On Fuzzy Logic Applications For Automatic Control, Supervision ...

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

221

On Fuzzy Logic Applications for Automatic Control, Supervision, and Fault Diagnosis Rolf Isermann, Member, IEEE

Abstract— The degree of vagueness of variables, process description, and automation functions is considered and is shown where quantitative and qualitative knowledge is available for design and information processing within automation systems. Fuzzy-rule-based systems with several levels of rules form the basis for different automation functions. Fuzzy control can be used in many ways, for normal and for special operating conditions. Experience with the design of fuzzy controllers in the basic level is summarized, as well as criteria for efficient applications. Different fuzzy control schemes are considered, including cascade, feedforward, variable structure, selftuning, adaptive and quality control leading to hybrid classical/fuzzy control systems. It is then shown how fuzzy logic approaches can be applied to process supervision and to fault diagnosis with approximate reasoning on observed symptoms. Based on the properties of fuzzy logic approaches the contribution gives a review and classification of the potentials of fuzzy logic in process automation. Fig. 1. Multilevel process automation.

I. INTRODUCTION

B

ECAUSE of the progress of automatic control theory and technology during the last 60 years, many technical processes operate automatically under certain operating conditions. There exists a large variety of design and implementation approaches of continuous feedforward and feedback control or discontinuous logic and sequence control for any kind of processes with all kind of behaviors. Most of the control algorithms are based either on differential resp. difference equations or Boolean logic and are suitable to solve most control problems quite sufficiently. The development of fuzzy logic theory [1]–[3] now stimulated alternative ways to solve automatic control problems. Based on these basic ideas of fuzzy logic Mamdani and Assilian [4] proposed fuzzy controllers which describe human control in linguistic form. Consequently the first applications of fuzzy control replaced a human operator. For about 20 years, contributions on fuzzy control were presented at conferences and in the control literature, but the field of fuzzy control did not obtain high attention for a longer period. It is only recently—since about 1990—that the interest in fuzzy control has increased strongly because of successes and advertisement of applications in Japanese consumer products such as washing machines and camcorders. Moreover, successful fuzzy control of industrial processes, such as, e.g., a cement kiln in Denmark [5], in train operation [6], or simulations for ship steering [64], was shown (see also [7] and [8]). Since then, a controversial discussion has Manuscript received February 27, 1995; revised February 5, 1997. The author is with the Darmstadt University of Technology, Institute of Automatic Control, 64283 Darmstadt, Germany. Publisher Item Identifier S 1083-4427(98)00137-4.

been going on concerning the merits of fuzzy control versus conventional control. This article tries to contribute to this discussion by considering not only basic control loops but also other functions in process control systems and process automation. II. VAGUENESS

IN

AUTOMATIC CONTROL

As fuzzy logic provides a systematic framework to process vague variables and knowledge the various automation tasks will first be discussed with regard to the sources and the degree of vagueness. The goal is to determine some automation functions for which fuzzy logic offers attractive possibilities and advantages. Fig. 1 shows a general scheme for the organization of process automation tasks in several levels. Usually measurement and manipulation takes place in the lower level. The next level comprises feedforward and feedback control. Here certain variables are adjusted according to conditions or reference variables. The second level comprises supervision which serves to indicate undesired or unpermitted process states (monitoring) and to take appropriate actions such as failsafe, shut-down, or triggering of redundance or reconfiguration schemes. The higher levels may perform tasks such as optimization, coordination or general management in order to meet economic demands or scheduling. The lower levels usually need a fast reaction and act locally whereas the higher levels are dedicated to tasks which do not require a fast response and act globally. Table I shows estimates of degrees of sharp and vague information for the main elements in process information: the

1083–4427/98$10.00  1998 IEEE

222

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

TABLE I DEGREES OF SHARPNESS AND VAGUENESS FOR MAIN ELEMENTS IN PROCESS AUTOMATION. INTENSITY OF A PROPERTY: +++ LARGE; ++ MEDIUM; + SMALL

input variables, the process classes, the automation functions and the output variables. Sharp information is labeled to be “crisp” and vague information to be “fuzzy.” A. Input Variables With regard to the type of input signals precise and imprecise values are distinguished. Usual measurement equipment is designed to deliver signals with precise mean values and low standard deviations. Low cost sensors may have imprecise outputs with biased values and large standard deviations. Nondirectly measurable variables like slip, material stress, or comfort or quality measures can only be calculated or estimated based on other measurement variables and analytical models and are therefore imprecise to some extent (This is also known as sensor fusion). Also linguistic values of a human operator may be processed, like the color of fire in a cement kiln, a brush fire of a dc motor or the type of noise of an engine. Hence, the degree of precision and imprecision of the input variables may vary within fairly broad ranges.

models based on physical laws result and one can distinguish the following cases of analytical process models (quantitative models): A.1) equations with known parameters; A.2) equations with partially unknown parameters; A.3) equations with unknown parameters. In practice, case A.2 appears more frequently than A.1 and A.3. If the structure is known, the unknown parameters can be determined experimentally by parameter estimation and identification methods, e.g., [9]. Hence, quantitative process models result from proper combination of theoretical and experimental modeling. The field of process modeling has reached a fairly mature status and is packaged in attractive software tools. However, if the internal behavior cannot be expressed by physical laws in the form of equations, as for “not well defined” processes like drying, vulcanizing, special chemical reactions or biological processes, some qualitative information on the causalities may be expressed as rules: (1)

B. Process Classes The description of the static and dynamic behavior in form of mathematical models plays a crucial role as well for the design of the technical processes as for a systematic design of high performance control systems. As an example continuous processes with continuous input and output signals are considered. If the internal process behavior can be described by basic physical principles, theoretical modeling can be applied by stating balance equations, physical-chemical state-equations (constitutive relations) and phenomenological laws. Then the structure of the mathematical model is obtained either as a lumped parameter or distributed parameter system (ordinary or partial differential equations). By this way quantitative

The condition part then contains as inputs the memberships of facts to the premise and possibly their logical connection by AND, OR and NEGATION, and the conclusion part describes the logical consequence on the output. Then one can distinguish heuristic process models (qualitative models): B.1) rules with known conditions, known conclusions and known relations between conditions and conclusions; B.2) rules with unknown logical connections within the condition, but known conclusion; B.3) rules with known condition and unknown conclusion. Hence, a gradual decrease of knowledge on the process can be observed from analytical process models, case A.1, to heuristic process models, case B.3. This is also shown in

ISERMANN: FUZZY LOGIC APPLICATIONS

Fig. 2. Reachable model accuracy in dependence on the degree of quantitative or qualitative knowledge of processes.

Fig. 3. Reachable model accuracy for different processes in dependence on the dimension of its elements.

