A P P L I C A T I O N O F F U Z Z Y LOGIC TO A P P R O X I M A T E R E A S O N I N G U S I N G LINGUISTIC SYNTHESIS E.H.

Mamdani D e p a r t m e n t o f E l e c t r i c a l and Electronic E n g i n e e r i n g Q u e e n Mary College (University of London) Mile End R o a d LONDON E1 4NS

Summary This p a p e r describes an a p p l i c a t i o n of fuzzy logic in d e s i g n i n g c o n t r o l l e r s for industrial plants. A Fuzzy Logic is used to synthesise linguistic control p r o t o c o l of a s k i l l e d operator. The m e t h o d has b e e n a p p l i e d to p i l o t scale plants as well as in a p r a c t i c a l industrial situation. The merits of this m e t h o d in its usefulness to control e n g i n e e r i n g are discussed. This work also illustrates the p o t e n t i a l for u s i n g fuzzy logic in m o d e l l i n g and d e c i s i o n making. A n avenue for further w o r k in this area is d e s c r i b e d w h e r e the needis to go b e y o n d a p u r e l y d e s c r i p t i v e a p p r o a c h and explore means by w h i c h a p r e s c r i p t i v e system may be implemented. Introduction The fact that m a t h e m a t i c s as a w h o l e is taken to be synonymous w i t h p r e c i s i o n has c a u s e d m a n y scientists and p h i l o s o p h e r s to show c o n s i d e r a b l e concern a b o u t its lack of a p p l i c a t i o n to real w o r l d problems. This concern arises because in logic as well as in science there is c o n s t a n t l y a gap b e t w e e n theory and the i n t e r p r e t a t i o n of results from the i n e x a c t real world. Many e m i n e n t thinkers have c o n t r i b u t e d to the d i s c u s s i o n on vagueness, o c c a s i o n a l l y h o l d i n g h u m a n subjectivity as the culprit. In an e x c e l l e n t analysis of the s u b j e c t B l a c k 1 says ... "that w i t h the p r o v i s i o n o f an adequate s y m b o l i s m the n e e d is r e m o v e d for r e g a r d i n g v a g u e n e s s as a d e f e c t o f language". In his p a p e r he s t r o n g l y argues that vagueness s h o u l d not be e q u a t e d w i t h subjectivity. Briefly, his a r g u m e n t may be s u m m a r i s e d by n o t i n g that the colour 'Blue', say, is vague b u t n o t subjective since its sensation a m o n g all h u m a n beings is roughly similar. It is p o s s i b l e to deal with colour p r e c i s e l y b y c o n s i d e r i n g the e.m. r a d i a t i o n p r o d u c i n g it b u t in d o i n g so the i m p o r t a n t h u m a n sensation of colour, as it happens to be vague, has to be sacrificed. Furthermore, it may be a r g u e d that vagueness is not a d e f e c t of language, but also an i m p o r t a n t source of creativity. A n a l o g i e s are extremely important to creative t h i n k i n g and vagueness plays a d o m i n a n t role in such thought processes. Black's m o t i v a t i o n to symbolise vagueness appears to be at the b a c k o f all i n v e s t i g a t i o n s o f " D e v i a n t Logics ''2. A n i m p o r t a n t c o n t r i b u t i o n in the p a s t iO years has b e e n that of Zadeh's fuzzy-settheory and fuzzy-logic 3 . In his recent w r i t i n g s Zadeh4, 5 states clearly his m o t i v a t i o n w h i c h is to use fuzzy sets to symbolise A p p r o x i m a t e R e a s o n i n g (AR). W h e r e a s there are many applications of fuzzy-settheory, this p a p e r describes one of the first results in the a p p l i c a t i o n of A R and linguistic synthesis. An Outline o f the Paper's C o n t e n t The intention in this p a p e r is to review the w h o l e p r o g r a m of i n v e s t i g a t i o n c o n c e r n i n g the a p p l i c a t i o n of A p p r o x i m a t e R e a s o n i n g and to analyse the findings in o r d e r to offer insightful comments and conclusions. The original w o r k in this p r o g r a m was done in e a r l y 19746 a n d first p u b l i s h e d later that y e a r 7,8. This was the control o f a p i l o t scale steam-engine using 196

