Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering http://pii.sagepub.

com/

Analysis of the bond graph model of hybrid physical systems with ideal switches
J Buisson, H Cormerais and P-Y Richard Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering 2002 216: 47 DOI: 10.1243/0959651021541426 The online version of this article can be found at: http://pii.sagepub.com/content/216/1/47

Published by:
http://www.sagepublications.com

On behalf of:

Institution of Mechanical Engineers

Additional services and information for Proceedings of the Institution of Mechanical Engineers, Part I: Journal of Systems and Control Engineering can be found at: Email Alerts: http://pii.sagepub.com/cgi/alerts Subscriptions: http://pii.sagepub.com/subscriptions Reprints: http://www.sagepub.com/journalsReprints.nav Permissions: http://www.sagepub.com/journalsPermissions.nav Citations: http://pii.sagepub.com/content/216/1/47.refs.html

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

SPECIAL ISSUE PAPER

47

Analysis of the bond graph model of hybrid physical systems with ideal switches
J Buisson*, H Cormerais and P-Y Richard Supelec, Campus de Rennes, Cesson Sevigne, France

Abstract: This paper deals with the modelling of hybrid physical systems. The bond graph technique is used to establish their knowledge model, based upon an ideal representation of the switches. These components are modelled either by ow or by eVort sources according to their state and therefore modify the circuit topology at switching times. The paper shows the usefulness of the implicit representation to derive a unique implicit state equation with jumping parameters, to analyse the model properties, to derive an implicit state equation with nilpotency index one for each con guration and to compute the discontinuities. Also, a comparison between the chosen ideal modelling approach and the more common non-ideal approach is carried out using singular perturbations theory. After a presentation of the whole study in the most general context, its results are applied to power converters, which constitute a particular class of hybrid physical systems where switches only commutate in pairs. Finally, an example is developed. Keywords: bond graph, hybrid systems, switching systems, ideal switch, power converters

NOTATION C e f GY I MGY MTF p q R Se Sf Sw TF 0 1 bond graph capacitor element bond graph eVort variable bond graph ow variable bond graph gyrator element bond graph inertial element bond graph modulated gyrator element bond graph modulated transformer element bond graph momentum variable bond graph displacement variable bond graph resistor element bond graph eVort source element bond graph ow source element non-classical bond graph switch element bond graph transformer element bond graph common eVort junction bond graph common ow junction

INTRODUCTION

From a general point of view, physical systems can be considered as evolving either continuously or discontinuThe MS was received on 20 November 2000 and was accepted after revision for publication on 30 March 2001. * Corresponding author: Supelec, Campus de Rennes, avenue de la Boulaie, BP 28, F35511 Cesson Sevigne Cedex, France.
I08300 IMechE 2002

ously depending on the modelling approach. There exists, however, a particular class of dynamic systems in which some elements, generically called switches or switching components, evolve much faster than the time scale at which the system behaviour needs to be analysed [1]. Such systems are widespread in physical and engineering elds (electrical systems containing diodes, relays or transistors, hydraulic circuits including valves, and mechanical systems with clutches). It is this class of hybrid physical systems that will be studied in the present paper. The commutation of switches can be either controlled (relays) or autonomous (diodes). An automaton can be used to model the discrete control and the autonomous changes of switches. The location (or discrete state) of the automaton de nes the con guration of the switches. The continuous part of hybrid physical systems can be modelled using two main bond graph approaches: nonideal switches with constant circuit topology [2, 3] or ideal switches with variable circuit topology [412]. The rst approach uses non-linear R elements or modulated transformers to model the switching components and leads to a unique state equation with variable parameters. It can create diYcult problems in simulation and in control. As a result, numerical considerations rather than physical considerations often dictate the choice of the values attributed to those parameters. The second approach is based on the notion of ideal switches. It leads to one diVerent set of state equations
Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

48

J BUISSON, H CORMERAIS AND P-Y RICHARD

for each location of the automaton, as well as to jump functions which represent the discontinuities of the state when the location changes [1319]. This approach permits both a simpli ed analysis with ideal elements when considering the global performance of the system, and a more precise analysis if a more realistic model of the switching device is known. It uses either controlled junctions or sources to model switches. In the rst case, the controlled junctions, which are 0or 1-junctions according to the topology of the system, have two states called on and oV in order to model ideal switches [17]. In the state on, the controlled junctions behave like normal junctions whereas, in the state oV, they inhibit the energy transfer. A 0-controlled junction is replaced by a zero eVort source and a 1-controlled junction is replaced by a zero ow source on each of the adjoining bonds. In the second case, the sources standing for ideal switches also have two states: the rst state denoted by E where they behave like zero eVort sources and a second state denoted by F where they behave like zero ow sources [8]. Only the latter model is considered in the present paper. Consequently, in the following, the term switch will refer to the ideal model of the switching component such as that de ned above. The paper is organized as follows. In Section 2, the notion of implicit junction structure matrices for switching bond graphs is introduced. In Section 3, the state equation is derived from this representation. In Section 4, some properties of the model are studied. Section 5 proposes transition matrices to compute the new implicit junction structure matrices after a commutation. Section 6 is devoted to the calculus of the discontinuities that occur at commutation times. In Section 7, a comparison with the non-ideal approach is established using the singular perturbation theory. Section 8 analyses the case of power converters in which the switching components commutate in pairs. Finally, an example illustrates these diVerent points.

Fig. 1

A general bond graph

modi ed SCAP [21], in order to avoid unity gain causal loops, causality can be assigned for any allowed conguration of the switches. Moreover, it is assumed that, for any causal loop including transformers or gyrators, the values of those elements are such that the loop gain is never unity. Figure 2 represents the block diagram that is deduced from the causal bond graph. The following key variables are used: 1. The state vector X is composed of the energy varii ables in integral causality ( p on I elements and q on C elements), and the vector Z is composed of power i variables (e on C elements and f on I elements). 2. A so-called semistate vector X is composed of the d energy variables in derivative causality (p on I elements and q on C elements), and the vector Z is d composed of the corresponding power variables (e on C elements and f on I elements). 3. The vectors D and D represent the variables going i o out of and into the R eld, with D , D nr . i o 4. The vector U is composed of the variables imposed by sources, and V is de ned such that VTU is the power provided by sources. 5. T is composed of the variables imposed by switches i in the chosen con guration: the ow for the switches which operate in the F state and the eVort for the switches which operate in the E state. T is composed o of the variables imposed on switches: the eVort for the

AN IMPLICIT FORMULATION OF THE JUNCTION STRUCTURE RELATION

2.1 Block representation of a switched bond graph Figure 1 represents a general bond graph where a eld with the switching components Sw is distinguished. The convention for the direction of power is from the sources Se and Sf towards the junction structure and from the junction structure towards the I, C, R and Sw elements. Power ow into the switches is always 0, except possibly on commutations, as will be seen in Section 6. Then, with this convention, it becomes positive. A con guration of switches is allowed if it respects KirchhoV s laws. Using a Sequential Causality Assignment Procedure (SCAP) [20] or if necessary a
Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