Fig. 2 where the degree of knowledge ranges from precise to vague. Precise quantitative models can be obtained for mechanical and electrical processes, fairly precise, and less precise models for thermal and chemical processes. Less vague qualitative models correspond, e.g., to biological processes and fairly vague and vague models to socioeconomic processes, compare Table I. However, the model accuracy also depends on the size of elements or components, i.e., the dimension of a process. With increasing dimensions, in general, the model accuracy decreases (see Fig. 3). If processes with only qualitative knowledge are of high dimension, complex processes result which naturally have a low reachable model accuracy. C. Automation Functions The functions of process automation are usually organized according to the levels shown in Fig. 1. Feedforward control systems are characterized by an open signal flow and can be divided in continuous and discontinuous signal flow systems. In the last case one distinguishes between logic control for which input and output signals, according to Boolean algebra, storages and time functions are related to each other and sequence control for which the switching from one step to the following depends on the switching condition. The outputs are usually binary variables (on-off-signals). Hence, crisp inputs and outputs are processed.

223

Feedback control systems have a closed-loop structure to adjust control variables to given reference variables. Usually both the control variables and the manipulated variables are crisp values. The known controller equations such as PIDcontrollers or state-controllers with observers lead to precise algorithms. Because the knowledge on the processes is vague in most cases the tuning of PID-controller parameters is frequently performed by tuning rules. Only for designing sophisticated and safety critical control systems precise mathematical process models are required usually. For supervision a basic task is to check if variables exceed tolerances. These tolerances are generally a compromise between small values to early detect faulty behavior and large values to avoid false alarms subject to normal process fluctuations. Hence, in reality these tolerances are not crisp but fuzzy variables. Moreover, in many cases fault diagnosis means to treat vague process knowledge and therefore to apply rules. The process management includes optimization with regard to economical performance or coordination of several processes. Here both the knowledge on the process and the criteria are frequently not precise and therefore a good part of qualitative knowledge has to be treated. The man-machine interface is designed for human interaction. Basic tasks of the human operator are higher level tasks as, e.g., supervision, management and redundancy for all control functions. The actions of human operators are based on qualitative knowledge and may be better described by rules than by precise algorithms. Hence, the kind of information in process automation generally changes from using more precise algorithms at the lower levels to using more rules at the higher levels, as summarized in Table I. D. Output Variables Actuators with electrical, pneumatic, or hydraulic drives need crisp inputs. Also on-off actuators like switches have crisply defined switching points. Outputs to the human operator through the man-machine interface can be either crisp, if numbers are indicated or fuzzy, if they are linguistic. In summary this discussion and Table I show that vagueness appears in different form and degree dependent on the type of variables, process classes and automation tasks. For the different automation functions Table II considers the degree of vagueness if a fuzzy logic approach is applied. The vagueness shows up at the input and output variables and in the rules. Because the degree of vagueness increases from the left to the right, it can be stated that the vagueness (qualitative knowledge) in general increases with increasing automation level. III. INTELLIGENT CONTROL SYSTEMS After the recent development of model-based and adaptive control systems the present development in control systems can be summarized with “intelligent” control. This new development now amplifies the treatment of qualitative knowledge. Various definitions do exist for intelligent control (see, e.g., [10]–[16]). In comparison to human beings first developments approach a low degree of intelligence. A “low-degree intel-

224

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

TABLE II DEGREES OF VAGUENESS FOR PROCESS AUTOMATION FUNCTIONS WITH A FUZZY LOGIC APPROACH.

11 VAGUE; 1 LESS VAGUE;  CRISP; 0 NOT DEFINED

ligent” controller has, e.g., the “ability to model, reason and learn about the process and its control within a given frame,” [17]. The intelligent controller then adapts the control to the mostly nonlinear process behavior (adaptation) and stores its controller parameters in dependence on situations (learning), supervises all relevant elements and performs a fault diagnosis (supervision) to request for maintenance, to shut off faulty components, or to fail safe (decision on actions). Fig. 4 shows the organization of an “advanced intelligent” automatic control system. It consists of multicontrol levels, a knowledge base, inference mechanisms, and communication interfaces. Both the knowledge base and the inference mechanism treat quantitative and qualitative knowledge. The processing of qualitative knowledge may be performed by methods of soft-computing [3], including fuzzy logic, neuronal networks and genetic optimization algorithms. This figure indicates that intelligent control systems can be considered as on-line, real-time expert systems. It can also be stated that, in general, the hierarchical structure of an intelligent system shows a decreasing precision with increasing intelligence [11]. This also means that with decreasing quantitative knowledge and increasing qualitative knowledge in general more intelligence is required. Hence, as well for conventional automation systems (section II) as for intelligent control systems it holds that the potentials of fuzzy logic approaches in general increase with higher automation levels. IV. FUZZY-RULE BASED SYSTEM Fig. 5 shows a general scheme of a fuzzy IF-THEN rulebased system [4], [8], [18], [19] consisting of the following modules: • fuzzfication: (converting of crisp facts into fuzzy sets described by linguistic expressions);

Fig. 4. Advanced intelligent automatic system with multicontrol levels, knowledge base, inference mechanism, and interfaces.

• inference: The fuzzy IF-THEN rule expresses a fuzzy implication relation between the fuzzy sets of the premise

ISERMANN: FUZZY LOGIC APPLICATIONS

225

Fig. 5. General scheme of a fuzzy IF-THEN rule-based system.

Fig. 6. Signal flow in a fuzzy rule-based system with two levels of rules and max/min operations and singletons as inputs.

and the fuzzy sets of the conclusion. If the rule is interpreted by the Mamdani-implication and the compositional rule of inference by Zadeh [3] is applied, the following steps can be distinguished: 1) matching of the facts with the rule premises (determination of the degrees of fulfillment of the facts to the rule premises fit values); 2) if the rule contains logic connections between several premises by fuzzy AND or fuzzy OR the evaluation is performed by t-norms or t-conorms (The result gives then the degree of fulfillment of the facts to the complete premise); 3) the evaluation of the resulting conclusion follows the max-min composition [3]. (The degree of fulfillment of the premise restricts the fuzzy set of the conclusion.) • accumulation of all conclusion fuzzy sets if several rules contribute to the output (a union operation yields a global fuzzy set of the output). • defuzzification to obtain crisp outputs (various defuzzification methods can be used, as, e.g., center of gravity, maximum-height, and mean of maximum, to obtain a crisp numerical output value). The computational effort of a fuzzy-rule-based system increases strongly if the number of fuzzy sets and the number of

rules grow. By special assumptions the computations can be simplified considerably [20], [21] for example by singletons at the input and output or universal sets at the output. Fig. 6 shows the signal flow for a fuzzy-rule-based system with two levels of rules and singletons as inputs. In this case fuzzification and matching reduces to directly determining the degree of fulfillment of the premises. The evaluation of the rule conclusion then consists of applying the respective norms and -conorms (e.g., min-max or prod-sum operation) for the logic operations in the premises and determination of the fuzzy sets of the conclusion. The chaining of rules can be performed with or without defuzzification. In Fig. 6 no defuzzification is considered between the two levels of rules. According to the knowledge base the rules can be represented by a network, building up several levels. For fuzzy control usually only one level of rules is required. However, fault diagnosis leads to tree structures with several levels. The contributions of all rules to the last level have then to be accumulated, e.g., by a union operation. This usually results in combined fuzzy sets. A crisp output as a final result is obtained by defuzzification, if required. In the next two sections different kinds of using fuzzy logic approaches in control systems are considered for both normal and special operating conditions.

