Re: MagLev System References

Marjan Golob (
Sat, 12 Jun 1999 02:01:19 +0200 (MET DST)

This is a multi-part message in MIME format.

Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Farooq Azam wrote:
> Hi,
> I am looking for references on magnetic levitation system (control of a suspended ball)
> modeling and control using Fuzzy logic or neural networks. I was only able to find
> Passino's paper in SMC Transactions on the subject so far.
> --
> Farooq-e-Azam

I recommend you to try:

Analysis and m-Based Controller Design for an Electromagnetic Suspension
System, J.-L. Lin and B.-C. Tho, IEEE Trans. on Education, Vol. 41, No.
2, May 1998, pp. 116-129.

Precision Motion Control of a Magnetic Suspension Actuator using a
Robust nonlinear Compensation Scheme, S. Mittal and C.-H. Menq,
IEEE/ASME Trans. on Mechatronics, Vol 2, No. 4, Dec. 1997, pp. 268-280.

GOLOB, Marjan. Decomposition of a fuzzy controller based on the
inference break-up method. In: R. Roy, T.Furuhashi, P. K.Chawdhry
(Eds). Advances in soft computing : engineering design and
manufacturing. London; Berlin; Heidelberg: Springer, 1998, cop. 1999,
pp. 215-227. On-line version is reachable on:

Best regards,

Marjan Golob

Content-Type: text/html; charset=iso-8859-1; name="IC-10.htm"
Content-Transfer-Encoding: 8bit
Content-Disposition: inline; filename="IC-10.htm"
Content-Base: "


Decomposition of a Fuzzy Controller Based on Inference Break-up Method

Decomposition of a Fuzzy Controller

Based on the Inference Break-up Method

Marjan Golob, Laboratory for Process Automation, Faculty of Electrical Engineering and Computer Science, University of Maribor, Smetanova 17, 2000 Maribor, SLOVENIA, Tel: +386 62 2207161, Fax: +386 62 211-178 , Email: , URL:


A concept called the decomposition of multivariable control rules is presented. Fuzzy control is the application of the compositional rule of inference and it is shown how the inference of the rule base with complex rules can be reduced to the inference of a number of rule bases with simple rules. A fuzzy logic based controller is applied to a simple magnetic suspension system. The controller has proportional, integral and derivative separate parts which are tuned independently. This means that all parts have their own rule bases. By testing it was formed out that the fuzzy PID controller gives better performance over a typical operational range then a traditional linear PID controller. The magnetic suspension system and the contact-less optical position measurement system have been developed and applied for the comparative analysis of the real-time conventional PID control and the fuzzy control.


1. Introduction

2. A Theoretical Approach to Fuzzy Control Based on the Inference Break-up Method

3. Fuzzy PID Controller Design

4. The Real-Time Application of the Fuzzy Control for a Magnetic Suspension system

5. Experiments and Results monstrations

6. Conclusions

On-line Demonstrations

1. Introduction

Fuzzy set theory was introduced by Zadeh [1] in 1965 and has been evolved as a powerful modelling tool that can cope with the uncertainties and non-linearity’s of modern control systems. Fuzzy controllers have become popular in recent years because they do not necessarily require a theoretical model of the plant which is to be controlled. Therefore, in order to develop a fuzzy controller, one needs to first have access to a human expert, find quantifiable means to present the expert’s experience, and determine a mapping from states of the plant to the fuzzy measures with which the expert’s knowledge is quantified. In [2] a supervisory fuzzy logic based controller is applied to the magnetic suspension system. The control architecture consists of two loops; a pole placement controller is utilised in the internal loop, and a supervisory fuzzy controller with a proportional-derivative structure is embedded in the other loop to enhance the transient response of the system. This control design is compared in simulation studies with a classical pole placement controller. Several magnetic suspension systems have been developed and applied for magnetically levitated transit systems by Japanese and American corporations during the last years [3], [4]. In most cases the control system and energy supply requirements to levitate the vehicle have a higher level of complexity. The non-linear nature of the system dynamics coupled with non-linear characteristics of the actuators complicate the controller design. The classical controller development approach relies on a linearization of the system dynamics and on the application of a PID controller to compensate the effects of the non-modelled non-linearity’s. By this approach certain system is stabilized close to its nominal operating point. Problems could be existent in the case when the set point is changeable within wide operating range. The method of robust stabilisation based on control theory was applied and tested in a laboratory prototype system. The results are presented by Kazuo et al. in [5]. A method called inference break-up is presented in this paper and used to model the multivariable fuzzy controller using the decomposition of the control rule base into a set of simple rule bases. We provide a comparative analysis of the conventional PID control and the fuzzy control for the magnetic suspension application. Our main objective is to make an initial assessment of what advantages a fuzzy control approach has over conventional control approaches. The structure of this paper is as follows. A theoretical approach to the fuzzy control based on inference break-up method is described in the next section. The development of the fuzzy controller is presented in Section 3. A real-time laboratory model with contact-less position measurement system is described in the section 4. Comparative analysis of the real-time conventional PID control and the fuzzy control are discussed in Section 5. The concluding remarks are given in Section 6.