Fig. 2

The information graph corresponding to the bond graph

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

49

switches which operate in the E state and the ow for the switches which operate in the F state T , T nSW . i o All the above vectors are organized so that the scalar product between the input and output vectors of each eld yields the power exchanged with the junction structure. When switches commutate, their causality changes and it is necessary to extend this change to other bonds. Some energy storage elements can lose or recover the integral causality and the state equation is changed. Consequently, the constitution of all the inputoutput vectors (U and V excepted ) depends upon the chosen con guration. 2.2 The standard implicit form Since there is no unity gain causal loop [21], each output of the junction structure (X , Z , D and T ) can be d o o i related to all its inputs (X , Z , D and T ) with the d i i i implicit equation [15]

guration, the elements of T corresponding to the i switches that have commutated will no longer be null but the corresponding elements of T will be null. New o inputoutput vectors can be de ned with the relation I T T i = i (3) I T T o o is a square diagonal matrix nSW nSW . Its diagonal elements are either 1 if they correspond to switches that have commutated, or 0 otherwise. If k is the number of switches that have commutated, the new con guration can also be de ned by k nSW which is extracted from by suppressing all its null rows. It is easy to verify that the following two properties hold:

A B A

BA B

T=I T = (4)

A B A
I S 0 0 0 0 0 12 ST 24

A B

S 0 11 ST I X i 12 = ST 0 X d 13 ST 0 14

13 0

0 0 I 0

S S S S

14 24 34 44

0 0 0 I

S S S S

15 25 35 45

33 ST 34

d D i D o T i T o U

(1)

In the following, this relation is called the standard implicit form. I and 0 denote structural identity and zero matrices; the other elements are composed of 0,1 and coeYcients of gyrators and transformers. The rst row of this relation de nes X , the second Z , the third D i d o and the last T . o S , S and S are skew symmetric. Those properties 11 33 44 are due to energy balance (there is no power stored or generated in the junction structure). Since the number of storage elements in integral causality is maximized with respect to the chosen con guration of switches, there is no relation between Z and D (or D and X ). d d i o Otherwise, a causal path between an element in derivative causality and a resistor port could be inverted, leading to one additional storage element in integral causality. For the same reason, there is no relation between X and Z . d d The output V on the sources is de ned by V=ST Z +ST X +ST D +ST T +S U (2) 35 i 15 i 25 d 45 i 55 This equation will only be used in Section 6 to compute the energy provided by sources. At this point, it is important to note that the implicit equation that was found for one con guration of the switches is true whatever the state of the switches is as long as no value aVects T or T . When the switches are i o in the chosen con guration, T =0. In any other coni
I08300 IMechE 2002

2.3 The reference con guration Some con gurations of the switches maximize the number of storage elements in integral causality. For all of them, S =0, which means that there is no causal 24 relation between a switch and an element in derivative causality. In order to nd such con gurations, a modied SCAP has to be applied considering switches as causally non-constraint elements. The rst stage consists in assigning the xed causality of sources and propagating it through the junction structure; this may de nitely impose a derivative causality to some storage elements. The second stage consists in assigning preferred integral causality to the remaining storage elements. At the end of this stage, there may be some of the latter in derivative causality again, but there exists a reversible causal path between each one and a storage element in integral causality. Thus the choice for the set of elements in derivative causality is not unique; only their number is an intrinsic property. Moreover, when inverting a causal path between two storage elements, causal constraints on adjacent bonds always remain unchanged, so that no other terminal element of the bond graph is aVected by causality changes. The result is that, for any choice of elements in derivative causality, the set of all the compatible con gurations of switches is the same. One
Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

50

J BUISSON, H CORMERAIS AND P-Y RICHARD

of these con gurations, associated with one arbitrary basic set of elements in derivative causality, will be called the reference con guration. It is postulated in the following that all the allowed con gurations can be reached from this reference con guration without changing the causality of the basic set of elements in derivative causality.

Otherwise, some storage elements lose their integral causality.

UNIQUE STATE SPACE EQUATION

2.4 The allowed con gurations Generally, not all of the con gurations of switches are allowed in a system. Some of them do not respect KirchhoV s laws. Using SCAP, those situations correspond to con icts at step one. Considering equation (1) established for a reference con guration, a new conguration de ned by equations (3) and (4) is valid if and only if it veri es the relation rank[ (ST ST S T)] =k 14 34 44 (5)

In the case of linear constitutive laws for the R, I and C elds, equation ( 1) allows an analytical determination of the state equation under an implicit form which is available in any allowed con guration. The third row of equation (1) permits the algebraic loop due to causality coupling between the resistive elements to be solved. This case occurs when S 0, which corresponds to a class-2 33 zero-order causal path [22]. Let the constitutive law for the R eld be D =LD (8) i o If H =L(I S L) 1 exists, which is particularly true 33 when L is a symmetric and positive de nite matrix, then the third row of equation (1) gives D =H(ST Z +S T +S U ) i 13 i 34 i 35 Eliminating D and D leads to the equation i o (9)

where k is the number of switches that have commutated. Proof. Considering the implicit standard form that corresponds to the reference mode, the fourth row of equation (1) (with S =0 de nes T : 24 o T =ST Z +ST D +ST T +S U o 14 i 34 i 44 i 45 (6) Once the con guration has changed, k variables in T o become inputs, and equation ( 6) which always holds must de ne k new output variables. Multiplying equation (6 ) by and reordering give

(ST ST S T) 14 34 44 A

AB
Z i D i T X i (7)

= T ST (I )T S U o 44 i 45 B

i S +S H S 15 13 35 Z d + S U 25 T i S ST HS 45 34 35 T o The constitutive relation for I and C written:

A BA B A ABA
I S 0 0 0 12 ST 24 X i X d = Z

S S H ST 0 11 13 13 ST I 12 ST +ST HST 0 14 34 13

S +S H S 14 13 34 S 24 S ST HS 44 34 34

0 0 I

(10)

elds can be

The right-hand term of this expression consists of the input T = 0 and ( I )T =0. Equation ( 7) is a o i linear system AX =B. The con guration is valid if k variables of X can be de ned, that is to say if equation (5 ) is veri ed. Otherwise, the new con guration is not allowed. This rank test speci es causality changes when the con guration is allowed. If rank (S T)= k, then 44 this equation allows the output T to be de ned, i the causality is only modi ed on switches. If rank ( S T) <k but rank [ (ST S T)] =k, then the 44 44 44 causality is changed on switches and resistive elements.
Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering

A B A
Z

F i = i FT Z d

F F d

BA B
X X i d

(11)

The square matrix involved in this relation is symmetric and positive de nite. In order to derive a state equation, the constitutive relation for the switches has to be added. According to equation (3), it can be written T =( I )T + T =0 i i o (12)

By adding equation (12) to equation ( 10) and using equation (11) in order to eliminate Z and Z , the followi d ing implicit state equation is obtained:
I08300 IMechE 2002

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

51

I 0 0 0

S 0

12

0 0 0 0

0 0 0 0

ST 24 0

BA B A

KF X i i ST F F T X d 12 i = (ST ST HST )F T i 14 34 13 i 0 T o

KF ST F F 12 d 0 0

S +S HS 14 13 24 S 24 S ST HS 14 34 34 I

0 0 I

S +S H S 15 13 35 S 25 + U S ST H S 45 34 35 0

where K=S S HST . This is a standard implicit 11 13 13 equation corresponding to the general form EX =AX+BU (14)

BA B
X X i d T i T o (13)

This model has been studied by several workers [2325]. It is an association of dynamic and algebraic modes and it can exhibit both impulsive and discontinuous phenomena at the origin. One of the main results is the following. If dim(X )=n, and rank( E )=q<n, then there are q dynamic modes and nq algebraic modes. Moreover, if rank

system is modelled by a unique state equation with a jump on a small part of its parameters ( ). However, its nilpotency index, as de ned in reference [24], is greater than one, which makes this form not convenient for simulation.

