 108-113 1988, AM Sogeno and KY Mratar “Ae experimental study on fuzzy Ding contol using a ods ct” in nda Aplewion of Fag Comal. MiSigene, ES.” Amstrdams” Nowell, 1988, pp. 125-138, LM Sopeno, An introductory sarey of zy conta” ifr Se Yh 36, pp 9-83. 98S WM Sogene sti GT. Kane “Strate idenifiation of fy Imodel Fay St Sp, al 2. I pp S33, Sugiyama “Anabss and symbesi of the rle used, el Cgansing cooler" PHD thes. Queen Mary College, Uni st London, 18h {7 Tata! ane M. Sugeno, "Dereon of fzey cond rales from human operator’ contrl con” in Pre. of the EAC Simp. on Fas Ifameaton ‘peenation and De ‘hoa, Marsels,Fraee July 1985, pp 98-8 Patz Wentiication of tems and its oplcations to iadcing and conta” EEE. Trans Sy Ma Chere SSMC15 nT pp. 16132, 1988 RUT Tap and RJ Matloland, “Comparing tuzy logic with Sasa eotoler desig” IEEE Trans Ss Man Cte. 0 SMCIT, no. 6 pp ORS" 108, 1987. FR Tansihec and EM. Schart,"Experinens withthe use of 2 Ihle-bedsiLorgannng into for robot spite. ‘us Sox St wo 2h pp 195-216, 1988 [M Top and Watante, "Expert ster on 3 chip A engine for reutime apprsimate reasoning” IEEE Expert Sut Mag, ol I pp 8-28. Mi Togat and 8. Chit, “A fuzz accelerator and a programming Eroonment for reastime Tuzy contro” Poe nd TFSA {Congres Tokyo, Japan uly 198, pp 87-13. Re'NE Tong “Amie of fay Comolli ung the elation matin” Ie) Mon Mach. Stas, sl 0.6.2 79885, 19%, “A consol eginetng review of fazey systems” GOTTA opp 889-809 1977, Sythe of fray mls for industrial process Gem yp wot kp S10 197 Or Ste elution of ary mode derived fam experimen: (BF ta” Puy Ses Shr 8 pp 112.100. RLM. Toog MLB Beek, and A taten "Fuzy cont ofthe ‘ied slags wastewater tentment proces fata, a 16, no 6 pp. 695-701, 198. RM. Tong "A retrrpective ew of fuzzy control stems,” iad Say Siw pp 19-10, RE "An annotated blography of fuzz conto" Ins ‘ral Apples of Fussy Conn. M. Sigeno. Eu Auta ‘ovieoland 8, pp. 29-208, Tolls an. Vaherde, “On mode ad implication in appro fate reasoning” in. Apposinate Renn tEspe ser MN Gupta A. Randel We Banden and JB. Keka, Es ‘Amuterdan: Non Holand, 18S, pp. 1872168. Ti yhokamora, "An approach to azy reasoning metho” in Zavaces in Bay Sey and dpc, MM Gop Oe aden and RR Yager Eide Amerdams Next Auton tr} ten) wn wal as) 29) ws) as tas usa) sn) usa ro) hiss) lisa) list uss) 9) (vo) bay bay iss Holland 19%, 1G Unbor and P. 5. King, “AM ais of humandecsion trating im cement kn contol andthe implications for automa. lon J. Man bch, Stads vole 2p m1 pp e231, Mt" Urgam M Mionnoo, and Taman, "Fuzzy robo feotels Cher a 6. pp. 38-08, 17. Ht Watanabe and W. Det. "Reconfigrable fuzzy loge pro {sane fl custom sil VLSI" inf. Warihop an Puy Stem Aicason, Haak, Japan, Aug. 198, pp. 49-80 1B. Willa “Optimal conta of fz ssc” i Po. I Ein ed So Rea and Cet, Aca Deca making with ossiblitie dynamite.” in ‘Froc20m TEEE Conf. om Dion and Conta, San Diep, Dew Willys and N. Miah, "Use of frzy model for proces otra In Prac TEBE Tn Conf. om Cssemencs and Soe, 1978, p, 942-86, een Contrion of the fuzy se theory 0 man-machine ‘aie fn Adc oF Set Phew and Aplains, MM. Gupia, Raga, and RR ager, es. ‘or Holand, 199, pp. 800, COW. Xu and ¥. Za, "Fursy rode! eniication and sl Teaming for dame tems" TEE Tras St Man Cite So SMC no. pp 889. AT RR Yager, §Ovehinions Ru M. Tong, and. Nase, icy, et ad Appleton Slted Papert by L, Bah New York: Wileynterscence, 1997. 