226

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

is a singleton at

by a crisp input

becomes

(3) at position Hence, a singleton of height results. Then it makes no difference any more, if as norm the minimum- or product-operator is applied. For the accumulation the maximum of all singletons is taken and defuzzification follows via the center of singletons Fig. 7. Signal flow of a fuzzy controller with two inputs (2 from 6 rules are shown).

(4)

V. FUZZY-FEEDBACK CONTROL FOR NORMAL OPERATING CONDITIONS With regard to automatic feedback control normal operating conditions are understood where automatic operation is possible without manual control. A. Fuzzy Control in the Basic Level (Direct Fuzzy Control) Before discussing more general aspects of fuzzy control applications some practical aspects of fuzzy control are first considered. For the design of a fuzzy controller the signal flow of a general fuzzy-rule-based system as shown in Fig. 5 reduces to a rule activation block with one rule level, as shown in Fig. 7. The two inputs and are, e.g., the control deviation and its derivative (2) and are assumed to be crisp values. The membership functions of the inputs (attributes) are usually symmetric triangles with equal distribution over the universe of discourse or ramps, such that the sum is always one. The formulation of the attributes of the output variable shows frequently more variety [22]. Now a special controller configuration is considered. The effect of choosing different output attributes like rectangles, triangles, ramps, etc., has been investigated in [21]. For two ramps as inputs and and center of gravity defuzzification, the analysis of the controller’s transfer behavior yielded the following results. 1) In selecting the output attributes the position of their center of gravity is mainly important. 2) The shape, the width of the output attributes and the inference method (max-min or max-prod.) are of minor importance. They influence the interpolation between the centers. 3) The width of rectangles and triangles has no influence, as long as the membership functions do not overlap each other. 4) Overlapping triangles or ramps (mountains) generate Sshaped nonlinear characteristics which are, however, not much different from a linear one. 5) Formulation of the linguistic output attributes as singletons leads to a drastic reduction of the computational effort. The activation of an output attribute which

A similar procedure follows for multiple inputs, leading to simple vector multiplications. 6) Drastic nonlinear characteristics can mainly be reached by distributing the output attributes unequally over the universe of discourse. B. Criteria for the Application of Fuzzy Control in the Basic Level To discuss the potentials of fuzzy controllers in the basic level (direct control) one has to consider the following basic control engineering viewpoints. 1) The proper design of the static and dynamic behavior of the process is the most important prerequisite for reaching a good control performance. 2) The engineering design includes the right specification and design of the actuator and the measurement equipment, such that the process can at least be controlled manually. 3) The solution of these problems presupposes the existence of models of the process, at least rough static characteristics. Hence, simple models frequently exist for important processes. 4) In most cases linear PI- and PID-controllers are applied and the parameters are tuned by tuning rules. Mathematical models are not required for the selection of the controller type and the parameter tuning. A rough understanding of the process’ control behavior is sufficient in most cases. 5) With modern microprocessor-based controllers it is no problem to generate a static nonlinear behavior like nonlinear characteristics, dead zones, direction dependent control actions, etc. Fuzzy-rule-based controllers as in Fig. 7 and Fig. 8(a) generate a three-dimensional static mapping of the output in dependence on the two inputs (look-up-table). Depending on the selection of the membership functions this mapping may become nonlinear and depending on the number of inputs multidimensional. By adding dynamic elements at the input or the output (e.g., derivative or integrating terms) the fuzzy controller gets dynamic properties.

ISERMANN: FUZZY LOGIC APPLICATIONS

227

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

Fig. 8. Control system structures with different fuzzy control components (hybrid classical/fuzzy control). F: fuzzification; I: inference; and D: defuzzification.

Nonlinear characteristic surfaces can of course also be generated directly. The advantage of the fuzzy-rule-based approach lies in the formulation of the input-output behavior in form of linguistically expressed rules. By this way the practical experience of human operators can be implemented in another form than by (simple) equations, which may be easier to understand for some operators and nonexperts.

The control engineering aspects for the application of fuzzy controllers are still under investigation. Some advantages and disadvantages can be summarized as follows. 1) Nonlinear processes. For processes which can be linearized and processes with analytically known nonlinear behavior, such as nonlinear characteristics or hysteresis, many solutions do exist with classical control sys-

228

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

tems like PID-, state space controllers with observers, cancellation-, deadbeat-, minimum-variance and predictive controllers [23]. In most cases only rough process model knowledge is required, especially, if self-tuning or adaptive control methods are applied [9]. For processes where no sufficient control performance can be reached with PI- and PID-controllers and parameter adjustment by tuning rules, fuzzy control may be an alternative. This is especially valid for processes with difficultly understandable behavior, if only qualitative knowledge is available and strongly nonlinear properties do exist. Some examples are milling plants, glass melting, cement kilns, food processing. 2) Knowledge of human operators. If manual control of a difficult process is possible the operator knowledge may be formulated in fuzzy rules with linguistic inputs and outputs (This is in many cases not an easy task). The parametrization can be improved especially by trial and error and also additional inputs can be used, e.g., in the sense of feedforward actions. However, the fine parametrization may take a long time, since much more parameters have to be adjusted than for PID-controllers. Another application is the imitation of human controllers in the loop, as “fuzzy control involves a representation of what is basically a human solution in the form of fuzzy if-then rules,” [3]. This kind of application is, e.g., interesting for car driving, commanding aircraft, operating a crane or workshop activities. Here the actions of a human can be formulated in rules and can be adapted by identification methods [21], [24]. Hence, fuzzy controllers can be used to investigate human’s behavior. 3) Fast prototype design. A further advantage of the fuzzy control concept is a fast prototyping by some few rules, if a classical standard controller is not suitable. However, the fine tuning for reaching a good control performance usually lasts too long. A still holding disadvantage of direct fuzzy control is the difficulty of stability analysis. In fact, the proof of stability for arbitrary fuzzy controllers is not a simple task due to their various structures and nonlinear characteristics. Several approaches for stability analysis have been presented in recent years. [25]–[27] give an overview of existing methods. Most of them are adopted from nonlinear control theory, e.g., Lyapunov method, harmonic balance, Popov criterion etc. The examples show that both time domain and frequency domain approaches are applied. It can be distinguished between methods considering the fuzzy controller as nonlinear input-output mapping, e.g., [8], and those regarding the internal structure of the fuzzy controller such as [28]–[30]. The latter approaches can also be used in a constructive manner in order to design robust fuzzy controllers. The next sections show other forms of applying fuzzy concepts for control systems.