PROPERTIES OF THE SPACE MODEL IN ONE MODE

A B
E 0 A E

=n+q h

then among the dynamic modes there are q h nite modes and h in nite modes. It is usual, when analysing the properties of equation (14), to de ne equivalent forms by pre-multiplying it by a non-singular matrix P, and by operating a variable change Q in order to obtain a new equivalent implicit state equation: (PEQ) (Q 1X)=PAQ(Q1X ) +(PB)U (15) Using equation ( 15), two equivalent canonical forms can be de ned:

In this section, the system is assumed to be in one conguration, for which the standard implicit form (1) has been established. Then T =0 and equation (13) can be i simpli ed to give

I 0

S 0

12

BA B A

KF KF X i i = ST F F T ST FF X d 12 i 12 d S +S H S X i 15 13 35 U S X d 25 (18)

A BA

This equation veri es the property rank

A
and

qh 0

0 N

BA B A

A x 1 = 1 0 x 2

0 I n+ h q

BA B A B
x x

B 1 + 1 u B 2 2 (16)

A B
E 0 A E

=dim(X ) +rank ( E)

A A B x x 1 1 + 1 u 11 12 (17) = A 0 A B x x 2 2 21 22 2 where I denotes an identity matrix, q q q, and N is q a nilpotent matrix of index h. In the rst alternative, known as the Kronecker form, the subscript 1 subsystem corresponds to the nite dynamics while, in the second alternative, known as the Smith form, it corresponds to all the dynamics. When N=0, then h=0, so both forms are equivalent and there is no impulsive mode. Thus, as will be shown in Section 6, some entries of X and X may be discontinuous, while some entries i d of T and T may be discontinuous or even impulsive. i o Equation (13) is very synthetic; the commutating

A BA B A
I q 0 0

BA B A B

proving that there is no impulsive mode; state variables are not impulsive on commutations, which is coherent with physical principles; only a nite energy can be instantaneously transferred. To go further in the analysis of the implicit equation (18), it is pre-multiplied by the non singular matrix P=

K(F+F S ) i 12 I

BA

I 0

0 R

(19)

where R =(ST F S +FTS +ST F+F ) 1 exists due 12 i 12 12 12 d to F and F being positive. De ning also the variable i d change Q

A B A B
X X 1 = i X X 2 d

(20)

where Q=

I 0

12 I

BA

I R(ST F +FT) 12 i

0 I

(21)

I08300 IMechE 2002

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

52

J BUISSON, H CORMERAIS AND P-Y RICHARD

leads to the following implicit state equation:

A BA B A A
I 0 +

X 1 0 0 X 2 KF K(F +F S )R(ST F +FT) i i 12 12 i = 0

0 I

S +S H S +K(F +F S )RS 15 13 35 i 12 25 U RS 25

BA B B
X 1 X 2 (22)

D i , i I D D o o D i (25) I D o The causality change can also be de ned by a matrix V , which is built from in exactly the same way as was built from . Similar properties hold: VV T=I V TV = (26) The causality of some storage elements may also have changed. Some elements can have lost the integral causality but, according to the above postulate, it is possible to nd a causal assignment such that all the elements in derivative causality in the reference con guration keep their initial causality. Thus X can be split into two parts, namely X and i i X , which denote the variables that remain in the state id vector after the commutation and the variables that go into the semistate vector respectively. In a similar fashion, Z is divided into Z and Z . In that case, there id i i exist two matrices T and T such that 1 2

A B A A

I D i = D o I =

BA B A B BA B
D

Equation (22) is equivalent to an ordinary state equation: X =[KF K(F+F S )R( ST F +F T)]X 1 i i 12 12 i 1 +[S +S H S +K(F+F S )RS ]U (23) 15 13 35 i 12 25 where the state X is continuous at the origin, associated 1 with an algebraic equation X =RS U 2 25 5 RECOVERING THE IMPLICIT FORM (24)

5.1 General study The preceding section has shown the usefulness of the standard implicit form, which allows a tractable state equation to be derived for the corresponding con guration. This section examines how to recover the implicit form for any allowed con guration de ned by equations (3) and (4), given its knowledge in the reference conguration de ned by equation (1) with S =0. 24 The causality of some R elements may have changed at the commutation. In order to describe this change, a diagonal matrix is used. Each diagonal element of corresponds to a particular R element and is equal either to 1 if its causality has changed, or to 0 otherwise. This allows the new inputoutput vectors to be de ned:

A B A
X X i = d

1 0

T 2 0

0 I

[ T T ] is a permutation matrix whose dimension equals 1 2 the number of energy storage elements in integral causality in the reference con guration. T speci es which of 1 these elements keep their integral causality and T speci2 es which of them lose their integral causality. As the reference implicit form is still true after the commutation, substituting equations (3), (25) and (27) in equation (1) yields

AB
X i X id X d

(27)

T 1 0 0 0

T S 2 12 0 0 0 0 0 0

BA B
X i X id X d 0 I 0 0 S (I ) 13 0 S (I ) 33 ST (I ) 34 S 13 0 (I )+S 33 ST 34 S (I ) 14 0 S (I ) 34 S (I ) 44 S 13 0 S 34 (I )+S 44 S 15 S 25 S 35 S 45

S T S T 11 1 11 2 ST T ST T 12 1 12 2 = ST T ST T 13 1 13 2 ST T ST T 14 1 14 2

Z i Z id Z d D i D o T i T o U (28)

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

I08300 IMechE 2002

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

53

Now let

A B
X id X d be replaced by X , and d

A B
Z id Z d by Z . In order to recover the new implicit form ( 1), d both sides of equation (28) just have to be multiplied by the inverse of a composite matrix W, whose successive columns are left-hand side matrix, rst column, and the respective opposites of the right-hand side matrix, second, third, fth and seventh columns: W T S T 1 11 2 0 ST T 12 2 = 0 ST T 13 2 0 ST T 14 2

0 I 0 0

S 13 0 I S 33 ST 34

S 14 0 S 34 I S 44 (29)

neither implies any causal con ict, nor any unity gain causal loop. It can be seen that W depends on , , T and T . In 1 2 practice, only is known a priori, since it speci es the commutation. As a consequence, starting from the reference standard implicit form and given a matrix , the search for the new standard implicit form rst requires that of three matrices , T and T such that 1 2 W(, T , T ) be invertible. In fact, T and T are not 1 2 1 2 independent, as seen previously. Hence and T just 2 have to be determined under the constraint that the ranks of these matrices be minimal (in order to ensure consistency with SCAP). Once and T have been found, T 2 1 is straightforwardly deduced and the matrix relationships (3) and (25), which express all the causality changes resulting from the commutation, are fully determined. Substituting them in the reference standard implicit form and multiplying each side of the resulting equation by W1, the new standard implicit form is obtained. In the following, a formally simpli ed computation of W 1 is presented. Then a systematic procedure that allows determination of matrices and T (and conse2 quently W ) is elaborated. 5.2 Formally simplifying the inversion of W The matrix W can be decomposed the following way:

