2 141 Fall 2002 Modeling and Simulation of Dynamic Systems Volume III

2.
141 Fall 2002

Modeling and Simulation of Dynamic
Systems
Volume III
Neville Hogan
2.141 Fall 2002
Modeling and Simulation of Dynamic
Systems
Neville Hogan
MIT OpenCourseWare
Cambridge, MA
http://ocw.mit.edu/
2006 Massachusetts Institute of Technology
The material in this book is provided under a Creative Commons license that grants
you certain privileges to use, copy, or adapt the contents for non-commercial
educational purposes as long as you give credit to MIT OpenCourseWare and to the
faculty author, and you make any derivative works freely and openly available to
others under the same terms as our license. Please refer to the full text of the
license and other notices at the back of this book.
Printed and bound in the United States of America.
MIT OpenCourseWare
Building 9-213
77 Massachusetts Avenue
Cambridge, MA 02139-4307
USA
ISBN <<ENTER ISBN>>
2006XXXXXX
10 9 8 7 6 5 4 3 2 1
What is MIT OpenCourseWare?
MIT OpenCourseWare (OCW) is a remarkable story of an institution rallying around
an ideal, and then delivering on the promise of that ideal. It is an ideal that flows
from the MIT Faculty's passionate belief in the MIT mission, based on the conviction
that the open dissemination of knowledge and information can open new doors to
the powerful benefits of education for humanity around the world.
Available online at http://ocw.mit.edu, MIT OCW makes the MIT Faculty's course
materials used in the teaching of almost all of MIT's undergraduate and graduate
subjects available on the Web, free of charge, to any user anywhere in the world.
MIT OCW is a large-scale, Web-based publication of educational materials.
With 1300 courses now available, MIT OCW delivers on the promise of open sharing
of knowledge.
Educators around the globe are encouraged to utilize the materials for
curriculum development;
Self-learners and students may draw upon the materials for self-study or
supplementary use.
Course materials contained on the MIT OCW Web site may be used, copied,
distributed, translated, and modified by anyone, anywhere in the world. All that is
required of adopters of the materials is that the use be non-commercial, that the
original MIT faculty authors receive attribution if the materials are republished or
reposted online, and that adapters openly share the materials in the same manner as
OCW.
MIT OCW differs from other Web-based education offerings:
It is free and open;
It offers a unique depth and breadth of content; and,
It takes an institutional approach to online course publication.
MIT OCW is not a distance-learning initiative. Distance learning involves the active
exchange of information between faculty and students, with the goal of obtaining
some form of a credential. Increasingly, distance learning is also limited to those
willing and able to pay for materials or course delivery. MIT OCW is not meant to
replace degree-granting higher education or for-credit courses. Rather, the goal is to
provide the content that supports an education, for use by educators, students, and
self-learners to supplement their teaching and learning activities.
Truly a global initiative, the MIT OCW site has received users from more than 215
countries, territories, and city-states since the launch of the MIT OCW pilot site on
September 30, 2002. Materials have already been translated into at least 10
different languages.
MIT is committed to this project remaining free and openly available. MIT OCW is not
a degree-granting initiative, and there will not be a registration process required for
users to view course materials now, or in the future.
Contents
Highlights of this Course ...................................................................... 1
Course Description.............................................................................. 1
Syllabus ............................................................................................ 2
Calendar............................................................................................ 4
Lecture Notes ..................................................................................... 8
Assignments .................................................................................... 11
Assignment 3: Gas-charged Accumulator, Solution and
Commentary
Assignment 4: Control through Singularities, Solution and
Commentary
Assignment 5: Simple Convection and Throttling, Solution and
Commentary
Projects ........................................................................................... 75
Suggested Term Project Topics
Sample Student Projects
Study Materials................................................................................135
Introductory Modeling Notes
Old Quiz for Study
Highlights of this Course
This course features an extensive collection of lecture notes
plus background study materials on modeling.
Course Description
This course deals with modeling multi-domain engineering systems at
a level of detail suitable for design and control system implementation.
Topics covered include network representation, state-space models;
multi-port energy storage and dissipation, Legendre transforms,
nonlinear mechanics, transformation theory, Lagrangian and
Hamiltonian forms and control-relevant properties. Application
examples may include electro-mechanical transducers, mechanisms,
electronics, fluid and thermal systems, compressible flow, chemical
processes, diffusion, and wave transmission.
Course Meeting Times
Lectures:
Two sessions / week
1.5 hours / session
Level
Graduate
This bond graph models the free-flight and contact behaviors of a ball bouncing off of
a another ball. (Image courtesy of Prof. Neville Hogan.)
1
Syllabus
Prerequisites
2.151 or equivalent exposure to physical system modeling -- see me
to clarify.
Textbook
Required
Brown, Forbes T. Engineering System Dynamics. New York: Marcel
Dekker, Inc., 2001. ISBN: 0824706161.
Course Description
This course is about Modeling multi-domain engineering systems at
a level of detail suitable for design and control system implementation.
It also describes Network representation, state-space models,
Multiport energy storage and dissipation, Legendre transforms,
Nonlinear mechanics, transformation theory, Lagrangian and
Hamiltonian forms, Control-relevant properties. The application
examples may include electro-mechanical transducers, mechanisms,
electronics, fluid and thermal systems, compressible flow, chemical
processes, diffusion, and wave transmission.
2
Grading Policy
Homework
About 6 homework problems will be assigned throughout the term at
approximately two-week intervals.
Term project
Students will be required to select a term project by lecture 8. A brief
interim progress report will be required by lecture 16. Term projects
must be completed by the last week of the term, at which time a final
report will be due. A brief oral presentation of each term project will
also be required.
ACTIVITIES PERCENTAGES
Homework 60%
Term Project 40%
Ethics Policy
Collaboration
Collaboration on and discussion of homework assignments is
encouraged but each student must submit an individual solution.
Collaboration on term projects is encouraged provided some means for
clearly identifying individual contributions is proposed and approved by
the instructor.
Use of Material from Previous Years
Use of material from prior offerings of this subject in preparing
homework assignments defeats the purpose of the assignments and is
forbidden.
3
Calendar
LEC # TOPICS KEY DATES
Week 1
1
Introduction
Multi-domain Modeling
Syllabus, Policies and Expectations
Assignment 1
out
Week 2
2
Review: Network Models of Physical System
Dynamics
Bond-graph Notation
3
Review (cont.): Equivalent Behavior in
Different Domains
Block Diagrams, Bond Graphs, Causality
Week 3
4
Thvenin and Norton Equivalent Networks
Impedance Control and Applications
5
Energy-storing Coupling between Domains
Multi-port Capacitor
Maxwell's Reciprocity
Assignment 1
due
Assignment 2
out
Week 4
6
Energy, Co-energy, Legendre
Transformation, Causal Assignment
Intrinsic stability
4
7
Review Revisited: Magnetism and Electro-
magnetism
Week 5
8
Electro-magnetic-mechanical Transduction
Use of Co-energy Functions
Term project
proposal due
9 Linearized Energy-storing Transducer Models
Assignment 2
due
Assignmnet 3
out
Week 6
10
Cycle Processes
Work-to-heat Transduction
Thermodynamics of Simple Substances
11
Causal Assignments and Co-energy
Functions
Second Law for Heat Transfer
Multi-port Resistors
Week 7
12
Nonlinear Mechanical Systems
Modulated Transformers and Gyrators
Assignment 3
due
Week 8
13
Lagrangian Mechanics
Coordinates, State Variables and
Independent Energy Storage Variables
14 Nonlinear Mechanical Transformations and Assignment 4
5
Impedance Control out
Week 9
15
Hamiltonian Mechanics
Stable Interaction Control
Canonical Transformation Theory
16 Term Project Progress Report Discussion
Term project
progress
report due
Week 10
17
Identification of Physical and Behavioral
Parameters
Model structure
18
(Internally) Modulated Sources
Non-equilibrium Multi-port Resistors
Nodicity
Assignment 4
due
Assignment 5
out
Week 11
19
Amplifying Processes
Small-signal and Large-signal Models
20
Thermodynamics of Open Systems
Convection and Matter Transport
Lagrangian vs. Eulerian Frames
Week 12
21
Power Conjugates for Matter Transport
Second Law for Non-heat-transfer Processes
6
Throttling and Mixing
22
Bernoulli's Incompressible Equation
The "Bernoulli Resistor" and "Pseudo-bond-
graphs"
Assignment 5
due
Week 13
23
Chemical Reaction and Diffusion Systems
Gibbs-Duhem equation
Week 14
24
Control-relevant Properties of Physical
System Models
Causal Analysis, Relative Degree,
Passivity and Interaction Stability
25
Transmission Line Models
Term Project Presentations
Week 15
26
Wrap-up Discussion
Term Project Presentations (cont.)
7
Lecture Notes and Topics
Lecture # Topic
Lecture 1 Introduction; Multi-domain
Modeling
Lecture 2 Review: Network Models of
Physical System Dynamics; Bond
Graph Notation, Block Diagrams,
Causality
Lecture 3 Review (cont.): Equivalent
Behavior in Different Domains
Lecture 4 Thevenin and Nortan Equivalent
Networks; Impedance Control and
Applications
Lecture 5 Energy-storing Coupling between
Domains; Multi-port Capacitor;
Maxwells Reciprocity
Lecture 6 Energy, Co-energy, Legendre
Transformation, Causal
Assignment; Intrinsic stability
Lecture 7 Review Revisited: Magnetism and
Electro-magnetism
Lecture 8 Electro-magnetic-mechanical
Transduction; Use of Co-energy
Functions
Lecture 9 Linearized Energy-storing
Transducer Models
Lecture 10 Cycle Processes; Work-to-heat
Transduction; Thermodynamics of
Simple Substances
8
Lecture 11 Causal Assignments and Co-energy
Functions; Second Law for Heat
Transfer; Multi-port Resistors
Lecture 12 Nonlinear Mechanical Systems;
Modulated Transformers and
Gyrators
Lecture 13 Lagrangian Mechanics;
Coordinates, State Variables and
Independent Energy Storage
Variables
Lecture 14 Nonlinear Mechanical
Transformations and Impedance
Control
Lecture 15 Hamiltonian Mechanics; Stable
Interaction Control; Canonical
Transformation Theory
Lecture 16 Term Project Progress Report
Discussion
Lecture 17 Identification of Physical and
Behavioral Parameters; Model
Structure
Lecture 18 (Internally) Modulated Sources;
Non-equilibrium Multi-port
Resistors; Nodicity
Lecture 19 Amplifying Processes; Small-signal
and Large-signal Models
Lecture 20 Thermodynamics of Open Systems;
Convection and Matter Transport;
Lagrangian vs. Eulerian Frames
Lecture 21 Power Conjugates for Matter
Transport; Second Law for Non-
heat-transfer Processes; Throttling
and Mixing
9
Lecture 22 Bernoullis Incompressible
Equation; The Bernoulli Resistor
and Pseudo-bond-graphs
Lecture 23 Chemical Reaction and Diffusion
Systems; Gibbs-Duhem equation
Lecture 24 Control-relevant Properties of
Physical System Models; Causal
Analysis, Relative Degree, Passivity
and Interaction Stability
Lecture 25 Transmission Line Models; Term
Project Presentations
Lecture 26 Wrap-up Discussion; Term Project
Discussions (cont.)
10
Massachusetts Institute of Technology
Department of Mechanical Engineering
2.141 Modeling and Simulation of Dynamic Systems
Assignment #3 Out: 10/3/02
Due: 10/17/02
Gas-charged accumulator
Gas-charged accumulators are commonly used in hydraulic systems, primarily to reduce the
magnitude of the pressure transients resulting from abrupt changes in flow rate. The attached
paper considers the energy-dissipating effects of work-to-heat transduction and heat transfer in a
gas-charged accumulator. The authors point out that the common polytropic model (PV
n
=
constant) used to characterize the pressure-volume relation in the gas fails to account for
"thermal damping" and present a nonlinear model, a linear model and experimental data to
support their point. You are to critique this paper by modeling and simulating the accumulator
system.
Background reading: A. Pourmovahed & D. R. Otis (1984) "Effects of Thermal Damping on the
Dynamic Response of a Hydraulic Motor-Accumulator System", J.D.S.M.C., 106:21-26.
1. Formulate a model of the thermofluid capacitive subsystem of P. & O. '84, fig. 5 and
represent it by a bond graph. Assuming the flow rate Q
a
as an input and that nitrogen may be
modeled as an ideal gas, derive nonlinear dynamic equations suitable for simulating figs. 1, 3
& 4 of P. & O. '84.
2. P. & O. '84 devote much of their paper to a linear analysis.
2a. Linearize your model about similar operating conditions. Comment on the usefulness of
the "anelastic model" of P. & O. fig. 2.
2b. Develop a bond graph corresponding to your linearized model using single-port storage
and dissipative elements and idealized transducers (e.g. gyrators, transformers) and show
that it yields the same linearized equations.
2c. Does your linearized model describe entropy production? Is entropy production an
essentially nonlinear phenomenon?
3a. Using the two models developed above, determine parameters for simulations corresponding
to P. & O.'s experimental results and simulate the accumulator to obtain cross plots
corresponding to P. & O. '84 figs. 1, 3 & 4.
3b. Comment on the degree of (dis)agreement between these models and P. & O.'s experiments
and simulations; that is, how plausible is the ideal gas assumption in this case? (Bear in
mind that your nonlinear model is different from the one P. & O. used -- see their reference
#3.)
2.141 2002 page 1 assignment #3
11
4. Develop a model of the complete hydraulic system of P. & O. '84, fig. 5 and represent it by a
bond graph. Assuming the flow rate Q
s
and the force F as inputs and that nitrogen may be
modeled as an ideal gas, derive nonlinear dynamic equations suitable for simulating the
transient response of the system. Hint: Keep in mind that your choice of state variables can
influence the complexity of your equations.
5. Assuming that both inputs are zero (Q
s
= 0, F = P
o
A
m
) simulate the pressure in the
accumulator in response to an initial velocity of the mass in the direction which would
compress the gas and large enough to cause a volume change comparable to that of P. & O.
'84 fig. 1. Assume the gas is initially at equilibrium with ambient conditions. Choose
parameters so that the undamped natural frequency of the system is 0.01Hz (as in figs. 1, 3 &
4.) Given an accumulator volume of 2 liters, the piston diameter of a corresponding linear
actuator would probably be on the order of a couple of centimeters (i.e., a half to one inch).
Simulate three cases:
5a. isothermal conditions gas temperature constant at ambient temperature
5b. adiabatic conditions no heat transfer from the gas
5c. the conditions corresponding to P. & O.'s experimental data.
5d. Now choose parameters so that the undamped natural frequency of the system is ten times
higher (i.e., 0.1Hz) and repeat 5c. above.
Based on this analysis, what would you conclude about the importance of the "thermal damping"
phenomenon?
12
2.141 Assignment #3: GAS-CHARGED ACCUMULATOR
The figure below (after Pourmovahed & Otis, 1984) is a schematic diagram of the hydraulic and
mechanical system.
gas
oil
motor
m
F
y
piston or
diaphragm
Q
m
Q
s
A
m
P
Q
a
P
P, v
accumulator
Our first goal is to model the thermofluid subsystem assuming the flow rate Q
a
as an input. The
following bond graph is the simplest representation of the accumulator subsystem.
R
T
w
dS
w
/dt
C
P
dV/dt
dS/dt
S
f
S
e
Q
a
(t):
hydraulic domain thermal domain
T
:T
w
The two-port capacitor on the left represents the reversible energy storage and work-to-heat
transduction in the ideal gas in the accumulator. The two-port resistor on the right represents
irreversible heat conduction through the wall of the accumulator with the concomitant entropy
production. The outside of the wall is assumed to remain at a constant temperature (i.e., ambient
temperature).
The direction of the power bonds has been chosen to follow the usual convention that power is
positive into the storage element. However, the sign convention for the two-port resistor follows
a more intuitive power in = power out convention but assumes for convenience that positive
heat flow is from the wall to the gas.
Nonlinear equations
Causal assignment indicates that both ports of the capacitor may be given the preferred integral
causal form. Note that the two-port resistor assumes its preferred causal form with temperatures
as inputs. Thus we may be confident that the model will properly reflect the second law.
We must choose state variables. There are many possible choices, but they are not equally
simple.
Gas-charged accumulator solution page 1 Neville Hogan
13
Hydraulic side:
We could choose either pressure, P, or volume, V, as state variable but volume seems to be the
most sensible choice as the corresponding state equation is simply
V = Q
a
(t)
The negative sign is due to the fact that positive flow rate compresses the gas.
Thermal side:
We might choose the energy variable, total entropy, S, but temperature, T, is more common. We
could also use total internal energy, U, as a state variable. To choose between these, consider the
available information.
Two-port capacitor constitutive equations are derived assuming nitrogen may be modeled as an
ideal gas. The ideal gas equation is
PV = m RT
where m is the mass of the gas and R is the gas constant for nitrogen (not the universal gas
constant). The second constitutive equation is derived by assuming
dU = m c
v
dT
where c
v
is the specific heat at constant volume. Integrating:
U - U
o
= m c
v
(T T
o
)
where subscript o denotes reference state.
The two-port resistor equations assume Fouriers-Law (i.e., linear) heat conduction.
Q = h A
w
(T
w
T)
where h is a heat transfer coefficient and A
w
and T
w
are the area and temperature of the wall.
Consider using S and V as state variables.
T
S =
Q = h A
w
(T
w
T)
hence
S = h A
w
(
T
w
T
1)
However, I need T = T(S,V) and P = P(S,V) to form state equations. From class notes,
T = T
o
\
|
.
|
|
V
V
o
R
c
v
exp
\
|
.
|
|
S S
o
mc
v
P = P
o
\
|
.
|
|
V
V
o
|
.
|
|
R
c
v
+ 1
exp
\
|
.
|
|
S S
o
mc
v
These are output equations. State equations are
14
S = h A
w
\
|
.
|
|
|
T
w
T
o
\
|
.
|
|
V
V
o
R
c
v
exp
\
|
.
|
|
S
o
S
mc
v
1
V = Q
a
These are highly nonlinear, coupled differential equations. Note that the thermal time-constant
is far from obvious.
Consider using U and V as state variables.
Pressure and temperature are determined as follows by rearranging the expression for internal
energy.
T =
U
m c
v
+ T
o
P =
m RT
V
=
m R
V
\
|
.
|
|
U
m c
v
+ T
o
These are output equations. State equations may be found from an energy balance equation (the
first law):
U =
Q P
V = Q
a
U = h A
w
\
|
.
|
|
U
m c
v
+ T
o
T
w
+
m R
V
\
|
.
|
|
U
m c
v
+ T
o
Q
a
These equations are also nonlinear and coupled. However, the thermal time constant
1
t
=
h A
w
m c
v
is clearly identified as may be seen by re-arranging.
U =
\
|
.
|
|
h A
w
m c
v
U + h A
w
(T
w
T
o
) +
m R
V
\
|
.
|
|
U
m c
v
+ T
o
Q
a
Note that choosing absolute zero for the reference state would simplify this equation to
U =
\
|
.
|
|
h A
w
m c
v
U + h A
w
T
w
+
U
V
\
|
.
|
|
R
c
v
Q
a
Finally consider using T and V as state variables.
(Note that this is loosely analogous to choosing displacement and velocity the Lagrangian
choice as the state variables to describe a mechanism.)
U =
Q P
U = m c
v
T = h A
w
(T
w
T) +
m RT
V
Q
a
15
T =
h A
w
m c
v
T +
h A
w
m c
v
T
w
+
R
c
v
T
V
Q
a
V = Q
a
These state equations are still coupled and nonlinear, but look a little simpler. Note that the
thermal time constant is again obvious but now the choice of reference temperature is less
important. The output equation for pressure is
P =
m RT
V
The equations using energy as a state variable are similar to those using temperature as a state
variable. That shouldnt be surprising given the simple relation between temperature and internal
energy for and ideal gas. The point to remember is that the integral causal form does not
constrain your choice of state variables. State variables should be chosen for convenience and to
maximize insight. I will use temperature and volume as state variables.
Linearized equations
The second task is to linearize the model and compare to P&Os linearized model. Linearizing
the state equations is always helpful. In this case it can provide insight into the physical system
and the model and parameters used by P&O.
The two-port capacitor constitutive equation is derived from the ideal gas equation PV = m RT.
To linearize, we need to write this in a causal formany of the four causal forms will serve. Use
the causal form with temperature and volume as (integrated) inputs, i.e.
C
P
dV/dt
dS/dt
T
P =
m RT
V
Denote the operating point by subscript o and linearize.
P P
o
-
m R
V
o
(T T
o
)
m RT
o
V
o
2
(V V
o
)
Write AV = V V
o
, AT = T T
o
and AP = P P
o
A -
m R
V
o
AT +
m RT
o
V
o
2
(AV)
Remember that the positive work compresses the gas, hence AV is the appropriate displacement
variable on the hydraulic side.
This indicates that the hydraulic side of the linearized model may be represented by a
transformer between thermal and hydraulic domains connected to a fluid capacitor by a 1-
junction (common flow).
transformer modulus:
m R
V
o
16
(inverse) fluid capacitance:
m RT
o
V
o
2
The second constitutive equation is in the form S = S(T,V) It may be derived from the internal
energy equation and the first law.
dU = m c
v
dT = T dS P dV
dS =
m c
v
T
dT +
m R
V
dV
S =
m c
v
T
o
T
m R
V
o
(
V)
Again, remember that
V is the appropriate flow variable of the hydraulic side given the power
sign we have assumed.
This indicates that the thermal side of the linearized model may be represented by a thermal
capacitor connected to a transformer between thermal and hydraulic domains by a 0-junction
(common effort).
transformer modulus:
m R
V
o
(as before)
thermal capacitance:
m c
v
V
o
A bond graph of the linearized two-port capacitor is as follows.
TF
dV/dt T
1 0
C C
: m c
v
/T
o
m R/V
o
: V
o
2
/m R T
o
Note that this linearized model indicates that the strength of the coupling between the thermal
and mechanical domains is proportional to the mass of gas and inversely proportional to the
volume it occupiesa useful insight.
The two-port resistor equations describe entropy flow as a function of temperature. For the gas
S =
Q
T
= h A
w
\
|
.
|
|
T
w
T
1
Linearizing
S
o
-
\
|
.
|
|
h A
w
T
w
T
o
2
(T
o
T)
S - h A
w
\
|
.
|
|
T
w
T
o
1
\
|
.
|
|
h A
w
T
w
T
o
2
(T
o
T)
17
Assume the operating point (and also the reference thermal state) is at equilibrium with the wall
temperature.
T
o
= T
w
S -
\
|
.
|
|
h A
w
T
o
(T
o
T)
Under the same operating conditions, the entropy flow rate from the wall is identical
S
w
=
Q
T
w
S
w
-
\
|
.
|
|
h A
w
T
o
(T
o
T)
A bond graph of this linearization of the two-port resistor is as follows.
S
e
dS/dt
1
R
: T
o
/h A
w
:T
w
Note that this linearized model does NOT reflect the second law of thermodynamics
linearization has eliminated entropy production due to heat transfer. This is due to the particular
choice of operating point. By choosing T
o
= T
w
we make the entropy flow from the wall
identical to the entropy flow to the gas.
S
net
=
S
w
= 0
However, entropy production is not an essentially nonlinear phenomenonits the choice of
operating point that matters. Linearization about equilibrium eliminates entropy production; a
model linearized about a non-equilibrium operating point will describe entropy production.
A bond graph of the complete linearized system is as follows.
S
e
dS/dt
1
R
: T
o
/h A
w
:T
w TF
dV/dt T
1 0
C C
m c
v
/T
o
:
m R/V
o
V
o
2
/m R T
o
:
S
f
Q
a
:
Using the causal assignment shown, linearized equations may be read from the graph.
A
T =
T
o
m c
v \
|
.
|
|
A
S
m R
V
o
A
V
18
A
S =
h A
w
T
o
AT
State equations
A
V = Q
a
A
T =
h A
w
m c
v
AT +
R
c
v
T
o
V
o
Q
a
Output equation
AP =
m R
V
o
AT
m RT
o
V
o
2
AV
To compare with P&Os linearized equations, take the Laplace transform.
AT
\
|
.
|
|
s +
1
t
=
R
c
v
T
o
V
o
Q
a
AV =
Q
a
s
AP =
m R T
o
V
o
2
\
|
.
|
|
|
R
c
v
+ s +
1
t
(s +
1
t
s)
Q
a
Substitute
m R T
o
V
o
2
=
P
o
V
o
R
c
v
+ 1 =
R + c
v
c
v
=
P
Q
a
=
P
o
V
o
t s + 1
s (t s +1)
This is the same as derived by P&O.
The complete hydraulic system
Adding the rest of the system is straightforward. The piston can be modeled as a transformer
linking the fluid domain with the mechanical domain. The piston itself is simply a mass with a
force acting on it (though with an unusual sign convention). A bond graph is shown below.
19
S
e
dS/dt
R
: F
:T
w
TF
P
T
0 1
I
C
: 1/M
A
m
S
f
Q
s
:
S
e
dV/dt
v
m
The translational inertia is an ideal linear element. It makes little difference whether momentum
or velocity is used as its state variable. For clarity I will use velocity with temperature and
volume as the gas state variables. State and output equations for the complete system are as
follows.
V = A
m
v
m
Q
s
T =
h A
w
m c
v
( )
T
w
T +
R
c
v
T
V
( )
A
m
v
m
+ Q
s
v
m
=
1
M
\
|
.
|
|
m R A
m
T
V
F
P =
m RT
V
where A
m
is the piston area, v
m
is the velocity of the inertia and M its mass.
20
Parameters for simulation
For P&O fig. 1:
P
o
= 1014.5 psia (operating pressure in the accumulator)
V
o
= 103 cu. in. (operating volume of the accumulator)
m = 0.293 lbm (mass of gas)
AV/V
o
= 0.25
For P&O fig. 3:
P
o
= 438.5 psia
V
o
= 122 cu. in.
m = 0.15 lbm
AV/V
o
= 0.05
For P&O fig. 4:
P
o
= 436.7 psia
V
o
= 122 cu. in.
m = 0.149 lbm
AV/V
o
= 0.01
In all cases:
T
w
= 540R (room temperature)
R = 660 in-lbf/lbm R (gas constant)
c
v
= 1900 in-lbf/lbm R
t =
m c
v
h A
w
= 15.3 seconds
For the complete hydraulic system
To estimate the effective mass I assumed isothermal conditions and used the linearized model.
Isothermal conditions keeps things simple because it essentially eliminates the thermal domain
from participating in the dynamics. The undamped natural frequency is given by
e
n
=
P
o
A
m
2
M V
o
hence the estimated mass is
M =
P
o
A
m
2
V
o
e
n
2
21
Assuming a piston diameter of 0.75 inches, A
m
is 0.44 inches squared. Using the pressure and
volume of P&O fig.1 and an undamped natural frequency of 0.01 Hertz yields
M =
1014.5 x 0.44
2
102.97 x (2 t x 0.01)
2
= 487
lbf
in
2
in
4
in
3
sec
2
rad
2
Convert to more meaningful units:
1
lbf sec
2
in
= 386.4 lbm
M = 1.88 x 10
5
lbm.
Thus the mass required to yield an undamped natural frequency that low is unbelievably large
84 tons!
However, the mass required to yield an undamped natural frequency of 0.1 Hz is a little more
reasonable1,880 pounds.
To choose an initial velocity I assumed undamped oscillations and used
v
m
=
V
A
m
V = AV sin e
n
t
v
m
(0) =
AV e
n
A
m
Simulations
Nonlinear simulations of P&O figs. 1, 3 and 4 are as follows.
-0.25 -0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
P&O fig. 1, nonlinear
normalized volume
n
o
r
m
a
l
i
z
e
d