2. A Theoretical Approach to Fuzzy Control Based on the Inference Break-up Method

The control algorithm is represented by fuzzy rules. A multivariable fuzzy system with three inputs and one output is considered. The linguistic description of the process is given by

where Xk(i) is the fuzzy set of the k-input variable defined in the universe of discourse Xk, k=1,2,3, and Y(i) is the output fuzzy set defined in the universe of discourse Y. m is the number of rules. The fuzzy relation R of the system is expressed as follows


where s the aggregation operator and is the implication operator. For each rule a fuzzy relation R(i) has to be constructed. To obtain the fuzzy controller relation R, fuzzy relations R(i) are aggregated. Dimension of relation matrix R is dim[R] = dim[X1]´dim[X2]´dim[X3]´dim[Y]. To obtain the new fuzzy output Y’, given the current fuzzy inputs X1, X2 and X3 the compositional rule of inference is used:


where  is the composition operator of fuzzy relations. Because of the multidimensionality of the fuzzy relation (2) the composition rule of inference (3) is difficult to perform or analytical solutions can usually not be obtained. To overcome this difficulty, it is proposed to breaks up the inference of a multidimensional rule base into the three rule bases of which the inference is easier to perform or has analytical solutions. Demirli and Turksen [6] showed that rules with two ore more independent variables in their premise can be simplified to a number of inferences of rule bases with simple rules (only one variable in their premise). Their proof is only valid for sub-min composition operator and the min operation for conjunction in the premise, and is based on the fact that S and R implications are non increasing with respect to their first argument. When the rule:

is considered and the implication operator min is used, the rule can be broken into a set of three rules:


The proof can be generalised to cases with more then three variables in the rule premise. Normally a rule base consist of more then just one rule. The inference of a rule base with more than one rule also can be broken up into the inference of a number of rule bases with simple rules. When considering a rule base with two rules, the rule base break up results in:


The result of the inference break-up can easily be extended to a large number of rules or more complex rules. The number of parts resulting from the inference break-up can become quite large (in the case of n variables in the premise of the m rules, the number of cross-products is nm), but many simplifications are possible. First simplification of the inference of a simple rule base can be achieved when the rule base can be divided into a set of simple rule bases which do not interact. The interaction of rule bases means overlap of the rules premises. When the premises of the rules are assumed to take fuzzy sets which form a fuzzy partition (fuzzy sets are convex and normalized) with no more then two overlapping fuzzy sets, then it can be derived thet the number of necessary simple rule bases is 2N-1, where N is the number of fuzzy sets defined on the domain of the output variable. At this point, a crucial simplifying argument is made. From all products over nm cross-products only three terms are selected. The inference (5) reduces to