Note that the existence of W1 is assured as soon as the new con guration resulting from the commutation

W=

T 1 0 0 0

T 2 0 0 0 A

0 I 0 0

0 0 I 0

0 0 0 I

BA

I+S

11 ST 12 ST 13 ST 14

0 0 0 0

13 0

S S

14 0

I+S

33 ST 34

34 I+S B

44

BA B A
T 2 0 0 0 0 0 0 VT 0 0 0 T TT S T 1 14 TT S T 2 14 0 S T 34 (I +S ) T 44

0 0 0

I 0 0

0 0 0 D

0 V 0

0 0

B
(30) (31)

The inversion lemma (ABCD) 1=A 1+A 1B(C1DA 1B)1DA1 will be used on this decomposition with C=I. It follows that

TT 1 TT 2 W 1= 0 0 0

A BA
0 I 0 0 0 0 0 0 I 0 0

TT (I +S )T 1 11 2 0 TT (I +S )T 11 2 2 0 + ST T 12 2 0 ST T 13 2 I ST T 14 2 0 S 11 S 13 S 33 ST 34

TT S V T 1 13 TT S V T 2 13 0 ( I+S ) V T 33 ST V T 34 S 14 S 34 S 44

I08300 IMechE 2002

CA BA
TT 2 0 0 0 V 0 0

ST 13 ST 14

BA

2 0 0

0 VT 0

0 0 T

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering

BD A

1 TT 2 0 0

0 0 0

0 V 0

(32)

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

54

J BUISSON, H CORMERAIS AND P-Y RICHARD

Let V be the matrix de ned by TT 2 V= 0 0

matrix V only depends upon 13 S S S 14

The inversion of W is obviously equivalent to that of V. Now, comparing equations (29) and (31), it can be observed that the dimension of the matrix to be inversed has signi cantly decreased. Indeed, the dimension of the new matrix equals the number of switches that commutate, plus the number of resistances and storage elements that change their causality, whereas the dimension of W equals the total number of switches, resistances and storage elements.

A BA A B
0 V 0 0 0 T 2 0 0 0 VT 0 0 0 T

ST 13 ST 14

11

S S

33 ST 34

34

44

B
(33)

(ST ST S T) 14 34 44 which is coherent with the result of Section 2.

DISCONTINUITIES ON THE VARIABLES AT THE COMMUTATION

6.1 Discontinuity of the state variables At the commutation time, the discontinuity of the eVort or the ow associated with switches implies that some state variables will be discontinuous [1319]. Equations (22) and (23) allow these discontinuities to be computed. The value of the state vector just after the commutation (denoted +) is derived from its value just before (denoted ) by the relations X+ =X 1 1 and (36)

5.3 An algorithm for the computation of matrices and T 2 Resulting from the above development, the search for and T such that W is invertible is equivalent to the 2 search for V and T such that V is invertible. In order 2 to ensure consistency with SCAP, the proposed method consists of three stages: 1. If S T is a full-rank matrix, then 44 V= S T (34) 44 V and T are empty matrices and T =I. This means 1 2 that the commutation does not causally aVect any element in the bond graph model but the involved switches (which therefore necessarily commutate by pairs). 2. In return, if S T is a singular matrix, a second 44 stage has to be carried out. A minimal rank V matrix is iteratively searched such that the rank of V as de ned by equation (33) either be full or at least maximal: V S V T V S T 33 34 (35) ST V T CS T 34 44 If the full rank of V has been reached, then T is an 2 empty matrix and T =I. This means that the com1 mutation necessarily aVects some resistances, but no energy storage element. This situation corresponds to the case where rank [ ( ST S T)] =k. 34 44 3. If the full rank of V has not been reached, keeping V as built previously, a minimal rank T matrix is 2 iteratively searched such that matrix V as de ned by equation (33) is invertible. It means that some storage elements will lose their integral causality. If the choice of V and T is made that way, the invertibility of 2 V=

(37) X+ =RS U 2 25 Using the variable change de ned by equation (20), it is easy to compute the discontinuity of the original state variables: X+ =X +S (X+ X ) d i i 12 d and X+ =R[(ST F +F T) (S X X ) +S U ] d 12 i 12 d i 25 6.2 Amplitude of the pulses at the commutation Another equivalent approach to compute what happens at a commutation uses the standard implicit form (1) de ned for the con guration after the commutation. In this equation, only T can be impulsive at the commu0 tation: T =T (t t ). A switch which does not commuo c tate may have an impulsive ow if it operates in the E state (then its eVort remains null ) or an impulsive eVort if it operates in the F state (then its ow remains null ) and a switch which commutates may have an impulsive ow if it turns to the E state (then its eVort becomes null ) or an impulsive eVort if it turns to the F state (then its ow becomes null ). Indeed, the integration of the rst and fourth rows of equation (1) at the commutation time yields 0 X+ X i i = (40) 0 ST T X+ X d d 24 The rst row of the latter relation is identical with equation (38) whereas the second row gives the amplitude of the pulses in the switches: I S 12 T = ST (X+ X ) (41) d 24 d By another way, using the second row of equation (1)
I08300 IMechE 2002

(38)

(39)

BA

B A B

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

55

together with the constitutive relationships and equation (40), equation ( 39) is recovered. 6.3 Energy transfer at the commutation Since some variables may be discontinuous or impulsive at the commutation, some energy is instantaneously transferred between sources, storage elements and switches [26 ]. The energy transferred to the storage elements is E = 1 (Z+ TX+ Z TX) 1 2 which can be expressed as (42)

approach is proposed. It will be established, using the singular perturbation theory [2730], that the ideal approach can be interpreted as a limit case of this nonideal modelling [12, 31]. In order to simplify the presentation, this comparison will be made under restrictive assumptions: 1. Only one commutation ( possibly with several switches commutating simultaneously) is considered. 2. In the initial mode, all the storage elements have integral causality and the power variables imposed by the switches are identically null, for both modelling approaches. Consequently, using either the ideal approach or the semi-ideal approach results in exactly the same model formulation for this mode. 3. The simultaneous commutation of all the switches from their initial position makes many storage elements lose integral causality when the ideal approach is used. It implies that no causal path exists between switches or between switches and resistive elements. Since no causal path exists between switches, there can be no causal relation between T i and T , hence S =0. o 44 7.1 State equation in the initial mode According to the previous hypothesis, X can be split i into two parts, namely X and X , which denote the ii id state variables that remain after the commutation and those that disappear when the ideal modelling approach is used respectively. In a similar fashion, Z is divided i into Z and Z . Thus the following is obtained for the ii id initial con guration: Z s s 12 0 s 13 0 0 I 0 s s 14 0 0 0 I S S S S 15