p
r
e
s
s
u
r
e
22
-0.05 0 0.05
410
420
430
440
450
460
470
normalized volume
p
r
e
s
s
u
r
e

(
p
s
i
)
-0.1 -0.05 0 0.05 0.1
380
400
420
440
460
480
500
normalized volume
p
r
e
s
s
u
r
e

(
p
s
i
)
Note that in all cases these simulations are quite close to P&Os even though those authors used
a much more elaborate model of the gas. For 10% volume changes, the ideal gas model is
essentially identical to their data. For 5% volume changes the ideal gas model is essentially
linear.
23
Transient responses
Simulations of the transient responses for the large mass assuming thermally damped, adiabatic
and isothermal conditions are as follows.
0 50 100 150 200 250 300 350 400
-4
-2
0
2
4
Transient response of complete nonlinear system
time (seconds)
v
e
l
o
c
i
t
y

(
i
n
c
h
/
s
e
c
)
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25
-0.4
-0.2
0
0.2
0.4
normalized volume
n
o
r
m
a
l
i
z
e
d

p
r
e
s
s
u
r
e
The plot shows the velocity of the piston mass. Note that the oscillation is steadily losing
amplitude as time increases. Also note that no mechanical losses such as friction have been
included in our model. This loss in energy is due to the irreversible entropy production due to
conduction in the gas-charged accumulator. Assuming only a reversible work storage in the gas
accumulator would not predict this behavior. Just like a bicycle pump, compression of the gas
generates heat that flows through the walls of the chamber. Not all of this heat energy is
recovered when the gas expands, so energy is lost and we have thermal damping.
The adiabatic case was modeled by adding an ideal flow source on the thermal side of the two-
port capacitor. Setting this flow source to zero sets the entropy flow to zero, i.e. zero entropy
flows which in this case means no heat flow. The result of this simulation shows oscillations as
in the previous case. However, the amplitude of the oscillations does not decrease. We have
essentially removed the heat lost through the wall leaving only the reversible work storage in
compressing the gas. The system without thermal damping behaves like an undamped spring-
mass system.
24
0 50 100 150 200 250 300 350 400
-4
-2
0
2
4
Adiabatic transient response
time (seconds)
v
e
l
o
c
i
t
y

(
i
n
c
h
/
s
e
c
)
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25
-0.4
-0.2
0
0.2
0.4
normalized volume
n
o
r
m
a
l
i
z
e
d

p
r
e
s
s
u
r
e
The isothermal case was modeled by setting dT/dt = 0 and setting the initial gas temperature
equal to T
w
. This is equivalent to assuming no thermal resistance across the wall of the
accumulator. Hence there is no entropy production and again, no thermal damping.
0 50 100 150 200 250 300 350 400
-4
-2
0
2
4
Isothermal transient response
time (seconds)
v
e
l
o
c
i
t
y

(
i
n
c
h
/
s
e
c
)
-0.3 -0.2 -0.1 0 0.1 0.2 0.3
-0.4
-0.2
0
0.2
0.4
normalized volume
n
o
r
m
a
l
i
z
e
d

p
r
e
s
s
u
r
e
The isothermal and adiabatic cases are undamped, as expected. Note that the isothermal
frequency of oscillation is lower than the damped case which in turn is lower than the adiabatic
case. This is also as expected.
25
Finally, a simulation of the transient response for the smaller (more reasonable) mass is as
follows.
0 20 40 60 80 100 120 140 160 180 200
-40
-20
0
20
40
Transient response for reduced piston mass
time (seconds)
v
e
l
o
c
i
t
y

(
i
n
c
h
/
s
e
c
)
-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25
-0.4
-0.2
0
0.2
0.4
normalized volume
n
o
r
m
a
l
i
z
e
d

p
r
e
s
s
u
r
e
Note that even though the normalized pressure-volume plot shows very little evidence of energy
dissipation (i.e., the area inside the loop is small) the effect on the transient response is
substantial. I conclude that thermal damping is likely to be a real phenomenon in practical
dynamic systems.
26
Not used:
U = mc
v
T
o
(
(
(
\
|
.
|
|
V
V
o
R
c
v
exp
\
|
.
|
|
S S
o
mc
v
1 + U
o
Choose an operating point at thermal equilibrium with the environment. Assume no forcing, i.e.,
Q
a
= 0
T
o
= T
w
For P&O fig. 4:
P
o
= 69.95 bar = 1014.5 psia (operating pressure in the accumulator)
V
o
= 1687cc = 103 cu. in. (operating volume of the accumulator)
Gas properties of nitrogen @ 300K (room temperature):
c
p
= 1041 Joules/kg K (specific heat at constant pressure)
= 1.399 (polytropic index)
R = 297 Joules/kg K (gas constant)
Remember: c
v
=
c
p
= c
p
- R
c
p
= 2560 in-lbf/lbm R
= 2560/1900 = 1.35
27
MEMORANDUM
To: 2.141 class
From: Professor Neville Hogan
Date: November 26, 2002
Re: Assignment #3
All of you did well on this assignment. Maybe Im making them too easy...
One common difficulty was to conclude that entropy production is an essentially nonlinear
phenomenon. This is a subtle problem: if you did the obvious (as I didsee my solution) and
linearized the two-port resistor about conditions with the same (ambient) temperature on both
sides, you linearized about equilibrium. That corresponds to assuming a reversible process and
hence the linearized model will not describe entropy generation. However, if you linearize about
non-equilibrium conditions (i.e., different temperatures on the two ports of the resistor,
corresponding to a non-zero heat flux) the resulting linearized model will describe entropy
production.
Class statistics on this part: mean 93%; standard deviation 4%.
28
Due: 11/7/02
Control Through Singularities
Motivation: One common and productive use of modeling and simulation is to study and
understand proposed engineering designs. In this assignment you will test the behavior of a
proposed robot controller.
A common form of robot motion control specifies a workspace position or trajectory (e.g., a
desired time-course of tool position in Cartesian coordinates) and transforms that specification to
a corresponding configuration-space position or trajectory (e.g., a time-course of joint angles).
However, most robot mechanisms have kinematic singularities, configurations at which the
relation between workspace and configuration-space becomes ill-defined. As a result, most robot
motions controllers do not operate at or near these singular points.
However, an energy-based analysis of mechanics shows that the transformation of positions and
velocities is well-defined in one direction while the transformation of efforts and momenta is
well-defined in the other. This implies that a controller that takes advantage of these facts should
be able to operate at and close to mechanism singularities. A simple impedance controller
attempts to impose the workspace behavior of a damped spring connected to a movable virtual
position.
( ) ( ) ( ) ( ) ( ) { } J x B L x K J
&
& + =
o o
t
where u is a vector of generalized coordinates, t a vector of (conjugate) generalized efforts, x
and x
o
are vectors of actual and virtual Cartesian tip coordinates, L() and J() are linkage
kinematic equations and Jacobian respectively, and K and B are stiffness and damping
respectively. Note that this controller requires neither the inverse of the kinematic equations nor
the inverse of the Jacobian.
In this assignment you are to test whether this simple impedance controller can operate at and
close to mechanism singularities.
Tasks:
Assume a planar mechanism open-chain mechanism with two links of equal length L = 0.5 m,
operating in the horizontal plane (i.e., ignore gravity) and driven by ideal controllable-torque
actuators, one driving the inner (shoulder) link relative to ground, the other driving the outer
(elbow) link relative to the inner link. Sensors mounted co-axially with the actuators provide
measurements of joint angle and angular velocity.
29
1. Write kinematic equations relating generalized coordinates to the position coordinates of the
tip, expressed in a Cartesian coordinate frame with its origin at the axis connecting the inner
link to ground.
2. Find the corresponding Jacobian (to relate generalized velocities to tip Cartesian velocities).
3. Identify the set of singular configurations for this linkage (at which the relation between tip
Cartesian coordinates and joint angles becomes ill-defined) and show that they include the
center of the workspace as well as at the limits of reach.
4. Formulate a dynamic model of the mechanism relating input actuator torques to output
motion of the tip in Cartesian coordinates. Assume the links are rods of uniform cross section
and mass m = 0.5 kg and that the joints are frictionless. (Hint: be careful in your choice of
generalized coordinates.)
5. Simulate the behavior of this mechanism under the action of a simple impedance controller.
Assume uniform tip stiffness and damping (i.e., the stiffness and damping matrices have the
form: and where k and b are constants).
(
=
1 0
k K
0 1 0 1
(
=
1 0
b B
5.a Choose the stiffness and damping matrices so that when the mechanism is making small
motions about a configuration with the inner link aligned along the Cartesian x-axis and
the outer link aligned parallel to the Cartesian y-axis the highest-bandwidth transfer
function between virtual and actual position has critically-damped poles with an
undamped natural frequency of 2 Hz. (Hint: it may be easiest to first transform the
stiffness and damping to generalized coordinates.)
5.b Simulate the response to the following virtual trajectories and plot at least the path of the
tip in Cartesian coordinates:
1) Starting from rest at {x = L, y = 0} ending at rest at {x = 2L, y = 0}; trapezoidal speed
profile with acceleration to peak speed in 250 ms, constant speed for 1.5 sec and
deceleration to rest in 250 ms.
2) Starting from rest at {x = -L, y = 0} ending at rest at {x = L, y = 0}; trapezoidal speed
profile with acceleration to peak speed in 250 ms, constant speed for 1.5 sec and
3) Starting from rest at {x = -L, y = L/20} ending at rest at {x = L, y = L/10}; trapezoidal
speed profile with acceleration to peak speed in 250 ms, constant speed for 1.5 sec and
6. Repeat the simulation of 5.b.3 with k and b chosen to yield critically-damped poles with an
undamped natural frequency of 20 Hz.
7. Repeat the simulation of 5.b.3 with the outer link 10% shorter than the inner link. How does
this change the set of singular configurations?
30
8. Comment briefly on (a) whether and (b) how well this controller operates near
mechanism singularities.
Tent-Pole Instability
The complex behavior of multi-body mechanical systems stems from kinematically modulated
transformation of energy. It can give rise to some highly counter-intuitive phenomena, one of
which you are to explore in this assignment.
A tent-pole is to be supported in the upright position by guy-ropes. However, the support is not
completely rigid as the guy ropes inevitably have some elasticity. It may seem natural to increase
the tension in the ropes to stiffen the support but in fact, this can have the opposite effect. To
gain insight you are to develop a model and numerical simulation to explore this phenomenon.
Assume the tent pole is a rigid rod of height, h = 2 m, and that it is mounted on a hinge at its base
so that it can only move in a plane. Assume that two elastic cords are attached to its top and to
the ground in the plane of motion at a horizontal distance, w = 1 m, on either side. Assume the
cords have negligible mass, linear elasticity (i.e., obeying Hooke's law) with stiffness, k, and
pretension, F
p
.
1. Develop a model to describe the mechanical system comprised of the pole and the two
cords.
To understand this system, simulate the following cases:
2. Assume the cords have pretension but zero stiffness; assume the pole starts in the upright
position; and assume no gravity (for now). Show that this mechanical system is unstable
for any value of F
p
even in the absence of gravity.
3. Assume the cords have non-zero stiffness but no pretension; assume the pole starts in the
upright position; and assume no gravity. Test whether this system is stable for (a) small
perturbations (e.g., initial conditions 1 from upright) and (b) large perturbations (e.g.,
initial conditions 45 from upright).
4. Now that you understand what is going on, include both stiffness and pretension in the
cords and include gravity and any other physical behavior relevant to system stability.
Using the numerical simulation, show that for a given cord stiffness there is a maximum
pretension for stable behavior and find how it varies with stiffness (a few representative
values will suffice).
This system is simple enough to be amenable to symbolic (rather than numerical) analysis.
5. Derive an algebraic expression for the relation between the maximum pretension for
stable upright behavior (i.e., corresponding to 4. above). How well do your numerical
simulation results agree with your analytical results?
31
Thomas A. Bowers
2.141 Fall 2002
Assignment 4: Kinematics
Two-Link Planar Robotic Mechanism
1. The configuration being analyzed in this problem is shown below in Figure 1.
u
2
u
1
L
1
L
2
(x,y)
(x
c1
, y
c1
)
(x
c2
, y
c2
)
Figure 1: Cartesian and Generalized Coordinates of 2-Link Robotic Mechanism
The equations relating the Cartesian coordinates to the generalized coordinates, u
1
and u
2
,
are as follows:
x = L
1
cosu
1
+ L
2
cosu
2
y = L sinu + L
2
sinu
2 1 1
L
1
c1
= cosu
1
2
L
1
y = sinu
1
2
c1
= L
1
cosu +
L
2
cosu
2
2
x
c2 1
y = L
1
sinu +
L
2
sinu
2
2
c2 1
The kinematic equations relating the generalized and Cartesian coordinates can be found
from derivation.