fuzzy-logic to i n t e r p r e t linguistic rules w h i c h q u a l i t a t i v e l y express the control strategy. This w o r k is b r i e f l y r e v i e w e d in the n e x t section of this paper. Since the p u b l i c a t i o n of the above w o r k several researchers elsewhere have also i m p l e m e n t e d the approach using d i f f e r e n t p i l o t scale plants. This together w i t h the c o n t i n u i n g work as p a r t of this p r o g r a m m e have p r o d u c e d results w h i c h throw more light on the u s e f u l n e s s of a p p l y i n g fuzzy-logic to l i n g u i s t i c synthesis. Below, comments are o f f e r e d on some Of the key findings o f these studies. One o f the comments that has b e e n made a b o u t fuzzy-logic is that in its p r e s e n t form it is essentially d e s c r i p t i v e a n d does not o f f e r a p r e s c r i p t i v e a p p r o a c h to reasoning. In the first place, it s h o u l d be n o t e d that fuzzy-logic, like any o t h e r form of l o g i ~ can only be a system for i n f e r r i n g c o n s e q u e n c e s from p r e v i o u s l y s t a t e d p r e m i s e s and only from these premises. A p r e s c r i p t i v e s y s t e m is possible, however, if a h i e r a r c h i c a l d e c i s i o n m a k i n g a p p r o a c h is u s e d so that the s t r a t e g y at a lower level is d e r i v e d as a c o n s e q u e n c e of a d e s c r i p t i o n at a h i g h e r level. Early results o f such an e x p e r i m e n t of a p r e s c r i p t i v e m e t h o d (some m i g h t term this a learning or an a d a p t i v e approach) is d i s c u s s e d later in this paper. To conclude this p a p e r the last section examines the future trend in this field in the light of e x p e r i e n c e b e i n g g a i n e d from c u r r e n t i n v e s t i g a t i o n s d e s c r i b e d here. An E x p e r i m e n t in L i n g u i s t i c Synthesis A B r i e f R e v i e w of F u z z y Logic The p o i n t of v i e w a d o p t e d h e r e is that the v a r i a b l e s are a s s o c i a t e d w i t h u n i v e r s e s of discourse w h i c h are n o n - f u z z y sets. These variables take on specific l i n g u i s t i c values. These linguistic values are e x p r e s s e d as fuzzy subsets o f the universes. G i v e n a subset A of X ( A C X ) A can be r e p r e s e n t e d b y a c h a r a c t e r i s t i c function: XA: X ~ O , i } . If the above m a p p i n g is from X to a c l o s e d interval [O,i] then we have a fuzzy subset. Thus if A w e r e a fuzzy s u b s e t of X it could be r e p r e s e n t e d b y a m e m b e r s h i p function: ~A: X[O, i] Note that X is a n o n - f u z z y s u p p o r t set o f a universe of discourse, say h e i g h t of people. A can then be e q u a t e d to a linguistic value such as tall people. Given two such linguistic values A 1 and A 2 on the same s u p p o r t set X, logical combinations: Ai; AiAA2; ALVA2; can be f o r m e d as: A2 is f o r m e d b y taking (i-~A 2) as its m e m b e r s h i p v a l u e

at each e l e m e n t of the s u p p o r t set. A i A A 2 is formed by taking min (~A I, ~A2) at each

element of the s u p p o r t set, a n d A l V A 2 is f o r m e d b y taking m a x element of the support set. It is in the d e f i n i t i o n o f i m p l i c a t i o n that this (~Ai,~A2) at each

logic m a y be found to differ from o t h e r logics. Given AB (If A then B), then it can happen that A and B are linguistic values o f two disparate universes of support, say X and Y. Note here that the i m p l i c a t i o n is b e t w e e n individual values and not the u n d e r l y i n g variables. Thus the relation R b e t w e e n A and B is a fuzzy subset of the universe of support X x Y, the c r o s s - p r o d u c t of X and Y. ~R: X M Y+[O,1]. ~R(x,y) is r e l a t e d to ~A(X) and ~B(y) (in the p r e s e n t application) by the following: ~R(x,y) = min(~A(X) , ~B(y)). xeX, yEg