E = 1(ZT S +Z T +UTST ) (X+ X ) (43) d d 1 2 i 12 d 25 As T is impulsive at the commutation [(T =T(tt )], o o c and since T+ =0, the energy dissipated into the switches i is E = 1 TT(T+ +T ) = 1 TTT 2 2 i i i 2 = 1 (X+ X )TS T d 24 i 2 d The second row in equation (1) gives S T =ST Z +Z S U 24 d 12 i d 25 Using equation (39)

(44)

(45)

E = 1(ST Z +Z S U )T(X+ X ) d d 2 d 25 2 12 i = 1 (ST Z +Z S U )TR(ST Z +Z S U ) d 25 12 i d 25 2 12 i "0 (46) Using the output relation (2), V can be impulsive on commutation:

V=ST (X+ X ) (47) d 25 d and; supposing that U is continuous, the energy provided by sources is E =(X+ X )TS U d d 25 3 And the energy balance is veri ed since E +E =E 1 2 3 7 COMPARISON WITH NON-IDEAL APPROACH (48)

AB A
I 0 I 0 0 0 0 0

A B

X ii X id

11 sT 12 = sT 13 sT 14

sT 23 sT 24

s 23 S 33 0

24 0

25 35 45

ii

id D i D o T i T o U

(50)

(49)

As formulated in the introduction, one of the usual approaches in dealing with switches consists in using variable resistive elements. In that section, a comparison between the ideal modelling and the non-ideal resistive
I08300 IMechE 2002

The right-hand side of this new standard implicit form exhibits a matrix with a structural zero block on its second column, which shows that there is no causal relation between X and Z . This property comes from id id the fact that no causal path can exist between two storage elements that lose the integral causality after the commutation. Otherwise, they should recover this causality by inverting their mutual causal path. The reference con guration is thus characterized by the state space equation s s HsT s s H sT X Z ii = ii 11 13 13 12 13 23 X sT s HsT s HsT Z id id 12 23 13 23 23 s +s H S 35 U + 15 13 (51) s +s H S 25 23 35

A B A

BA B

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

56

J BUISSON, H CORMERAIS AND P-Y RICHARD

It must be completed by the constitutive relationship that describes the storage eld behaviour. Since linearity is assumed, this relationship can be written in matrix form, the latter being symmetric positive de nite, as F F Z X ii ii = ii (52) FT F Z X id id id In further development, the situation after the simultaneous commutation of all the switches will be studied, considering successively their ideal modelling approach and their semi-ideal approach. 7.2 Ideal approach Results of Section 5 allow to derive implicit state space equations of the system after the commutation to be derived: I s s 1 X ii 14 24 0 0 X id I s s 1 0 14 24 = 0 0 I

7.3 Semi-ideal approach In this alternative approach, switches are modelled as variable resistances. Their constitutional relationships can be expressed in a global way as T =E1T , where i o E1 denotes a diagonal matrix whose elements are either resistances or conductances, depending on the causality of individual switches. Also, this causality does not change any longer but always remains the same as in the reference con guration. The inverse notation may seem quite arti cial. In fact, it is only introduced for further consistency with singular perturbation formalism. It is now assumed that all the parameters of the switches (either resistances or conductances) have the same positive value denoted by 1, which can be written 1 T= T i o (56)

A B A

BA B

BA B

s +s HS 15 13 35 + s +s H S U (53) 25 23 35 S 45 As in Section 4, now consider the change in variables de ned by

CA

s s HsT 11 13 13 sT s H sT 12 23 13 sT 14

BD
id

s s H sT 12 13 23 s H sT 23 23 sT 24

A B
Z Z ii id

Since T 0 in the reference con guration, then i !+?. The corresponding state space representation is the same as in the ideal approach. It is given by equation (53). After the commutation on the other hand, will take a very small but non zero value (otherwise the approach would be equivalent to the ideal approach). By substitution of equation (56) in equation (50),

A B A

A B A A B A
X1 X Z 2 =

I 0 I

s s 1 14 24 I s s 1 14 24 I

1 = 0 Z 2

BA B BA B
Xii X Z id Zii

s s HsT s s H sT X ii = 11 13 13 12 13 23 sT s HsT s HsT X id 12 23 13 23 23 s +s H S 1 s 35 U+ 14 + 15 13 s s +s H S 25 23 35 24 Z ii +S U (sT sT ) 14 24 Z 45 id

BA B
Z Z ii id (57)

A C

B A B A B D

7.4 Analysis using the singular perturbation method (54) This section analyses the model obtained with the nonideal approach using singular perturbation theory. Using the same change in variables (54) as in the ideal approach in equation (57) yields the equivalent system [25]

Applying this in equation (53) gives I 0 X 1 0 0 X 2 I s s 1 s +s H S 0 14 24 15 13 35 + = S U 45 0 0 s +s HS I 25 23 35 I s s1 14 24 + 0 0 s s H sT s s HsT 11 13 13 12 13 23 sT s H sT s HsT 12 23 13 23 23 0 I s s 1 Z 14 24 1 (sT sT ) (55) 14 24 I 0 0 Z 2 This implicit equation is expressed in the Smith form.
Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering

A BA B CA BA CA B A AB DA

B AB D B BA B

s +s HS 0 15 13 35 + S U 45 s +s HS s 25 23 35 24 I s s1 s s HsT s s HsT 14 24 11 13 13 12 13 23 + I 0 sT s H sT s HsT 12 23 13 23 23 0 I s s1 Z 1 14 24 (sT sT ) (58) 14 24 s 0 I Z 24 2 Now Z and Z are algebraic functions of (X , X ), the 1 2 1 2 = I 0

A B CA CA

X 1 X 2

s s1 14 24 I

BA BA

B A B D BA B

A B

DA

I08300 IMechE 2002

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

57

expressions of which are given in the linear case by

id X 1 (59) 0 X 2 Hence equations (58) and (59) de ne a singularly perturbed model [28], which can be formulated in the most general case as X (t)= f (X (t), X (t), U(t), ) 1 1 2 1 X (t) = f (X (t), X (t), U(t), ) 2 2 1 2 (60)

A B A
Z Z

I 1 = 0 2

s s1 14 24 I

s s1 14 24 I

BA B BA B
F ii FT F F

the initial condition on X and X is given. In the new 1 2 time scale, the slow variables X and X behave like con1 2 stants. Then it can be shown [28] that X satis es the 2 boundary layer equation 14 s 24 id s1 X ()+s S [U()U(t )] (64) 0 24 2 24 45 This equation shows that the boundary-layer solution is stable, since the matrix that characterizes the storage eld constitutive relationship is positive de nite. In such conditions, Tikhonovs theorem [30] can be applied. It states that the slow solution of the singular perturbation problem makes a good approximation of the general problem, insofar as the latter uniformly converges towards the former. F F dX ii 2 ()= s (sT sT ) 24 14 24 FT d

BA B
s

Setting =0, the associated reduced model, whose solution is denoted by (X , X ) is obtained. The bar on the 1 2 variables indicates a slow motion representation. In the particular present case, the reduced model is X 1 0 =

A B CA CA
I 0 +

THE CASE OF POWER CONVERTERS

s s 1 14 24 0 I

s s1 14 24 0

BA B

s +s HS 0 15 13 35 + S U 45 s +s HS s 25 23 35 24

B A B D B BA B

s s H sT s s HsT 11 13 13 12 13 23 sT s H sT s HsT 12 23 13 23 23 0 I s s1 Z 1 14 24 (sT sT ) (61) 14 24 s 0 I Z 24 2 Using its second row, which corresponds to the algebraic relation, Z can be expressed as an explicit function of 2 Z and U, namely 1 (62) Z = s (sT s +sT s )1(sT Z S U) 2 24 14 14 24 24 14 1 45 Substituting equation (59) in equation (62), an equivalent explicit expression for X in terms for X and U is derived. 2 1 After some algebraic manipulations,

A A B

DA

In this section, the emphasis is put on a particular class of switching systems, namely static converters, which are extensively used in power electronics. Generally, in normal operating conditions, the commutations that aVect such systems only imply causality changes on switches. This means that an even number of switches commutate simultaneously and that each allowed conguration is a reference con guration. As a consequence, it will be seen that a unique standard implicit form can be established, which is parametrized according to the mode. Then from this form a unique state equation can be derived which is parametrized in the same way.