x = L u sinu L
2
u sinu
1 1 1 2 2

y = L u cosu + L u cosu
1 1 1 2 2 2
32
L
1
u
1
x
c1
= sinu
1
2
L
1
u
1
y = cosu
1 c1
2
L
2
u
2
x
c2
= L
1
u sinu sinu
2 1 1
2
y = L
1
u cosu +
L
2
u
cosu
2 c2 1 1
2
2. The Jacobian matrices for the tip and the centers of mass of the mechanism can be
found directly from the kinematic equations:
L
1
sinu L
2
sinu
2
(
= J
tip
L
1
cosu
1
L
2
cosu
2
(
1
(
L
2
1
sinu
1
0
(
L
1
cosu
1
0
(
(
2 (
0 (
J =
1
L
2
(
cm
L
1
sinu sinu
2
(
1
2
L
1
cosu
1
L
2
cosu
2
(
(
2 (
0 1 (

3. To find the singular values, the determinant of the Jacobian for the tip of the
mechanism is set to zero. For these angles the inverse of the Jacobian is not defined so
there is no transformation from Cartesian to generalized coordinates.
det J = L L
2
(sinu cosu
2
sinu
2
cosu
1
) = sin(u u
2
) = 0
1 1 1
u u = , 0 t....kt
1 2
This corresponds to all configurations where the links are parallel or anti-parallel.
4. To formulate the dynamic model of the system the inertances of the links must be
determined. The links each have a translational and rotational inertia:
I = m
mL
2
J =
12
The mass matrix for the system is then:
33
2
( m
1
0 0 0 0 0
0 m
1
0 0 0 0
2
L m
1
0 0
1
0 0 0
12
(
(
(
(
(
(
(
(
(
M =
0 0 0 m
2
0 0
0 0 0 0 m
2
0
L m
2
2
0 0 0 0 0
2
12
This can be converted to the generalized inertia of the system.
T
MJ J I =
T
L
1
L
1
( (
sin
1
u m
1
0 0 0 0 0 u sin
1
0 0 (
2
(
(
(
(
(
(
(
(
(
(
(
u
2
u
(
(
(
(
(
(
(
(
(
(
(
2 2
(
(
(
(
(
(
(
(
(
0
1
0 0 0 0
L
1
m
L
1
2
1
1 0
u u cos
1
0 0 cos 2
L m
1 1
0 0 0 0 0
1 0
12
0 0 0 m 0 0
2
0 0 0 0 m
2
0
=
1
u
L
2
2
u
1
L
2
2
u sin
2
L
1
L
1
sin sin sin
L L
L m
2
2 2
2
2
2
2
1
0 1
u u
1
0 1
u cos
2
L
1
L
1
cos cos cos
0 0 0 0 0
12
2 2
L
1
L m
1 1
u (
L L
2 1
sin
1
sin
2
( ) (
2
sin )
2 2 2
1 1
L L
1 2
u sin
1
u
(
u u
)
1
u m
1
u sin
2
u
1
L m
2 2
u cos L m
1 2
+ + + + +
m
1
cos cos cos
2
4 12 2
=
2
L
2
4
(sin
2
)
2
u
2
L L
1
u cos
1
u
(
2
u
2
L L
1
u cos
2
sin + + + m cos m
2 2
2 12
2 2
(
(
(
(
(
L m
1 1
L m
1 1
u (
1
=
(
(
(
(
)
2 2 2
m cos
2
m cos
2
u
2
L m
1 2
L m
1 2
+ +
2 2
3 2 3 2
=
2
2
2
2
L L
1
L m
2
3
L L
1
L m
2
2
2
3
u (
1
u
2
)
2
cos m
2
2
m cos
2
2
This expression of the generalized inertance is used to find the co-energy of the system:
e
1
T *
) ( e u E I =
k
2
2
L m
1
L L
1 1
L m
(
2
2
m cos
2
+
| u
2
u
u
(
(
(
(
2
(
(
1
2
2 1 2
u |
1
3 2
=
2
2
L L
1
L m
2
2
2
3
2
cos m
2
(
2 2
|
u
|
|
|

1
|
u
|
|
|
2

2
u

1
( L m
1
3
|
\
m
2
L L
2 1
2
L L
2 1
L m
2
3
1
2
u
|
|
.
2
|
\
u
|
|
.
1
+
2 1
L m
1 2
cos
2
cos
2
+ +
u
(
(

2
(
m =
2
2
. \ .
| |
2 2
|
u
|
|
|
2
1
u
|
|
|
2
2
2

L m
1
3
L m
2
3
1
( L L m
2 2 1
)

u u
\
\
2 1
L m
1 2
\
.
2
+ + + cos =
1 2
2
.
(
(
(
(
(
)
|
|
.
2
2
34
2
1
2
1
1
The angular momentum of the system is related to the torque applied and the kinetic co-
energy as follows:
d I d
dt dt
( q ( )u u ) E
*
k
c
= t =
u c
2
L m
1
L L
1 1
(
2 2
m cos
2
m L
1
+
2
u
u
(
(
(
(
(
(
2 2
3 2
= q

2
L L
1
L m
2
cos m
2
2 2
2
2
3
L m
1
2
L L
2 1 1
(
2
cos
2
L
1
+

1

2
u
u
(
(
(
(
m m
(
(
\
u

2
(
(
L L
2 1
2
d
dt
q |
|
.
2
2 2
u
1
( )
3 2
u
sin
2

1
= m
2 2
L L
1
L
2
m
2
cos m
2
2
2
2
3
|
|
|
2 2
| | |
u
|
|
|
2

2
2
L m
1
3
L
2
3
1 m
2
m
1 2
) ( u
|
|
.
* 2 1
m
\

u u E
2
L
1
L L
1
+ + + = cos
k 2 2 1 2
2
. \ .
m
2
L L
2 1
( |
\

u u
|
|
.
2

u u
|
|
.
sin
(
(
(
(
* 1 2
E
u c
c
2
k
=
m
2
L L
2 1
2
|
\
sin
2
1 2
To express the flow variables in terms of the torques the momentum equation must be
inverted. At this point it is easier to substitute values for the variables.
(
1
|
2
L L
2 1
( L L
2 1
|
\
m
|
|
.

u u
|
\
|
+
(
(
(
(
|
|
.
2
L m
1
L L
1 1
(

u u u

2
2
|
|
.
2

u u
sin
2
sin
2 2
m cos L
1
m
(
(
(
(
m
t
1
2 1 2 2 1 2

u ( (
2 2
2 2
3 2

u
2
=
(
2
(
2
L L
2 1
L L
2 1
L L
1
L
2
|
\
|
|
.
u (
2
1
m
2
m cos
2
2
2
3
sin
2
sin
2

u u
2 1
m m
2 1 2 2 2
2 2
\
1
( 1
8
1
16
+
|
\
|
|
.

u u
\
|
|
u
|
|
.
2
2
1 1 (
t
t
1
2
sin
2
sin
2 cos
2
(
(
(
(
=
(
(
(
(
(
1 2
6 16
u
1 1 1 1
16
|
\
t
(
(
(
(
1
2
|
|
.

u u u
|
|
.
2
1
16
2
24
sin
2
sin
2
+ cos
1 2
8
2
cos
2
cos 144 96 ( (
+
|
\
1
16
|
|
.
(

2 u u
2 1 (
(
(
(
)
)
u

2
2
sin
2
u

(
2 2
2
2
cos
2
16 16 cos
u

=
(
cos 144 384
\
1
| |
|
.
(

2 u u
2 1
u
2
1
sin
2
2 2
cos
2
16 16 16
The motion of the tip can be found from the kinematic relationship between the
generalized coordinates and the tip Cartesian coordinates.
5a. To simulate the behavior of the system under the action of the impedance controller,
the tip stiffness and damping matrices must be transformed to generalized coordinates.
For small motions about a configuration where the inner link is aligned with the x-axis
and the outer link is parallel to the y-axis, sinAu
1
= Au
1
and cos(t/2 + Au
2
) = Au
2
.
(
(
(
(
)
)
(|
|
|
|
|
|
.
35
T T
A t = F J = x K J
t |(
L
1
sinu
1
L
1
cosu
1
(
(
k
L
2
cos
|
A + u |
(
0 L
1
( L
2
Au ( 0 L
2
2
(Au
1
(
t =
\
2
2
2
(
L sinu
2
L
2
cosu
2

L
1
sin Au
.
(
L
2
0
(
L
1
Au
1
2
(
= k
L
1
0

Au
2
(
2
1
2
( L
2
K( ) = k
L
0
1
2
0

(
u
Similarly, the damping matrix is transformed to generalized coordinates using the
Jacobian.
T T T
L
1
sinu
1
L
1
cosu
1
( L
1
sinu L
2
sinu
2
(
2
0 (
t = F J = x B J = J BJ u = b
L sinu
2
L
2
cosu
2
(
L
1
cosu
1
(
u = b
L
1
2
(
u
L
2
cosu
2 2 1
0 L
2
2
0 (
B( ) = b
L
1
2
(
u
0 L
2
The desired natural frequency for the system is 2 Hz. This allows the stiffness constant, k,
to be evaluated:
u
1
e
n
= K( ) I (u )
1
2
(
2
L
2
u k = (e )I ( )
L
0
1
2
0

( n
L m
1
2
+ m
2
L
1
m
2
L L
2
(
1 (
2 1
I ( ) =
cos
2 (

6
0
(
1
3
( = u
L L
2
2
2
1
m
2
cos
2
m
2
L
2 (
0
1
(
(
2 3
( 24
1 ( 2 (
0
3
(
2
k = ( ) I (u )
0
2
L
2
2
(
(
1
= 16t

6
0
(0 4(
= 16t 4t
2
L
1
0

0
1
(
(
4 0
(
2
1
0
(
(
24 6
8
2
k = t
max
3
If the stiffness coefficient is larger than k
max
then the system will oscillate faster than 2Hz
in the horizontal direction, which is the lower impedance direction for this configuration.
The damping coefficient must be found to provide critical damping to the system in this
configuration. The critical damping coefficient is found from the following expression:
36
B( ) = 2 K( ) ( ) u u u I
1 (
2
0 L
2
2
(
6
0
(
2
| L
1
0
2
(
(
|
|
|
2
= 4
8
t
L
1
2
0

0
\
0 L
2 .
3
(
1
(
(
24
1(1 ( 1 (
1
1(
b
2
32t
2
0
4
(
6
0
(
16
0
(
32t
2
0
6
(
= =
(
1
(
1
(
3
2
(
3
1
0(0 ( 0 ( 0(
4 24 16 3
b
4
critical
=
3
t
5b. The system is driven by a simple impedance controller with the following
characteristic:
u
T
( u (
0

t = J( ) ( X K L( )) + V B J(u )u))
0
These torques are the torques corresponding to changes in the angles u
1
and u
2
, which are
not the same as the motor torques on the links. If the motor torques are needed in order to
control actual hardware, the following relationships are used. The equivalent torque
acting on each of the links is depicted below in Figure 2.
u
1
-u
2
u
1
L
1
L
2
t
1
t
2
t
2
Figure 2: Uncoupled 2-Link Mechanism Indicating Torque Input
From this figure it is evident that the torque corresponding to u
1
is the difference between
the motor torques: t t
2
. Also, the torque corresponding to u
2
is the torque
1
corresponding to u
1
minus the torque corresponding to
2
, which results in t
1
.
1) The first simulation describes the motion of the two link mechanism from the
Cartesian position (L, 0) to (2L, 0) with a trapezoidal speed command. Figure 3 shows
the position of the mechanism at 50ms intervals during its translation.
37
0 0.2 0.4 0.6 0.8 1
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
x position (m)
y

p
o
s
i
t
i
o
n

(
m
)
Time Plot of Mechanism Position for First Simulation
Figure 3: Time Plot of Mechanism for Virtual Trajectory from (L,0) to (2L,0)
2) The second simulation describes the motion of the mechanism as it passes through the
singular point (0, 0). The virtual trajectory of the robot is from (-L, 0) to (L, 0). Figure 4
demonstrates the behavior of the mechanism for this input.
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
-0.1
0
0.1
0.2
0.3
0.4
0.5
x position (m)
y

p
o
s
i
t
i
o
n

(
m
)
Time Plot of Mechanism Position for Second Simulation
Figure 4: Time Plot of Mechanism for Virtual Trajectory from (-L,0) to (L,0)
38
It is clear from this plot that the compliance in the system allows the outer link to rotate
around its pivot since there is less inertia in this direction than in the direction of the
virtual trajectory. This is seen as the mechanism begins to move and again after it passes
the singular configuration. There is some overshoot at the end of the movement due to the
inertance of the mechanism.
3) The third simulation attempts to pass the tip of the mechanism near the singular
configuration at the origin. The virtual trajectory is from (-L, L/20) to (L, L/10). This
results in the following behavior for the mechanism.
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
x position (m)
y

p
o
s
i
t
i
o
n

(
m
)
Time Plot of Mechanism Position for Third Simulation
Figure 5: Time Plot of Mechanism for Virtual Trajectory from (-L,L/20) to (L,L/10)
Figure 5 demonstrates that the mechanism passes through the singular configuration
instead of following the virtual trajectory. In order for the mechanism to pass near the
singular configuration without actually passing through it, the links of the mechanism
would need to quickly rotate through 180
o
as the mechanism approached the singular
configuration. This requires very large torques on the mechanism as it passes near the
singular point.
6. The natural frequency of the controller is now adjusted to 20Hz to determine the
impact this has on the system. The corresponding k and b are found as follows:
2
max
2
3
800
0
6
1
3
2
0
1600
t
t
=
(
(
(
=
k
k
39
t
t t
3
40
0
3
2
6
1
0
3
3200
16
1
0
0
16
1
24
1
0
0
6
1
0
4
1
4
1
0
3
3200
2
1
2
2
=
(
(
(
=
(
(
(
(
(
(
(
(
(
critical
b
b
Incorporating these into the model of the controller, yields the following output for the
conditions specified in the third simulation.
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
x position (m)
y

p
o
s
i
t
i
o
n

(
m
)
Time Plot of Mechanism Position for Third Simulation with 20Hz Natural Frequency
Figure 6: Time Plot for Trajectory from (-L,L/20) to (L,L/10) with 20Hz Critically Damped Poles
Figure 6 demonstrates the need for the mechanism to rotate 180
o
as it passes near the
singular configuration if it is stiffly linked to the virtual trajectory.
7. Finally, simulation three is repeated with the outer link 10% shorter than the inner link
(.45m instead of .5m). With this discrepancy in the link lengths the mechanism can no
longer reach points within .05 meters of the origin. Additionally, it cannot reach beyond a
circle of radius .95 meters. The resulting behavior of the mechanism is shown below in
Figure 7.
40
-0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
x position (m)
y

p
o
s
i
t
i
o
n

(
m
)
Time Plot of Mechanism Position for Third Simulation with 20Hz Natural Frequency
Figure 7: Time Plot for Trajectory from (-L,L/20) to (L,L/10) with Uneven Link Lengths
This figure shows that the low rotational inertia of the outer link again causes it to rotate
away from the virtual trajectory. As the mechanism approaches the origin in this
simulation it is unable to pass within .05 meters. This causes it to rotate away from the
origin and the prescribed virtual trajectory. After passing through the singular
configuration at (0,.05) it converges back onto the virtual trajectory and overshoots its
final destination again due to the inertance of the rotating links.
8. Despite the fact that the Jacobian of the mechanism is undefined at its singular
configurations, the simple impedance controller allows the mechanism to operate at its
singular configurations. In the first simulation, the final desired position of the
mechanism was the singular position corresponding to both links fully extended to the
maximum attainable radius. With the impedance controller the mechanism was able to
reach this configuration. However, due to the low translational inertance that corresponds
to this position and relatively high rotational inertance the mechanism rotated beyond the
desired position pulling the tip of the mechanism back towards the origin. If the time
scale is extended, it is seen that the mechanism continues to oscillate around this position
with decreasing frequency. At 200 seconds the mechanism still has 1
o
oscillations with a
period of 36 seconds. The actual trajectory and virtual trajectory of the mechanism are
shown below in Figure 8, which includes only the 2 second command interval.
41
Actual Trajectory and Desired Trajectory for First Simulation
0.1
0.08
0.06
0.04
V
e
r
t
i
c
a
l

P
o
s
i
t
i
o
n

(
m
)