8"Vapanin, ©. Hoh and Nt Sopeo, “Application of fey ‘Rasoning to the mater puretion process "tend Ap (tons of Pas Conrl, M. Sugeno, Ed” Amsler: Now Hota 1985, pp. 19 ‘Te'Yamatawa tnd TMK, “The carent mode. fey lope imuegrted uit abiatd bythe standard CMOS proces IEEE Trans. Comper, vol C35, 0.2 pp. 11-162, 188 {T Yasataos and K. Susi. "Ferry memory devin Poe 2nd TESA Congress, Tobyo, Japan, July 198. pp 881-555. ‘Tamaki speed fry contr bear sem.” Pres Dud Pay Sem Symp Tapa. 1986, pp 132-130 Fuse controler hardware stm 10 Proc. 2nd IFSA Cag Toye, pan ay 8 "SASnpie fue) computer hardware stem employing in and max operations chllenge to gonersion om puter in Pro Dd TSA Congres, Toke, Span. ly 18 JP Samana, "Fanry meroprceuors Rte chip abd fiz fer chp lr Weksop on Fasy Sem Aponons. Haak, [pam Rug 98 pp ste? ‘Tr Yamazahtand'M. Sugeno, “A. mictoprocestor based frzy Contr fvindestial purposes” Padua Apceons of Fray Con, Me Soper ES Anterdame ert Hold 19S, p. 281-20 SiYatinbs, S Miyamoto, and HL. thars, “Fuzzy coata for Sawmatietsin operation stem” it Pc. fh IEAC/ IFIP TEORS ti Compo Conia on Transporation Stes, Ben Baden, Api 1 S°Yisunabu and S. Mismoo, “Automatic tin operation ty Dretiethe fan sono in Inde! Appian of Fay Com foi, M. Sogeno, Ed." Amsterdam Noh Holland, 198, pp ie SS Yasonobe and . Hasegawa, “Etalution of an automatic Sumer cane peraten sytem based on prediive fz coe {tol Conol Thay ah. Teal Wh. 2.00.3. pp. 19-82 i, spree fuzzy conto and its pplication or automatic tance crane operation tem Pr. Dad TFS Cones ‘Tlyor pan al 1987. pp 3932 S Yosandba,§Sehiny ad. Hasegaa, “Automatic tain Speration and automatic cane opetaton sims based on pe ‘Ste fez contain Pres 2nd IFSA Crs, TOR. pa Sai 19 pp. nS IM Yonebi, The application af fz Ses theory 1 he te perature of hx anealng furnace sing simulation techniques pre th TEAC Word Congress. Tye 38. LA” Zadeh “Pony ste Ifoomat Conta 8p 38-383, ir BES Runyan.” Ifo, Coa 2, pp. 94-12 “Toward hooy of ary systems" In Ages of Nektu ss) us) usr ues) as) us0) Ler 0 8 STAG, Ma A RENN, SOK 2, 1/1 and Sytem Theo, R. & Kalman and N. DeClars, Ed. New ‘You Hot, Rincaet snd Wieson. 97, pp 8-430, IOUS relation and zy orderings.” Iyorma, Si SoU pp 177-200 1971 NEN ational for fury consol” Tans. ASME, J. Dam. Sa Men. Con v0.98, pp 3-4 17D see SOutine of 8 ew approach 10 the ana complex ‘ens and decom rose” IEEE Tran St Man Coes SOU SMC, pp 28-4, 1973. ee NSTHe concept linguistic yrable and ite appicaton to ‘Sppreinate semoning UH I" formar Soh ma BP TBD pp SUL-BST ol 9 pp 49-80, 1S HJ, Zimmermana, Fuzy Set Theory aris Appleains Boston: lower Academie Publishes 9, HLL Zinmermann, Fas Se, Decision Mating, and Exper ‘Sema. Howor: Kiower Academie Pubes, 1987. CC Lee, "Fany loge n control systems: Fury ope consol pantie IEEE Pont, Sy Man Cybem. ve 20. 0. 2 pp 19-435, Mar/Apr 1950 ‘Choon Chien Lee (SW received the BS. de free in cool engineering fom the National Nerang Univers. Tawa, 189 and the rs MS egite in eectocal engineering trom the = Unosty of Caoria, Bercy, 1987 He 7 furemly'»eandiate fr the PRD depres i = the Departnent of Eetreat Engineering the (Comer Senos a the Univer af Calor ‘nar Bere ‘Sine 19M he hasbeen 2 Research Asstant in the Elem Reseach Laboratory a the Universi of California erie. He spat the summer of 198 tthe NASA Anes Research Comer working on fiz) coral sens project Hie current researc interest ein the ares of earning Bescon cot, oy contol gains cm abo ‘eurl nctwork appeonmatetetoning, 09 ria lige ‘Mi c's a menr of PhiTau Ph He abo stugeot member of tne American Asseation for ArGbeatInligenee, the IEEE. Com pier Soc the IEEE Control Sooty, and the TEEE Systems, Man. nd Cybernetics Soe