(b) The decision m a k i n g phase is i n v o k e d at each s a m p l i n g i n s t a n t during r u n - t i m e w i t h the e x a c t m e a s u r e d values A 1 and B 1 s u p p l i e d to it. This p h a s e then is n o t h i n g b u t the use o f a c o m p o s i t i o n a l rule o f inference to derive C 1 as follows: C 1 = AIo(BIoR). Note that A I , B 1 can be non-fuzzy, w h e r e a s since C 1 is a fuzzy subset of the set of all p o s s i b l e actions a p r o c e d u r e is r e q u i r e d to determine the actual action to be taken from the knowledge of C I . A l s o there is a certain advantage in d e f e r r i n g the computation of R until the s e c o n d phase. B e c a u s e then this p r o v i d e s a means o f a l t e r i n g the control strategy on-line b y a l t e r i n g the data structures c o n t a i n i n g the rules during run-time. However, w h a t n e e d concern us at p r e s e n t is the results o b t a i n e d from the application of this m e t h o d to the p i l o t scale plant. In r e p e a t e d trials it was found that the results c o m p a r e d favourably w i t h those from a s t r a i g h t a p p l i c a t i o n of classical methods from control e n g i n e e r i n g p r a c t i c e (i.e. 2 or 3 t e r m controllers). C o m m e n t s on F u z z y - l o g i c C o n t r o l l e r Studies

If the relation R represents a "nested" implication (i.e. If A then (If B then C) or AB-~C), then R will have a c o r r e s p o n d i n g higher order c r o s s - p r o d u c t support set. Now if some relation R b e t w e e n A and B is k n o w n and so is some value A 1 then the idea is to infer B I from R and AI; B 1 = AIOR, w h e r e A 1 is c o m p o s e d w i t h R. This has the effect of r e d u c i n g the d i m e n s i o n a l i t y of the s u p p o r t set of R to that of B I . In this work, the compositional rule of inference used to relate ~B 1 to ~R and ~A 1 is: ~Bl(y) = m i x min(~Al(X), ~R(x,y)).

A p p l i c a t i o n to a F u z z y - c o n t r o l l e r As s t a t e d e a r l i e r the linguistic synthesis approach was initially a p p l i e d to control a p i l o t scale steam-engine, a more d e t a i l e d d e s c r i p t i o n of w h i c h is given elsewhere6,7, 8. A concise summary of this work is p r e s e n t e d here. One aspect of control in this system is the regulation of p r e s s u r e in the boiler around a p r e s c r i b e d set-point. The control is a c h i e v e d b y m e a s u r i n g the p r e s s u r e at r e g u l a r intervals and inferring from this the h e a t setting to be used during that interval. The essence of this w o r k is simply that if an e x p e r i e n c e d o p e r a t o r can p r o v i d e the p r o t o c o l for achieving such a control in q u a l i t a t i v e linguistic terms, then fuzzy logic as d e s c r i b e d above can be u s e d to i m p l e m e n t successfully this strategy. The p r o t o c o l o b t a i n e d from the o p e r a t o r in this case considers p r e s s u r e error (PE) and change in the p r e s s u r e error (CPE) to infer the a m o u n t of change in the h e a d (HC). The p r o t o c o l consists of a set of rules in terms o f specific linguistic values of these variables and is shown in figure i*. N o w it can be seen that these rules are in the form of If...Then statements (implications) and thus, from above, each rule i will translate into a relation R i. The overall p r o t o c o l is then a r e l a t i o n R formed b y 'oring' together the Ri's: R = R i V R 2 . . . V R i . . . V R n. L e t us say now that each rule R i represents an implication AiBi-~Ci . The decision m a k i n g a l g o r i t h m that is i m p l e m e n t e d contains two phases: (a) The initial setting up p h a s e w h e n the p r o t o c o l R is formed from two sets of data: (i) The individual linguistic values Ai, Bi, C i given as fuzzy subsets. (ii) The rules as in fig. 1 w h i c h specify the actual c o m b i n a t i o n o f these values to form each R i. *The a b b r e v i a t i o n s u s e d for these linguistic values here are: ZE-zero; P Z - p o s i t i v e zero; P S - p o s i t i v e small, P M - p o s i t i v e medium; P B - p o s i t i v e b i g a n d the same for n e g a t i v e values NZ, NS, N M and NB. Change in E r r o r n e g a t i v e is taken as m o v e m e n t towards s e t - p o i n t and p o s i t i v e as away from set-point.