0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
0.5 0.6 0.7 0.8 0.9 1
Horizontal Position (m)
Figure 8: The Actual and Virtual Trajectory for Desired Movement from (L,0) to (2L,0)
For the second simulation, the mechanism was commanded to pass directly through the
singular position at the origin. Because the impedance controller does not require the
inverse of the Jacobian, the mechanism was able to pass through this point. Indeed the
mechanism prefers to pass through this point, as shown below in Figure 9, since it is the
lowest energy path through the origin.
Actual Trajectory and Desired Trajectory for Second Simulation
0
i
l
i

(
m
)

-0.1
-0.08
-0.06
-0.04
-0.02
0.02
0.04
0.06
0.08
0.1
V
e
r
t
c
a

P
o
s
t
i
o
n
-0.4 -0.2 0 0.2 0.4 0.6
Figure 9: The Actual and Virtual Trajectory for Desired Movement from (-L,0) to (L,0)
42
This is also seen on the third simulation, which attempts to pass the mechanism near the
origin. Rather than follow the virtual trajectory, which passes just above the origin, the
mechanism deviates from the virtual trajectory to pass through the lowest energy path:
directly through the origin. This is clearly evident in Figure 10, which shows the virtual
trajectory passing through (0, .0375) and the actual trajectory passing through (0, 0).
0
l i i i l ion
i
l
i
t
i
(
m
)

-0.1
-0.08
-0.06
-0.04
-0.02
0.02
0.04
0.06
0.08
0.1
Actua Trajectory and Des red Trajectory for Th rd S mu at
V
e
r
t
c
a

P
o
s
o
n

-0.4 -0.2 0 0.2 0.4 0.6
Figure 10: The Actual and Virtual Trajectory for Desired Movement from (-L,L/20) to (L,L/10)
When the stiffness of the virtual linkage was increased, the mechanism was constrained
more closely to the virtual trajectory. This is seen clearly in Figure 11.
Actual Trajectory and Desired Trajectory for High Stiffness Simulation
0
i
l
i
t
i
(
m
)

-0.1
-0.08
-0.06
-0.04
-0.02
0.02
0.04
0.06
0.08
0.1
V
e
r
t
c
a

P
o
s
o
n

-0.4 -0.2 0 0.2 0.4 0.6
Figure 11: Actual & Virtual Trajectory from (-L,L/20) to (L,L/10) with Increased Stiffness
43
While the increased stiffness resulted in better performance away from the singular
position, near the singular position the mechanism required quick rotations and huge
torques in order to maintain the desired trajectory. This was seen previously in Figure 6.
The mechanism performs well, but requires much more energy to reach the same desired
end state.
V
e
r
t
i
c
a
l

P
o
s
i
t
i
o
n

(
m
)

The final simulation demonstrated how a larger singular region around the origin affected
the behavior of the robot. With unequal link lengths the robot could not attain any
position within .05 meters of the origin. However, the desired trajectory of the robot was
through the point (0, .0375). Because this configuration could not be attained, the
mechanism passed through the closest point with the least resistance, (0, .05). This is seen
in Figure 12, which shows the virtual trajectory and actual trajectory of the mechanism.
This figure also demonstrates that when the singular configuration is closer to the desired
trajectory, the mechanism can more accurately follow the desired path. This is especially
apparent when comparing Figure 12 to Figure 10.
Actual Trajectory and Desired Trajectory for Uneven Link Simulation
0.1
0.08
0.06
0.04
0.02
0
-0.02
-0.04
-0.06
-0.08
-0.1
-0.4 -0.2 0 0.2 0.4 0.6
Figure 12: Actual & Virtual Trajectory from (-L,L/20) to (L,L/10) with Uneven Link Lengths
44
Tent Pole Instability
1. The mechanical system of a tent pole and guy-ropes is shown below in Figure 13.
1 m 1 m
u
k k
l
1
l
2
2 m
Figure 13: Tent-Pole & Guy-Rope System
The relationship between the variables l
1
, l
2
and u can be determined from the Law of
Cosines:
l
1
= 5 cos 4 u
l
2
= 5 cos 4 ( u t )
The Jacobian for the transformation between coordinates is found by differentiating each
of these terms with respect to u.
sin 2 u

l
1
= u
5 cos 4 u
sin 2 ( u t )
u
l
2
5 cos 4 ( u t )
sin 2 u (
5 cos 4 u
(
J =
sin 2 ( u t )
(
5 cos 4 ( u t )
(

The inertia of the pole is simply mL
2
/3. The co-energy, therefore, is dependent only on e.
This results in the following relationship between torque and angular acceleration:
3t

u =
2
mL
The torque on the pole is proportional to the forces from each of the guy-ropes, according
to the following equation:
45
F J
T
= t
The forces from the guy-ropes are found from the next equation, where k is the elasticity
of the rope and F
0
is the pretension.
0
F l k F + A =
There is also a torque component due to gravity, which results in the final expression for
torque:
u t cos
2
mgL
F J
T
+ =
2. The first simulation is intended to show the mechanism is unstable for any pretension
if there is no elasticity in the guy-ropes. Gravity is also not to be considered. With these
assumptions and a pretension of only 10N the following behavior is exhibited.
-1 -0.5 0 0.5 1 1.5 2 2.5
-1
-0.5
0
0.5
1
1.5
2
2.5
V
e
r
t
i
c
a
l

P
o
s
i
t
i
o
n

(
m
)
Tent-Pole Simulation 2
Figure 14: Time Plot of Tent-Pole with 10N Pretension and Zero Elasticity
As the pretension increases in the guy-ropes the system becomes more unstable due to the
higher torques generated.
3. The next simulation removed the pretension in the guy-ropes and added in the
elasticity. Gravity effects were again omitted in order to emphasize the effect of the
elasticity on the system. Initially, the system was only perturbed 1
o
to show that it was
stable. This produced the following result where the pole oscillated between 1
o
.
46
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
2
2.5
V
e
r
t
i
c
a
l

P
o
s
i
t
i
o
n

(
m
)
Tent-Pole Simulation 3a
Figure 15: Time Plot of Tent-Pole with Elasticity and No Pretension for 1
o
Perturbation
The next task was to see if the tent-pole was still stable for perturbations up to 45
o
. Figure
16 demonstrates the stability of the pole with an initial perturbation of 45
o
.
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
-1
-0.5
0
0.5
1
1.5
2
2.5
V
e
r
t
i
c
a
l

P
o
s
i
t
i
o
n

(
m
)
Tent-Pole Simulation 3b
Figure 16: Time Plot of Tent-Pole with Elasticity and No Pretension for 45
o
Perturbation
47
4. With stiffness, pretension, and gravity all considered in the model of the system, the
following behavior is witnessed. Figure 17 shows the case where the system is stable, and
Figure 18 demonstrates the unstable case.
Tent-Pole Simulation 4 with F0<2.23607k
-1
0
1
2
i
l
i

(
m
)

-0.5
0.5
1.5
2.5
V
e
r
t
c
a

P
o
s
t
i
o
n
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2
Figure 17: Time Plot of Complete Tent-Pole System with Stable Behavior
Tent-Pole Simulation 4 with F0>2.23607k
-1
0
1
2
i
l
i

(
m
)

-0.5
0.5
1.5
2.5
V
e
r
t
c
a

P
o
s
t
i
o
n
-1 -0.5 0 0.5 1 1.5 2 2.5
Figure 18: Time Plot of Complete Tent-Pole System with Unstable Behavior
48
5. To analytically solve for the stability of the tent-pole system the equivalent stiffness of
the joint is found. The expression for the joint stiffness is:
K =
c J
T
F + kJ
T
J
cu
The system becomes unstable at the point where there is no restoring torque on the
system for a small displacement Au. For this analysis the mass of the system is neglected.
This yields the following equation for the torque on the tent pole.
T
t = J k
Al
1
(
(
J
T
F = 0
Al
2
sin 2 u (
(
sin 2 u sin 2 (t u )
(

5 cos 4 u
(Au =
(
F
0
(
(
5 cos 4 u 5 cos 4 (t u )
(
(
sin 2 (t u )
(

5 cos 4 u 5 cos 4 (t u )
(
F
0
(

5 cos 4 (t u )
(

|
4k
| sin 4
2
u sin 4
2
(t u ) |
|
|Au = F
0
|
+
\
5 cos 4 u 5 cos 4 (t u )
.
\
5 cos 4 u

5 cos 4 (t u )
.
|
|
For stability about the vertical position, u = t/2, where there are small displacements, Au,
this expression becomes:
2 2
4k
| cos Au cos Au
|
|
|
Au = F
0
|
cos 2 Au cos 2 Au |
+
|
|
\
5 4Au 5+ 4Au
. \
5 4Au 5+ 4Au
.
10
|
|
Au = 2F
0
cos Au
|
5+ 4Au 5 4Au
|
|
4k cos
2
Au
|
2
2
\ 25 16Au .
\
25 16Au
.
|
F
0
=
20Au cos Au
2
k
25 16Au ( 5+ 4Au 5 4Au )
Using LHopitals Rule:
cos 20 Au 20Au sin Au 20
lim = = = 5
x 0
| 2 2 | | 4 | 16Au
2
( 5+ 4Au 5 4Au )+ 2516Au
\
5+ 4Au
+
5 4Au
.
|
| 5
|
|
2
\
5
.
Therefore, the maximum value for the pretension is k 5 or 2.23607k. This result can be
compared to the ratios found through simulation, which are shown in Table 1:
2516Au
Table 1: Ratio of F
0
to k for MATLAB Simulation of Tent-Pole System without Gravity
F
0
k F
0
/k % error
50 N 22.36 N/m 2.23614 -0.003
100 N 44.72 N/m 2.23614 -0.003
150 N 67.08 N/m 2.23614 -0.003
2000 N 894.43 N/m 2.23606 -0.0003
49
It is clear from the simulation data, that MATLAB accurately predicts the stability point
for the system. The precision on the ratio becomes greater as the value of the pretension
is increased. When the mass is considered as well, the stability point decreases as the
angle from vertical increases. The mass has a larger effect on the stability of the system
for lower values of pretension.
50
MEMORANDUM
To: 2.141 class
Date: December 9, 2002
Re: Assignment #4
Performance on assignment #4 was generally excellent.
In the problem on control in and near kinematic singularities, the main challenge was to
deal with the nonlinear kinematics. It is substantially easier to use configuration variables
(angles) measured with respect to an inertial reference frame to formulate the kinetic
energy and derive state equations. The sensor measures relative angles and the motors
generate relative torques, but the transformation between relative angles and inertial
angles is simple as is the transformation of relative torques to generalized torques.
To evaluate the controller parameters you could either transform the inertia tensor from
configuration space to end-point coordinates or transform the controller stiffness and
damping from end-point coordinates to configuration space. In the configuration
specified (links at right angles) the Jacobian is particularly simple. The natural frequency
and damping ratio for small motions is the same in either frame.
In the tent-pole instability problem, pencil-and-paper analysis up front can do a lot to
guide numerical simulation. Some of you need to look more closely at your numbers; the
simulations were generally correct, but non-negligible errors can and did occur. In this
problem analysis shows that the relation between stiffness and preload should be a
straight line. Thats a good way to check your simulations.
51
Due: 11/21/02
Simple convection and throttling processes
The main point of this assignment is to give you experience with modeling transport
processes in a simple thermofluid system. A secondary purpose is to show you that even
simple systems involving matter transport can exhibit interesting behavior.
Two air cylinders are connected by a valve, which is initially closed. One cylinder
contains air at 20 psi, the other, air at ambient pressure, 14.7 psi. Both cylinders are
initially at ambient temperature, 70F. You are to model and simulate the transient that
follows when the valve is abruptly opened.
70F,
20 psi
70F,
14.7 psi
Assume air may adequately be described as an ideal gas. Assume each cylinder is 10
inches in diameter and 30 inches long. Assume the valve orifice has diameter 0.1 in.
Model the heat transfer between the cylinders and their surroundings, but you may
neglect any heat transfer across the valve or between the valve and its surroundings. Vary
the heat transfer coefficient over a range of values to represent adiabatic conditions,
isothermal conditions and at least one value in between. I suggest K (BTU/secR) = 0,
0.0001, 0.001 and (or, more realistically, 1).
1a. Draw a bond graph of your system model, clearly identifying key quantities and
power fluxes.
1b. Choose state variables and develop (nonlinear) state equations.
2a. Simulate the transient changes in (a) pressure (b) temperature and (c) mass flow
rate into the right cylinder.
2b. An interesting aspect of this problem is that since thermal systems are composed
only of capacitors and resistors, and we might intuitively expect their transient
responses to exhibit no overshoot. You should be able to show that this is not
always true. Comment and explain.
2c. Check the veracity of your simulation by computing the equilibrium conditions in
each tank after the transients have expired. (Calculating equilibrium states before
and after is standard engineering thermodynamics.)
52
3. A much more common situation (as any SCUBA diver knows) is to fill a cylinder
from a (nominally) constant-pressure supply. One way to model this situation is to
assume the left-hand cylinder in the diagram above is extremely large (so that
withdrawing air from it changes its pressure very little). Use the model you
developed to simulate this situation. What is the minimum set of properties needed
to specify the air supply?
4. Modify your model to include heat transfer across the valve due to any temperature
difference between the two cylinders. (Just show me a bond graph. Equations and
simulation are optional.)
5. The pressure difference above was chosen to keep valve flow in the non-choked
flow regime. Of course, realistic pressure differences are much higher. (A typical
SCUBA tank is routinely filled to 3,000 psi.) Revise your model and simulation to
cover this situation. In this case, how valid is the assumption of no heat transfer at
the valve?
53
2.141, Assignment#5 21 November, 2002
Matter-Transport and Convection
Will Booth
54
W. Booth
08/18/04
1
1 Nomenclature
Arabic
A
1
[in
2
] Surface area of tank 1
A
2
[in
2
] Surface area of tank 2
A
t
[in
2
] Valve (throat) cross-sectional area
C
d
none Coefficient of discharge for air expansion on valve exit
C
v_air
[ft-lbf/lb-mol-R] Specific heat of air at constant volume
D
1
[in] Diameter of tank 1
D
2
[in] Diameter of tank 2
D
t
[in] Diameter of valve (throat)
h [ft-lbf/sec-in
2
-R] Heat transfer coefficient (tank environment)
L
1
[in] Length of tank 1
L
2
[in] Length of tank 2
M
air
[1/mol] Molecular weight of air
N
1
[lb-mol] Mass of air in tank 1
N
2
[lb-mol] Mass of air in tank 2
P
1
[psi] Pressure in tank 1
P
2
[psi] Pressure in tank 2
R
u
[ft-lbf/lb-mol-R] Universal gas constant
S
1
[ft-lbf/R] Total entropy of tank 1
S
2
[ft-lbf/R] Total entropy of tank 2
dS
env1
/dt [ft-lbf/sec-R] Entropy flux to the environment from tank 1
dS
env2
/dt [ft-lbf/sec-R] Entropy flux to the environment from tank 2
dS
vlv
/dt [ft-lbf/sec-R] Entropy flux across valve
T
1
[R] Temperature of tank 1
T
2
[R] Temperature of tank 2
T
amb
[R] Ambient temperature
V
1
[in
3
] Volume of tank 1
V
2
[in
3
] Volume of tank 2
Greek
none ratio of specific heats (c
p
/ c
v
)
1
[lb-mol/in
3
] density in tank 1
2
[lb-mol/in
3
] density in tank 2
55
W. Booth
08/18/04
2
2 Results
Part 1 Bond graph and state equation derivation
The bond graph for the two-tank / valve system is shown below. State variables S
1
, N
1
, S
2
, N
2
are chosen.
C R
0
0
0
0
C
R
R
0
S
e
s
u
.
s
d
.
s
2
.
s
1
.
s
env2
.
s
env1
.
N
1
.
N
2
.
N
u
.
N
d
.
T
1
T
2
T
amb
: :
1

2
::
:
h
h :
S
f
: 0 S
f
: 0
P
1
P
2
Figure 2-1: Bond graph for 2-tank system with throttling through a valve and heat transfer to environment.
State equations are derived as follows:
Looking at the throttling process with respect to the throat, relative to the upstream
conditions:
Following the logic outlined in the notes, the non-bulk flow terms can be neglected, because
we are satisfying continuity with:
t
m
in
d
d t
m
throat
d
d
i.e. no leakage. Therefore, the remaining terms in a power balance are the "flow work rate",
and the "kinetic energy transport rate", to use the suggested terminology.
Thus, a power balance from input to throat is:
P
u
M
air

u
v
u
2
2
+
|
\
|
.
t
m
u
d
d
P
t
M
air

u
v
t
2
2
+
|
\
|
.
t
m
t
d
d
56
W. Booth
08/18/04
3
where the units of
are [lb-mol / in
3
] and the units of M
air
are [1 / mol]. Subscript '
u
' denotes
upstream parameters, and subscript '
t
' denotes parameters at the throat.
Cancelling mass flow rates, and assuming there is negligible upstream velocity, re-arranging,
gives:
v
t
2
M
air
P
u
u
P
t
\
|
.
i.e. in terms of molar mass, the mass flow-rate at the throat is:
t
m
t
d
d t
M
air
N
( )
d
d
M
air

t
A
t
2
M
air
P
u
u
P
t
\
|
.

and making the suggested assumption that all the flow work goes into speeding up the flow,
and not compressing the air:
where the units of N are [lb-mol] and the units
of P are [psi]
t
N
d
d
A
t
2
u
M
air
P
u
P
t
( )

A problem arises, since from the ideal gas law:
P
u
P
t
T
u
T
t
Therefore, to get mass flow, we require P
u
> P
t
, which implies T
u
> T
t
, and therefore, u
u
>
u
t
, and h
u
> h
t
which disagrees with the common assumption that throttling is isenthalpic.
To reconcile this inconsistency, it is recognised that we measure the flow, pressure and
temperature downstream of the throttle, therefore mixing will occur between throat-flow and
the downstream gas. If this mixing is assumed to occur at constant pressure, and continues
until T
u
= T
d
then, the process will be isenthalpic, as commonly assumed.
To determine equations for entropy flux, per the definition of specific entropy:
s
S
N
therefore
The entropy flux in a particular tank is, given the sign conventions in the system bond graph
t
S
d
d
S
N t
N
d
d
t
S
env
d
d
(in terms of total entropy)

i.e. the entropy flux is equal to the specific entropy multiplied by the mass flux minus the
change in entropy due to heat transfer to the environment.
Specifically:
t
S
1
d
d
S
1
N
1
t
N
1
d
d
t
S
env1
d
d
and
t
S
2
d
d
S
2
N
2
t
N
2
d
d
t
S
env2
d
d
57
W. Booth
08/18/04
4
P
N
V
R
u
T
and since the volume of each tank is fixed:
(where R
u
is the universal gas constant)
P R
u
T
The pressure in the chamber can be calculated from the ideal gas law, since we are
assuming that we can model air as an ideal gas, therefore:
(where T
i
is the initial temperature, and S
i
is the reference
entropy)
T
T
i
exp
S S
i
M N c
v
\
|
.
Therefore:
M N c
v
dT T dS
So:
(where M is the molar weight, and N has units of lb-mol)
U M N c
v
T
From the definition of internal energy:
(since the tank is of fixed volume)
dU T dS P dV T dS
Therefore, in terms of total quantities:
(specific quantities)
du T ds Pdv
Recalling the expression of the 1st law from assignment 3:
But, to determine the heat lost to the environment and the pressure in each tank, we need an
expression for the temperature of each tank, which can be derived as follows:
since all the mass which leaves tank 1 arrives in tank 2 (assumption of no leakage)
t
N
1
d
d t
N
2
d
d
Choosing state variables, S