Using the inclusion operator, the otput Y from inference (5) can be written as Y’ Y’s. The inclusion sign means that the output fuzzy set Y’ is contained in the set obtained by neglecting the terms in the inference (5). In many practical cases this neglected term may not be very significant. In the Section 3 the development of the simple fuzzy PID controller based on the clasic discrete PID controller is presented. The output of the the regular fuzzy controller is a nonlinaer function of three inputs (error, change-of-error and intgral-of-error). The brak-up method of regular fuzzy controller inference leads to similar result described by equation (5), with intersections of three basic simple inferences for each input and intersections of many simple cross-inferences. The simple fuzzy PID controller inference is obtained by neglecting all simple cross-inferences and the output Y’s is a intersection of outputs of three basic simple inferences. Comparing this result with the output of the discrete PID controller, where yPID is the linear algebraic equation, the discrete PID controller can be seen as a subset of the simple fuzzy controller. When designing a simple fuzzy controller and applying specific choices for the shape of membership functions, logical operators and the scalong of inputs and outputs, the simple fuzzy controller can amulate a linear control. Now define the fuzzy relation Rk from (5) as


and the following is obtained from (6):


To understand better the linguistic description (1) and its mathematical presentation (8), a block-diagram form is proposed in Figure 1.

Figure 1: Block diagram of multivariable fuzzy system

The multivariable structure of the fuzzy system which has n inputs contains n functional blocks and an intersection block. Advantages of the block diagram presentation of a multivariable fuzzy system are in the fact that this kind of presentation allows evaluation of the contribution of each component to the overall performance of the system. An interesting possibility concerning the inference break-up is a processor which performs composition of a fuzzy set and a two-dimensional fuzzy relation. A speed-up of the inference can be achieved by using optimised code to perform the composition of a fuzzy set and a fuzzy relation. Because of the inference break-up, the code optimisation can be applied to software implementation.

3. Fuzzy PID Controller Design

For a PID-like fuzzy controller the number of rules increases as the third power of the number of membership functions. In this section we would explain how to minimise the number of rules in the case of fuzzy PID controller.

The basic idea of the discrete PID-controller is to choose the control law by considering error e(kT), change-of-error de(kT):= (e(kT) - e((k-1)T))/T and the numerically approximated integral of error ie(kT):=ie((k-1)T)+Te((k-1)T)). The PID control law is


where KP is a proportional constant, KD is a differential constant and KI is a integral constant. For a linear process the control parameters KP, KD and KI are designed in such a way that the closed-loop control is stable. The corresponding analysis can be done by means of the knowledge of process parameters taking into account special performance criteria. In the case of non-linear processes which can be linearized around the operating point, conventional PID-controllers also work successfully. However, the PID-controller with constant parameters in the whole working area is robust but not optimal. In this case, tuning of the PID-parameters has to be performed.

Fuzzy PID controller start from the same assumptions which are decisive for the conventional PID-controller:

The output of the fuzzy controller u(kT) is given by


where N(x) is a non-linear function determined by fuzzy parameters.

In the case when we assume the structure of fuzzy PID controller with three input variables and the one output variable with five base fuzzy sets on each fuzzy variable, we get one rule base with maximum 125 rules. To minimise the number of rules the simplification of the fuzzy PID structure is proposed. The basic idea is to decompose of multivariable control rule base into three sets of one dimensional rule bases for each input. The output of the decomposed fuzzy controller u(kT) is given by


where NP(×), ND(×) and NI(×) are non-linear functions determined by three separated rule bases. The structure and principal components of the fuzzy logic based PID controller is shown in Figure 2.

Figure 2: The structure of the fuzzy logic based PID controller

The linguistic description of the knowledge base is given by three inference rules: The output signal is sum of defuzzified proportional, differential and integral action as given in equation (12):


4. The Real-Time Application of the Fuzzy Control for the Magnetic Suspension System

The functional diagram of the real-time application of the simple suspension system is shown in Figure 3.

Figure 3: Scheme of the real-time application of the simple suspension system

Our fuzzy logic controller based on the personal computer (PC) is extended by the OMRON FB-30AT fuzzy board (with the fuzzy processor FP 3000). An eight channels analog to digital (A/D) converter and two channels digital to analog (D/A) converter with 12 bit resolution is realised on the plug-in PC board. The first cannel of the A/D converter is used to measure input of the control system, namely: ball position which is output of the optical position measurement system. The fuzzy controller software was implemented in ANSI-C. The real-time sampling frequency of 1 kHz was attained. The realisation of the model is presented on figure 4.