8.1 Determination of the diVerent standard implicit forms If the standard implicit equation (1) has been established in one of the allowed modes, it is easy to derive the standard implicit equation for all the other con gurations using the results of Section 5. Since the causality changes only involve switches, the expression for W in equation (27) can be simpli ed to

F ii (sT sT ) X =s 2 24 14 24 FT

F ii X S U (sT sT ) 1 45 14 24 FT

A A B

F F

id

BA BD D
s s 14 24

1 (63)

The result is that the singularly perturbed model de ned by equations (58) and (59) is put in standard form [30]. At this stage, an essential remark has to be made; multiplying the second row of equation (63) by s1 , the Smith 24 form already obtained in the ideal case is recovered. It shows that the slow solution exactly coincides with the solution of the ideal approach. Considering again the general system (58), the fast vari able X =X X is introduced. The time scale is also 2 2 2 changed by =(tt )/, where t is the time at which 0 0
I08300 IMechE 2002

W=

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

A A

I 0 0 0 I 0 0 0

0 I 0 0 0 I 0 0

0 0 I 0 0 0 I 0

S 1 14 0 S 34 S 1 14 0 S 1 34 1

(65)

58

J BUISSON, H CORMERAIS AND P-Y RICHARD

where =I (I +S ) 44 and 1=I (I +S ) T(S T)1 44 44 The general standard implicit form is

Simpler expressions for 1 and 1 are given by 1=rS1 +( 1r)I (69) 44 1= rS 1 (70) 44 The unique state equation depends on only one scalar Boolean parameter. Also, if the sequence of modes is Z i Z d D i D o T ( ) i T ( ) o U

A B A
I S 0 0 0 0 0 0 12

A B

S ( ) 11 ST X i = 12 ST ( ) X d 13 ST ( ) 14

0 I 0 0

S ( ) 13 0 S ( ) 33 ST ( ) 34

0 0 I 0

S ( ) 14 0 S ( ) 34 S ( ) 44

0 0 0 I

S ( ) 15 S 25 S ( ) 35 S ( ) 45

(66)

where the displayed submatrices are de ned as follows: S ( )=S S 1ST 11 11 14 14 S ( )=S S 1ST 13 13 14 34 S ( )=S T 14 14 S ( )=S +S 1S 15 15 14 45 )=S +S 1ST S ( 33 33 34 34 S ( )=S T 34 34 S ( )=S +S 1S 45 35 35 34 S ( )= 1[S (I ) ] 44 44 S ( )= 1S 45 45 8.2 Determination of the implicit state equation Using the results of Section 4, equation (15) leads to a unique implicit state equation depending on : periodic with a very small period with respect to the time constants of the system, an average model can be established. If r=1 for a duration 01 of this period, then the average model will be obtained by setting 1= S 1 (1 )I 44 1= S1 44 9 EXAMPLE (71) (72)

The purpose of this section is to apply the previous results on an example. It consists of a d.c. motor powered by a boost converter and driving a load by means of a clutch such as that depicted in Fig. 3. The motor is modelled by a gyrator with a coeYcient a that operates

X [S ( )S ( )HST ()]F [S ( ) S ( )H ST ( )]F 13 13 11 i 13 13 i 11 = ST F FT ST F F 0 0 X 12 i 12 d d S ( )S ( )HS ( ) 35 13 + 15 U S 25 where H=L[I S ( )L]1. 33 Using both equation (38) and the second row of equation (66), which does not depend on the mode, it can be veri ed that the state vector is continuous on commutations. In the particular case where only two modes are allowed, non-commutating switches, if any, are eliminated from the model; since the values of the corresponding entries in T are always null, the associated columns i and rows in the matrices of equation (1) are cancelled. Then S is invertible and can be written 44 =rI (68)

12

BA B A

BA B
X X i d (67)

with r=0 or 1.
Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering

Fig. 3

A boost converter supplying a d.c. motor with load

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

59

the electromechanical conversion associated with a mechanical inertia I and an electrical resistance R . The 3 5 power converter is constituted of two switches T and 7 T associated with a voltage source E and an inductance 8 L . The load is an inertia I associated with a viscous 2 4 friction R . Finally, the clutch is modelled as a mechan6 ical switch T . 9 Figure 4 represents a bond graph for that system. The causality has been assigned with T in the F state, and 8 T and T in the E state (the load is disconnected from 7 9 the motor), which de nes a reference con guration for the system.

where =1 if T has commutated with respect to coni i guration C . The result is that six con gurations are 0 allowed: =0, =0, =1 7 7 7 C : =0, C : =1 C : =0, 8 1 8 2 8 0 =0, =1, =0 9 9 9 =1, =1, =1 7 7 7 C : =1, C : =0, C : =0 3 8 8 8 4 5 =1, =0, =1 9 9 9 but only four of them, C to C , correspond to normal 0 3 operating conditions for the system, and two of them, C and C , are reference con gurations. 0 2

G G

G G

G G

9.1 Determination of the standard implicit form The standard implicit form in that reference con guration (denoted C ) yields 0

f f f

2 3

1 0 0 0 0 0 0 0

0 1 0 0 0 0 0 0

0 0 1 0 0 0 0 0

0 a

a 0 0 0 0 0 0 1

0 0 0 0 1 0 0 1

1 0 0 0 0 0 0 0

0 0 1 0 0 0 0 0

0 0 0 1 0 0 0 0

0 0 0 0 1 0 0 0

1 0 0 0 0 0 1 0

0 0 0 0 0 1 0 0

0 1 1 0 0 0 0 0

0 0 0 0 0 1 0 0

0 0 0 0 0 0 1 0

0 0 0 0 0 0 0 1

0 0 0 0 0 0 1 0

AB

p 2 p = 3 p 4

0 1 0 1 0 0

4 e 5 e 6 f 5 f 6 e 7 f 8 e 9 f 7 e 8 f 9 e 1

(73)

9.2 Determination of the allowed con gurations The results of Section 2 are applied to determine the allowed modes for the present system. Equation ( 5) gives 7 0 rank 0

0 0 9

0 0 9

0 0 0

0 0 0

0 7 8 0

7 8 0 0

7 0 =rank 0 0 0 0

B A

0 8 0

0 0 9

(74)

Fig. 4
I08300 IMechE 2002

A causal bond graph in a reference con guration

60

J BUISSON, H CORMERAIS AND P-Y RICHARD

9.3 Recovering the implicit form Consider the transition from con guration C to con0 guration C . By applying equation (30), 1 TT 2 V= 0 0

The state equation in that con guration is deduced in a straightforward fashion:

B A BA BA
0 V 0 0 0 (0 0 1) a 0 0 0 0 a 0 1 0 0 0 0 1 0 0

1 0 0 0 0

0 0 0 0 0

0 1 1 0 0 0

In order to make this matrix invertible, V must be chosen empty and T =(0 0 1)T; then 2 V=

A A BB
T 2 0 0 0 VT 0 0 0 1 0 0 1 0

A BA BA B
1 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 0 0 0 0 0 1 1

1 0

B A BA
0 0 0 0

B
(75)

A BA B
1 0 0 0 1 0 0 p 1 0 p 3 p 4 2 1 0 0 1 0

R 5 L 2 a L 2 0

a L 3 R 6 L 3 1 L 3

0 0 1 I 4

AB
p 2 p 3 p 4 (77)

9.4 Discontinuities on the variables at the commutation When the system is in con guration C , for instance, the 1 state variables are de ned by equation (40). In the present case, the following expression is obtained: p+ p 0 2 2 p+ p = 0 (78) 3 3 p+ p T 0 0 4 4 Using the third row of equation (76) and constitutive laws, 1 1 p+ =p 2 2 (79)

A BA B A B
A A B B

0 1

L 3 p+ = ( p +p ) 3 3 4 L +I 3 4 I 4 p+ = ( p +p ) 4 3 4 L +I 3 4 9.5 Determination of the average model

(80)

(81)

The matrix is determined using equation (29). Multiplying both sides of equation (73) by the matrix W1, the standard implicit form in the new con guration is obtained:

Assume some operating conditions where the only available modes are C (r=0) and C (r =1). This implies 1 3 f f f 2 3

1 0 0 0 0 0 0 0

0 1 0 0 0 0 0 0

0 1 0 0 0 0 0 1

0 a

a 0 1 0 1 0 0 0

0 0 1 0 0 0 0 0

1 0 0 0 0 0 0 0

0 1 0 0 0 0 0 1

0 0 0 1 0 0 0 0

0 0 0 0 1 0 0 0

1 0 0 0 0 0 1 0

0 0 0 0 0 1 0 0

0 0 1 0 1 0 0 0

0 0 0 0 0 1 0 0

0 0 0 0 0 0 1 0

0 0 0 0 0 0 0 1

0 0 0 0 0 0 1 0

AB

p 2 p = 3 p 4

0 1 0 1 0 0

4 e 5 e 6 f 5 f 6 e 7 f 8 f 9 f 7 e 8 e 9 e 1

(76)

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

I08300 IMechE 2002

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

61

that the load always remains connected, whereas the power converter behaves in a normal fashion. Ignoring switch T leads to the simpli ed standard implicit form 9 f f 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 a 0 1 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 f 2 3

AB

a 0 1 0 1 0

p 2 p = 3 p 4

4 e 5 e 6 f 5 f 6 e 7 f 8 f 7 e 8 e 1

(82)

where S is invertible. Then, applying the results from Section 9, a general standard implicit form [such as in equation 44 (66)] is obtained: f 2 f 3 f 4 e 5 e 6 f 5 f 6 T (r) i T (r) o e 1 (83)

1 0 0 0 0 0 0

0 1 0 0 0 0 0

0 1 0 0 0 0 0

a 0 1 0 1 0 0

0 0 1 0 0 0 0

1 0 0 0 0 0 0

0 1 0 0 0 0 0

0 0 0 1 0 0 0

0 0 0 0 1 0 0

1r 0 0 0 0 0 2r 1

r 0 0 0 0 12r 0

0 0 0 0 0 1 0

0 0 0 0 0 0 1

r 0 0 0 0 r 1r

AB

a 0 1 0 r 1 r

p 2 p = 3 p 4

where (1 r)e +rf 7 7 =0 T (r) = i (1r) f +re 8 8 ( 1r) f +re 7 7 T (r)= o (1r)e +rf 8 8

A A
0 1 0

B B

Then the parametric model for both modes can be deduced: R 5 L 2 a L 2 0 a L 3 R 6 L 3 1 L 3 0 0 1 I 4

A BA B
1 0 0 0 1 0
I08300 IMechE 2002

p 2 p = 3 p 4

A B AB

p r 2 p + 0 E 1 3 p 0 4

(84)

If the system periodically commutates between modes C (r =0) and C (r =1), at a high frequency with respect to 1 3 the time constants of the system, with a duty cycle , then an average model can be deduced by formally replacing r by in equation ( 84).
Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

62

J BUISSON, H CORMERAIS AND P-Y RICHARD

10

CONCLUSION

As reported in reference [32], hybrid systems modelling always implies a trade-oV between conceptual and computational complexity. This paper discusses some properties of ideal models for the continuous part of hybrid linear physical systems using the notion of an implicit junction structure relation. Although such an approach increases conceptual complexity, some solutions are proposed in order to deal with it. A unique implicit state equation with jumping parameters is rst established, which constitutes a very synthetic representation of the global discontinuous behaviour but is not convenient for simulation. Then, it is demonstrated that the system can be modelled by non-impulsive implicit linear state equations in each con guration. An explicit solution is proposed in order to compute variables at the commutation as well as the weight of the pulses, which allows the energy instantaneously dissipated in the switches to be determined. A solution is also proposed to compute the implicit junction structure and then the implicit state equation for each mode. Those two last points yield an alternative for the simulation of hybrid systems. Rather than using non-linear elements to model switches, which can lead to numerical diYculties, it seems better to use ideal components, with some simple extra formal computing to deduce the diVerent models and to change the model at commutation times. Using singular perturbation theory, it is shown that both approaches lead to similar results. Some of the previous outcomes, although established in the linear case, can be extended to systems with nonlinear constitutive laws or MTF and MGY (provided that the modulating variables are continuous on commutations). The results of Section 5 fully apply. Non-linear implicit state equations can be deduced. As for discontinuities, there is no general explicit solution, but a nonlinear set of algebraic equations can be written using storage element constitutive laws, i.e. the second rows of equations ( 1) and (40). This method has been successfully applied to the simulation of a large system consisting of an asynchronous motor connected to a starter [33].

5 6

8 9 10

11

12

ACKNOWLEDGEMENT
13

The work reported here has been partly supported by GDR Automatique, from the French CNRS.

REFERENCES
1 Mosterman, P. and Biswas, G. A theory of discontinuities in physical system models. J. Franklin Inst., 1998, 335B(3), 401439. 2 Dauphin-Tanguy, G. and Rombaut, C. Why a unique causProc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering

14 15

ality in the elementary commutation cell bond graph model of a power electronics converter. In Proceedings of the 1993 International Conference on Systems, Man and Cybernetics: Systems Engineering in the Service of Humans, Le Touquet, France, 1993, Vol. 1, pp. 257263. Ducreux, J.-P., Castelain, A., Dauphin-Tanguy, G. and Rombaut, C. Modelling and analysis of power electronic networks by bond graph. In Proceedings of the IMACS Conference (TCI90), Nancy, France, 1990, Vol. 2, pp. 405416. Asher, G. M. The robust modelling of variable topology circuits using bond graphs. In Proceedings of the 1993 International Conference on Bond Graph Modeling and Simulation, Simulation Series, Vol. 25 (Eds J. J. Granda and F. E. Cellier), La Jolla, California, January 1993, pp. 126131 (Society for Computer Simulation, San Diego, California). Buisson, J. Analysis of switching devices with bond graphs. J. Franklin Inst., 1993, 330(6), 11651175. Stromberg, J.-E. and Soderman, U. Variable causuality in bond graphs caused by discrete eVects. In Proceedings of the 1993 International Conference on Bond Graph Modeling and Simulation, Simulation Series, Vol. 25 (Eds J. J. Granda and F. E. Cellier), La Jolla, California, January 1993, pp. 115119 (Society for Computer Simulation, San Diego, California). Cellier, F. E., Otter, M. and Elmqvist, H. Bond graph modeling of variable structure systems. In Proceedings of the 1995 International Conference on Bond Graph Modeling and Simulation, Simulation Series, Vol. 27 (Eds F. E. Cellier and J. J. Granda), Las Vegas, Nevada, January 1995, pp. 4955 (Society for Computer Simulation, San Diego, California). Stromberg, J.-E. A mode switching modelling philosophy. Thesis, Linkoping, Sweden, 1994. Soderman, U. Conceptual modelling of mode switching physical systems. Thesis, Linkoping, Sweden, 1995. Cormerais, H. De lApplication des Bond Graphs a ` Causalite Variable dans la Modelization Multimodele des ` Sytemes Dynamiques Hybrides. Thesis, Rennes, France, ` 1998. Mosterman, P. and Biswas, G. Hybrid modeling speci cations for dynamic physical systems. In Proceedings of the 1997 International Conference on Bond Graph Modeling and Simulation, Simulation Series, Vol. 29 (Eds J. J. Granda and G. Dauphin-Tanguy),Phoenix, Arizona, January 1997, pp. 162167 (Society for Computer Simulation, San Diego, California). Edstrom, K. Switched bond graph. Simulation and analysis. Thesis, Linkoping, Sweden, 1999. Lorenz, F. Discontinuties in bond graphs: what is required? In Proceedings of the 1993 International Conference on Bond Graph Modeling and Simulation, Simulation Series, Vol. 25 (Eds J. J. Granda and F. E. Cellier), La Jolla, California, January 1993, pp. 137142 (Society for Computer Simulation, San Diego, California). Borutzky, W. Discontinuities in a bond graph framework. J. Franklin Inst., 1995, 332B(2), 141154. Buisson, J. and Cormerais, H. Descriptor systems for the knowledge modelling and simulation of hybrid physical ` systems. J. Eur. Systemes Automatises, 1998, 32(910), 10471072.
I08300 IMechE 2002

Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011

ANALYSIS OF THE BOND GRAPH MODEL OF HYBRID PHYSICAL SYSTEMS

63

16 Buisson, J. and Cormerais, H. Modeling hybrid linear systems with bond-graph using an implicit formulation. Bond Graph Dig., 1997, 1, 17. 17 Mosterman, P., Biswas, G. and Otter, O. Simulation of discontinuities in physical system models based on conservation principles. In Proceedings of the 1998 Summer Computer Simulation Conference, Reno, Nevada, 1998. 18 Mosterman, P. and Biswas, G. Deriving discontinuous state changes for reduced order systems and the eVect on compositionality. In Proceedings of the 13th International Workshop on Qualitative Reasoning, Scotland, 1997, pp. 160168. 19 Lorenz, F. and HaVaf, H. Combination of discontinuities in bond graphs. In Proceedings of the 1995 International Conference on Bond Graph Modeling and Simulation, Simulation Series, Vol. 27 (Eds F. E. Cellier and J. J. Granda), Las Vegas, California, January 1995, pp. 5664 (Society for Computer Simulation, San Diego, California). 20 Karnopp, D. C., Margolis, D. L. and Rosenberg, R. C. System Dynamics: A Uni ed Approach, 2nd edition, 1990 ( WileyInterscience, New York). 21 van Dijk, J. On the role of bond graph causality in modelling mechatronicsystems. Thesis, Twente, The Netherlands,1994. 22 van Dijk, J. and Breedveld, P. C. Simulation of systems containing zero-order causal paths. I: classi cation of zeroorder causal paths. J. Franklin Inst., 1991, 328(56), 959979. 23 Gantmacher, F. R. Theory of Matrices, 1959 (Chelsea, New York). 24 Brenan, K. E., Campbell, S. L. and Petzold, L. R. Numerical Solution of Initial-Value Problems in DiVerential-Algebraic Equations, 1989 (North-Holland, Amsterdam). 25 Dai, L. Singular control systems. In Lecture Notes in Control and Information Sciences, 1989 (Springer-Verlag, Berlin).

26 Cormerais, H. and Buisson, J. Energetic balance in hybrid systems using bond graphs. In Proceedings of the 16th IMACS World Congress 2000 on Scienti c Computation, Applied Mathematics and Simulation, Lausanne, Switzerland, 2125 August 2000, pp. 16. 27 Kokotovic, P., Khalil, H. and OReilly, J. Singular Perturbations Methods in Control: Analysis and Design, 1986 (Academic Press, New York). 28 Borne, P., Dauphin-Tanguy, G., Richard, J. P., Rotella, F. and Zanvettakis, I. Modelization et Identi cation des Processus, Vol. 2, 1992 (Technip, Paris). 29 Fossard, A. J. and Norman-Cyrot, D. Systemes non lineaires, ` Vol. 3, 1993 (Masson, Paris). 30 Levine, W. S. The Control Handbook, 1996 (CRC Press, Boca Raton, Florida). 31 Buisson, J., Richard, P. Y. and Cormerais, H. Ideal versus non-ideal approaches in bond graph modeling of switching devices: a comparison based on singular perturbation theory. In Proceedings of the 4th International Conference on Automation of Mixed Processes: Hybrid Dynamic Systems (ADPM2000), Dortmund, Germany, September 2000, pp. 257264. 32 Breedveld, P. C. The context-dependant trade-oV between conceptual and computational complexity illustrated by the modeling and simulation of colliding objects. In Proceedings of the 1996 International Multiconference on Computational Engineering in Systems Applications, Symposium on Modelling, Analysis and Simulation (CESA96), Lille, France, July 1996, pp. 4854. 33 Buisson, J., Cormerais, H. and Quenechdu, Y. Bond graphs and the modelling of the continuous part of hybrid dynamical systems. In Proceedings of the 1996 International Multiconference on Computational Engineering in Systems Application, Symposium on Modelling, Analysis and Simulation (CESA96), Lille, France, July 1996, pp. 303308.

I08300 IMechE 2002

Proc Instn Mech Engrs Vol 216 Part I: J Systems and Control Engineering
Downloaded from pii.sagepub.com by Sekar Anindyajati on June 28, 2011