1
, N
1
, S
2
, and N
2
, the above equations are 2 of the state
equations, and the other 2 are the equations for mass flowrate through the throttle above,
where
An attempt at linearization of the equations for the 3-port capacitor with the mechanical work port closed, is shown
in Appendix A. This system could then be coupled with the linearized four port resistor model shown in the class
notes, to provide some insight to the system. I could not successfully linearize the capacitor model, however,
because the cross-partial terms representing the couple between domains werent equal, therefore not satisfying
Maxwells reciprocity conditions.
Part 2a Simulation results
Matlab m-files for all the simulations are contained in Appendix B. Results for simulations with h=0, 0.0001, 0.001,
0.01 and 1 ft-lbf / sec-in
2
-R are shown in Figure 2-2. I used values for h suggested in the problem statement
(directly in ft-lbf/sec-in
2
-R as opposed to BTU/sec-R), but when multiplied by the surface area of the tanks, the
result is approximately equivalent hence the results show the range of responses intended by the suggested values.
58
W. Booth
08/18/04
5
Figure 2-2: Simulation results for part 2a with various values of h
One aspect of note, regarding the simulation, came about in discussion of this problem set with Tom Bowers. He
had found that the solution of the problem was more efficient and less noisy when using a stiff ODE solver, like
ode15s, rather than a non-stiff solver like ode45. Matlab does not offer a definition of stiffness, so I went looking
on the web for a definition of a stiff vs. non-stiff system. One reference indicates there is no exact definition of
stiffness, but that when using Runge-Kutta solvers of order 4 or 5, symptoms of stiffness are non-efficient numerical
solution (small time steps) and oscillations in the response. In addition, if I had been able to linearize my system
equations, I could have plotted the eigenvalues for the initial time, and then at some later time, and looked to see that
they move significantly, indicating the system is stiff.
When I solve my system equations with no heat transfer (h = 0) and using ode45, and vary the relative tolerance I
get the following results:
RelTol time (sec) successful steps failed attempts
1.e-3 1.1 57 1
1.e-5 8.74 565 7
1.e-7 93.21 5587 8
whereas with ode15s solver:
RelTol time (sec) successful steps failed attempts
1.e-3 0.44 23 5
1.e-5 3.57 245 152
1.e-7 1.54 110 55
59
W. Booth
08/18/04
6
These results, suggest symptoms of a stiff problem, and the plot of ode45 results for the first two values of relative
tolerance shows that when the tolerance is large, the response is oscillatory, whereas if the tolerance is much
reduced, the solver is less efficient, but produces a smooth response.
Figure 2-3: Simulation results for ode45 solver with different relative tolerances
The initial conditions of the system are shown in the Table 2-1, with corresponding units given in section 1.
N
1
(init) T
1
(init) P
1
(init) S
1
(init) N
2
(init) T
2
(init) P
2
(init) S
2
(init)
0.0048 529.67 20 174.5244 0.0035 529.67 14.7 129.9494
Table 2-1: Table of initial conditions
For the adiabatic process (h=0), by definition there is no heat loss to the environment, therefore, the initial pressure
difference between the two tanks drives mass flow from left to right, until the pressures are in equilibrium, at which
point there is no further changes. The mass of air in tank 1 remains greater than the mass in tank 2, whereas the
temperature in tank 2 is greater than the temperature in tank 1, i.e., N
1
> N
2
, and T
1
< T
2
. What tank 1 loses in
pressure, it must also lose in temperature, from the ideal gas law. Comparing the equilibrium conditions in Table
2-2 with the initial conditions in Table 2-1, the total entropy decrease in tank 1, is equal to the total entropy increase
in tank 2. As expected, the ideal system conserves entropy.
Part 2b & 2c Transient Overshoot and Tank Equilibrium Conditions
Figure 2-2 shows that as the heat transfer between the tanks and the environment increases, the following occurs:
1. Settling time of pressure transient increases
2. Temperatures settle to ambient (if h = 0) and settle faster as h increases.
3. Total entropy of the tanks return to their initial values (while the entropy of the environment increases)
4. Mass of air in the two tanks equalizes, as the tank temperatures reduce to ambient
60
W. Booth
08/18/04
7
I had not expected results 1,3 or 4. I expected the pressure transients to settle independent of the heat transfer
coefficient. The reason this is not the case, is because of the coupling between the pressure, temperature and mass
flowrate, i.e. the fact that this is a 3-port capacitor. The pressure difference between the tanks dominates the initial
response, but thereafter, it is the heat transfer to the environment, which governs the settling time of the system.
Some of the initial entropy produced is lost to the environment, therefore the temperature of tank 2 does not initially
increase as rapidly as in the adiabatic case, nor does tank 1s temperature decrease as rapidly. Therefore, the
pressure difference between the two tanks is maintained for longer, until the response is governed primarily by the
heat transfer to the environment, and thus, the pressure difference can only decay as quickly as the pressure
difference (see settling times for h=0.001 and h=0.01 in Table 2-3.
The pressure difference between the two tanks is plotted in Figure 2-4.
Figure 2-4: Pressure difference between tanks 1 and 2
Given that the R elements all have temperature in, entropy flow rate out (i.e. conductance) causality on both
ports, and through-power flow, these elements guarantee satisfaction of the 2
nd
law of thermodynamics. It is not the
absolute value of entropy which can never decrease, it is the rate of entropy production that must never decrease,
therefore my initial surprise that the entropy of tank 1 initially decreases and subsequently returns to its initial value
is expected. This is due to the fact that when T
1
> T
amb
then the entropy of tank 1 will decrease (disregarding mass
transfer for the time being), but if T
1
< T
amb
the entropy of tank 1 will increase.
An interesting result occurs when the heat transfer coefficient is low, i.e. h=0.0001 ft-lbf/sec-in
2
-R. The pressure
difference decays to the adiabatic equilibrium condition of ~17.5psi almost as quickly as the adiabatic case itself, i.e.
the settling time for the initial temperature/entropy and pressure/mass transients is 22sec. After which, the heat
drain to the environment causes the pressure in both tanks to decay to the isothermal equilibrium conditions (for
h=1) over a much longer timer (>600sec). At first glance, there is no pressure difference to cause this final mass
flow to equilibrium. However, the gradual entropy generation as the temperature reduces to ambient, decreases the
pressure more in tank 2 than in tank 1 (since T
2
> T
1
) which results in mass flow between tank 1 and tank 2, tending
to equalize the masses of air in both tanks. This is as expected, since at equilibrium, even though not shown in
Figure 2-2, both tanks will be at the same temperature and pressure.
h increasing
61
W. Booth
08/18/04
8
This case (h=0.0001) is the perfect example of the 2-stage process alluded to earlier, since the stages are distinct.
The first stage is mass flow due to significant pressure differences between the tanks. The second stage is gradual
change of tank temperature to ambient, and the equalization of air masses in each tank, due to heat loss to the
environment. The time constant of the second stage is much longer than the time constant of the first stage, hence
the prolonged pressure settling times as the heat transfer coefficient increases (see Table 2-3).
The temperature and entropy overshoot is a result of the trade-off between the time constants of the 2 processes, the
initial mass flow due to significant pressure differences between the tanks and the subsequent change in tank
temperatures due to heat transfer to the environment. In the extreme cases for h=0 (adiabatic) or h=1 (isothermal),
there is no overshoot, since there is only a single process, due to mass flow as a result of AP.
As the heat transfer coefficient increases to isothermal conditions (h=1) the temperature and entropy transients will
tend to settle faster, and in the isothermal limit, there is no transient since temperature and entropy in each tank are
always equal to their initial conditions.
The following shows the equilibrium conditions for the mass, temperature, pressure and entropy in each tank at the
end of the transients, for the various heat transfer coefficients (with units given in section 1).
h N
1
T
1
P
1
S
1
N
2
T
2
P
2
S
2
P
diff
0 0.0047 470.30 17.54 172.35 0.0036 621.03 17.53 132.15 0.0088
0.0001 0.0046 480.30 17.29 172.79 0.0037 585.87 17.29 131.41 0.0087
0.001 0.0042 525.99 17.34 174.41 0.0041 532.52 17.33 130.04 0.0042
0.01 0.0042 529.67 17.35 174.52 0.0042 529.67 17.35 129.95 0.0014
1 0.0042 529.67 17.35 174.52 0.0042 529.67 17.35 129.95 -0.0011
Table 2-2: Table of equilibrium conditions
h
Temperature /
Entropy
settling time
[seconds]
Pressure /
Mass settling
time
[seconds]
0 20 20
0.0001 22, > 600 22, > 600
0.001 600 400
0.01 200 230
1 0 190
Table 2-3: Table of approximate settling times
????????????DO STUDY WITH T1 = T2 < TAMB ?????
????????????OR T1 < T2 = TAMB = 100F ????????
Part 3 Tank 1 = Constant Pressure Supply
Tank 1 is now given dimensions of 40 diameter and 60 length, to simulate a constant pressure supply. The results
are shown in Figure 2-5. The pressure of tank 1 is more or less constant, as desired, and its temperature does not
change from its initial condition. The entropy and mass of the air in tank 1 are significantly greater than that in tank
2 therefore, only results for tank 2 are plotted. The results show that the qualitative response of tank 2 is identical to
that shown in Figure 2-2, except that the tank equilibrium pressure is now = to the supply pressure (20psia) since the
supply pressure does not change.
The state equations assume knowledge of the initial entropy of the supply air, as well as its initial pressure, so both
these quantities must be specified in order to determine the transient response of the system.
62
W. Booth
08/18/04
9
Figure 2-5: Simulations for Tank1 = constant-pressure supply
Part 4 Model heat transfer across the valve
Adding heat transfer across the valve, introduces an additional R element into the bond graph, shown in Figure 2-6.
My initial calculations for a heat transfer coefficient, kA/L for a 0.005 coating of cadmium on a one meter long
interconnecting tube, set kA/L = 2.062E-5 ft-lbf/sec-R.
Initial simulations with this value of kA/L gave no change in the transient response, because the entropy change due
to heat transfer across the valve was minimal, even without heat transfer to the environment. Therefore, to show the
effects of heat transfer across the valve, kA/L was artificially inflated by a factor of 10
4
, and the simulation results
are shown in Figure 2-7. As expected, even with no heat transfer to the environment, the temperatures in the two
tanks will equilibrate over time.
The pressure transient shows an interesting response, due to the slow changing temperature and tank air mass
responses.
63
W. Booth
08/18/04
10
C R
0
0
0
0
C
R
R
0
S
e
s
u
.
s
d
.
s
2
.
s
1
.
s
env2
.
s
env1
.
N
1
.
N
2
.
N
u
.
N
d
.
T
1
T
2
T
amb
:
:
:
1

2
::
:
h
h
kA/L
:
S
f
: 0 S
f
: 0
P
1
P
2
s
vlv1
s
vlv2
. .
R
Figure 2-6: System bond graph with added model of heat transfer across the valve
64
W. Booth
08/18/04
11
Figure 2-7: Simulations with heat transfer across valve (kA/L = 2.062E-1 ft-lbf/sec-R), and h=0 (no heat
transfer to environment)
Part 5 Choked flow
Using the following more accurate equations for subsonic flow:
t
N
d
d
C
d
A
t
2
1

u
P
u
P
d
P
u
|
\
|
.
2
P
d
P
u
|
\
|
.
1 +
(
(
(
(

and the following equation for choked flow:
t
N
d
d
C
d
A
t
2
1 +
|
\
|
.
1
1

u
P
u

2
1 +

The system equations are modified, and the results, for P
1
(init) = 200psi are shown in Figure 2-8. 200psi was
chosen as representative of choked flow, since P
d
/ P
1
< 0.528. The results are not very satisfactory, because:
1) The ode15s solver no longer converged, so I had to resort to ode45, which gives oscillatory response at
equilbrium
2) More importantly, the temperature of tank 2 can get incredibly high, if there is no heat transfer to the
environment, therefore there will be a large temperature differential across the valve due to the high pressures
involved, even without considering the effects of the shock. Thus I conclude that I need a better model of the
heat transfer across the valve, because the one for the cadmium-coated interconnecting pipe did not give me an
accurate coefficient of heat transfer.
65
W. Booth
08/18/04
12
Figure 2-8: Simulations for choked flow, with P
1
(init) = 200psia
References:
Hogan, N. 2.141 Lecture Notes MIT. Fall, 2002.
Massey, B.S. Mechanics of Fluids 6
th
ed. London: Chapman & Hall, 1992.
Rogers, G.F.C and Mayhew, Y.R. Thermodynamic and Transport Properties of Fluids 5
th
ed. Oxford: Blackwell,
1995.
http://thermal.sdsu.edu/testcenter/indexjavaapplets.html
66
W. Booth
08/18/04
13
APPENDIX A: Linearization of multi-port capacitor equations
Given the 3-port capacitor, with the mechanical port closed, we must choose a
causal-assignment to determine the form of the linearized equations, hence, choose form as
above, with S and N as inputs, and T and G as outputs on each port, respectively. Hence,
the form we are looking for is the following:
oT
oG
|
\
|
.
S
T
d
d
S
G
d
d
N
T
d
d
N
G
d
d
|
\
|
|
|
.
oS
oN
|
\
|
.
(evaluated at S
o
, N
o
operating point)
The expression for T is as used in the analysis. The expression for G is as follows:
G U T S P V +
where T(S) and P(S):
Evaluating the matrix of partials:
S
T
d
d
T
i
M c
v
N
o
exp
S
o
S
i
M c
v
N
o
\
|
.
N
T
d
d
S
o
S
i
M c
v
\
|
.
T
i
N
o
2
exp
S
o
S
i
M c
v
N
o
\
|
.
S
G
d
d
M N
o
c
v
S
T
d
d
|
\
|
.
T
i
exp
S
o
S
i
M c
v
N
o
\
|
.
S
o
S
T
d
d
|
\
|
.
V
S
P
d
d
+
substituting for dT/dS and evaluating dP/dS and factoring common factor, gives:
S
G
d
d
T
i
exp
S
o
S
i
M c
v
N
o
\
|
.
R
c
v
S
o
M c
v
N
o
\
|
.
N
G
d
d
M c
v
T S ( ) M c
v
N
o
N
T
d
d
|
\
|
.
+ S
o
N
T
d
d
|
\
|
.
V
N
P
d
d
|
\
|
.
+
substituting for dT/dN and evaluating dP/dN and factoring common factor, gives:
N
G
d
d
T
i
exp
S
o
S
i
M c
v
N
o
\
|
.
M c
p
S
o
M c
v
N
o