Two m a i n conclusions have b e e n drawn from this work. First, that the results vindicate the approach a d v o c a t e d b y Zadeh and d e m o n s t r a t e its potential. Second, it can be a s s e r t e d that the m e t h o d can easily be a p p l i e d to m a n y p r a c t i c a l situations. This assertion is s u p p o r t e d b y c o n s i d e r i n g a p r a c t i c a l instance, that of cement kiln operation, in w h i c h a similar control p r o t o c o l applies. In a b o o k on cement kilns, Peray and W a d d e l l 9 list a c o l l e c t i o n of rules for c o n t r o l l i n g a kiln. E x a m p l e s of these rules are shown in figure 2. F r o m this it is i m m e d i a t e l y a p p a r e n t that the m e t h o d as described, can be u s e d for t r a n s l a t i n g these rules. In recent months, this m e t h o d has b e e n further t e s t e < b y various research w o r k e r s II-17 on other p i l o t scaie'plants such as b a t c h chemical reactors 12, heatexchangers 15 and so on. In some of these studies 19 the p l a n t is c h a r a c t e r i s e d by time lag b e t w e e n the a p p l i e d action a n d the o b s e r v e d o u t p u t n e c e s s i t a t i n g a m o d i f i c a t i o n o f the m e t h o d d e t a i l e d above. The c o n t r o l l e r is u s e d in c o n j u n c t i o n with a fuzzy p l a n t m o d e l w i t h o u t time lag to determine the actual action to be a p p l i e d to the p l a n t itself. In one study d i f f e r e n t control p o l i c i e s are c o m p a r e d analogous to the classical 3 term controller. Worth n o t i n g is the a p p l i c a t i o n o f this m e t h o d to a real p l a n t - a sinter p l a n t at the B r i t i s h Steel C o r p o r a t i o n 16,17. Results show that fuzzy control achieves useful reduction in the s t a n d a r d deviation of the m e a s u r e d o u t p u t when c o m p a r e d with a manual o p e r a t i o n of the plant. Furthermore it also compares favourably w i t h a classical 2term control of the plant. In m a n y of these studies, rules e x a c t l y as those given in fig. 1 are u s e d w i t h only m i n o r changes. This is n o t s u r p r i s i n g as the rules indicate the relationship b e t w e e n error, change in e r r o r and control action that exists in m o s t dynamical plants. This r e l a t i o n is m a i n l y one o f m o n o t o n i c i t y b e t w e e n the outputs of a p l a n t and the i n p u t a p p l i e d to it. W h a t is more of i n t e r e s t is that in m o s t studies it is found that this form of c o n t r o l l e r is far less sensitive to p a r a m e t e r changes w i t h i n the p l a n t than the classical 2-term controller. A t this stage only a q u a l i t a t i v e e x p l a n a t i o n can be o f f e r e d for this. It appears that the former is a reasonable c o n t r o l l e r as it relies on the u n d e r l y i n g relationships b e t w e e n the p l a n t outputs and inputs w h e r e a s the latter is a p e d a n t i c c o n t r o l l e r in w h i c h the action is c o m p u t e d as a linear c o m b i n a t i o n