TABLE III CONTROLLER STRUCTURE IN DEPENDENCE ON THE PROCESS BEHAVIOR AND SIGNALS

auxiliary controller and a main controller [Fig. 8(c)] [23]. The auxiliary controller can, e.g., be a classical P- or PI-controller and the main controller can be a fuzzy controller which is especially designed for the nonlinear behavior of process part Another interesting possibility is a cascade control system with a fuzzy main controller for two different control variables [Fig. 8(d)]. Here fuzzy rules decide which one of both control variables is in action with a smooth transition from one to the other. An example is the velocity and distance control in car driving [31]. D. Fuzzy Feedforward Control If an external disturbance of a process can be measured before it acts on the output variable the control performance with respect to this disturbance can be improved by feedforward control. Then the fuzzified measured disturbance is just added to the premise of the rules of the fuzzy feedback controller [Fig. 8(e)]. Thus, proportional acting nonlinear feedforward control can be realized. E. Variable Control Structures with Fuzzy Weighting The selection of classical controllers depends among others on process properties such as the transfer behavior of the plant, the size of control deviations and the kind of disturbances [23]. Therefore the type of controller (and its parameters) should be altered, if such properties change. This is the case if for example the size of control errors change drastically or if wide process operating ranges have to be governed. Table III shows some examples for changing (classical) controller structures. Usually only one controller type with robust properties is implemented or two controller types are switched on and off alternatively. A fuzzy logic approach now offers a gradual change from one controller type to another by weighting the control deviation by for controller 1 and for controller 2 based on the (e.g., identified) process behavior [Fig. 8(f)]. Dependent on the process behavior rules can be established to weight the controllers action (and to change its parameters). One example is IF disturbances have higher frequencies AND control deviation is medium THEN decrease weight of PI-controller AND increase weight of PD-controller (see also [32]).

C. Fuzzy Cascade Control Systems

F. Fuzzy-Tuning of Classical Controller Parameters

The control performance of a single loop system can be improved by expanding it to a cascade control system with an

In practice the parameters of P-, PI- and PID- controllers are mostly adjusted by tuning rules based on

ISERMANN: FUZZY LOGIC APPLICATIONS

closed loop oscillation experiments with a P-controller (Ziegler–Nichols rule) or open loop transfer function measurement (Chien–Hrones–Reswick rule). In [24] and [33], a parameter tuning procedure with if-then fuzzy rules [Fig. 8(h)] is shown for PI- and PID-controllers. The controller starts with small values of the parameters. Measured features of the closed loop transient function such as the overshoot and the ratio rise time/settling time are fuzzified and rules for the changes of the controller parameters allow to adjust the controller parameters after some few (2–8) step response experiments. One example of the 24 tuning rules for a PI-controller is: IF overshoot is too small AND settling ratio is alright THEN increase gain by 30% The tuning rules hold for a large variety of process behaviors and also for stochastic disturbances. G. Fuzzy Adaptive Control Adaptive control systems adapt their control behavior to the changing properties of the controlled processes and their signals. Most adaptive control systems can be divided into feedforward and feedback adaptive controllers [9]. Fig. 8(g). shows a scheme for a feedforward adaptive control which is based on the fact that the changing properties of the plant can be grasped by the measurement of a signal on the process (e.g., mass flow, speed, load). If rules are known how the controller parameters of a classical (or fuzzy controller) must be changed in dependence on the measured signal, a fuzzy rule adaptation can be realized. If the process behavior cannot be determined directly by the measurement of external process signals, feedback adaptive control has to be used [Fig. 8(h)]. The changing properties of the process have to be determined by the measurement of different internal control loop signals. Classical feedback adaptive control is then based on the identification of a mathematical process model (model identification adaptive control) or a comparison with a reference model (model reference adaptive control) with the goal to design a controller automatically. A fuzzy feedback adaptive controller does not use a mathematical process model but rules for the adaptation of the controller based on measured closed loop signals. For example the controlled and manipulated signals are used to determine measures for the control performance. After the fuzzification of these measures tuning rules of the controller determine the adjustment of the parameters of a classical or a fuzzy controller. A further possibility results from using a fuzzy model of the process. The expert knowledge is here used to model the process by rules and measured signals to refine the model with identification procedures [34]–[36]. The controller design then is performed automatically. The controller types may be classified as predictive controller [35], PID-controller [36], feedback linearization, etc. or fuzzy controllers. This results then in adaptive control with fuzzy models. H. Fuzzy Control of Nonmeasurable Quantities Sometimes a control variable is not measurable or a sensor should be saved because of economic reasons. Then one can

229

try to use already available other sensor signals as inputs [see Fig. 8(i)]. If analytical relations are known the control variable can be calculated, e.g., a heat stress from measured temperature variations. However, if mathematical models are unknown or too complicated, one can formulate rules. An example is the room temperature heating control without indoor and outdoor sensors. The basic water temperature is determined by the average temperatures over the year and during a day (yearly and daily temperature profile). This rough forecast is corrected by AND connections in the premise with measured inputs as: average energy consumption of the previous day and present energy consumption [37]. I. Fuzzy Quality and Comfort Control The quality of products is mostly based on several properties. Examples are as follows. 1) Metal products: geometrical precision, roughness of surface, texture, stiffness, hardness. 2) Ceramic products: geometry, color, surface, acoustic sound. 3) Textile products: elasticity, strength, texture, color, pattern, homogenity, bulking, skin reception, washing, and iron properties. 4) Cement: particle size and its distribution, specific surface (Blaine number), sieve residues, early strength and strength after 28 days of binding. 5) Tyre: stiffness, traction, durability, noise, rain and aquaplaning sensitivity, temperature sensitivity. The various product properties which accumulate to an overall quality measure can be influenced during processing by changing of reference values of control loops, reaction times, load variables, ingredients, etc. As frequently no analytical models are known, but only rules, a multivariable quality fuzzy controller with quality attributes as inputs and manipulated process variables as outputs can be formulated [Fig. 8(b)]. A similar situation can be found for the control of comfort where the human perception plays a role. Examples are as follows: 1) The comfort of the climate in air conditioning where the temperature, humidity and kind of labor of human beings follow certain rules. 2) The comfort of the passengers during car driving where the received frequency spectra of accelerations can be formulated as rules with the adjustable shock absorbers as outputs and road conditions and accelerations as inputs. In [38] the simulation of a fuzzy controller is shown where rules are given for the adjustment of the shock absorber dependent on the relative speed of the damper and the speed of the chassis. 3) The comfort of the start-up of automobiles with manipulation of the clutch and gear shifting [39]. J. Neuro-Fuzzy Control An increasing activity can be observed in the field of neurofuzzy control systems [40]–[42]. The combination of fuzzy logic and neural networks integrates the advantages of both approaches in one controller. On the one hand, the controllers

230

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

can be designed in the fuzzy sense using heuristic knowledge, on the other hand, they can be trained from learning data by means of algorithms developed for neural networks. Various training algorithms have been presented for both on-line and off-line training. Two general strategies in neuro-fuzzy control can be distinguished: The most simple proposals use two separate components (one fuzzy system, one neural network) in a parallel control structure. However, approaches utilizing the structural equivalence between certain types of fuzzy systems and artificial neural networks [43]–[45] seem to be more promising. One single controller can be trained from data (interpretation as neural network) and still remain transparent due to the internal fuzzy-rule structure (interpretation as fuzzy systems).

VI. FUZZY-CONTROL FOR SPECIAL OPERATING CONDITIONS AND LARGE-SCALE DYNAMIC EFFECTS Many processes are automatically controlled for normal operating conditions with sufficient control performance. With regard to feedback control systems normal operating conditions mean that stable or not too sluggish control is possible with a fixed controller or an adaptive controller (mostly feedforward adaptation of controller parameters). This is for example the case for the control of flow or pressure, heat exchangers or steam boilers within a range of about 20% to 100% of the load. Especially for small loads closed loop operation with the normal controller is frequently not possible as the gain of the process and its time constants increase too much or the valves operate in nearby closed position with undefined characteristics. Other examples of abnormal process behavior are 1) the start-up and shutdown phase of engines, machines, industrial processes; 2) mode changes in chemical processes; 3) spinning of cars outside of normal tire/road slip (unstable steering behavior); 4) large slip of tires and blocking of wheels after heavy braking; 5) stalling of an aircraft after falling short of speed; 6) control of processes after failures (emergency control). In these cases the process behavior is frequently not well known because some physical laws do not hold any more or are not known (e.g., change from turbulent to laminar flow in heat exchangers, rotating of combustion engines with irregular speed, chattering in metal cutting processes). Hence, there is need to control processes with large uncertainties and largescale dynamic effects [46]. The usual way for governing the abnormal process is manual control with experienced operators by bringing the process back into normal or safe conditions. Now a special fuzzy controller can be designed by making use of the rules gained by the operator [Fig. 8(j)]. It is typical for these abnormal operating conditions that the manipulation has to be performed differently in comparison to the normal operation, e.g., by using fast and large actions as in the case of spinning cars or the recovery of an aircraft from the poststalling phase. The special fuzzy controller may then use an

additional manipulated variable and controlled variable [Fig. 8(j)]. The discussion in Sections III and IV has shown that the potentials for fuzzy logic approaches in the feedback control domain are mainly 1) fuzzy control in the basic level for nonlinear processes with only qualitative knowledge or if imitation of human controllers is required; 2) addition of fuzzy control to classical control (cascadecontrol, feedforward control, variable control structures); 3) fuzzy control in the adaptation level; 4) fuzzy control for nonmeasurable quantities, quality or comfort measures; 5) fuzzy control for abnormal operating conditions and large-scale dynamic effects. Hence, fuzzy control approaches may be especially advantageous as an addition to classical control systems, forming hybrid classical fuzzy control systems. Then different fuzzy control components can be specified, as shown in Fig. 8. VII. SUPERVISION, FAULT DETECTION AND DIAGNOSIS A. Knowledge-Based Fault Detection Within automatic control of technical systems supervisory functions serve to indicate undesired or unpermitted process states and to take appropriate actions in order to maintain the operation and to avoid damages or accidents. The following functions can be distinguished: a) monitoring: measurable variables are checked with regard to tolerances and alarms are generated for the operator. b) automatic protection: in the case of a dangerous process state, the monitoring function initiates automatically an appropriate counteraction. c) supervision with fault diagnosis: based on measured variables features are calculated, symptoms are generated via change detection, a fault diagnosis is performed and decisions are made for counteractions. Fig. 9 shows an overall scheme of knowledge based fault detection and diagnosis. The main tasks can be subdivided in fault detection by analytic and heuristic symptom generation and fault diagnosis [47], [48]. 1) Analytic Symptom Generation: The analytical knowledge on the process is used to produce quantitative, analytical information. Based on measured process variables a data processing has to be performed to generate first characteristic values by 1) limit value checking of directly measurable signals. Characteristic values are exceeded signal tolerances. 2) signal analysis of directly measurable signals by use of signal models like correlation functions, frequency spectra, autoregressive moving average models (ARMA). Characteristic values are, e.g., variances, amplitudes, frequencies or model parameters. 3) process analysis by using mathematical process models together with parameter estimation, state estimation and parity equation methods. Characteristic values are parameters, state-variables or residuals.

ISERMANN: FUZZY LOGIC APPLICATIONS

231

Fig. 9. Scheme of a knowledge-based fault detection and diagnosis.

In some cases special features can be extracted from these characteristic values, e.g., physical defined process coefficients or special filtered or transformed residuals. These features are then compared with the normal features of the nonfaulty process and methods of change detection and classification are applied. The resulting changes (discrepancies) of the described direct measured signals, signal models or process models are then analytic symptoms of faults. 2) Heuristic Symptom Generation: In addition to the symptom generation with quantifiable information heuristic symptoms can be produced by using qualitative information from human operators. Through human observation and inspection heuristic characteristic values in form of special noise, color, smell, vibration, wear and tear, etc. are obtained. The process history in form of performed maintenance, repair, former faults, and lifetime, load measures constitute a further source of heuristic information. Statistical data (e.g., MTBF, fault probabilities) achieved from experience with the same or similar processes can be added. By this way heuristic symptoms are generated which can be represented as linguistic variables (e.g., small, medium, large) or as vague numbers (e.g., around a certain value). B. Model-Based Fault Detection Methods Different approaches for fault detection by using mathematical models were developed in the last 20 years. The task consists in the detection of faults in the technical process including actuators and sensors by measuring the available input and output variables and . The process is considered to operate in open loop. A distinction can be made between static and dynamic, linear and nonlinear process models.

Three important model-based fault detection methods and their residuals are: 1) parameter estimation: changes of parameter estimates ; 2) state estimation: changes of state estimates or output errors ; 3) parity equations: output error or polynomial error . For more details, see [47]. The measured or estimated quantities like signals, parameters, state-variables or residuals are usually stochastic variables with mean value and variance (5) as normal values for the nonfaulty process. Analytic symptoms are then obtained as changes

(6) with reference to the normal values. Fig. 10(a) shows an exIf a fixed threshold ample for a stochastic symptom is used, a compromise can be made between the detection of small faults and false alarms because of short term exceeding [see Fig 10(b)]. Methods of change detection which estimate the change of the mean value and relate it to the standard deviation may improve the (binary) decision. Examples are likelihood-ratio test, Bayes-decision, run-sum test, two probe -test. Now a triangle-membership function is specified with the center as the mean and the lower width or because for a Gaussian distribution 68% or 95% of all values lie within those intervals [Fig. 10(c)], thus obtaining a combined symptom representation of the mean and the

232

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

(a)

(b)

()

(c)

Fig. 10. Crisp and fuzzy threshold for a stochastic variable S t with standard deviation  . (a) Time history with a change of mean by (b) Membership function of a crisp threshold. (c) Membership function of a fuzzy threshold and matching with current value.

standard deviation. By matching the current value the symptom’s membership function “increased”

1 S for t > Tf .

with

one obtains a gradual measure for exceeding a threshold, a fuzzy threshold.

(a)

(b)

C. Fault Diagnosis Methods The task of fault diagnosis consists of determining the type of fault with as much details as possible such as the fault size, location and time of detection. The diagnosis procedure is based on the observed analytical and heuristic symptoms and the heuristic knowledge of the process, as shown in Fig. 9. In this section the heuristic part of the knowledge and inference mechanisms for diagnosis are described in order to build up on-line expert systems for fault diagnosis. 1) Unified Symptom Representation: For the processing of all symptoms in the inference mechanism it is advantageous to use a unified representation. One possibility is to present the analytic and heuristic symptoms with confidence numbers and treatment in the sense of probabilistic approaches known from reliability theory [49]. Another possibility is the representation as membership function of fuzzy sets [50] (Fig. 11). By these fuzzy sets and corresponding membership functions all analytic and heuristic symptoms can be represented in a unified way within the range [0, 1]. These integrated symptom representations are the inputs for the inference mechanism (Fig. 9). 2) Heuristic Knowledge Representation: For establishing heuristic knowledge bases for diagnosis several approaches do exist. In general specific rules are applied in order to set up logical interactions between observed symptoms (effects) and unknown faults (causes). The propagation from appearing faults to observable symptoms in general follows physical cause-effect relationships where physical properties and variables are connected to each other quantitatively and also as functions of time. However, the underlying physical laws are frequently not known in analytical form or too complicated for calculations. Therefore heuristic knowledge in form of qualitative process models can be expressed in form of IF-THEN rules.

(c) Fig. 11. Examples for membership functions of features S resp. symptoms S represented as fuzzy sets (a) symptom increases, (b) symptom decreases, and (c) symptom increases or decreases with linguistic terms (labels).

1

The condition part (premise) contains facts in the form of observed symptoms as inputs and the conclusion and faults as a logical cause part includes events of the facts. This procedure results in fault-symptom trees (“directed graphs”), relating symptoms to events and faults. Then symptoms or events are associated by AND and OR connectives and the rules become of form

(7) In the classical fault-tree analysis the symptoms and events are considered as binary variables with for happened and for not happened [51]. 3) Diagnostic Reasoning: Based on the available heuristic knowledge in form of heuristic process models and weighting of effects different diagnostic forward and backward reasoning strategies can be applied. Finally the diagnostic goal is achieved by a fault decision which specifies the type, size and location of the fault as well as its time of detection. A review of recent developments in reasoning oriented approaches for diagnosis was given in [52]. By using the strategy of forward

ISERMANN: FUZZY LOGIC APPLICATIONS

233

chaining a rule, the facts are matched with the premise and the conclusion is drawn based on the logical consequence (modus ponens). Therefore with the symptoms as inputs the possible faults are determined using the heuristic causalities. In general the symptoms have to be considered as uncertain facts. Therefore a representation of all observed symptoms as membership functions of fuzzy sets in the interval [0, 1] is feasible, especially in unified form. 4) Approximate Reasoning with Fuzzy Logic: With the structure of a fault symptom tree a fuzzy-rule-based system with multiple levels of rules can be established, as shown in Fig. 6. The symptoms of the features are now represented by fuzzy sets with linguistic meanings like “normal,” “less increased,” “much increased,” etc. Then the following analogies to Fig. 6 hold: — symptoms — events — faults In contrast to fuzzy control in the basic level fuzzy fault diagnosis differs by 1) inputs are mostly no crisp measurements, but detected symptoms represented as fuzzy sets; 2) not only one level of rules does exist, but mostly several levels; 3) frequently it is difficult to specify the membership functions of the intermediate events because of very vague knowledge. The approximate reasoning follows the steps described in Section IV: fuzzification or matching, rule activation with evaluation by the compositional rule of inference, accumulation and defuzzification. The dimensions of the fuzzy-rule-based system is given by — number — number — number — number

of of of of

symptoms levels rules per level faults

The overall dimension may therefore blow up strongly even for small components or processes. Therefore the software implementation is important. Two procedures to perform the reasoning are known. Sequential rule activation from level to level (horizontal procedure): 1) matching of symptoms with the premises of rules (including their logical operations) to of level determine the events ; 2) matching of events premises of rules of level ; 3) matching of events

with the to determine the events with the faults fuzzy subsets

; 4) accumulation of fault fuzzy subsets to obtain the global possibility for each fault;

5) fault decision by maximal possibility; 6) defuzzification to obtain a crisp size of the most possible fault. Multiple rule activation from all symptoms to all faults (vertical procedure) [53]: 1) chaining by multiple composition to obtain overall relations in form of a matrix ; 2) matching of with all current symptoms ; 3) accumulation to obtain the global possibility for each fault; 4) fault decision by maximal possibility; 5) defuzzification to obtain a crisp size of the most possible fault. In both approaches, a defuzzification can be performed after each level or not. To reduce the computational effort, simplifying assumptions may help. The following simplifications are possible: 1) singletons for the symptoms: matching is reduced to direct fuzzification of crisp values; 2) singletons for the events and faults: the conclusions are singletons with a height of the matching degree between facts and rule premise; 3) universal fuzzy sets for the events and faults: the conclusion is a universal set with height of the matching degree between facts and rule premise. But then the size of the event or fault cannot be determined. In the cases of singletons or universal sets for events and faults the conclusions are identical with the membership degree of the facts with the logical operation in the premise. This means the whole procedure reduces to an aggregation of the facts with appropriate t-norms and t-conorms, and chaining of the results [47]. Fuzzy fault diagnosis can be expanded to time dependent faults and symptoms, including incipient and intermittent faults and multiple faults [53]. 5) Neuro-Fuzzy Systems for Diagnosis: The knowledge acquisition and tuning for fault diagnosis usually takes much effort and time. Therefore adaptation mechanisms of the establishment of rules and membership functions are required. Considering now the similarity between fuzzy logic and neuronal networks adaptive neuro-fuzzy systems can be developed [54]–[59]. In [54] it is shown that a combination of the bounded operator for fuzzy-AND and the bounded sum for fuzzy-OR leads to an adjustable ANDOR-operator ( operator), which can be approximated by a sigmoidal activity function. Furthermore, by applying Gaussian membership functions adjustable symptom membership functions can be obtained. By measurement of faults and symptoms the neurofuzzy diagnosis system is adapted by a backpropagation method, see also [55]. VIII. CONCLUSION Fuzzy logic provides a systematic framework to process vague variables and vague knowledge. As the vagueness in automation generally increases with increasing functional