Figure 4: The realisation of the laboratory model.

The electromagnet used in this study has 1200 coils of copper wire with diameter 1,5 mm. The ferromagnetic core has standard E form. Nominal parameters are: the coil resistance R is 2,8 , the electromagnet inductance L is 520mH and the maximum air gap D is 25mm. Electrical and mechanical time constants and gains can be calculated from mentioned nominal parameters and obtained from the measured static characteristic Fm = f(I,X). Nominal parameters, electrical and mechanical time constants and gains calculated with Equation (8) are shown in Table 1.

Table 1: Nominal system parameters

Mass of the steel ball m
0,147 kg
Maximum air gap D
Number of coils n
Coil resistance R
2,8 W
Electromagnet inductance L
520 mH
0,185 s
0,0505 s

The measurement of the ball position with the simple optical contact-less position measurement system is realised. The transmitter D1 (LED NLPB-500) with emission angle of 15 degrees is used to generate blue light. The homogenous light wave is provided by optical system of two optical lenses (L1, L2). The light detector (D2) provides an output voltage proportional to the position of the ball in the light wave. The room-light influence is compensated by differential structure of the light detector. By use of the 12 bits A/D converter the error between the set point and the current ball position the integer values from 0 to 4095 are mapped. The same normalisation procedure is used for the derivative input and for the integral input. Since ball position measurement is made in the fixed time frame the differential control (determines the rate of error) was implemented by simply subtracting the value of the current error from the previous error in ball position. All fuzzy inputs are divided into five ranges: Negative Large (NL), Negative Small (NS), Zero (ZE), Positive Small (PS) and Positive Large (PL). Triangular membership functions with 50% overlap are applied to inputs of the fuzzy controller, as shown in Figure 5.

a. .


Figure 5: Input and output membership functions

The output of the controller is the output voltage (ranges of 0 to 10 V) from the D/A converter. The final control law is a sum of proportional, differential and integral action. Singleton memberships functions are applied to fuzzy outputs, as shown in Figure 5.

The knowledge base is given by three inference rule bases: the proportional rule base, the differential rule base and the integral rule base. Five rules of the rule base for the proportional part of the fuzzy controller are described in table 2.

Table 2: The rule base of the proportional part of the fuzzy PID controller


The differential part and the integral part of the fuzzy logic controller are realised with the same type of rule base. The inference is performed using the minimum operator and the composition is done using the maximum operator (Mamdani type of the inference engine). Two defuzzification methods are implemented: Standard methods such as Mean of Maxim and Centre of Gravity (COG). The Center of Gravity defuzification method is used in experiments presented in the next Section.

5. Experiments and Results

The real-time implementation of the control for a magnetic suspension system allowed to perform several interesting experiments where the fuzzy controller proved to be very efficient. At first, the ball position control by the discrete PID controller with optimal parameters is considered. The optimisation of parameters for the set-point h=16 mm was done. The obtain performance (see Figure 6) of PID controller at high set-points (14 - 18 mm) was good and highly oscillatory at low set-points (4 - 8 mm).

Figure 6: The performance of the discrete PID controller

For a more serious comparison of the conventional discrete PID controller and the fuzzy PID controller, a discrete PID controller with gain scheduling was created. The gain schadulling use the set-point to choose suitable parameters which have been optimised at different set-points. The responses of the adaptive controller are shown in Figure 7.

Figure 7: The performance of the discrete PID controller with gain schedding

Results shown in Figure 8 prove that the fuzzy PID controller is similar to the discrete PID controller with gain scheduling in their performances.

Figure 8: The performance of the fuzzy PID controller

Differences between up-down step responses in Figures 5 to 9 are not evident. It is the reason that results have been compared by use of the performance index IAE and the performance index ISE. Results are presented in Table 3. The dis-crete PID controller has been optimized for the set point at 14 mm. Increased effect of the differential part of the PID controller become problematical in the case of greater changes of the set point. By changing of the set point at 6 mm parameters of the discrete PID controller become nonoptimal, which is expressed by greater overshoot and by oscillation.