197

of the m e a s u r e m e n t s and thus more susceptible to p a r a m e t e r changes. These results and o b s e r v a t i o n s have m o t i v a t e d a further analysis of the fuzzy c o n t r o l l e r 19. It is easy to see that because of the incremental nature of the action (heat change i n s t e a d of actual h e a t value) the fuzzy c o n t r o l l e r is analogous to a p r o p o r t i o n a l plus integral controller, (i.e. integral control is a c h i e v e d w i t h o u t the n e e d for m e m o r y in the controller w h i c h c o u l d n o t be easily a c c o m m o d a t e d in p u r e l y fuzzy logic context). It has also been n o t e d t h a t just as a classical c o n t r o l l e r can be t u n e d by a d j u s t i n g the gains of the terms, the tuning of the fuzzy c o n t r o l l e r can be a c h i e v e d b o t h by a d j u s t i n g the gains in a similar way and also by a d j u s t i n g the p r o t o c o l (see below). This has p r o m p t e d m a n y to s u g g e s t that in a fuzzy c o n t r o l l e r one has a 'nonlinear' c o n t r o l l e r . W h i l e such analDgies are useful~ in the view of this author they can also sometimes mislead. For example, the term 'non-linear' in a control sense implies that one views the model of the p l a n t as an i n t e g r o - d i f f e r e n t i a l system. In a truly system sense the w o r d 'non-linear' can have no meaning. Partly to emphasise this, the fuzzy c o n t r o l l e r has been a p p l i e d to a system which is not conmnonly m o d e l l e d u s i n g i n t e g r o - d i f f e r e n t i a l equations - a traffic junction 21. In this a p p l i c a t i o n the c o n t r o l l e r considers the w a i t i n g queue in one leg of the junction and the rate of flow in the o t h e r to determine the extension of the green phase of the traffic lights. R e s u l t s i n d i c a t e that the m e t h o d is o b v i o u s l y applicable in areas other than industrial p l a n t controllers. To p u t i t succinctly therefore, w i t h complex h u m a n i s t i c systems as w i t h d i f f i c u l t plants (difficult in the sense that they are d i f f i c u l t to model a c c u r a t e l y u s i n g integrodifferential equations), h e u r i s t i c decision making and control can be used very effectively. The p o i n t therefore is that fuzzy logic is a p o w e r f u l m e t h o d for i m p l e m e n t i n g heuristics. To i m p l e m e n t h e u r i s t i c s using this m e t h o d the likely source of difficulty, then, is that the q u a l i t y of decision is only as g o o d as the r e l a t i o n R from w h i c h it is inferred. R is in turn a f f e c t e d by three factors. First, it is a f f e c t e d by the set of rules in the protocol. W i t h more c o m p l e x situations a g o o d p r o t o c o l is n o t easy to,~erive. W h a t is r e f e r r e d to as h u m a n factors r e s e a r c h is d e v o t e d to exactly such m a t t e r s 18. Unlikely as it may seem the human b e i n g does n o t always find it easy to verbalise his c o n s i d e r a t i o n s d u r i n g d e c i s i o n making. As a h y p o t h e t i c a l example an o p e r a t o r may say he took an action b e c a u s e the system felt 'sticky'. Thus he m a y attribute an action to variables that c a n n o t be r e l a t e d to p h y s i c a l q u a n t i t i e s let alone m e a s u r a b l e ones (colour of the flame in some p r o c e s s for a n o t h e r example). The h u m a n o p e r a t o r also finds it difficult to control either too fast a s y s t e m (because he c a n n o t integrate his action) or too slow a system (because he c a n n o t determine rates of change) or too complex a system (because he has too m a n y v a r i a b l e s to control). Thus it is i m p o r t a n t that the future w o r k in this area m u s t c o n s i d e r the derivation o f the initial p r o t o c o l very carefully. The second factor a f f e c t i n g the q u a l i t y of d e c i s i o n (though not R itself explicitly) is the u n d e r l y i n g range of elements in the s u p p o r t set w h i c h p r o v i d e s the c o n t e x t for i n t e r p r e t i n g the linguistic rules. This can b e i l l u s t r a t e d by n o t i n g that 'tall people' in a land of pygmies is likely to have the support set of range of h e i g h t from 3 to 5 ft. 6in, whereas the more normal range o f h e i g h t may be from 4ft to 7ft. Such c o n s i d e r a t i o n s are i m p l i c i t in any 198