234

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART A: SYSTEMS AND HUMANS, VOL. 28, NO. 2, MARCH 1998

hierarchical level the potentials of fuzzy logic approaches naturally grow with higher levels. Fuzzy control in the basic level may be attractive for processes where only qualitative knowledge is available and strongly nonlinear properties do exist, or if human control has to be imitated. Interesting possibilities for fuzzy control exist in the higher control levels, as e.g., cascade control, variable control structures with fuzzy weighting, fuzzy controller parameter tuning, fuzzy adaptive control, fuzzy control of nonmeasurable quantities and fuzzy quality and comfort control. Another application area is the fuzzy control for special (abnormal) conditions where today mostly only manual control is applicable. Thus hybrid classical/fuzzy control system structures result. The supervision of processes requires the treatment of quantitative and qualitative knowledge. Here fuzzy approaches are especially attractive for symptom generation with fuzzy thresholds, linguistically described observed symptoms and the approximate reasoning with multilevel fuzzy-rule-based systems for fault-symptom tree structures. Further applications of fuzzy logic methods may arise within the qualitative knowledge processing of intelligent or multiobjective control systems including higher level control functions and learning. Especially the combination with nonlinear approximation (identification) and classification by artificial neural networks may open many new possibilities. REFERENCES [1] L. A. Zadeh, “Outline of a new approach to the analysis of complex systems and decision processes,” IEEE Trans. Syst., Man, Cybern., vol. SMC-3, pp. 28–44, 1973. [2] , “A rationale for a fuzzy control,” J. Dyn. Syst., Meas., Contr., vol. 94, Series G, pp. 3–4, 1972. [3] , “The role of fuzzy logic in modeling and control,” Model., Identif. Contr., vol. 15, no. 3, pp. 191–203, 1994. [4] E. H. Mamdani and S. Assilian, “An experiment in linguistic synthesis with a fuzzy logic controller,” Int. J. Man-Mach. Stud., vol. 7, pp. 1–13, 1975. [5] L. Holmblad and I. I. Ostergaard, “Control of a cement kiln by fuzzy logic,” in Fuzzy Information and Decision-Processes, M. M. Gupta and E. Sanchez, Eds. Amsterdam, The Netherlands: North-Holland, 1982, pp. 389–399. [6] S. Yasunobu and S. Miyamoto, “Automatic train operation by predictive fuzzy control,” in Industry Applications of Fuzzy Control, M. Sugeno, Ed. Amsterdam, The Netherlands: North-Holland, 1985, pp. 1–18. [7] H. P. Preuß, “Fuzzy-control-heuristische Regelung mit unscharfer Logik,” Automatisierungstechnische Praxis, vol. 34, pp. 176–184, 1992. [8] D. Driankov, H. Hellendoorn, and M. Reinfrank, An Introduction to Fuzzy Control. Berlin, Germany: Springer-Verlag, 1993. [9] R. Isermann, K.-H. Lachmann, and D. Matko, Adaptive Control Systems. Hertfordshire, U.K.: Prentice-Hall, 1992. [10] G. N. Saridis, Self Organizing Control Of Stochastic Systems. New York: Marcel Dekker, 1977. [11] G. N. Saridis and K. P. Valavanis, “Analytical design of intelligent machines,” Automatica, vol. 24, pp. 123–133, 1988. ˚ om, “Intelligent control,” in Proc. European Control Conf., [12] K. J. Astr¨ Grenoble, France, 1991 [13] Handbook of Intelligent Control, D. A. White and D. A. Sofge, Eds. New York: Van Nostrand Reinhold, 1992. [14] P. Antaklis, Defining intelligent control. IEEE Trans. Contr. Syst. Technol., pp. 4–66, June 1994. [15] M. M. Gupta and N. K. Sinha, Intelligent Control Systems. New York: IEEE, 1996. [16] Advances in Intelligent Control, C. J. Harris, Ed., London, U.K.: Taylor and Francis, 1994. [17] R. Isermann, “Toward intelligent control of mechanical processes,” Contr. Eng. Practice, vol. 1, no. 2, pp. 233–252, 1993. [18] W. Pedrycz, Fuzzy Control And Fuzzy Systems, 2nd ed. New York: Wiley, 1993.

[19] R. Yager, Essentials of Fuzzy Modeling And Control. New York: Wiley, 1994. [20] B. Kosko, Neural Networks and Fuzzy Systems. London, U.K.: Prentice-Hall, 1992. [21] B.-M. Pfeiffer, Einsatz von Fuzzy-Logik in lernf¨ahigen digitalen Regelsystemen. Fortschr.-Ber. VDI-Reihe 8 Nr. 500. D¨usseldorf: VDI-Verlag, 1995. , Selbsteinstellende klassische Regler mit Fuzzy-Logik,” Automa[22] tisierungstechnik, vol. 42, pp. 69–73, 1994. [23] R. Isermann, Digital Control Systems. New York: Springer-Verlag, 1989, vol. 1–2. [24] B.-M. Pfeiffer and R. Isermann, “Fuzzy logic applications in learning digital control systems,” in Proc. 3rd European Congr. Fuzzy Intelligent Techniques Soft Computing, Aachen, Germany, 1995, pp. 847–852. [25] G. Bretthauer and H.-P. Opitz, “Stability of fuzzy systems—a survey,” in Proc. European Congr. Fuzzy Intelligent Technologies, Aachen, Germany, 1994, pp. 283–290. [26] J. P. Marin and A. Titli, “Comparative analysis of stability methods for fuzzy controllers,” in Proc. European Congr. Fuzzy Intelligent Technologies, Aachen, Germany, 1994, pp. 1183–1187. [27] W. J. Jansen et al., “Stability properties of fuzzy controllers,” J. A, vol. 36, no. 3, pp. 27–37, 1995. [28] H. Kiendl and J. J. R¨uger, “Stability of fuzzy control systems using facet functions,” Fuzzy Sets Syst., vol. 70, no. 2–3, pp. 275–285, 1995. [29] H. O. Wang, K. Tanaka, and M. Griffin, “An analytical framework of fuzzy modeling and control of nonlinear systems: Stability and design issues,” in Proc. American Control Conf., Evanston, IL, 1995, pp. 2272–2276. [30] T. A. Johansen, “Fuzzy model based control: Stability, robustness, and performance issues,” IEEE Trans. Fuzzy Syst., vol. 2, pp. 221–234, 1994. [31] S. Germann and R. Isermann, “Nonlinear distance and cruise control for passenger cars,” in Proc. IFAC-Workshop Advances Automotive Control, Ascona, Switzerland, Mar. 13–17, 1995, pp. 203–208. [32] R. Stenz and U. Kuhn, “Automation of a batch destillation column using fuzzy and conventional control,” IEEE Trans. Contr. Syst. Technol., vol. 3, no. 2, pp. 171–176, 1995. [33] B.-M. Pfeiffer and R. Isermann, “Selftuning of classical controllers with fuzzy logic,” Math. Comput. Simul., vol. 37, pp. 101–110, 1994. [34] L. X. Wang, Adaptive Fuzzy Systems and Control. Englewood Cliffs, NJ: Prentice-Hall, 1994. [35] R. Babuska, J. Sousa, and H. B. Verbruggen, “Model-based design of fuzzy control systems,” in Proc. 3rd European Congr. Fuzzy Intelligent Technologies, Aachen, Germany, 1995. [36] M. Fischer, O. Nelles, and D. F¨ussel, “Tuning of PID-controllers for nonlinear processes based on local linear fuzzy models,” in Proc. 4th European Congr. Fuzzy Intelligent Technologies, Aachen, Germany, 1996. [37] D. Pfannstiel, “Einsatz adaptiver und fuzzybasierter Regelungsstrategien in der Heizungstechnik,” Automatisierungstechnische Praxis, vol. 37, pp. 42–48, 1995. [38] A. Heinrich, J. Kraft, F. D¨orrscheidt, M. Boll, “Fuzzy-Regelung eines semiaktiven Pkw-Fahrwerks,” Automatisierungstechnische Praxis, vol. 36, pp. 23–30, 1995. [39] K. Pfeiffer and R. Isermann, “Automatic control of vehicle starting for emission tests on engine dynamometers,” in Proc. American Control Conf., Seattle, WA, 1995. [40] H. Takagi, “Fusion technology of fuzzy theory and neural networks—survey and future directions,” in Proc. Int. Conf. Fuzzy Logic Neural Networks, Iizuka, Japan, 1990, pp. 13–26. [41] M. Brown and C. Harris, Neurofuzzy Adaptive Modeling and Control. Hertfordshire, U.K., Prentice-Hall, 1994. [42] H.-P. Preuß and V. Tresp, “Neuro-fuzzy,” Automatisierungstechnische Praxis, vol. 36, no. 5, pp. 10–24, 1994. [43] J. S. R. Jang and C. T. Sun, “Functional equivalance between radial basis function networks and fuzzy inference systems,” IEEE Trans. Neural Networks, vol. 4, no. 1, pp. 156–158, 1993. [44] K. Hunt, R. Haas, and R. Murray-Smith, “Extending the functional equivalence of radial basis function networks and fuzzy inference systems,” IEEE Trans. Neural Networks, vol. 7, no. 3, pp. 776–781, 1996. [45] S. Halgamuge, Advanced Methods for Fusion of Fuzzy Systems and Neural Networks in Intelligent Data Processing, Fortschr.-Ber. VDI Reihe 10 Nr. 401. D¨usseldorf, VDI-Verlag, 1996. [46] J. Perkins, “Trends in process systems engineering,” Model., Identif. Control, vol. 15, no. 3, pp. 171–177, 1994. [47] R. Isermann, “Integration of fault detection and diagnosis methods,” in Proc. IFAC Symp. Fault Detection, Supervision, Safety for Processes, Espoo, Finland, June 13–16, 1994, pp. 597–612.