Table 3: The IAE performance index and ISE performance index

Set Point
Per. Index
6 mm
Adaptive PID
Fuzzy PID
14 mm
Adaptive PID
Fuzzy PID

Performance indexes are satisfactory small in the case of the case of the set point 14 mm and much greater in the case of the set point of 6 mm. The adaptive discrete PID controller has also at the set point of 6 mm a nonoscilatory response and a great overshoot. IAE and ISE are suitable small. A step response of the fuzzy controller is similar to step response of the adaptive discrete PID control-ler. It should be mentioned that results of the simple fuzzy controller have been obtained by the basic shape of membership functions. It could be expected that fine tuning of membership functions will results in better performance (smaller overshoot). Fifteen rules in three rule basis have been used in the mentioned test. Note that the rule base can easily be extended with some multivariable control rules.

6. Conclusions

An approach of a fuzzy real-time controller based on the decomposition into the multivariable rule base is presented. The multidimensional fuzzy equation has been decomposed on the set of one-dimensional fuzzy equations. Calculated outputs of the decomposed fuzzy equations are a little more fuzzified then the output of the original multivariable fuzzy equation. This is due to the simplifi-cation of the multivariable structure and the lack of a good mapping property of the multivariable fuzzy equation. A loss of accuracy must be accepted.

The model of the magnetic suspension system has been realised by the simple electromagnet and the steel ball. It is evident from test results that the perfor-mance of the discrete PID controller depends heavily on the operating parame-ters of the system. The implementation of the gain scheduling algorithm in the PID structure causes a better performance. The same or better performance has been achieved with the fuzzy PID controller. Advantages of an simplified fuzzy PID structure are:

An interesting possibility concerning the inference break-up is a processor which performs composition of a fuzzy set and a two-dimensional fuzzy relation. A speed-up of the inference can be achieved by using optimised code to perform the composition of a fuzzy set and a fuzzy relation. Because of the inference break-up, the code optimisation can be applied to software implementation. Further researchs will be directed to the control of theoretic aspects of simplified fuzzy PID control in vague environments like stability analysis or robustness.


1 Zadeh L A 1965 Fuzzy sets. Information and Control 8:338-353.

2 Tzes A, Chen J C, Peng P Y 1994 Supervisory fuzzy control design for a magnetic suspension system. Proc. of the 3th IEEE Conf. on Fuzzy Systems, Vol 1, pp. 138-143.

3 Rule R , Gilliland R 1980 Combined magnetic levitation and propulsion: the mag-transit concept. IEEE Trans. On Vehicular Technology, Vol. VT-29, No. 1. pp. 41-49.

4 Ion B, Trica A, Papusoiu G, Nasar S A 1988 Field tests on a maglev with passive guideway linear inductor motor transportation system. IEEE Trans. On Vehicular Technology, Vol. VT-38, No. 4, pp. 230-236.

5 Kazuo S, Tadashi T, Hidenori K 1991 Robust stabilisation of a magnetic levitation system. Proc. of the 30th IEEE Conf. In Decision and Control, vol 3, pp. 2368-2371.

6 Demirli K, Turksen I B 1992 Rule break up with compositional rule of inference. Proc. of the 1th IEEE Conf. on Fuzzy Systems, San Diego, pp. 949-956.

On-line Demonstrations

A Slide-show presentation of a paper "Decomposition of a Fuzzy Controller Based on Inference Break-up Method":

One-line and Real Time control of a ball position in magnetic field:


This message was posted through the fuzzy mailing list.
(1) To subscribe to this mailing list, send a message body of
"SUB FUZZY-MAIL myFirstName mySurname" to
(2) To unsubscribe from this mailing list, send a message body of
(3) To reach the human who maintains the list, send mail to
(4) WWW access and other information on Fuzzy Sets and Logic see
(5) WWW archive: