Professional Documents
Culture Documents
and HALLVARDENGJA
Dicision 9fMarine Engineering, The Norwegian Institute ~f’TechnologJ>,
N-7034 Trondheim-NTH, Norway
I. Zntvoduction
In many engineering systems compressible fluid flow in pipes and passages plays
an important role in overall system performance. In modelling such inherently
distributed dynamic phenomena, the techniques usually’employed involve the use
of finite difference or finite element methods. This approach results in a large
number of equations which are difficult to interpret and removed from the physical
structure.
Over the years, the bond graph method has been developed into a powerful
unified approach to modelling interconnected dynamic systems. It is therefore a
strong motivation for developing bond graph models for compressible flow. A
general system dynamic approach to fluid flow problems has been treated very
extensively. References (1) and (2) are exceptions ; however, neither of these contain
satisfactory documentation of boundary treatment, which is essential in fluid flow
modelling.
In what follows, the one-dimensional gas dynamic conservation equations are
applied to a finite size control volume. The resulting equations are then interpreted
in a bond graph model, which assures a close connection between the bond graph
and the original conservation statements. By connecting several control volumes
together with a proper boundary treatment, an accurate and simple fluid line model
results.
The use of the developed model will be demonstrated with results from laboratory
experiments, standard benchmark problems and computer simulation.
With unsteady flow in pipes and passages the distributed mass of the working
fluid is as important as the fluid compressibility because of momentum effects that
cannot be neglected. When such a situation occurs, the pressure is not the same
throughout the passage, and both the fluid mass and fluid compressibility must be
considered as distributed characteristics. Similarly, the density, velocity, and flow
rate differ from one point to another in the system as well as varying from one
time to another at each point.
Thus, in obtaining a mathematical model of the pipe flow, it is natural to make
use of the following conservation principles :
l conservation of mass
l conservation of momentum (Newton’s law)
0 conservation of energy.
(;’ L
A?
at s
0
I.
p dx = (/~~4),=“-(puA),=r
1.
pu dx = (~u~A+PA),=~-(~~~A+PA),=~- 7~Dt,. dx (2)
A;t: s 0 s0
L L
3
Am p(e++u’) dx = [puA(h+:u2)],=o-[puA(h+:u’)l.,=~-A ~4 dx
at s 0
(3)
where p is fluid density, u is fluid velocity, e is specific internal energy of the fluid,
T,, is the frictional shear stress tensor, 4 is the rate of heat added per unit mass, and
h (=e+P/p) is specific enthalpy. These conservation equations together with a
thermodynamic equation of state,
P = P(P>T) (4)
L
I- 4
FIG. I, A cylindrical pipe with constant cross-sectional area.
782
Bond Graph Interpretation of One-dimensional Fluid Flow
AZ
I-l
1 2 3 i-l i it1 N-i A
I ‘I
I
I
k-4 t-4
I- I
I * I I I
I gi_l I gi 1 Y*+l 1
I
I
XI-2 X
I
I
k X
I ,x
I-l I Iii
FIG. 3. The constant value approximation of the flowfield properties distribution in a pipe
control volume.
dm, = (P4,,
at
,-_(w‘c,
;; (mu); = $ p,u,lujl
where m, is the total mass in the control volume and e, represents the total rate of
heat added to the control volume. In the momentum equation, the frictional shear
stress tensor, T,,, is here replaced by :.f~ul~l and ;1 = 4.f; where f and 1. are the
Fanning and Darcy friction coefficients. The partial derivative with respect to time
is replaced by the total derivative, since time is now the only independent variable.
These equations express the time rate of change of mass, momentum and energy
in control volume i, in terms of the flowfield properties in control volume i and at
its boundaries. Such equations will result for each of the N elements/control
volumes. The flowfield properties at the common boundary for adjacent control
volumes must be equal when the control volumes are interconnected to form the
entire pipe. It is therefore reasonable to express them in terms of the properties in
the adjacent control volumes. A simple approximation is given in Eq (9) for
properties at the boundary between control volume i and control volume i+ 1
g/ = 12(g,+g,+
I). (9)
This mean value approximation is simple, but it does not take into account the
distinctive physical characteristics of fluid flow problems. One such characteristic
feature is the convective nature. Convection is the transport of mass, momentum
and energy, with a given velocity, often referred to as the convection velocity. When
referring to one location in the pipe, it is obvious that the condition at that location
will be more affected by the condition upstream of the location than downstream
of the location, for a given velocity. This skew property will increase with the
magnitude of the convection velocity, and for subsonic flows, the speed of sound
is the upper limit of the convection velocity. Changes in the downstream condition
in that case will not affect the condition at the location where this velocity is reached.
This upwind nature of fluid flow can be taken into account by a modification of
the mean value approximation. When relating this upwind nature to the local Mach
number, M, a positive relation between this upwind nature and the Mach number
will exist. Positive in the sense that the higher the Mach number, the more upwind
is the flow. The Mach number is the ratio between the local convection velocity
and the local speed of sound, and is consequently equal to 1 for sonic (U = c)
velocities. The following Mach-modulated approximation, originally proposed in
(3), takes into account this upwind nature of fluid flow,
Y/ = ti(sj+sitI)+~.f(M,,)sign(u,).(,~,-_g,+ J (10)
where u,, and M, are the velocity and Mach number in the control volume upstream
of the boundary and f e [0,11. The Mach number is here assumed to be a positive
number. The flow direction is taken care of by the sign function, which is equal to
1 for positive velocities and - 1 for negative velocities. The power law function is
a general function which satisfies these constraints
(12)
d
drp~=rir,~,u,~,+P,~,A--ril,~,-P,A-~rn,~~,u~l (13)
and
(16)
(17)
where m, pi, and E, will be the state variables for control volume i. The mass flow
rate, puA, is replaced by the symbol r+r. Inclusion of physical boundary conditions
may be imposed at the left boundary of control volume I, i.e. at xc,, and at the
right boundary of control volume N, i.e. at x,~.
Proper treatment of boundary conditions is very important when formulating
flow problems. Analysis of the general nature of hyperbolic type problems has
shown that the number of boundary conditions to be specified at a physical
boundary equals the number ofcharacteristics entering the domain at the boundary
(4). For subsonic (u < c) flow in pipes with open ends, there are two entering
characteristics, u and u + c, at the boundary with inflow, and one, u ~ c, at the
boundary with outflow. At open ends with sonic outflow (U = c), the number of
boundary conditions is one. This means :
In true bond graphs for thermodynamic systems, temperature and pressure are
identified as efforts and volume flow rate and entropy flux as flows. Authors such
as Thoma (6) and Breedveld (7) have done work of theoretical importance in
modelling compressible fluid flow using these variables. In the opinion of the
authors, these formulations are still difficult to comprehend from an engineering
viewpoint though they may be preferable from a philosophical point of view.
In this paper an extension of the pseudo-bond concept proposed by Karnopp
(8) will be presented.
One-dimensional fluid flow problems differ from problems represented within
this pseudo-bond concept, in the way that there are three quantities (mass, momen-
tum and energy) to be conserved in each control volume, versus two (mass and
energy) of the original concept. Due to this close relation, and to the fact that
Bond Graph Interpretation of One-dimensional Fluid Flow
systems modelled with this pseudo-bond concept often are connected to pipes, it
is natural to use this bond concept, when proposing a bond graph representation
for one-dimensional fluid flow problems. When extending the original scheme with
one extra bond, carrying the information connected to the momentum exchange,
the pseudo-bond concept shown in Fig. 4 results. The variables of the upper two
bonds are those of the original Karnopp scheme, and the variables on the third
bond are those commonly used in bond graphs representing mechanical systems.
As for the original pseudo-bond concept, the variables on each of these bonds are
not independent. There is, however, one major advantage inherent in this scheme.
The exchange of the conserved quantities, mass, momentum and energy, are each
related to a bond, and in this way a close connection between the bond graph and
the original conservation statements is assured. A direct implication of this choice
of bond graph variables is also that the conserved quantities become state variables.
Both the quantities, mass and energy, are time integrals of the corresponding flow
variables, and are therefore generalized displacement variables in bond graph
terminology. The third conserved quantity, momentum, may be considered as a
generalized momentum variable, where the time rate of change of momentum is a
force. This scheme is now used for the bond graph interpretation of the fluid
dynamics in one pipe element/control volume.
Bond graph constitutive laws are now to be recognized in the governing
equations. However, when looking at Eqs. (15), (17) and (1 S), it appears that no
single bond graph element or field may represent these relations. The convection
velocity, uir for instance, is a flow variable, and is, through Eq. (15) related to both
a displacement and a momentum variable, the mass, m,, and the momentum, p!,
respectively. So, neither a C-element nor an I-element may alone represent this
equation. The same conclusion results for Eqs. (17) and (18) : In Eq. (17), the effort
variable, T,, is expressed in terms of the displacement variables, mass, m,, and
energy, E,, and the momentum variable, p,. Pressure, P,, which is also an effort
variable, is likewise related to the mass, energy and momentum, through Eq. (18).
This leaves only one possible conclusion, a bond graph field having the properties
of both the C-fields and I-fields. Such mixed energy-storing fields are consequently
denoted by both the letters I and C (see Fig. 5). The generalized bond graph
constitutive laws for a 3-port IC-field, representing Eqs (I 5), (17) and (18) are thus
P
I
lil
T
E
F
u I
i-l i it1
1 I
pi_*’ ‘i-1
-t- PI, Til Pi’ ui ~ p, 11,
:+l’ I+1
I
-- -- I I
I
I-l’ 1I
4
FIG. 5. (a) Pipe control volume i schematic. (b) A micro-bond graph of fluid flow in control
volume i.
P, = 6,~:Cm,,P, , -6)
T = 4i.i (m,,p,, 4).
These are the standard integral forms of these laws, and occurs in a natural way
for pressure/temperature boundary conditions. Such IC-fields represent the internal
relations of any internal control volume, and elements/fields representing the
approximations of internal boundary terms may now be sought. The mass flow
rate between control volume i and if 1 is :
where density, p,. and velocity, u, are expressed from the respective properties in
adjacent control volumes, through Eq. (IO). Equation (22) relates only effort and
flow variables at the boundary. since density is determined from pressure and
temperature through an equation of state. Consequently, this equation is identified
with the generalized laws of restrictive fields. This is also the case for Eq. (23).
expressing the total energy flow rate across the same boundary :
where ~3, is given in Eq. (22). The force due to pressure and convected momentum
at the boundary is given by the following equation :
F, = PIA+n?,u,. (24)
This equation also relates effort and flow variables at the boundary, and none of
these three equations relate only one effort and one flow variable, but they involve
simultaneously several of the flow and effort variables of the proposed pseudo-bond
concept. This implies that these internal boundary relations may be represented by
a restrictive field, embodying all three equations. Consequently, the generalized
R-field laws are :
F, = ~R,‘(P,,P,+I~T~,TI+I~u,,~,+,). (27)
These equations are written in the form in which they naturally appear, and in that
sense are they possessing the natural causality for the R-field. This R-field does not
represent dissipative energy mechanisms, but rather represents the exchange of
mass, energy and momentum between the adjacent control volumes. Such R-fields
will result at all internal boundaries, and the resulting micro-bond graph for control
volume i is shown in Fig. 5.
The l-port R-element connected to the common flow junction at the bottom of
the graph is representing the frictional force. The l-port flow source element
connected to the common effort (temperature) junction is representing the total
heat flow, e,, added to the control volume. This element will vanish for the
adiabatic case. The overall bond graph model results, when interconnecting a
finite number of such micro-bond graphs, and including the appropriate boundary
conditions at the ends.
For the case when pressure and temperature are boundary conditions at both
ends, the pseudo-bond graph model representing the flow in open-open pipes
(independent of flow direction) is shown in Fig. 6. The third bond (the momentum
bond) is terminated in the R-fields at the ends, because there are no natural
continuation of this bond into the adjacent subsystem. The two remaining bonds
are those of the original Karnopp scheme. Thus, interconnection of this pipe flow
model to other submodels established within this scheme is simplified.
FIG. 6. An overall pseudo-bond graph model for fluid flow in an open-open pipe
FIG. 7. The Riemann shock tube solution: (a) pressure, (b) velocity, (c) temperature,
(d) density, (e) entropy
790
Bond Graph Interpretation of One-dimensional Fluid Flow
tube problem. This solution is obtained for non-viscous flows, using a quasi-steady
approach. In Fig. 7, the Riemann solution is compared with the corresponding
solution obtained for a control volume based model (the delta marks indicates the
condition in each control volume). The curves show the distribution of certain
flowfield properties 22 msec after the diaphragm has broken.
It appears from Fig. 7 that the solution obtained from the control volume based
model is a fairly good approximation to the Riemann shock tube solution, even
though the smearing of the discontinuities, especially the contact discontinuity, is
significant. The pressure and velocity solutions are very close to the exact solutions,
and the over- and undershoots, which are usually present in finite difference
schemes, are not present. The speeds of the propagating waves are also reproduced
exactly.
Finally the model was compared with experimentally obtained data. The test
facility shown in Fig. 8 was set up to provide such data.
Allowing the ellipsis-shaped symbol to represent the pipe flow model, the bond
graph model shown in Fig. 9 may represent this laboratory system.
In Fig. 10,the model is compared with pressure data recorded at three different
locations in the pipe (see Fig. 8). The solid lines represent the stimulated results
:: 0 0. 02 0.0. 0. oe 0. oe 0. IO
FIG. 10. Results from laboratory experiments and calculation at three locations in the pipe.
Non-choked flow.
using the proposed model. As shown experimental and simulated results compare
very well.
V. Conclusions
References
(1) D. L. Mar&ohs. “Bond graph fluid line models for inclusion with dynamic systems
simulations”, J. Franklin Inst., Vol. 308, No. 3, pp. 219-234, 1979.
(2) N. Nervik, “Bond graph model of compressible flow in tubes”, IFMM Report No. 84
07, Division of Marine Engineering, The Norwegian Institute of Technology, 1984.
(3) K. Strand, “A system dynamic approach to one-dimensional fluid flow”, Dr.lng. Thesis,
Division of Marine Engineering, The Norwegian Institute of Technology. 1986.
(4) M. Sandberg, “Instationlr striimning av vatska og gas i ror med variabel tvcrrsnitts-
area”, Bulletin M79:22, The National Swedish Institute for Building Research, 1979.
Bond Graph Intrrpretcrtion qf’One-dimensional Fluid Flovt
(5) J. Oliger and A. Sundstrom, “Theoretical and practical aspects of some initial boundary
value problems in fluid dynamics”, Siam J. Appl. Math., Vol. 35, Nov. 1978.
(6) J. Thoma, “Network thermodynamics with entropy stripping”, J. Franklin Inst., Vol.
303, No. 4, pp. 319-328. 1977.
(7) P. C. Breedveld, “Physical system theory in terms of bond graphs”. Ph.D. Thesis,
Twente University of Technology, Enschede, The Netherlands, Feb. 1984.
(8) D. C. Karnopp, ‘State variables and pseudo bond graphs for compressible thermofluid
systems”, J. Dynamic Systems Measurements Control, Vol. 101, No. 3. Sept. 1979.