a p p l i c a t i o n and are e q u i v a l e n t to w h a t a control engineer w o u l d term the gains a p p l i e d to each variable. Finally, R is a f f e c t e d by the m e m b e r s h i p values in the fuzzy subsets d e f i n i n g the linguistic values. This is p e r h a p s the least important of all the factors because the degree of change p e r m i t t e d here is l i m i t e d as too m u c h change in the m e m b e r s h i p v a l u e s of a fuzzy subset is likely to affect the linguistic m e a n i n g a s c r i b e d to it. This is i l l u s t r a t e d in figure 3 in w h i c h the e f f e c t of a g i v e n linguistic value (bold line) is a l t e r e d by using a d i f f e r e n t linguistic value (as in a), i n c r e a s i n g the gain, thus d e c r e a s i n g the range of the s u p p o r t set (as in b), and lastly in a m i n o r way by a d j u s t i n g the d e f i n i n g values of the fuzzy subset. A Recipe for a P r e s c r i p t i v e A p p r o a c h As m e n t i o n e d earlier, the m a i n d i f f i c u l t y that is likely w i t h this m e t h o d is that a g o o d d e c i s i o n requires that a g o o d set of rules are d e s c r i b e d at the beginning. A n d y e t the goal in any a p p l i c a t i o n and a set of assumptions r e g a r d i n g that a p p l i c a t i o n are m u c h easier to state. In the control situation the goal is simply to regulate the p l a n t o u t p u t a r o u n d the s e t - p o i n t and the only a s s u m p t i o n is that the p l a n t inp u t and o u t p u t are m o n o t o n i c a l l y related. If the o u t p u t is h i g h then too m u c h input was a p p l i e d and vice versa. T h e r e f o r e the p r o p e r amount o f i n p u t r e q u i r e d can u s u a l l y be i n f e r r e d b a c k w a r d from the s t a t e d goal. A n early a t t e m p t at i m p l e m e n t i n g such a p r e s c r i p t i v e a p p r o a c h is d e s c r i b e d h e r e 20. The o v e r a l l schematic d i a g r a m for the control system is shown in figure 4*. The goal is e f f e c t i v e l y a b a n d w i t h i n w h i c h the o u t p u t is to be maintained. It is s p e c i f i e d as fuzzy rules w h i c h give the corrections n e e d e d to k e e p the o u t p u t w i t h i n the band. The input to these rules are time from s t a r t and the s e t - p o i n t deviation. The rules specify the change to b e made in the controller. E x a m p l e s of these rules are: (a) (b) IF Time is Small AND deviation is negative big T H E N desired change is big. IF TIME is big AND deviation is positive zero THEN desired change is zero.

The result of this set of rules is u s e d to alter the lower level control rules appropriately~ Since these control rules are of the form Ai+Bi-~C i the m o d i f i c a t i o n is a f f e c t e d by first finding the l i n g u i s t i n g values A i a n d B i that b e s t d e s c r i b e the p l a n t state for w h i c h a change in action is required. This search is simply done b y a s u p r e m u m o p e r a t i o n on the range of linguistic values. The action C i in the controller rules at Ai, B i value is then a l t e r e d b y the d e s i r e d amount. The results of a p p l y i n g this p o l i c y are shown in figure 5. The tables in fig. 5 are a m e t h o d of d i s p l a y i n g all the linguistic rules of the controller. The m e a s u r e m e n t s e r r o r and change in error are g i v e n on the axes and the table entries give the action to be applied. The a b b r e v i a t i o n s are as s t a t e d in the footnote on p a g e 2. The rules in figure 5(a) (those w i t h o u t asterisk) are the s t a t e d p r o t o c o l of an e x p e r i e n c e d operator. On a p p l y i n g the above p r o c e d u r e the extra rules m a r k e d w i t h an a s t e r i s k are c r e a t e d i m p r o v i n g the

*The system in this e x p e r i m e n t is a s i m u l a t i o n o f a b a t c h chemical reactor.

