Modeling Convective Flows Using Bond Graphs

Jurgen Greifeneder Francois E. Cellier

Fakultat Verf.-technik und Techn. Kybernetik Dept. of Electr. & Comp. Engr.
Universitat Stuttgart The University of Arizona
P.O.Box 210104
70550 Stuttgart Tucson, Arizona 85721-0104
Deutschland U.S.A.
J@Greifeneder.de Cellier@ECE.Arizona.Edu
http://www.greifeneder.de/jg/ http://www.ece.arizona.edu/~cellier

Abstract 2 THE C-FIELD (CF)

This article, together with two companion papers, of- 2.1 Choice of variables
fers a general methodology for modeling thermodynam- Analysing the energy balance of a compartment, three
ical phenomena using true, rather than pseudo-, bond dierent components can be distinguished: mass ow
graphs. This paper deals in particular with the diÆ- into and out of the compartment, heat exchange
cult problem of modeling mass ows in a thermody- between neighboring compartments, and mechanical
namic sense, i.e. with the formulation of mathemat- work. Three independent (state) variables are need-
ical models of convective ows. No quasi-stationary ed to describe a thermodynamical system { one more
or ow-equilibrium assumptions were made, such that than for mechanical or electrical systems. Tradition-
the models generated using the proposed methodology ally, temperature T , pressure p, and mass M are used
would be kept as general as possible. as basic state variables, from which the specic terms
of e.g. entropy s and enthalpy h are being calculated.
Keywords: Thermodynamics; Convection; Bond For each of the three state variables, T , p, and M ,
graphs. a corresponding adjugate variable can be found, such
that the product of the two adjugate variables repre-
sents a facet of (co-)energy stored in the compartmen-
1 INTRODUCTION t. These are the entropy S for the temperature, the
volume V for the pressure, and the free (or Gibbs')
Bond graphs enable the modeler to describe the dy- enthalpy g = h T
s for the mass.
namics of a physical system in a modular fashion using
energy storage, dissipative power ow, energy source, p
and transformation elements. The basic bondgraphic q
icon library enables the user to model, in a systemat-
ic fashion, physical systems, the dynamic behavior of Figure 1: pneumatic-hydraulic power track
which is governed by power ows alone.
Systems with macroscopic mass ows add addition- Since the pneumatic (hydraulic) bond is already de-
al complexity to the modeling task, since the mass that ned in mechanics, as shown in Fig. 1, as tempera-
ows through the system carries with it its stored in- ture clearly denes a potential and not a eld variable,
ternal free energy, which is thus transported from one whereas the mass intuitively has ow characteristics,
location to another in a non-dissipative fashion. it makes sense to use S_ , M_ , and q as ow variables,
In the most general sense, convection ought to be with T , g, and p being identied as the corresponding
described by distributed parameter models. Since (adjugate) eort variables (it makes sense, to use g
bond graphs are geared to be used for the descrip- and not the enthalpy h = g + T
s as the eort vari-
tion of lumped parameter models only, a simplify- able, as a separate T,s-bond has already been dened
ing assumption will be made, in that the system to [1]).
be modeled is compartmentalized, whereby each com-
partment is considered to be homogeneous. 2.2 Computation of variables
New bondgraphic macro-elements are introduced to Breedveld showed [2] that no inductance can exist in
describe the energy storage within a compartment as a thermal system, i.e. only ows can be accumulated.
well as the mass (and energy) ow between neighbor- Therefore from a bondgraphic perspective, it makes
ing compartments. sense to dene the ow variables as the (independent)
state variables, whereas the eort variables are indi- and converting it as follows, whereby cv is the 'con-
rect variables, i.e. they are dependent on the state stant volume heat capacity,'
variables. 
@T
 
@p
 
@T


@v
= @T @s
(2.7)
s v v
dM
dt
= M_ (2.1)      
@T @T
dS =
@u ; (2.8)
dt
= S_ (2.2) @s v
|
@u
{z v @s
} | {z } v

dV = c1v =T
dt
= V_ = q (2.3)
the equation for the temperature can be determined
as follows:
According to physics, none of the six variables T , T
 
@p
 
S , p, V , g, and Mcan ever become negative. A com- dT =
ds dv (2.9)
partment can only be emptied. Once empty, nothing c @Tv v
further can be removed from the compartment. How- For an ideal gas, the above equation can be con-
ever, this can happen in a simulation due to numerical verted to the well-known caloric state equation:
inaccuracies. It may therefore be better to dene e.g.
the mass of a system (analogous for volume and en- c
ds = v dT + R

dv
(2.10)
tropy) as T v
Unfortunately, a (thermal) equation of state is al-
Zt so needed. If available, the pressure can be directly
M = maxf; M_ dtg computed (p = p(M; V; T )). Lacking an equation of s-
tate, interpolation tables will have to be used instead.
t0
In this case, the total dierential is not needed, as
where  is a large enough number, such that divisions fT; pg = fT; pg(v; s) can be directly obtained by in-
by M will still result in stable solutions. terpolation1.
The three dierent aspects of stored (co-)energy Last not least, the equation for the free enthalpy must
must not be treated as independent storage elements. be determined. If a corresponding interpolation ta-
If this were true, i.e., if the the three aspects could ble is available, it may be easiest to use that table.
be modeled by three separate conventional capacitors, Otherwise, another total dierential may be used. Ul-
it would follow that: T = T (S ), p = p(V ), and timately, an equation g(v; s) is needed; however, since
g = g(M ). Evidently, at least the third of these e- T and p have already been computed, it may be more
quations is clearly in error, as a intensive variable (g) convenient to determine h(T; p), and then compute
must not depend on the mass. Assuming incompress- g = h T
s.
ible uids, the temperature equation would be correct,
as ds = cp dT . However, this would limit the gener- Z T1  
ality of the envisaged approach to modeling. For the h(p1 ; T1 ) = h(p0 ; T 0) + @h
dT +
pressure, the situation is similar. | {z
energy of f ormation
} T0
|
@T
{z
p0
}
The reader may recall that any thermodynamical =cp
system can be completely described by two intensive +
Z p1 
@h

dp
and one extensive variables. That is exactly what the p0 @p
proposed three state variables are. T1
(2.11)
Commonly a polynomial approach is used to deter-
T = T ( MS = s; M
V
= v) (2.4) mine the specic heat cp. The variables ; ; ; Æ are
mostly constant and tabulated for p = 1 bar.
This equation can be derived using the total dieren-
tial: cp (T; p
) = (p ) + (p )
T + (p )
T 2 + Æ(p )
T 3
   


@T
 
@T
 (2.12)
dT = @s
ds +
@v
dv (2.5)
v s
1 As most tables are provided in the form v,s(T,p), it may be
Using one of Maxwell's equations: necessary to rst built T (p; v) and T (p; s) and nd the value of
p, for which T (p; v ) = T (p; s). This problem has an unique solu-
   
@T @p tion. To save time, this conversion should be done o-line once
@v
= @s
(2.6) and for all, and not be implemented as part of the simulation
s v algorithm.
 For p 6= p, Eq. 2.11 should be replaced by

of an ideal gas, are as follows:
dM
Z T1 
@h

dt
= M_ dS = S_ dt
dV
dt
=q (2.1 2.3)
h(p1 ; T1) = h(p0 ; T 0) + dT +
T0 @T  
Z p   Z p1   T dv
+ @h
dp +
@h
dp dT =

cv
ds R

v
(2.15)
p0 @p T0 p @p T1
(2.13)
R
T
p= (2.16)
2.3 Overall model V
To represent the relationships between the storage
( ow) and potential (eort) variables, a new bond dh = cp dT (2.17)
graph element (Fig. 2) is introduced. It is called ca-
pacitive eld, or C-eld, and is introduced in the bond g =h T
s (2.18)
graph as the CF-element2.
It is important to mention that specic material prop-
cv = cp R (2.19)
0
.
S T
R = 0:2872 kgkJK

and cp = 1:004 kgkJK .

C
3 CONDUCTION (CD)
CF
C C Conductive heat ow Q_ can be represented in a bond
graph as a purely thermal phenomenon using bonds
g q
.
with T as eort and S_ as ow variables. If the C-eld
M p
0 0
is exposed to such a heat ow only, then M_ = q = 0
Figure 2: bond graph representation of the C-eld _ Q_
and dt = S = T . This is consistent with the second
dS
fundamental equation of thermodynamics (balance of
erties shall be modeled within the C-eld only. All entropy without mass ow).
other elements will be totally generic.
Analysing the fp=qg-bond properties, it may be no- Heat ow only occurs between two connected C-
ticed that a system loses energy for a positive volume elds at dierent levels of temperature. The heat ow
ow and gains energy for a negative one. In order to from the C-eld located at the higher temperature to
avoid this asymmetry, the authors of this paper de- the C-eld located at the lower temperature is pro-
cided to dene the pressure as negative. For example, portional to the dierence of temperatures and to the
the equation of state of the ideal gas is written as heat transfer coeÆcient :
R
T = p
V . In this way, the balance of the C-eld
becomes: Q_ = 
(T1 T2 )
(3.1)
dU
= T
S_ + p
V_ + g
M;
_ (2.14) The entropy ows can be graphically depicted as
dt shown in Fig. 3.
This equation can be graphically interpreted as the
sum over all bonds connecting to the C-eld. CF1 CF2

The kinetic energy of the CF-element has not been 3 3

modeled, as the element is supposed to be stationary Ø

T1
. 1
T2
. Ø
in space. The potential energy can easily be taken into S1 S1
account by adding Epot to g; however in this case, the . .
S 1 ∆T .
C-eld must `know' its own vertical position in space. T1 ∆ S1x . T2 ∆ S2x

Radiation can be included by modeling the entropy- S1

2 ∆T
cut of the C-eld accordingly. mGS
∆T 0 .
S1
mGS

The equations of a C-eld for air, using the assumption

2

Figure 3: modelling heat ows

2 CF-elds have been previously used in the bondgraphic lit-
erature, but their purpose was simply to describe a network of The mGS-element is a conductive source element
conventional, i.e., independent, capacitors. [3] with a thermal primary side. The m denotes the
fact that the element is modulated, as the conduction In addition to this forced volume ow, there also exists
G depends on the temperature: a volume ow that is driven by dierences between
the pressure of two connected C-elds. An example
   for this phenomenon would be two systems, separated
GCD = = = (3.2) by a movable wall or membrane. It can be concluded
T1 + T2 2
T1
T 2
T2 +
T
that there will take place only a single volume ow,
The notation is adopted from [1]. The authors i.e., that the volume leaving the rst system must be
prefer to use an mGS-element rather than an mRS- added to the second one.
element because of the xed causality of the autgoing Such a volume-exchange ow is proportional to the
bond. The secondary side of the mGS-element always pressure dierence and is of a dissipative nature (the
represents an entropy (i.e., ow) source, and never a power
p
q  0 will be converted to heat).
temperature (i.e., eort) source. The primary side of
each mGS-element computes Ohm's law as follows: CF1 CF2

S_ 1
2 = GCD

T (3.3)
3 3
p1 p2
Ø q 1 q Ø
whereas the secondary sides compute the power ow .
continuity across the mGS-elements: T1 ∆ S1x
q
q ∆ p .
T2 ∆ S2x
p
∆

S_ 1
2

T
mGS 0 mGS

S_ 1x = 2 T1 (3.4)
∆ p q
2

_ Figure 5: bond graph representation of the DVA-

S_ 2x = S21

TT (3.5) element
2
Heat ow is a dissipative phenomenon. In the pro- As in the case of conductive heat ow, the gener-
cess, additional entropy (but not heat) is being gener- ated entropy will be distributed equally to both sides
ated in the amount of
S_ 1x +
S_2x. (Fig. 5).
The reader may notice that the heat ow from the The overall eects of pressure equilibration can be rep-
rst to the second C-eld is indeed: resented by a new bondgraphic macro element, the
DVA-element.
Q_ = T1
(S_ 1
S_ 1x ) = T2
(S_ 1 +
S_ 2x ) = 

T
(3.6) 3
DVA
3

as expected.
The conductive heat ow of Fig. 3 can be represented Figure 6: Icon-representation of g. 5
by a new macro bond element, the CD-element, as
shown in Fig. 4.
The attentive reader may have discovered that a
T1 T2 triple bond, a so-called bus-bond3, (Fig. 7) was used in
. CD . the iconized representation (Fig. 6). As no mass ow
S1 S2 takes place during pressure equilibration, the mass
ow is set equal to zero.
The equations of the DVA-element can be summa-
Figure 4: Iconized conduction element rized as follows:
If one of the two C-elds is located outside the sys-
tem, that C-eld can be replaced by a one-port SE- =0 M_ (4.1)
element. Alternatively, the external C-eld can be ne- q = GDV A
(p2 p1 ) (4.2)
glected, replacing the CD-element by a one-port SF-
element.
3 The literature mentions vector-bonds and multibonds as
means to describe sets of bonds. However, these are used for
the compact treatment of threedimensional mechanics, where-
4 VOLUME WORK AND EQUILI- by each strand denotes one space dimension. In order not to
BRATION OF PRESSURE (DVA) confuse these dierent concepts, the authors of this paper de-
cided on a dierent notation, borrowed from digital electronics,
where multiple signal paths are connected into a signal bus. The
Ideal volume exchange does not pose any problem, as bus-bond connects multiple power paths of dierent type into a
it is reversible and therefore does not generate entropy. single power bus.
 T
For these reasons, the authors decided to model

{
.
S
mass ow phenomena as pure mass ows, i.e., with
3
)
p
g as eort, and introduce the bus-bond as a conve-
q
nient means to keep the accompanying volume and
g
. entropy ows together with the mass ow component
M
in a compact power bus.
Figure 7: Bus bond Will this make the use of activated bonds to de-
scribe information ows unnecessary? Unfortunately,
q
(p2 p1 ) this is not the case. Mass carries additional properties
S1 =
2
T1 (4.3) that may be needed in the computation of the mass
ow characteristics, such as the density, , and the
q
(p2 p1 ) total entropy contained in a C-eld, S . These infor-
S2 = (4.4)
2
T2 mation paths will emanate from a C-eld and end at
the neighboring R-eld downstream, i.e., they will stay
fairly local. This is consistent with the fundamental
5 CONVECTION: THE R-FIELD principle of network theory [5], and could provide a
hint for a future merger of these two ways of modeling
5.1 Introduction physical systems.
So far, the paper only discussed entropy and volume
ows, but did not use the third cut of the C-eld: the Before the concept or the R-eld is derived in detail,
mass ow. Unfortunately, mass ows are not indepen- two additional notations need to be explained:
dent (as entropy and volume ows could be), because ;
mass has a volume and is carrying enthalpy. Obvious- describes a bus-zero-junction. Dierent busses can
ly, it would have been possible, to model this by usingp connect to a bus-zero-junction, whereby the nor-
a single bond (eort: (internal energy) u = g +T
s+  ; mal rules of zero-junctions shall apply to each s-
ow: M_ ), but in that case,
a vector of information trang separately. If a regular bond (either thermic
consisting of T p  s would have to accom- of hydraulic/pneumatic) is connected to a bus-zero-
pany the power ow, such that the internal energy junction, it will be treated like a bus-bond with the
bond could be decomposed into its components again other two bus- ow variables set to zero.
further down the road.
0Ti
Another possibility would have been to adopt the This nomenclature is helpful for detail graphics. It
notation of Brown [4], who introduced a convection represents the temperature-zero-junction of the CF-
bond, whereby the power bond is given by fh,M_ g with element named CFi (p, g will work analogously).
0 0

Using these abbreviations, an RF-schematic is pro-

vided in Fig. 9.
Figure 8: The convection bond introduced by von
Brown [4] ρ,S
1 1
ρ,S
2 2
CF1 CF2
the pressure superposed as information path. In our
view, this approach carries three disadvantages: 3 3

 By superposing the pressure as an information

path, the concept of pure power bonds is com-
3 3
Ø RF Ø
promised.
 The decomposition of the enthalpy bond into a Figure 9: The resistive eld
thermal bond and a mass ow bond at the en-
trance of the next compartment while recalcu-
lating the corresponding volume ow necessitates
additional connection models. This complicates 5.2 The basic RF model
the modeling eort. For the time being, let us model the situation, where
 Keeping track of mass ows from the entrance to a xed mass is being transported through a pipe at
the exit point of a compartment without an ap- constant speed and without friction. The pipe is com-
propriate bondgraphic mass ow notation neces- partmentalized, such that each CF-element represents
sitates the introduction of additional information one section of the pipe. In this case, the following
paths within the compartment models. equations can be derived:
  Modeling volume changes within C-elds is more
practical, as in this way, material properties can
q =A
v (5.1) be mostly limited to individual C-elds, thereby
_M = q
 (5.2) minimizing the need for information ows (acti-
vated bonds) between neighboring bond-graph el-
Still missing is an equation for S_ . Let us consider ements.
an amount  of a given homogeneous element with
the mass M , the volume V , and the Entropy S (g. Last not least, the case of entropy ows across an
10). If a portion M~ of this homogeneous mass is being R-eld needs to be discussed. These ows constitute
cut o,~ that portion will obviously~ have the volume the most diÆcult case. Entropy certainly is not con-
V~ = M ~ M
M
V and the entropy S = M
S .
served. The R-eld, in general, describes dissipation,
Generalizing to continuous innitesimal ows, one and therefore, entropy is generated inside the R-eld
(g. 11).
~
Μ

Θ
T. 1 T2
1 .

Figure 10: Cutting a portion M~ o a given amount 

S S
p1 p2
q 1 q

can conclude: g1
.
M
1
g2
.
M
. .

M_
S ∆T q ∆ p M ∆g

= (5.3) RS RS RS
M
S_ = 
S
T2 T2 T2
(5.4) 0 T2
.
where  is the mass fraction. ∆S

To determine the ows leaving the R-eld, it is known Figure 11: The R-eld schematic
that
M_ out  M_ in (5.5) 3 3
Ø 1 Ø
because no mass is stored in the R-eld. More d- 3 3 3

iÆcult is the case of the volume ow, as volume does T2 .

not need to be preserved. Yet, it may make sense CF1 RS ∆S
CF2

to assume volume preservation across the R-eld, and

renegotiate the valume of an arriving ow at the next
C-eld downstream. Consequently, the volume ow is Figure 12: R-eld schematic in bus-bond notation
dened analogously to the mass ow:
If the two entropy ows S_ out and
S_ are combined,
V_out = V_in (5.6) and if one takes into consideration that the R-eld
In this way: should be designed such that it can be used in both
directions, a model as shown in gure 13 is obtained.
 Mass, being transported from A to B, will arrive
at B with the same volume that it had when leav- 5.3 The general model
ing A, leaving it up to the C-eld representing B Up to this point, ows were supposed to have constant
to renegotiate its volume. speed. As there is always friction, this was evidently
 The RF-element does not need to know, what vol- a simplication. Based on the balance of impulses
ume ow V_ the mass ow M_ will have, after it
got mixed with the mass at B.
p
A =
v
M_ (5.7)
 Modeling the case VCF = const: (as it holds e.g. a feedback loop of the pressure to the speed results,
for pipes) inside the RF-element would not make which is a function of the geometry of the pipe, the
much sense, as this would mean that there would viscosity of the uid, and the ow characteristics.
be no volume ow at all. Furthermore, this would This leads to another simple generalization which ac-
cause problems in section 5.3. counts for varying cross-sections and the eects of
 gravity.
p1
A1 p2
A2 =
v
M_ + cos
G (5.8)
where  is the vertical inclination of the ow.
g1 g2 p A *p A *p p
. 1 . 1 1 1 2 2 2
M M
q1
TF
1 q
1 q
TF
A 2
q2
RS A 1

q ∆ (Α∗p)
T. 1 T2
S1
1 0 1 .
S2 ∆ (Α∗p)
SE:
G * cos α 0 q
RS 0T
Sw 0 0 Sw R
T
RS
∆(Α∗p)
RS RS
.
p1 p2 GY: M
1 RS
q q

∆v F 1
Figure 13: The R-eld, considering both ow direc-
∆F v 2

tions. F1
0
F1
v 1
F2
v 2
v2
1

Figure 15: Part of the bond model of the generalized

R-eld.

5.4 Generalized Exchange Element

Merging the eects of the three previously introduced
elements (CD { conduction, DVA { pressure and vol-
ume exchange, RF { convection), the following set of
equations can be extracted from the combined bond
graph:

=A
v (5.1)
p p
0p 1
1
2
0p qRF
qDV A = RDV A
(p2 p1 ) (4.2)
1 q1 q1 2

q1
q = qRF + qDV A (5.9)
∆ p

0
∆p

q
RS
T2
.
SR
_M = qRF
 (5.2)
R

q ∆ p T2
. 0T
 
F

_Si = 1
T1
SF
p1 p2
2

.
T2 + q
(5.10)
2


Ti
M
GY: A RS

∆v F 1 ∆F v 2

F1 F1 F2
ST_ = qR
(p1 p2 ) + M_
(g1 g2 ) + v
(F1 F2 )

v1 0 v2 1 v2 (5.11)
Figure 14: Hydraulic part of the RF-model (

S1 + S_ 1 forM_ > 0;
S_ 1 = (5.12)


ST_ +S_ 2 T2 + S_
1 else


T1

(

S + S_ forM_ < 0;
S_ 2 =
ST_ 2+S_ 1 T12 _ (5.13)


+S
T2


2 else

5.5 Special case: C-eld with constant References
volume
Let us consider once more the example of the set  [1] Francois E. Cellier (1991), Continuous System
(g. 10) where a part of the system (M; ~ V~ ; S~) has Modelling, Springer-Verlag, New York.
been cut o. The remaining system will be character- [2] Peter C. Breedveld (1984), Physical Systems The-
ized by fT, p, V = V V~ , M = M M~ , S = S S~g.
  
ory in Terms of Bond Graphs, Ph.D. dissertation,
Thinking of a leaking pipe or gas cylinder, the volume University of Twente, Enschede, The Netherlands.
of which remains constant, a second setp is needed, in
which the volume V of the remaining gas is expanded

[3] Jean U. Thoma (1975), \Entropy and Mass
until it equals the volume V that the gas had occupied Flow for Energy Conversion," J. Franklin Institute,
initially. The energy needed for this expansion must 299(2), pp. 89{96.
come from the inside of the C-eld. This can be ac- [4] Forbes T. Brown (1991), \Convection bonds and
complished by making use of a controlled ow source bond graphs," Journal of the Franklin Institute,
that creates exactly the amount of volume that has 328(5/6), pp. 871{886.
left . Considering the remaining energy balance (no
mass ow) [5] Ernst Dieter Gilles(1997), \Netzwerktheorie ver-
fahrenstechnischer Prozesse" Chemie Ingenieur
Technik (69) 8|97 pp. 1053{1065, VCH Verlags-
0 = P + Q_ (5.14) gesellschaft mbH, 0009-286X/970808-1053.
it is found that the ow source must be connected [6] Jurgen Greifeneder, Francois E. Cellier (2001),
as a two port element to the temperature-entropy-cut \Modelling Multi-phase systems," ICBGM'2001,
of the C-eld. The energy needed will thus be gener- Phoenix, Arizona, U.S.A.
ated by a (reversible) entropy ow within the C-eld,
leading to a reduction of temperature in the pipe or [7] Jurgen Greifeneder, Francois E. Cellier (2001),
gas cylinder, which is consistent with physical knowl- \Modelling Multi-element Systems and Chemical
edge. Reactions, Using Bond Graphs," in preparation.

0
.
S T

0 .
. T Sq
∆S T
CF SF: q~ = q
C
~
C ∆q
p q
g C
=0
. p
g 0 M 0 q
. p
0 M 0

Figure 16: bond graph model for a constant CF-

volume.

6 CONCLUSION
The elements introduced in this paper enable the mod-
eler, to describe simple thermodynamical systems de-
scribed by conduction and convection. The methodol-
ogy introduced so far only deals with mono-elements
and mono-phase systems. The description of multi-
phase and/or multi-element systems is left to two com-
panion papers [6] and [7].