Introduction to bond

graph theory
Second part: multiport field and
junction structures, and
thermodynamics

Multiport fields
We will look at multiport generalizations of C, I and R elements.

C-fields

f2

e1
f1

e2

en
fn

qn

= f1
= f2
..
.
= fn

Allows easier displaying

of individual causalities

State variables q1 , q2 , . . . , qn
q1
q2

Used mainly for sets of bonds

with geometrical properties.
e

e1
e2

en

= 1 (q1 , . . . , qn )
= 2 (q1 , . . . , qn )
..
.
= n (q1 , . . . , qn )

Energy is computed as
H(t) = H(t0 ) +

Z tX
n

ek ( )fk ( ) d

t0 k=1

Changing t q yields the line integral

H(q) = H(q0 ) +

e(
q ) d
q

is any curve connecting q0 and q

However, this must be independent of the

particular curve connecting q0 and q!
Barring topological obstructions, this is equivalent to
j
i
=
,
qj
qi

i, j = 1, . . . , n

exactness of the 1-form given by e

Maxwell reciprocity condition

e = d

Linear case:

compliance form
q = Ce

stiness form
e = kq

all dierential

all integral

Mixed forms are also possible, but for a given system

some of the forms, including the compliance one, may not exist.
The above nomenclature extends to the nonlinear case as well.
In the linear case, exactness of e implies that
the matrices k and C, if the latter exists, are symmetric.
The available forms determine which causal patterns are admissible.

all-integral is possible
2 0 2
e 2 f2
all-dierential is not
k = 0 1 1 det k = 0
e1
e3
2 1 3
C
f1
f3
furthermore . . .

1
2

q1
e1
0 1
q2 = 0 1 1 e2
q3
e3
1 1 0

e 2 f2
e1
f1

e3
f3

is possible

C-fields given from the beginning as a set of

eort-displacement relations at n ports are called explicit.
Implicit C-fields are obtained when several C-elements
are assembled by way of a power continuous network.
Implicit C-fields can be reduced to implicit form. In the process, some
elements with dierential causality may be hidden from the port interface.

C3
P1

C1

C2

P2

C : C1

C : C3

C : C2

C3
P1

C1

C2

P2

P1

5
6

e1 = e6 = e3

f6 = f 5 = f7

e4 = e7 = e2

f3 = f1 + f6

e5 = e6 e7

f4 = f7 + f2

q3 = f3
e3 =

1
C1 q3

q5 = C3 e5
f5 = q5

q4 = f4
e4 =

q5 = C3

q4
q3
+
C1
C2

f1 = i 1

1
C2 q4

q5 = C3 e5 = C3 (e6 e7 ) = C3 (e3 + e4 ) = C3

f2 = i 2

q4
q3
+
C1
C2

P2

q4
q3
q3 = f3 = f1 + f6 = i1 + f5 = i1 + q5 = i1 C3
+
C1
C2

q4
q3
q4 = f4 = f2 + f7 = i2 + f5 = i2 + q5 = i2 C3
+
C1
C2

C3
C3
1 + C1
i1
q3
C2
=
C3
C3
q4
i2
1
+
C1
C2
q3
q4

1
C1 C 2 + C 2 C3 + C 1 C 3

C 1 C2 + C1 C 3
C2 C3

C1 C3
C1 C 2 + C2 C3

i1
i2

We introduce new state variables q1 , q2 such that q1 = i1 = f1 , q2 = i2 = f2 .

Using q3 = C1 e3 = C1 e1 , q4 = C2 e4 = C2 e2 , and integrating in time:

e1
e2

1
C 1 C2 + C2 C 3 + C 1 C3

C2 + C3
C3

C3
C1 + C3

q1
q2

e1
e2

1
C 1 C2 + C2 C 3 + C 1 C3

C2 + C3
C3

C3
C1 + C3

q1
q2

k
This is a 2-port C-field in stiness form.
e1
f1

The state variables q1 , q2 do not

correspond to physical charges.

e2
f2
They are just a convenient
parametrization of the R3 surface
q5 = C3 Cq31 + Cq42

The dependent state variable has been hidden away from the port interface.

I-fields

f2

e1
f1

e2
en
fn

State variables p1 , p2 , . . . , pn
p1
p2

pn

= e1
= e2
..
.
= en

f1
f2

fn

= 1 (p1 , . . . , pn )
= 2 (p1 , . . . , pn )
..
.
= n (p1 , . . . , pn )

Energy:
H(p) = H(p0 ) +

f (
p) d
p

j
i
=
,
pj
pi

i, j = 1, . . . , n

independence of

V2

Rigid bar with mass m, length L and

moment of inertia J respect to the CM.

V1
CM

We consider only vertical displacements

and small rotations around CM.

F2
F1

Under these assumptions, this can be described as an

explicit I-field, with constitutive relation

V1
V2

1
m
1
m

L
4J2
L
4J

1
m
1
m

L
4J2
L
4J

p1
p2

IC-fields
fI

e1
f1

eI e1 f
1

IC

eC

p1

pI
q1

fC
qC

=
..
.
=
=
..
.
=

e1

f1

eI
f1

fI
e1

fC

eC

=
..
.
=
=
..
.
=

1 (p1 , . . . , pI , q1 , . . . , qC )

I (p1 , . . . , pI , q1 , . . . , qC )
1 (p1 , . . . , pI , q1 , . . . , qC )
I (p1 , . . . , pI , q1 , . . . , qC )

Maxwell reciprocity equations

j
i
=
,
pj
pi

i, j = 1, . . . , I
j
i
=
,
qj
pi

j
i
=
,
qj
qi

i, j = 1, . . . , C

i = 1, . . . , I, j = 1, . . . , C

A typical example of an IC-field is an electrical solenoid transducer.

A more academic example is
1

P1

GY

..
m

= f1

p5

= e6

1
e6
m

P2

C:C

q2

I :L

To get an explicit IC-field, define

e6 = p
state variables q, p such that f1 = q,
p = p5 , q = q2 +

e1
f6

1
1
q
p
C
mC

1
1
1
q+
+ 2
p
=
mC
L m C

1
m p5

Maxwell condition

R-fields
f2

e1
f1

Onsager forms

e2

en
fn

resistance form

e = (f )

conductance form

f = 1 (e)

Mixed causality forms may also be possible

In the linear case, implicit R-fields without gyrators
or sources have Onsager forms with symmetric matrices.
If some (e, f ) pairs are interchanged in their causality from
an Onsager form, the corresponding matrix adquires antisymmetric
terms. Such contitutive relations are said to be in Casimir form.

P1

e1
f1

R : R3

GY : m

e2
f2

e1
e2

Several forms are possible by

switching the causality around.

R : R5

P2

R : R4

R3 + R4
m + R4

m + R4
R4 + R5

f1
f2

This Onsager form is not symmetric, due to the presence of a gyrator.

Junction structures
Assemblages of 0, 1, TF and GY elements which switch energy around.
Limiting cases of R-fields (without sources)which do not dissipate.
Unless modulated TF or GY elements are present, eort/flow
constitutive relations in a junction structure are always linear.
With an all-input power sign convention, the matrix
relating inputs to outputs must be antisymmetric.
Causality patterns are more restricted, though.
Junction structures without gyrators cannot accept
conductance or resistance causality on all ports.

P1

Pure conductance or resistance

forms are not possible.

m : TF

GY

0
f1
e2 = m
f3
m
r

:r

P2

P3

m
r

m
e1
0
m f2
m
0
e3

Multiport transformers are an special case of junction structures.

Through-power convention
P1

P3

P4

:m

TF

P2

TF
..
m2

With an all-input power

convention, (e1 , e2 , f3 , f4 ) would be
obtained from (f1 , f2 , e3 , e4 ) with
an antisymmetric matrix.

With the through-power convention, the matrix is symmetric

and can be decomposed into two matrices wich are transpose:

e1
e2

1
m1

1
m2

e3
e4

M
..

TF

f3
f4

1 m1
1 m2

f1
f2

The fact that the flow

transformation is given by M T
ensures the power continuity.

Multiport transformers need not have

the same number of inputs ans outputs.
Example: abc dq transformation in induction machines.
Junction structures are also necessary to connect the bond graph
formalism with port Hamiltonian and Dirac structure concepts.

Thermodynamics from
the bond graph point of view
Pure substance with no motion, constant mass and
no electromagnetic or surface-tension forces:
internal energy
per unit mass

u = u(s, v)
entropy
per unit mass

Gibbs equation:

volume
per unit mass

du = T ds p dv

absolute temperature

pressure

u
T =
s

du = T ds p dv

u
p =
v

Maxwell relation for a 2-port C-field

with a power-through convention

T and p are eorts

T = T (s, v)
p = p(s, v)

(p)
T
=
v
s

constitutive
equations

s and v are the corresponding flows

for all-integral causality

T
s

In thermodynamics, mixed and all-derivative causality

is implemented by means of Legendre transformations.

entalphy h

Gibbs equation

h = u + pv

dh = du + p dv + v dp = T ds + v dp
h = h(s, p)

constitutive equations
T
v

=
=

T (s, p)
v(s, p)

mixed causality

T
v

Maxwell condition

h
s
h
p

=
=

T
s

v
T
=
p
s

Helmholtz free energy f

Gibbs equation

f = u Ts

df = du T ds s dT = s dT p dv
f = f (T, v)

constitutive equations
s = s(T, v)
p = p(T, v)

mixed causality

p =

Maxwell condition

f
T
f
v
T
s

s
p
=
T
v

Gibbs free energy

= u + pv T s

d = du + p dv + v dp T ds s dT
= s dT + v dp

= (T, p)

Gibbs equation

constitutive equations
s =
v =

s(T, p)
v(T, p)

All-dierential causality

s =
v

Maxwell condition

(s)
v
=
T
p

T
s

Any of the four formulations gives constitutive equations

which guarantee conservation of energy.
The computation path depends on the causality pattern.
For instance, assume the entalphy is given, h = h(s, p)
p
T
s

s
p
v
T

s=

s dt

h = h(s, p)
2h
2h
s
v =
p +
2
p
sp

h
s
h
p

One can also give constitutive equations without

using any of the energy functions u, h, f or .
However, this can easily give impossible substances, which
violate the First Principle of Thermodynamics (energy conservation).
Ideal gas:
pv = RT
Since this is a pure substance of the type considered, another
constitutive equation is needed to specify the 2-port.
The remaining equation is related to the gas being mono- or diatomic.
Giving this second equation arbitrarily runs into the above problem.
It is better to start with two other relations and build
an energy function from them, incorporating pv = RT .

specific heat at constant pressure

h
cp =
T

this makes sense since p is an

input in the h(s, p) formulation

specific heat at constant volume

cv =

u
T

T is not a natural variable of h or u.

this makes sense since v is an

input in the u(s, v) formulation

Together with pv = RT , we assume that cv

is a constant, determined by the particular ideal gas.
pv = RT
h = u + pv

cp = cv + R

cv =

u
T

cv constant
h = u + RT

u = cv (T T0 )
h is also function
of T alone
h
cp =
T

reference
temperature

h = cp (T T0 ) + RT0

cp constant
due to cp = cv + R

dv
dv
du
dT
du = T ds p dv
+p
+R
ds =
= cv
T
T
T
v
cR
v
T
v
s
v
c
s = cv log
+ R log
T = T0 e v
T0
v0
v0
integration of
the 1-form

dh = T ds + v dp

ds =

dp
dp
dh
dT
v
v
= cp
T
T
T
T

v
1
v cp
dv
dv
dp
(p dv + v dp) dp = cp
= cp
+
1 dp = cp
+ cv
RT
T
v
T R
v
p
v
p
s = cp log
+ cv log
v0
p0

integration of
the 1-form

p = p0 e

s
cv

T = T0 e

s
cv

v
v0

ccp

v
v0

cR

p = p0 e

s
cv

v
v0

ccp

constitutive equations for

a perfect gas in all-integral form
s,
v
T, p

R, cv

Exercise: compute the constitutive equations

for the other three causality patterns.

Chemical engineering:
transport phenomena
quantities of substances vary with time
Requires an extension of the basic
thermodynamic bond graph framework

stirred reaction tanks

fuel cells

Universit Claude Bernard Lyon:

Bernhard Maschke & Christian Jallut