performance slightly. When the experiment is started with no rules at all then the rules of fig. 5(b) are created. The difference between the two strategies merely imply two feasible solutions within the prescribed band. Now it is possible to narrow this band but in the extreme case this results in a lack of convergence of rules, the quality of control however remains satisfactory. The convergence property itself is a matter of further investigation. Analysis suggests that the lack of convergence may be attributed to two factors. Firstly it can arise because the controller is required to make a decision based on not enough variables. Secondly, since the measured state of any system is the consequences of a whole history of past inputs it is not enough to alter the rule affecting only the single previous state. This is the familiar 'bootstrapping' notion found in learning control application. To conclude therefore, the advantage of the prescriptive method is that it reduces the difficulties with deriving a good protocol mentioned earlier providing the relevant variables can be identified. The protocol is then derived by a process of accumulation or integration of past experience. Present work is aimed at extension of this method to a multi-variable situation. Conclusions The prescriptive approach described above is very much an ad hoc implementation. It illustrates what needs to be done to advance beyond a simply descriptive system. What is desired is t~at such an approach should appear naturally within a suitably improved fuzzy logic theory itself. If, as is suggested here, hierarchical statements are a main requirement of such a theory then this means that fuzzy logic should have an auto-descriptive property found in multiple valued logics IO. From the application point of view both a learning situation described here as well as decision making in complex systems are best framed in terms of hierarchical structures. This is very much the direction in which the theory of Fuzzy logic and approximate reasoning is likely to go. The work described here demonstrates the great usefulness of applying AR using fuzzy logic to management and other humanistic systems. References i. M. Black, "Vagueness", pp. 427-455, (1937). s. Haack, DeViant Press, 1974.

121, pp. 1585-1588, Dec. 1974.

8.

E.H. Mamdani and S. Assilian, "An Experiment in Linguistic Synthesis with a Fuzzy Logic Controller", Int. J. Man-Machine Studies, 7, pp. 1-13, 1974. K.E. Peray and J.J. Waddell, The Rotary Cement Kiln, The Chemical Publishing Co., New York, 1972. M. Rescher, Many-valued New York, 1969.

9.

iO.

Logics, McGraw Hill,

ii.

W.J.M. Kickert and Van Nauta Lemke, "Application of Fuzzy Controller in a warm water plant", to appear in July 1976 issue of Automatica. P.J. King and E.H. Mamdani, "The application of fuzzy control systems to industrial processes".

12.

Proc. Workshop on discrete systems and fuzzy reasoning, Queen Mary College, London 1976.
13. R.M. Tong, "An assessment of a fuzzy control algorithm for a non-linear multi-variable system"

Proc. Workshop on discrete systems and fuzzy reasoning, Queen Mary College, London 1976.
14. A. Rutherford and G.C. Bloore, "The Implementation of fuzzy algorithm for control", control system centre report no. 279, UMIST, Manchester. To appear in Proc. IEEE. N.K. Sinha and J.D. Wright, "Application of a fuzzy control to a heat exchanger", Internal memoranda, Research group in Simulation, Optimisation and Control, McMaster University, Hamilton, Canada, 1975. D.A. Rutherford, "The implementation and evaluation of a fuzzy control algorithm for a sinter plant", Proc. workshop on discrete systems and fuzzy reasoning, Queen Mary College, London, 1976. G.A. Carter and M.J. Hague, "Fuzzy control of raw mix permeability at a sinter plant", Proc.

15.

16.

17.

Workshop on didcrete systems and fuzzy reasoning, Queen Mary College, London 1976.
18. L. Bainbridge, "The Process controller, in

The

Study of Real Skills, ed. W.T. Singleton,

Academic Press, 1975.

Philosophy of Science, 4,
19. W.J.M. Kickert et al., "Analysis of a fuzzy logic controller" Internal memorandum, Electrical Engineering Department, Queen Mary College, London 1976. E.H. Mamdani and N. Baaklini, "Prescriptive method for deriving a control policy in a fuzzy logic controller", Electronics Letters, Vol ii, p. 625, 1975. C.P. Pappis and E.H. Mamdani, "A fuzzy logic controller for a traffic junction", research report, Department of Electrical Engineering, Queen Mary College, London 1976.

2.

Logic, Cambridge University Inf. Contr., 8, pp. 338-

3.

L.A. Zadeh, "Fuzzy Sets", 353, 1965.

20.

4.