ISERMANN: FUZZY LOGIC APPLICATIONS

[48] [49]

[50] [51] [52] [53] [54] [55] [56] [57] [58]

[59] [60]

, “Model based fault detection and diagnosis methods,” in Proc. American Control Conf., Seattle, WA, June 21–23, 1995. R. Isermann and B. Freyermuth, “Process fault diagnosis based on process model knowledge. Part I: Principles for fault diagnosis with parameter estimation. Part II: Case study experiments,” J. Dyn. Syst., Meas., Contr., vol. 113, p. 627, 1992. R. Isermann, “On the integration of fault detection methods via unified fuzzy symptom representation,” 1st European Congr. Fuzzy Intelligent Technologies, Aachen, Germany, Sept. 7–10, 1993, pp. 400–407. W. S. Lee, D. L. Grosh, F. A. Tillmann, and C.H. Lie, “Fault tree analysis, methods and applications—a review,” IEEE Trans. Rel., vol. R-34, no. 3, pp. 194–202, 1985. R. Isermann and M. Ulieru, “Integrated fault detection and diagnosis,” in IEEE/SMC Conf. Systems Engineering Service Humans, Le Touquet, France, 1993. M. Ulieru, “Fuzzy logic in diagnostic decision: Possibilistic networks,” Ph.D. dissertation, Tech. Univ. Darmstadt, Darmstadt, Germany, 1996. M. Ayoubi, “Fuzzy systems design based on a hybrid neural structure and application to the fault diagnosis of technical processes,” Contr. Eng. Practice, vol. 4, no. 1, pp. 35–42, 1996. R. Isermann and M. Ayoubi, “Fault detection and diagnosis with neuro-fuzzy-systems,” in Proc. 4th European Congr. Fuzzy Intelligent Technologies, Aachen, Germany, 1996. H. Berenji, “Learning and tuning fuzzy logic controllers through reinforcements,” IEEE Trans. Neural Networks, vol. 3, no. 5, 1992. R. Kruse, J. Gebhardt, and F. Klawonn, Foundations of Fuzzy Systems. New York: Wiley, 1994. T. Pfeufer and M. Ayoubi, “Fault diagnosis of electro-mechanical actuators using a neuro-fuzzy network,” in Proc. 3rd Workshop FuzzyNeuro-Systeme ’95, Gesellschaft f¨ur Informatik, Darmstadt, Germany, 1995. H. Ushida, T. Yamaguchi, K. Goto and T. Takagi, “Fuzzy-neuro control using associative memories, and its applications,” Contr. Eng. Practice, vol. 2, no. 1, pp, 129–145, 1994. J. H. Kim, K. C. Kim, and E. Chong, “Fuzzy precompensated PIDcontrollers,” IEEE Trans. Contr. Syst. Technol., vol. 2, pp. 406–411, 1994.

235

[61] H. Kiendl, “Fuzzy-control,” Automatisierungstechnik, vol. 41, pp. A1–A8, 1993. [62] G. V. S. Raju and J. Zhow, “Adaptive hierarchical fuzzy controller,” IEEE Trans. Syst., Man, Cybern., vol. 24, p. 973, July/Aug. 1993. [63] H. J. Zimmermann, Fuzzy Set Theory and Its Applications, 2nd ed. Boston: Kluwer, 1991. [64] J. van Amerongen, H. R. van Nauta Lemke, and J. C. T. van der Veen, “An autopilot for ships designed with fuzzy sets,” in IFAC/IFIP Conf. Digital Comp. Applications. Amsterdam, The Netherlands: North-Holland, 1977.

Rolf Isermann (M’92) received Dipl.-Ing degree in mechanical engineering and the Dr.-Ing. degree, both from the University of Stuttgart, Stuttgart, Germany, in 1962 and 1965, respectively. He became Professor at the University of Stuttgart in 1972. Since 1977, he has been Head of the Laboratory of Control Engineering and Process Automation at the Institute of Automatic Control, Technical University of Darmstadt, Darmstadt, Germany. His main area of research interest covers process modeling and identification, digital control systems, adaptive control systems, fault diagnosis, and mechatronic systems. He has also written several books and papers on these topics. Dr. Isermann was the Chairman of the international program committees of several IFAC Symposia and of the 10th IFAC-World Congress on Automatic Control in Munich, Germany, in 1987. He was awarded by the Dr. h.c. degree from L’Universit´e Libre de Bruxelles, Bruxelles, Belgium, in 1989 and from the University of Bucharest, Bucharest, Romania, in 1996, and received the highest scientific award from VDE in 1996. Since 1996, he has been IFAC Vice President (Executive Board).