L.A. Zadeh, "Outline of a New Approach to the Analysis of a Complex System and Decision ProcesseS'

IEE Trans. Systems, Man and Cybernetics, SMC-3,

1973. 5. L.A. Zadeh, "Calculqs of Fuzzy Restrictions",

21.

ERL Memorandum M474, Oct. 1974.

!

6.

Artificial Intelligence in the Control of Real Dynamical Systems, Ph.D. Thesis, London
S. Assilian, University, 1974. E.H. Mamdani, "Application of Fuzzy Algorithms for the Control of a Dynamic Plant", Proc. IEE, 199

7.

PRESSURE ERROR = PE, CHANGE IN PRESSURE ERROR = CPE AND HEAT INPUT CHANGE = HC

IF PE = (NB OR NM) THEN IF CPE = NS THEN HC = PM OR IF PE = NS THEN IF CPE = PS THEN HC = PM

OR

IF PE = NO THEN IF CPE = (PB OR PM) THEN HC = PM OR IF PE = NO THEN IF CPE = (NB OR NM) THEN HC = NM OR IF PE = PO OR NO THEN IF CPE = NO THEN HC = NO OR IF PE = PO THEN IF CPE = (NB OR NM) THEN HC = PM OR IF PE = PO THEN IF CPE = (PB OR PM) THEN HC = NM OR IF PE = PS THEN IF CPE = (PS OR NO) THEN HC = NM OR IF PE = PB OR PM) THEN IF CPE = NS THEN HC = NM

Fig. 1

A list of rules used in the steam-engine control system

BACK-END TEMPERATURE = BE, BURNING ZONE TEMPERATURE = BZ PERCENTAGE OF OXYGEN GAS IN THE KILN EXIT GAS = OX

CASE 1

CONDITION
BZ L W O OX L W O BE L W O

ACTION TO BE TAKEN
WHEN A, B,

BZ

is D R A S T I C A L L Y KILN FUEL SPEED

LOW

REDUCE REDUCE

W E BZ IS SLIGHTLY L W HN O C, INCREASE I.D. FAN SPEED D, INCREASE FUEL RATE

BZ L W O OX L W O BE OK
BZ L W O OX L W O BE HIGH

A. REDUCE KILN SPEED B. REDUCE FUEL RATE C. REDUCE I.D. FAN SPEED
A, REDUCE KILN SPEED B, REDUCE FUEL RATE C, REDUCE I,D. FAN SPEED

TOTAL OF 27 RULES
Fig. 2 Examples of rules used for controlling a cement kiln 200

(A) (B) ............... (C)

CHANGINGTHE SUBSET (I,E, THE RULE) CHANGINGTHE SUPPORTSET CHANGING THE MEMBERSHIPFUNCTION

.t,o

/ "->-f'--~'~.

i \~ III/I
//
o
Fig. 3

\
I

...

'"
100

~o
Modification of fuzzy control

TIME

PRESCRIPTIVE FUZZY BAND

SET POINT

RULE MODIFICATION ALGORITHM

FUZZY CONTROLLER
I

ACTION

ERROR

SYSTEM
Fig. 4 Schematic diagram for an adaptive controller

201

ERROR
NB NB NM
0 ~J

NM

NS

NZ

PZ

PS

PM

PB

PB* PM* PM PB PB

PB*

NS NS

NM NM NS ZE PS PM
PM

PM PM PS ZE NS NM
NM

PS PS ZE NM NM NB
NB NB

NB*

NS

PM PB PB

ZE PM PM PB

NM NB NB

NM NB NB

-- ZE
Z

= PS PM PB

PB

PB

PB

NB

(A)

ERROR
NB NB NM
0 L~ z

NM PB

NS NB

NZ PB

PZ

PS

PM

PB

PB PB PS PB

PS PB PM PS ZE
NM

PB NS NM NM NB
NS

ZE PS ZE PS PS ZE

ZE ZE ZE PB ZE PS

NB NB

NS

-- ZE
Z

< PS PM PB PB

PS

PB

NS (B)

NB

NB

NB

Fig.

Rules

resulting

from a l e a r n i n g p r o c e s s

202