( )
S
o
S
i
M c
v
\
|
.
1
N
o
2
+
(
(
(
Therefore, it appears that my linear model does not satisfy Maxwells reciprocity conditions, because dG/dS does
not equal dT/dN.
I thought I had kept track of all my variables in terms of the state variables, S, and N, but I still do not see where I
have made my mistake?
67
W. Booth
08/18/04
14
APPENDIX B: Matlab M-Files
%Assignment #5 -- 2.141 W.Booth
%due: 21 Nov 02
%Assumptions:
% 1) ...
% 2) ...
global T1_init N1_init S1_init M_air Cv_air V_tank1 V_tank2 R_universal T2_init N2_init S2_init At h Asurf1
Asurf2 Tamb Question gamma_air Cd
%Change as required, to simulate various parts of problem.
Question = 5
%given
D_tank2 = 10; %in
L_tank2 = 30; %in
if Question == 3
D_tank1 = 40; %in
L_tank1 = 60; %in
else
D_tank1 = 10; %in
L_tank1 = 30; %in
end
if Question ~= 5
P1_init = 20; %lbf/in^2
else
P1_init = 200; %lbf/in^2
end
P2_init = 14.7; %lbf/in^2
T1_init = 70 + 459.67; %R -- need calc. in absolute temperature
T2_init = 70 + 459.67;
Tamb = 70 + 459.67;
Dt = 0.1; %in
%known
R_universal = 1545; %ft-lbf/lb-mol-R -- p.26 of Rogers' and Mayhew's "Thermodynamic & Transport Prop. of
Fluids"
M_air = 28.94; %1/mol -- http://www.rwc.uc.edu/koehler/biophys/8a.html
gamma_air = 1.4; %p.16 of Rogers and Mayhew
Cd = 0.5 ; %discharge coefficient
%calculations
V_tank1 = (1/4)*pi*(D_tank1^2)*L_tank1; %in^3
V_tank2 = (1/4)*pi*(D_tank2^2)*L_tank2; %in^3
At = (1/4)*pi*(Dt^2); %in^2 -- throttle area
Asurf1 = pi*D_tank1*L_tank1; %in^2 -- surface area of tank
Asurf2 = pi*D_tank2*L_tank2; %in^2 -- surface area of tank
Cv_air = R_universal/(gamma_air-1); %ft-lbf/lb-mol-R
%initial conditions
N1_init = P1_init *V_tank1/ ( 12 * R_universal *T1_init);
%lb-mol = lbf/(in^2)*(in^3)/ ((in/ft)*(ft-lbf/lb-mol-R)*R )
68
W. Booth
08/18/04
15
N2_init = P2_init*V_tank2/(12*R_universal*T2_init); %lb-mol
%Found values of specific entropy for initial conditions at
% http://thermal.sdsu.edu/testcenter/indexjavaapplets.html
%to calc. total entropy must multiply by initial mass in tank 1 (m1 = M_air * N1), hence:
S1_init = 1256.7 * M_air * N1_init;
%ft-lbf/R = %ft-lbf/lb-R * (1/mol)* (lb-mol)
S2_init = 1273.1*M_air*N2_init;
%matrix of initial conditions:
init = [N1_init T1_init P1_init S1_init N2_init T2_init P2_init S2_init];
if Question == 5
tMAX=150; %run for less time b/c numerical problems with divide by zero ~200sec ??
else
tMAX=600; %seconds
end
colour1=['b' 'g' 'r' 'c' 'k']';
hv=[0 0.0001 0.001 0.01 1]'; %heat transfer coefficients
%tab = [];
%matrix of equilibrium conditions
equil = [];
%Run model for various heat transfer coefficients
for i=1:length(hv),
%Run model with various ODESET parameters (see CODE REFERENCE [1] below)
%rtol=[ 1.e-3 1.e-5 1.e-7 ];
%for i=1:length(rtol),
%op=odeset('RelTol',rtol(i),'Stats','on');
%legend('ode45, RelTol=1.e-3 (tank1)','ode45, RelTol=1.e-3 (tank2)',...
% 'ode45, RelTol=1.e-5 (tank1)','ode45, RelTol=1.e-5 (tank2)')
h = hv(i);
%h = hv(1)
clear N1 S1 N2 S2 T1 P1 T2 P2 choked
%execute model:
%tic;
if Question ~= 5
[t,X]=ode15s(@throttling,[0 tMAX],[N1_init S1_init N2_init S2_init]);%,op);
else %ode15s could not converge with choked flow, therefore use ode45
[t,X]=ode45(@throttling,[0 tMAX],[N1_init S1_init N2_init S2_init]);%,op);
end
%tt = toc;
%tab = [ tab ; rtol(i) tt ]
N1=X(:,1); %molar mass of air in tank 1
S1=X(:,2); %entropy of tank 1
N2=X(:,3); %molar mass of air in tank 2
S2=X(:,4); %entropy of tank 2
69
W. Booth
08/18/04
16
T1=T1_init.*exp(((S1-S1_init).*M_air)./(M_air.*N1.*Cv_air));
P1 = 12*N1.*R_universal.*T1./V_tank1;
T2=T2_init.*exp(((S2-S2_init).*M_air)./(M_air.*N2.*Cv_air));
P2 = 12*N2.*R_universal.*T2./V_tank2;
Pdiff = P1-P2;
%equilibrium conditions, assuming tMAX > settling time
equil = [equil ; h N1(length(t)) T1(length(t)) P1(length(t)) S1(length(t)) N2(length(t)) T2(length(t)) P2(length(t))
S2(length(t)) Pdiff(length(t))];
%1=choked, 0=not choked
choked=zeros(length(P1));
choked=(P2<=0.528*P1);
if (Question == 5)
num_sub_plots = 5;
else
num_sub_plots = 4;
end
figure(1)
subplot(num_sub_plots,1,1),plot(t,P1,colour1(i),t,P2,strcat(colour1(i),':')),ylabel('Pressure [psi]'),axis([0 t(length(t))
P2_init P1_init]),hold on
subplot(num_sub_plots,1,2),plot(t,T1,colour1(i),t,T2,strcat(colour1(i),':')),ylabel('Temp [degR]'),hold on
subplot(num_sub_plots,1,3),plot(t,S1,colour1(i),t,S2,strcat(colour1(i),':')),ylabel('Entropy [ft-lbf/R]'),hold on
if (Question == 3); axis([0 t(length(t)) 120 140]), ylabel('Entropy of tank2 [ft-lbf/R]'); end
subplot(num_sub_plots,1,4),plot(t,N1,colour1(i),t,N2,strcat(colour1(i),':')),ylabel('Mass [lb-mol]'),xlabel('Time
[sec]'),hold on
if (Question == 3); axis([0 t(length(t)) 3.5e-3 5e-3 ]),ylabel('Mass of tank 2 [lb-mol]'); end
if (Question == 5)
subplot(num_sub_plots,1,4),xlabel('')
subplot(num_sub_plots,1,5),plot(t,choked,colour1(i)),ylabel('Choked [1=YES / 0=NO]'),xlabel('Time
[sec]'),axis([0 t(length(t)) -1 2]),hold on
end
figure(2)
plot(t,Pdiff,colour1(i)),ylabel('Delta P = P1-P2 [psi]'),xlabel('Time [sec]'),hold on,grid
end %of heat transfer coeff. for-loop
figure(1)
legend(strcat('h = ',num2str(hv(1)),' -- tank1'),...
strcat('h = ',num2str(hv(1)),' -- tank2'),...
strcat('h = ',num2str(hv(5)),' -- tank2'))
if Question == 5
legend off %legend doesn't print properly ?
end
70
W. Booth
08/18/04
17
figure(2)
legend(strcat('h = ',num2str(hv(1))),...
strcat('h = ',num2str(hv(2))),...
strcat('h = ',num2str(hv(5))))
%CODE REFERENCE [1]
%CODE TO VARY ODESET PARAMETERS AND LOOK AT STIFFNESS OF DIFF'L EQUATIONS
%FROM PAPER ON STIFF DIFF'L EQUATIONS FROM
http://www.maths.uq.edu.au/~gac/math3201/mn_ode2.pdf
%tab = [ ] ;
%for atol=[ 1.e-5 1.e-7 1.e-9 ]
%op=odeset('reltol',1.e-15,'abstol',atol,'stats','on');
%tic ;
%[t,Y]=ode23('rob',[0,1],[1;0;0],op);
%tt = toc ;
%tab = [ tab ; abstol tt ]
%end
71
W. Booth
08/18/04
18
%throttling.m -- for use with ode45 function
function xdot=throttling(t,x)
global T1_init N1_init S1_init M_air Cv_air V_tank1 V_tank2 R_universal T2_init N2_init S2_init At h Asurf1
Asurf2 Tamb Question gamma_air Cd
%x = [N1 S1 N2 S2]'
xdot=zeros(4,1); %column vector of derivatives
N1=x(1);
S1=x(2);
N2=x(3);
S2=x(4);
%"1" stands for left tank, initially at higher pressure, "2" stands for right tank
%calculating temp from M.Cv.N.dT = T.dS (similar to assign #3)
%Units on S-S0 is ft-lbf/R (total entropy), so need to multiply by M_air (1/mol) so (S-S0)/N*Cv expression is
unitless
T1=T1_init*exp(((S1-S1_init)*M_air)/(M_air*N1*Cv_air));
%calculating pressure from ideal gas law: PV = N*Ru*T
P1 = 12*N1*R_universal*T1/V_tank1;
%calculating density from definition: ro = N/V
ro1 = N1/V_tank1;
%same for tank 2
T2=T2_init*exp(((S2-S2_init)*M_air)/(M_air*N2*Cv_air));
P2 = 12*N2*R_universal*T2/V_tank2;
ro2 = N2/V_tank2;
if (Question ~=5)
if (P1 >= P2) %If P1 > P2, then flow is passing from left to right, and equations are as derived in notes:
N1dot = -At * sqrt(2*ro1*(P1-P2)/M_air); %mass flow leaving tank 1
N2dot = -N1dot; %same mass flow arrives at tank 2
else %P2 > P1 and flow is reversed, hence, N1dot is +ve and N2dot is -ve
N1dot = At * sqrt(2*ro1*(P2-P1)/M_air); %mass flow arriving at tank 1
N2dot = -N1dot; %same mass flow leaves tank 2
end
else %QUESTION 5
if (P1 >=P2) %If P1 > P2, then flow is passing from left to right
if (P2 <= 0.528*P1) %Subsonic flow -- better model

N1dot = -Cd * At * sqrt((2*gamma_air/(gamma_air-1))*ro1*(1/M_air)*P1*(((P2/P1)^(2/gamma_air))-
((P2/P1)^((gamma_air+1)/gamma_air))));

N2dot = -N1dot;

else %CHOKED FLOW
72
W. Booth
08/18/04
19

N1dot = -Cd * At * ((2/(gamma_air+1))^(1/(gamma_air-1))) *
sqrt((2*gamma_air/(gamma_air+1))*ro1*(1/M_air)*P1);

N2dot = -N1dot;

end

else %P2 > P1 and flow is reversed, hence N1dot is +ve and N2dot is -ve

if (P1 <= 0.528*P2) %Subsonic flow -- better model

N1dot = Cd * At * sqrt((2*gamma_air/(gamma_air-1))*ro2*(1/M_air)*P2*(((P1/P2)^(2/gamma_air))-
((P1/P2)^((gamma_air+1)/gamma_air))));

N2dot = -N1dot;

else %CHOKED FLOW

N1dot = Cd * At * ((2/(gamma_air+1))^(1/(gamma_air-1))) *
sqrt((2*gamma_air/(gamma_air+1))*ro2*(1/M_air)*P2);

N2dot = -N1dot;

end

end
end
%entropy flux to environment due to convective heat transfer:
Sdot_env1 = h * Asurf1 * (T1-Tamb)/T1 ;
Sdot_env2 = h * Asurf2 * (T2-Tamb)/T2 ;
%entropy flux between tanks (across valve):
Sdot_vlv1 = 2.062E-5*(T1-T2)/T1; %assuming 0.005" covering of Cadmium on 1m long interconnecting tube
%conductivity of cadmium = 96.9J/m-sec-K
Sdot_vlv2 = 2.062E-5*(T1-T2)/T2; %Sdot_vlv1 and Sdot_vlv2 will both be >0 if T1>T2, but Sdot_vlv1 >
Sdot_vlv2
%i.e. entropy lost at valve by tank 1 is > entropy gained at valve by tank 2
%irrespective of flow direction, the entropy fluxes can be defined:
if (Question ~= 4)
S1dot = (S1/N1)*N1dot-Sdot_env1;
S2dot = (S2/N2)*N2dot-Sdot_env2;
elseif (Question == 4)
S1dot = (S1/N1)*N1dot-Sdot_env1-Sdot_vlv1;
S2dot = (S2/N2)*N2dot-Sdot_env2+Sdot_vlv2;
end
%assigning state variable derivates to xdot-vector:
xdot(1)=N1dot;
xdot(2)=S1dot;
xdot(3)=N2dot;
xdot(4)=S2dot;
73
MEMORANDUM
To: 2.141 class
Date: December 9, 2002
Re: Assignment #5
Performance on assignment #5 was mixed. The effort put in was excellent but there were
some difficulties.
State variables are associated with energy-storage elements. In the two-tank throttling
problem, as the tank volumes are fixed, you need at most two state variables for each
tank, four in all. However, the system doesnt leak so the total mass in the system is
constant, hence only three of the four state variables are independent. In principle you
could work with excess state variables but its inefficient and inelegant (if you care
about that sort of thing) and you run a risk of generating numerical problems, depending
on how you formulate your state equationsthink of computing the rate of change of a
constant to be numerically indistinguishable from zero instead of recognizing that it is
identically zero.
Most of you exhibited significant numerical artifacts in your simulations (some more
than others). You should be very suspicious about abrupt rates of change in a physical
system model unless you are quite certain you deliberately introduced them. If you arent
sure, try changing parameters of either the numerical integration routine or the model.
You should be able to change the integrator parameters (especially in the direction of
greater precision) with no significant quantitative effect on the results. Because physical
system behavior usually varies smoothly with model parameters, small changes of model
parameters should result in no significant qualitative changes, only small quantitative
changes.
Another potential cause of difficulty is your choice of state variables. Most of you used
total entropy and total number of moles in each chamber but the exponentials in the
resulting state equations can cause numerical difficulties. Alternative choices (e.g.,
temperature instead of entropy) can make things simpler.
74
2.141 Term Project
Pump Fault Detection and Diagnosis (FDD) Based on Electrical Startup Transient
2.141\termprj\MPL5rpt.doc 2002.12.12
The repeatability of start transients for a typical 3-phase, single speed HVAC pump is
illustrated in Figure 1. The plot shows six smoothed
1
time histories of phase-to-neutral
power observed on the A-phase. We wish to model the system (or as much of it as
necessary) to explore the possibility of using this relatively easily observed signal
[Laughman] as a basis for fault detection and diagnosis.
Figure 1. 50-hp chilled water pump startup, A-phase power, 120Hz sampling, 6 repetitions
A common approach to fault detection and diagnosis is to model the target system, then
simulate its behavior under various conditions. After the correctly functioning system has
been characterized, various faults can be modeled and the simulations repeated. At this
point there are two possible paths. One is to find distinctive features of the responses that
can be associated with each fault. This may be difficult. The other approach is to use the
simulation results to see if system parameters, changes in which might be associated with
various faults, can be estimated from the responses. This is also difficult.
In either case, the first step is to model the system in order to better understand it, to find
out how it responds under various conditionsspecial excitations as well as typicaland
to learn if the responses contain information that could be used to identify faults or
nascent failure by one of the two diagnostic paths. At some point, real data is needed to
test detection and diagnostic procedures. Since we already have measured responses of a
real system, the main goals of this term project have been to understand the system and
develop a verifiable model, i.e. a model that reproduces the measured responses.
System Overview. The main elements of the system to be modeled are the power source,
motor, pump, and hydraulic load.
Inertances and resistances are likely to be important determinants of startup behavior. For
example, on the electrical side, inrush current is governed by source [Shaw, 2000] and
stator resistances and inductances. The inrush current time constant is likely to be small
compared to other system time constants. The flow transient is governed, in the fluid
1
by a spectral envelope filter (Appendix A)a low-pass filter found [Leeb] useful in categorizing the
start transient signatures of individual electric loads of buildings and other energy-using facilities.
Children's Court 17 June P4, 6 reps
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
22000
1 13 25 37 49 61 73 85 97 109 121 133 145 157 169 181 193
Time (1s/tick)
s
i
g
n
a
l

(
A
-
p
h
a
s
e

A
)
1
2
3
4
5
6
75
Armstrong 2/29
domain, by inertances and resistances in the piping circuit; it may, depending on coupling
and relative time scales, be significantly affected by motor and pump rotor inertances.
The non-linear natures of motor, pump, and load static performance are, of course, very
basic to the process because the entire range of operation is traversed during startup.
Ringing, evident on the tail in Figure 1
2
, could stem from compliances in the motor field,
in mechanical coupling of motor to pump, or in the hydraulic circuit
3
. In the latter case,
compliances could be near to or remote from the pumpor distributed over the circuit.
Model 1. We can represent the motor, pump, and load by their non-linear static behaviors.
The source/nature of ringing can then be explored by inserting plausible compliances,
inertances, and damping. Typical motor and pump static performance curves, taken from
9.3.3-4 of [Brown], are shown in Figures 2 and 3.
Figure 2. Typical induction motor torque-speed curve.
The pump characteristics are given in terms of non-dimensional shaft torque, static
pressure, and flow rate. The general impeller pump similitude relations [Sabersky] are:
|
|
.
|
\
|
=
|
|
.
|
\
|
=
L
Q
L
Q
g
L
P
L
Q
L
Q
f
L
T
e e
e e
, '
, '
3 2 2
3 2 5
where
T = impeller shaft torque,
P = static pressure between the pumps discharge and suction ports,
Q = volumetric flow rate,
= fluid viscosity,
= fluid density (constant for incompressible flow),
L = impeller outside diameter
4
, and
= shaft speed.
2
Very similar ringing profiles have been observed consistently in a wide range (5-60 hp) of 3-phase
induction motor driven loads including belt-driven fans as well as direct-coupled pumps
3
It could also result from electric side inertance which transforms, via motor gyrator action, to mechanical-
side compliance; our initial sense is that the transformed inductances typically encountered would be too
small to account for the rather low resonant frequency observed.
4
In the flow coefficient, Q/L
3
, L
3
may by replaced by WL
2
where W is the impeller discharge width.
0 20 40 60 80 100 120 140 160 180 200
0
5
10
15
f
t
-
l
b
torque (ft-lb)
shaft speed (rad/sec)
76
Armstrong 3/29
Figure 3. Radial-flow pump efficiency, torque, pressure versus dimensionless flow Q/L
3
e.
The effect of Reynolds number, Re = Q/L, is customarily neglected even though
viscous effects tend to be significant at low Re (low Q). Our justification for using the
simpler form (Eulerian similitude) is that torque and speed are dominated by inertial
effects early in the start transient when Q is low. Thus we have:
|
|
.
|
\
|
= =
|
|
.
|
\
|
= =
e e
e e
3 2 2
3 2 5
Pr
L
Q
g
L
P
essure nd
L
Q
f
L
T
ndTorque
The system can be readily modeled with a shaft coupling compliance and separate motor
and pump rotor inertias. Including the hydraulic inertia is also straightforward. The bond
graph representation is shown in Figure 4 and the governing equations are:
0 2 4 6 8 10 12 14
0
0.5
1
pumpEfficiency=eta
0 2 4 6 8 10 12 14
0.5
1
1.5
2
ndTorque=f'(flowCoeff)
0 2 4 6 8 10 12 14
0
0.1
0.2
ndPressure=g'(flowCoeff)
77
Armstrong 4/29
Figure 4. Bond graph of Model 1 with a shaft coupling compliance.
) ( ) , (
) , (
) (
Q P
Q
P I Q
Q
T
C
I
C
T I
L
p
p p Q
p
p p p p
m m m m
p m
=
=
=
=
e
e
e
e
e e
e e
where
m
,
p
I
m
, I
p
are angular velocities and inertances of the motor and pump rotors,
, C are the angular deflection and compliance of the motor-pump coupling,
Q, I
Q
are hydraulic flow rate and inertance,
T
p
(),P
p
() are pumps static head and torque characteristics,
T
m
() is the motors torque-speed curve, and
P
L
() is the hydraulic loads pressure-flow rate curve.
Note that the motor is treated as a black box in this initial model. Back emf and, to the
extent that they affect developed steady-state torque, the nonlinear self- and mutual-
inductance relations involving rotor position wrt the rotating fields
5
are embedded in this
black-box. But the torque-speed curve does not explicitly say anything about load current.
We hope, at most, to see if the ringing apparent in the electrical response might plausibly
be associated with mechanical and hydraulic inertances coupled by a compliance.
Model 1 Simulation. The equations governing Model 1 are coded for Matlab simulation
in Appendix D. The simulated system responses during a startup are shown in Figure 5a
with I
m
=0.008, I
p
=0.042 ft-lb/s
2
.
The effect of removing the compliance is estimated by the same model with the coupling
made very stiff (low compliance). Results of one such simulation are shown in Figure 5d.
Note that the panel labeled motor torque in all of the Figure 5 cases should be
interpreted as approximate electromagnetic torque, rather than torque available at the
shaft, since rotor inertia is modeled downstream of the static motor performance block.
5
note that the magnetic storage field is not modeled and hence the (modulated) effective mechanical
compliance, observed at the shaft of a real induction motor, does not appear in this model of the system.
T(e
m
) 1 0
C
c
I
m
1
I
p
T
p
(e
p
,Q)
P
p
(e
p
,Q)
1
I
fluid
P
L
(Q)
e
m
e
p
o Q
T(e
m
) 1 0
C
c
I
m
1
I
p
T
p
(e
p
,Q)
P
p
(e
p
,Q)
1
I
fluid
P
L
(Q)
e
m
e
p
o Q
78
Armstrong 5/29
Figure 5. Start transient of the Model1 with I
m
=0.008, I
p
=0.042 (ft-lb/s
2
), C=.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
100
200
pump speed [1/s]
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
100
200
motor speed [1/s]
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
0.2
0.4
flow rate [cfs]
time (s)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
2
4
6
x 10
-3
h
y
d
.
l
o
a
d

(
p
s
i
)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
5
10
15
m
o
.
t
o
r
q
u
e

(
f
t
-
l
b
)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
100
200
c
p
l
g
.
d
e
f
l

(
r
a
d
i
a
n
s
)
79
Armstrong 6/29
Figure 5d. Start transient with I
m
=0.0001, I
p
=0.0499 (ft-lb/s
2
).
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
2
4
6
x 10
-3
hyd.load (psi)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
5
10
15
mo.torque (ft-lb)
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
0
5
10
15
cplg.defl (radians)
time (s)
80
Armstrong 7/29
Model 2. The ringing obtained in the simulation of Figure 5, which involved shaft
compliance, is similar to the observed ringing depicted in Figure 1. However, it is not
quite plausible thateven during the brief period of startup when the motor develops peak
torquea shaft coupler would deflect 10s or 100s of radians! We expect that similar
responses might be obtained by inserting a suitable compliance at any one of several other
points in the system. In fact, we know that change in flux linkage with shaft position
appears as a mechanical compliance and a given compliance inserted at this bond-graph
location will produce the slowest and largest oscillations because it puts all of the
significant (mechanical and hydraulic) inertances downstream.
To model this variant of the system we must invert the torque characteristic. With the
shaft coupling made rigid, motor and pump rotor inertances are combined. The bond
graph representation is shown in Figure 6 and the governing equations are:
) ( ) , (
) , ( ) (
) (
Q P
Q
P I Q
Q
T
C
I I
C
L
p
p p Q
p
p p p m p
p m
=
= +
=
e
e
e
e
e
e
where
I
m
+ I
p
is the inertance of the motor and pump rotors combined,
m
(T) is the motors inverted torque-speed curve, and
, C are now the angular deflection and compliance of the stator-rotor coupling.
Figure 6. Bond graph representation of the system with a motor field compliance.
The inverted torque function is two-valued except at the torque peak. For the purpose of
simulating the start transient, this difficulty might be overcome by taking the value of
shaft speed that is higher than the value used in the previous time step (the initial value is
given as zero) or by taking the value closest to the pump shaft speed. More likely, the
causality problem is simply confirmation that an effort source cannot be easily tricked into
exhibiting compliance. In any event, other difficulties remain: 1) the electromagnetic
compliance is, in reality, far from linear and 2) the responses of main interestphase
currents and total motor powerhave not been addressed.
e
m
(T) 0
C
em
1
I
m+p
T
p
(e
p
,Q)
P
p
(e
p
,Q)
1
I
fluid
P
L
(Q)
e
p
o Q
e
m
(T) 0
C
em
1
I
m+p
T
p
(e
p
,Q)
P
p
(e
p
,Q)
1
I
fluid
P
L
(Q)
e
p
o Q
81
Armstrong 8/29
Induction Motor Model. Since model 2 faces a number of difficulties, the approach of
using a linear mechanical compliance to approximate the air-gap flux to rotor perturbation
relation was abandoned in favor of using a more realistic and informative motor model.
Standard 3-phase induction motor designs have multiple stator polls (most commonly 2 or
4 poles per phase) and a smooth rotor with a distributed winding such that the current in
each rotor bar is distinct. The apparent system order is depressingly large. If one
considers air gap flux the situation is somewhat better. However, a transformation
developed in the 1920s [Park] that allows the electrical elements to be modeled in terms
of one complex stator flow-effort pair (i
qds
,v
qds
) and one complex rotor flow-effort pair
(i
qdr
,v
qdr
) has been used ever since, with variations [Stanley, Kron, Krause, Karnopp,
Breedveld], to simulate dynamic performance of induction and synchronous machines.
The assumptions commonly made in the analysis of 3-phase, line-powered machines are
1) balanced construction and excitation,
2) linear magnetics (to be relaxed later), and
3) sinusoidal distribution of air-gap flux.
In the qd domain the very simple equivalent circuit model shown in Figure 7 is used to
evaluate input and output power when the standard assumptions hold. Although not
competent to model the significant transients expected on startup, an even simpler model
(essentially half of Figure 7; see Appendix C) applies under steady state conditions. These
circuits similarity to the usual transformer equivalent circuit can help one understand
most of the important behaviors, such as locked rotor currents and how rotor resistance
shapes the steady-state torque-speed curve. Note that the two circuits of Figure 7 contain
just four independent energy storage elements, e.g. the direct and quadrature components
of stator and rotor leakage flux. (The primes indicate rotor parameters referred to the
stator side by the stator-rotor turns ratio). Also note that there is no difficulty adding
source impedance elements to the left of the stator RL elements.
Figure 7. Magnetically superposed equivalent circuits in qd coordinates rotating at e.
Since machine behavior with rotor and air-gap flux transients is non-linear, we must
numerically integrate the state equations. Parks transformation, besides reducing the
number of electrical energy storage state variables, makes the treatment of the two
independent angular velocities manageable. The transformation is applied as shown in
Figure 8 [after Krause, slightly modified]. Each of the transformation matrices, K
s
and
82
Armstrong 9/29
K
r
, is a function of e and analogous to kinematics position modulated transformer. For
induction motors, wherin v
qdr
=0, synchronous speed (e=e
e
) is the most convenient choice
for generating state equations and results in v
qds
being constant rather than periodic.
Figure 8. Simulation block showing conjugate power variables and their causality
If the input voltages are given as boundary conditions (e.g., v
abcr
= [0 0 0]
T
and v
abcr
=
1.414sin(t)[277 277 277]
T
where [a b c]
T
is a column vector) the transformed voltages
may also be treated as given. Thus a simulation can be run with just the two center blocks.
The i
abcs
and i
abcr
currents, if needed, can be computed by a post processor.
The usual constitutive equation for inductive circuit elements, = Li = Xi, must be
expressed in matrix form when multiple windings share various portions of the magnetic
circuit. As it happens, the matrix resulting from the topology of Figure 7 has four
diagonal sub-elements such that it can be inverted by inspection, thus:
K
s
K
s
-1
K
r
K
r
-1
Flux Linkages
Currents
Torque
v
qds
i
qds
i
qdr
v
qdr
v
abcs
i
abcs
i
adcr
v
abcr
Rotor Speed
T
L
T
e
=
e
r
m lr rr m ls ss m rr ss
dr
qr
ds
qs
ss M
ss M
M rr
M rr
dr
qr
ds
qs
dr
qr
ds
qs
rr M
rr M
M ss
M ss
dr
qr
ds
qs
X X X and X X X X X X D where
X X
X X
X X
X X
D
i
i
i
i
i
i
i
i
X X
X X
X X
X X
+ ' = ' + = ' =
(
(
(
(
'
'
(
(
(
(
'
'
=
(
(
(
(
'
'
(
(
(
(
'
'
(
(
(
(
'
'
=
(
(
(
(
'
'
, ,
0 0
0 0
0 0
0 0
1
0 0
0 0
0 0
0 0
2
83
Armstrong 10/29
Model 3. The topology of Figure 8 maps rather directly to the bond graph depicted in
Figure 9.
We choose state equations that use the foregoing linear transformation, X, of the currents,
i, to flux linkage rates, , because use of flux states is both more convenient (one state
variable derivative in each equation) and (by using the net magnetizing flux instead of
separate rotor and stator contributions) more intuitive. The resulting state equations are:
Mechanical load can be modeled as a simple linear load by adding to the rotor moment of
inertia (increasing J) and letting T
Load
= Br, or the mechanical and hydraulic load static
coupler equations from model 1 or 2 can be added instead to model non-linear pump and
load characteristics.
Figure 9. Bond graph of Figure 8 with mechanical/hydraulic load elements added.
poles of number P and X X X Y where
T X Y P J
X X Y r v
X X Y r v
X X Y r v
X X Y r v
m rr ss
Load e qr ds dr qs m r
qr r ds m dr ss r dr e dr
dr r qs m qr ss r qr e qr
qs dr m ds rr s ds e ds
ds qr m qs rr s qs e qs
= ' =
' ' =
' ' ' ' = '
' ' ' ' = '
' ' =
' ' =
2 / 1 2
2
2
2
2
2
) (
/ ) ( ) 4 / 3 (
) ( )) ( (
) ( )) ( (
)) ( (
)) ( (
e e
e e e
e e e
e e
e e
S
e
1
I
m
+I
p
T
p
(e
p
,Q)
P
p
(e
p
,Q)
1
I
fluid
P
L
(Q)
e
p
= e
r
Q
IC
MTF
MTF S
e
3 2
2
3
e
84
Armstrong 11/29
Model 3 Simulation Results. The no-load start responses of a 50hp motor using
parameters suggested by Krause (from Cathay; reproduced in Appendix B) are shown in
Figures 10a-d. Figure 10a shows the 3-phase electrical power trajectory of interest. The
origin of the early chop observed in Figure 1 is now clearly evident. It is essentially the
load of a transformer with its secondary shorted and its mutual inductance modulated by
rotor slip. The response of rotor speed, Figure 10b, shows that air-gap torque is similarly
modulated even though only a small fraction of input power is being converted to
mechanical power in this early part of the motor startup.
Note that 3-phase motor power can be evaluated two ways 1) in machine coordinates: the
sum of three phase loads, each the product of phase-to-neutral voltage and the in-phase
component of current, and 2) in qd coordinates: the product of direct stator current and
voltage. Figure 10c shows the three phase currents in panel 1 and the stator direct and
quadrature currents in panel 2. The equivalent direct stator current (note scale change
from 10a) is also plottedand seen to deviate very little from the direct component of qd
current. Figure 11 shows the static torque-speed curve (red) and the torque-speed state
trajectory obtained in the simulation.
Figures 12 and 13 show the effect of reducing rotor resistance. The rotor resistance used
in the previous simulation (Figures 10-11) is typical of a 50hp motor produced before
1973. The trend since 1973 has been to market energy-efficient motors with lower
stator and rotor resistances for a given nominal power rating. Stator resistance affects
motor efficiency. Rotor resistance also affects motor efficiency but, in addition, has a
profound effect on the shape of the static torque-speed curve (Appendix C). In Figure 12
the rotor resistance is halved and in Figure 13 the rotor resistance is halved again.
The increased steepness of the torque speed curve above the torque peak is apparently
equivalent to an increased spring constant in the mechanical domain. This implies that
slip (rotor deviation from synchronous speed normalized by synchronous speed) is the
displacement variable associated with air gap flux storage. Note that induction and
synchronous motors are very different in this respect: torque angle (deviation from
synchronous position) is the equivalent of mechanical displacement in the latter. For a
given peak torque, an induction motor is much more compliant than the corresponding
synchronous motor.
The results of Figures 10-13 show that the ringing observed in the measured start
transients of Figure 1 (and all of the twenty-some other delta-connected pumps and fans
tested to date
6
) can be reproduced by adjusting the motor resistances and a single
parameter representing the system aggregate mechanical inertance.
6
7.5 to 50 hp for heating and cooling water pumps, supply, return, and cooling tower fans, in four buildings.
85
Armstrong 12/29
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0
50
100
150
200
250
300
350
400
Time (s)
R
o
t
o
r

S
p
e
e
d

(
r
a
d
/
s
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
-200
0
200
400
600
s
t
a
t
o
r

(
a
,
b
,
c
)

c
u
r
r
e
n
t
s

(
A
)
-1000
-500
0
500
1000
S
t
a
t
o
r

L
o
a
d

(
A
)
3-Phase Load: ids(red) iqs(blue) P/Vrms(green) |ids+jiqs|(blk)
total: black
re(ia): red
im(ia): blue
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
0
100
200
300
400
500
600
n
r
m
l
z
d

3
-
p
h
s

l
o
a
d

(
P
a
+
P
b
+
P
c
)
/
V
r
m
s

(
A
)
Figure 10. Simulated startup of a 50hp induction motor with no load.
86
Armstrong 13/29
0 50 100 150 200 250 300 350
-1000
-500
0
500
1000
1500
2000
Rotor Speed (rad/s)
T
o
r
q
u
e

(
N
-
m
)
Figure 11. Torque-speed curve (red) and corresponding state path during simulated startup
0 0.2 0.4 0.6 0.8 1 1.2 1.4
-100
0
100
200
300
400
500
600
n
r
m
l
z
d

3
-
p
h
s

l
o
a
d

(
P
a
+
P
b
+
P
c
)
/
V
r
m
s

(
A
)
p.rs = 0.087;
p.rr = 0.114;
p.J = 1.662;
p.Bl = 0;
Figure 12. Simulated startup of the 50hp induction motor with r'
r
= 0.114O and no load.
0 0.2 0.4 0.6 0.8 1 1.2 1.4
-100
0
100
200
300
400
500
600
700
n
r
m
l
z
d

3
-
p
h
s

l
o
a
d

(
P
a
+
P
b
+
P
c
)
/
V
r
m
s

(
A
)
p.rs = 0.087;
p.rr = 0.057;
p.J = 1.662;
p.Bl = 0;
Figure 13. Simulated startup of the 50hp induction motor with r'
r
= 0.057O and no load.
87
Armstrong 14/29
0 50 100 150 200 250 300 350
-600
-400
-200
0
200
400
600
800
1000
Rotor Speed (rad/s)
T
o
r
q
u
e

(
N
-
m
)
Figure 14. Transient and nominal torque-speed curves, 50hp motor, r'
r
= 0.057O, no load.
Thermal Model. The first order model considers i
2
R heating of the balanced stator
windings as a single lumped thermal capacitance with conductance of the resulting heat to
ambient. The rotor is the same but has its own parameters and separately tracked
temperature state. This two-node model can be tweaked to work well early in startup
before penetration of the heat waves into the stator and rotor core masses becomes
significant. With a thoughtfully elaborated conductance network, it can also give good
agreement with steady state operating temperatures. In reality, portions of the rotor and
stator windings do exhibit approximate first order behavior. Parts that lose most of their
heat via the iron core, however, should be modeled by a distributed parameter diffusion
model or some multi-node approximation thereof. The bond graph is shown in Figure 15.
The possibility of using distributed resistance-capacitance (RC) elements to model the
motors iron core elements is indicated by dashed lines surrounding the iron elements
currently represented as lumped RC elements. Notation is defined below:
u
s
i
2
r
s
= heat generated in fraction, u
s
, of stator winding that is embedded in the stator core,
u
r
i
2
r
r
= heat generated in fraction, u
r
, of rotor winding that is embedded in the rotor core,
(1-u
s
)i
2
r
s
= heat generated in fraction of stator winding that is exposed to cooling air,
(1-u
r
)i
2
r
r
= heat generated in fraction of rotor winding that is exposed to cooling air,
R(e) = one of several
7
thermal conductances to the cooling air, functions of rotor speed,
R = one of several
7
thermal resistances in the winding and core materials,
C
cs
= thermal capacitance of portion of the stator conductors embedded in the stator core,
C
csair
= thermal capacitance of portion of the stator conductors exposed to cooling air,
C
isid
= thermal capacitance of the inner (air gap-cooled) portions of the stator iron,
C
isod
= thermal capacitance of the outer portions of the stator iron and motor frame,
C
cr
= thermal capacitance of portion of the rotor conductors embedded in the rotor core,
C
crair
= thermal capacitance of portion of the rotor conductors exposed to cooling air,
C
ir
= thermal capacitance of the rotor iron, and
C
agap
= thermal capacitance of the cooling air (mainly to eliminate an algebraic equation).
7
thermal resistance subscripts have been omitted to reduce clutter in the figure; the function of a given
resistance is indicated by the subscripts on the capacitances associated with the zero junctions connected by
the resistance in question.
88
Armstrong 15/29
Figure 15. Thermal model involving four stator and three rotor capacitances; alternatively,
consider this one of many cross-coupled sections arrayed along shaft with R
agap
T
amb
ap-
pearing only in the first slice and RC
c*air
R only in the first and last (half to each) slices.
0 0.2 0.4 0.6 0.8 1 1.2 1.4
0
50
100
150
200
250
300
T
s
-
b
l
u
e
,

T
r
-
r
e
d

(
C
)
-100
0
100
200
300
400
500
600
700
(
P
a
+
P
b
+
P
c
)
/
V
r
m
s

(
A
)
Figure 16. Response of the 50hp motor start using lumped rotor and stator thermal masses.
0 C
isid
R(e) 0 0 C
ir
R(e)
0 C
cs
R(e) 0 C
cr
R(e)
R
R R
0 C
csair
R
0
0 C
crair
R
RR
(1-u
s
)i
2
r
s
(1-u
r
)i
2
r
r
u
r
i
2
r
r
u
s
i
2
r
s
R
R
T
amb
0
C
isod
C
agap
0 C
isid
R(e) 0 0 C
ir
R(e)
0 C
cs
R(e) 0 C
cr
R(e)
R
R R
0 C
csair
R
0
0 C
crair
R
RR
(1-u
s
)i
2
r
s
(1-u
r
)i
2
r
r
u
r
i
2
r
r
u
s
i
2
r
s
R
R
T
amb
0
C
isod
C
agap
0 C
isid
R(e) 0 0 C
ir
R(e)
0 C
cs
R(e) 0 C
cr
R(e)
R
R R
0 C
csair
R
0
0 C
crair
R
RR
(1-u
s
)i
2
r
s
(1-u
r
)i
2
r
r
u
r
i
2
r
r
u
s
i
2
r
s
R
RR
T
amb
0
C
isod
C
agap
89
Armstrong 16/29
A first order version of this model has been implemented. Results, for the 50hp motor
with reduced (r
r
= 0.057) rotor and default (r
s
= 0.087) stator resistances, are shown in
Figures 16 and 17. It is important to implement a more realistic model
8
of at least the
rotor because its electrical resistance has such a profound effect on the torque-speed curve
(compare Figure 17 to the corresponding result with no thermal model, Figure 14).
0 50 100 150 200 250 300 350
-600
-400
-200
0
200
400
600
800
1000
Rotor Speed (rad/s)
T
o
r
q
u
e

(
N
-
m
)
Figure 17. Deviation of transient from nominal torque stems largely from rotor heating.
Saturation Model. The simplest reasonably competent saturation model characterizes a
flux-current relation for the entire balanced magnetic circuit using the magnetizing current
in qd coordinates [Corzine]. The magnetizing reactance, X
m
in Model 3, can no longer be
treated as a constant, rather it is a function of magnetizing flux. The state equations of
model 3 become:
This introduces an algebraic loop because X
=
is a function of the direct component of the
stator-rotor common flux,
md
, which, in turn, is given in terms of X
=
:
8
Locked-rotor tests can be performed to estimate the thermal parameters if thermocouples are staked or
drilled into a rotor bar and external airflow is provided to simulate the rotor fan airflow developed at rated
speed. Responses to various faults that keep the rotor stalled might also be observed with the fan off.
1
1 1 1
) ( )) ) / / ( )( / ( (
) ( )) ) / / ( )( / ( (
)) ) / / ( )( / ( (
)) ) / / ( )( / ( (
=
=
=
=
=
|
|
.
|
\
|
+
'
+ =
' + ' ' ' ' ' + ' = '
' ' ' ' ' ' + ' = '
+ ' ' + =
' ' + =
m lr ls
qr r dr lr dr ls ds lr r dr e dr
dr r qr lr qr ls qs lr r qr e qr
qs ds lr dr ls ds ls s ds e ds
ds qs lr qr ls qs ls s qs e qs
X X X
X where
X X X X r v
X X X X r v
X X X X r v
X X X X r v
e e e
e e e
e e
e e
) ( (
qr qs ls md
X Y o ' =
=
90
Armstrong 17/29
The non-linear reactance, X
m
(
md
), could be implemented by table lookup but a safer
approach is to use the arctan function recommended by Corzine to avoid bad derivative
behavior. This simplest of saturation models ignores the fact that leakage inductances
become variable, increasing significantly at the onset of saturation.
Next Steps. Effects of saturation and thermal transient parameters need to be explored. A
test motor with known magnetic and thermal parameters is needed. A number of other
tasks should also be pursued.
For verification, the simulation results for A-phase direct and quadrature current and
power must be passed through the appropriate spectral envelope filter before comparing to
responses of the type depicted in Figure 1. We need to simulate a known
9
system and, if
possible, one in which at least the piping circuit inertance and static characteristics can be
varied, e.g. by two shunt valves, one close to and one remote from the pump.
Measurements of rotor speed and pump pressure during startup would be very useful.
Extension
10
of the system model to include resistance and inertance in the power supply as
shown in Figure 18, should be implemented. Source impedance can move downstream of
the MTF but dissipation from stator resistance must be separate from source dissipation.
Adequacy of the Eulerian similitude model [Sabersky] needs to be checked for operation
at low speeds and when the impeller is subjected to the much higher than normal torques
that are developed briefly during startup.
Faults, such as broken impeller vane, broken rotor bar, and nicked stator conductor,
should be introduced into a test pump-motor system to establish the extent to which such
parameter changes can be identified by parameter estimation based on electric-side start
transient measurements. (review bkn rotor bar literature: start excitation? thermal model?}
Figure 18. Bond graph of motor-pump-load system with source impedance added.
Discussion. A 6
th
order model, with rotor, impeller, and hydraulic inertances included,
was developed and shown to reproduce the main features of measured start transients.
The thermal behavior of the stator, that might raise stator resistance fast enough to explain
the observed diminishing inrush current while the motor is still acting like a transformer

9
for which the start transient has been accurately and repeatably measured and for which all the of the
system parameters referenced in the model have been measured or can be reasonably estimated.
10
This extension is not needed if the FDD system measures voltage waveform, as well as the phase of its
fundamental and current, at the motor terminals.
S
e
I
m
+I
p
T
p
(e
p
,Q)
P
p
(e
p
,Q)
1
I
fluid
P
L
(Q)
e
p
= e
r
Q
IC
MTF
MTF S
e
3 2
2
3
e
1
1
3
I
src
R
src
91
Armstrong 18/29
with its secondary shorted, has not been modeled yet. The state equations that account for
saturation have been developed but not yet coded in Matlab.
The proposed FDD system measures all three phase currents and voltages, making
detection of electrical imbalances in both the power supply and load very straightforward.
This study therefore focused on mechanical and hydraulic aspects of the system to explore
the extent to which they might be characterized and monitored by the electric-side
measurements. Two derived responses, total real and total reactive motor power, were
found, as expected based on the qd (Parks) transformation, to be useful, given balanced
electric-side conditions, for these purposes.
The effects of parameter changes were explored. The following parameters had effects
that were found to be important:
stator and rotor winding resistances
lumped mechanical/hydraulic inertance
dominant compliance (function of motor parameters)
stator and rotor thermal capacitances
When a compliant coupling (belt or flexible shaft coupling) is added between the motor
and fan or pump, the rotor inertance is decoupled from the downstream inertances by the
compliance. Preliminary simulations indicate that the additional poles are too highly
damped or too far from the real axis to be detected reliably from a typical start transient.
Future work should explore whether such poles might be detected from response to step
changes or impulses in excitation during normal run conditions, e.g. brief power
interruption with stator leads open or shorted.
There may be cases where compliances in the hydraulic system are significant to the
observed responses. Further field testing is needed to explore this hypothesis.
The most practical approach to FDD appears to involve a priori data. Blocked-rotor and
no-load run-up responses are easily obtained by 1) making modest extensions of the FDD
monitoring systems software and 2) calling for such tests as part of building
commissioning. The static torque curve is important, particularly the slope around
nominal steady-state load. It is not clear if the curves given by manufacturers are
sufficiently reliable and, if not, whether reliable electrical parameters (mutual inductance,
rotor and stator resistance and leakage inductance) can be obtained in the field. The
possibility of obtaining a reasonable static torque-speed curve during commissioning
seems tenuous but may be worth exploring.
Note that there are two important reasons to integrate the FDD system with the control
system. First, it can simplify building commissioning and, second, it enables an active
testing component of FDDi.e., extends passive-monitoring-based FDD by enabling
automated active testing processes. There is a third reason that does not necessarily apply
to pumps but would certainly apply to other pieces of the plant, e.g., chillers: model-based
FDD and model-based control can share the same model.
To aid further development and refinement of on-line-model based FDD, it may be
important to have the 3-phase, high bandwidth monitoring subsystem (under
development) output the unfiltered (or lightly filtered) 3-phase power signal, sampled at a
fairly high frequency (e.g. 900-1200 Hz), in addition to the spectral envelopes output by
the preprocessor stage in the current single-phase implementation.
92
Armstrong 19/29
The problem of tracking shaft speed during transients is a difficult one [Velez-Reyes,
Shaw 1999] that has not been addressed here. We have so far focused on FDD
measurements, models and approaches that do not require a measurement or estimate of
shaft speed. Belt driven shaft speeds, if needed, can be estimated in steady operation
from the spectral envelope data [5/01 CEC progress report] and might also be reliably
estimated for direct coupled pumps were the 3-phase power measurement, discussed
above, made available. It would be reasonable to measure transient shaft speeds during
commissioning if there is sufficient value in doing so. These are additional possibilities to
be explored in the future.
References.
Breedveld, P.C., 2001. A generic dynamic model of multiphase electromechanical
transduction in rotating machinery, Proc. WESIC.
Brown, F.T. 2001. Engineering System Dynamics, Marcel Dekker, New York..
Cathay, J.J., et al, 1973. Transient load model of an induction machine, IEEE Trans. PAS
92(1399-1406).
Corzine, K.A., et al, 1998. An improved method for incorporating magnetic saturation in
the q-d synchronous machine model, IEEE Trans. Energy Conversion, 13(3) pp.270-275.
Karnopp, D. 1991. State functions and bond graph dynamic models for rotary, multi-
winding electrical machines, J. Franklin Institute, 328(1) pp.45-54.
Krause, P.C., et al, 2002. Analysis of Electric Machinery and Drive Systems, 2
nd
Edition,
IEEE Press, Piscataway, NJ.
Kron, G., 1951. Equivalent Circuits of Electric Machinery, J. Wiley and Sons, NY.
Laughman, C.R., et al, 2002. Advanced non-intrusive monitoring of electric loads,
submitted October 2002 to IEEE CAP.
Leeb, S.B. and S.R. Shaw, 1994. Harmonic estimates for transient event detection, IEEE
UPEC.
Ong, C-M. 2998. Dynamic Simulation of Electric Machinery, Prentice-Hall, NJ.
Park, R.H., 1929. A two-reaction theory of synchronous machinesgeneralized method
of analysis, part I; AIEE Trans. 48(716-727).
Sabersky, R.H. and A.J. Acosta, 1971. Fluid Flow, MacMillan, NY.
Sen Gupta, D.P. and J.W. Lynn, 1980. Electrical Machine Dynamics, MacMillan.
Shaw, S.R., and S.B.Leeb, 1999. Identification of Induction Motor Parameters from
Transient Stator Current Measurements, IEEE Trans. Industrial Electronics, 46(1) pp.139-
149.
Shaw, S.R., et al, 2000. A power quality prediction system, IEEE Trans. Industrial
Electronics, 47(3) pp.511-517.
Tavner, P.J. and J. Penman, 1987. Condition Monitoring of Electric Machines, Wiley.
Velez-Reyes, M., et al, 1989. Recursive speed and parameter estimation for induction
machines, Proc. IEEE-IAS Annual Meeting, pp.607-611.
93
Armstrong 20/29
Appendix A: Spectral Envelope Filter (after Laughman)
The spectral envelope filter is an ideal demodulator for amplitude modulated signals. The
ideal demodulator assumes a known carrier frequency. The carrier of interest for analysis
of electric loads on the North American grid is, of course, 60Hz. For nonlinear loads,
useful information may be lost with the 60Hz filter. In this case, signals obtained by
spectral envelope filters applied at harmonics of 60Hz will contain additional information.
If i(t) is the current of a load driven by a voltage, e(t) = Vsin(t), the direct and quadrature
k
th
harmonic spectral envelopes are given by
e
e
e
2
) cos( ) (
2
) (
) sin( ) (
2
) (
=
=
=
)
)
T where
d k x
T
t b
d k x
T
t a
t
T t
k
t
T t
k
Additional smoothing can be obtained (albeit with additional loss of information) by using
integer multiples of T. {Matlab function to implement spectral envelope filter goes here.}
Appendix B. Motor and Pump Data
Table B-1. Bell and Gosset 1510 Series, Model 6E, 10.5 impeller. Data scaled from
B&G published performance chart at 1770 rpm.
Gpm ft.WC hp Eff
420 102 25 44
520 101.9 26.8 50
690 101.6 29.9 60
930 101 34.2 70
1070 99.6 36.5 75
1250 97.7 38.9 80
1465 94.2 41.6 85
1650 90.3 43.5 86
1895 82.5 45.9 87
2040 77 46.4 86
2125 73.5 46.7 85
2295 64 46.6 80
2415 57.5 46.3 75
Table B-2. Induction motor parameters; Cathay (1982, Table 4.10 in Krause), Ong (1998)
Machine Rating T
B
I
B
R
s
X
ls
X
M
X
lr
R
r
J
Hp V Rpm N-m A kg-m
2
V* X
M
hp
3 220 1710 11.9 5.8 .435 .754 26.13 .754 .816 .089
20 220 1748 79.2 49.7 .1062 .2145 5.834 .2145 .0764 2.8
50 460 1705 198 46.8 .087 .302 13.08 .302 .228 1.662
500 2300 1773 1980 93.6 .262 1.206 54.02 1.206 .187 11.06
2250 2300 1786 8900 421.2 .029 .226 13.04 .226 .022 63.87
94
Armstrong 21/29
Appendix C: Steady-State Motor Model:
Equivalent circuit, torque equation, and torque-speed curves.
In the qd domain a very simple equivalent circuit model, Figure B-1 (based on Figure 7)
can be used to evaluate input and output power under steady state conditions. Although
not useful for simulating certain details of transient response, its similarity to the standard
transformer equivalent circuit does help one understand most of the important motor
behaviors, such as locked rotor currents and how rotor resistance shapes the steady-state
torque-speed curve.
function [w,ssTe]=ssTcurve(p);
% [w,ssTe]=ssTcurve(p)
% given induction motor parameters (p=pIND*hp.m) return torque-speed
% curve, [w ssTe]; note that friction and windage are neglected
%p.{PP rs rr Xm Xls Xlr we J Bl vds vqs vqr vdr Tl Xss Xrr Y}
D=1/p.YY;
G=(p.rs^2 + p.Xss^2)/(p.rs^2*p.Xrr^2 - D);
sstep=.01;
w=[0:sstep:1];
w=p.we*w;
for i=1:1/sstep
s=1-(i-1)*sstep;
x=3*p.PP*p.Xm^2*p.rr*s*p.vds^2/2/p.we;
ssTe(i)=x/((p.rs*p.rr+s*D)^2 + (p.rr*p.Xss+s*p.rs*p.Xrr)^2);
end
ssTe(1+1/sstep)=0.0;
function [smax,Temax]=ssTmax(p)
% [smax,Temax]=ssTmax(p)
% given induction motor parameters (p=pIND*hp.m) return slip, smax
% at which maximum electromagnetic torque, Temax, is developed.
%p.{PP rs rr Xm Xls Xlr we J Bl vds vqs vqr vdr Tl Xss Xrr Y}
D=1/p.YY
G=(p.rs^2 + p.Xss^2)/(p.rs^2*p.Xrr^2 - D)
smax=p.rr*G
Temax=3*p.PP*p.Xm^2*G*p.vds^2/2/p.we/((p.rs-G*D)^2 + (p.Xss+G*p.Xrr)^2)
% pssIND1.m runs
[p pINDname] = pind50hp;% set motor parameters
[smax Tmax]=ssTmax(p);
[w Tss]=ssTcurve(p);
%subplot(2,1,1);
p.rs=.087;
p.rr=1.824;[w Tss]=ssTcurve(p);plot(w,Tss);hold on
p.rr=.912;[w Tss]=ssTcurve(p);plot(w,Tss);
%ylabel('Torque (N-m)')
%subplot(2,1,2);
p.rs=.0435;
p.rr=1.824;[w Tss]=ssTcurve(p);plot(w,Tss,'g');
p.rr=.912;[w Tss]=ssTcurve(p);plot(w,Tss,'g');
xlabel('Speed (rad/s)');ylabel('Torque (N-m)');hold off
95
Armstrong 22/29
0 50 100 150 200 250 300 350 400
0
100
200
300
400
500
600
700
800
900
1000
Speed (rad/s)
T
o
r
q
u
e

(
N
-
m
)
Figure C-1. Steady-state torque-speed curves for 50hp induction motor with no
mechanical losses and rotor resistances (left to right) of r
r
= 1.824, 0.912, 0.456, 0.228,
0.114, 0.058, 0.029, 0.0145; stator resistances are r
s
= 0.087 (blue) and r
s
= 0.087 (green).
(describe LRA and variation of Rr with slip to represent gyration to mechanical power.
give torque eqn and explain gyration).
96
Armstrong 23/29
Appendix D. Simulation script and derivatives function for Model 1
%simpump02.m 2002.11.11pra
pndPofndQ=[.174,0,-.000328,-.0000281];pndPofndQ=pndPofndQ(length(pndPofndQ):-1:1);
pEffofndQ=[0,.2,-.0238,.00158,-.000053];pEffofndQ=pEffofndQ(length(pEffofndQ):-1:1);
pTofW=[60,1.2,0,0,-1453e-9,182e-10,-635e-13];pTofW=pTofW(length(pTofW):-1:1);
%pPofQ=[.0005 0 30];%arg in cfs, result in psi
pPofQ=[.0005 0 0];%arg in cfs, result in psi
I.motor=0.02;I.pump=0.03;I.fluid=8310;% ft-lb/2; lb-s2/ft5
C.coupl=0.05;%1/10 is compliant, 1/100 is stiff [ft-lb/radian]
T0=.01*polyval(pndPofndQ,.01)/polyval(pEffofndQ,.01);%else Q/eff=undefined at speed=0
[t,y]=ode15s('motorpumpload02',[0 1.6],[0 0 0 0],[],1.94,1,I,C,T0,pTofW,pPofQ,pndPofndQ,
pEffofndQ);
subplot(3,1,1);plot(t,y(:,1));title('pump speed [1/s]');
subplot(3,1,2);plot(t,y(:,3));title('motor speed [1/s]');
subplot(3,1,3);plot(t,y(:,2));title('flow rate [cfs]');xlabel('time (s)');
function f=motorpumpload02(t,x,flag,rho,L,I,C,T0,pTofW,pPofQ,pndPofndQ,pEffofndQ)
%x(1),f(1) = pump shaft speed [1/s] and its time derivative [1/s2]
%x(2),f(2) = pump flow rate [cfs] and its time derivative [ft3/s2]
%x(3),f(3) = motor shaft speed [1/s] and its time derivative [1/s2]
%x(4),f(4) = shaft coupling deflection [rad] and its time derivative [rad/s]
f=[0 0 0 0]';
if x(1)>0; flowCoeff=x(2)/x(1)/L^3;else;flowCoeff=0;end
loadPressure=polyval(pPofQ,x(2))*144;%convert from psi to psf
ndPressure=polyval(pndPofndQ,flowCoeff*10000);%g'
pumpEfficiency=polyval(pEffofndQ,flowCoeff*10000);%eta
if flowCoeff>.01;
ndTorque=flowCoeff*ndPressure/pumpEfficiency;%10000f'
else;ndTorque=T0;end
moTorque=polyval(pTofW,x(3))/12;%convert from in-lb to ft-lb
f(1)=(x(4)/C.coupl - rho*L^5*x(1)^2*ndTorque/10000)/I.pump;
f(2)=(rho*(L*x(1))^2*ndPressure - loadPressure)/I.fluid;if f(2)<0; f(2)=0;end
f(3)=(moTorque - x(4)/C.coupl)/I.motor;
f(4)=x(3)-x(1);
97
Armstrong 24/29
Appendix E. Simulation script and derivatives function for Model 2
%simpump03.m 2002.11.12pra
pTofW=[60,1.2,0,0,-1453e-9,182e-10,-635e-13];pTofW=pTofW(length(pTofW):-1:1);
n=length(pTofW)-1;
d1=pTofW(1:n);d1=d1.*[n:-1:1]
p=d1;p=roots(p);k=0;
for i=1:n-1;%eliminate infeasible roots
if imag(p(i))==0 & real(p(i))>0; k=k+1; p(k)=p(i);end
end
wmax=min(p(1:k));maxT=polyval(pTofW,wmax);
I.motor=0.02;I.pump=0.03;I.fluid=8310;% ft-lb/2; lb-s2/ft5
C.coupl=0.05;C.emag=0.05;%1/10 is compliant, 1/100 is stiff [ft-lb/radian]
T0=.01*polyval(pndPofndQ,.01)/polyval(pEffofndQ,.01);%else Q/eff=undefined at speed=0
[t,y]=ode15s('motorpumpload03',[0 1.6],[0 0
0],[],1.94,1,I,C,T0,pTofW,maxT,pPofQ,pndPofndQ,pEffofndQ)
% function f=motorpumpload01(t,x,flag,rho,L,I,T0,pTofW,pPofQ,pndPofndQ,pEffofndQ)
subplot(3,1,1);plot(t,y(:,1));title('pump speed [1/s]');
subplot(3,1,2);plot(t,y(:,3));title('motor speed [1/s]');
subplot(3,1,3);plot(t,y(:,2));title('flow rate [cfs]');xlabel('time (s)');
function f=motorpumpload03(t,x,flag,rho,L,I,C,T0,pTofW,maxT,pPofQ,pndPofndQ,pEffofndQ)
%x(1),f(1) = pump shaft speed [rad/s] and its time derivative [rad/s2]
%x(2),f(2) = pump flow rate [cfs] and its time derivative [ft3/s2]
%x(3),f(3) = field-rotor deflection [rad] and its time derivative [rad/s]
%global wm;%need initial guess on first call (previous value is fine for subsequent calls)
%another ploy that might work: use pump speed as the initial guess
f=[0 0 0]';
if x(1)>0; flowCoeff=x(2)/x(1)/L^3;else;flowCoeff=0;end
loadPressure=polyval(pPofQ,x(2))*144;%convert from psi to psf
ndPressure=polyval(pndPofndQ,flowCoeff*10000);%g'
pumpEfficiency=polyval(pEffofndQ,flowCoeff*10000);%eta
if flowCoeff>.01;
ndTorque=flowCoeff*ndPressure/pumpEfficiency;%10000f'
else;ndTorque=T0;end
moTorque=12*x(3)/C.emag; %convert from ft-lb to in-lb %=polyval(pTofW,x(3))/12;
%use Matlab fcn ROOTS and fact that wm>wmprev during start transient
p=pTofW; p(7)=p(7)-moTorque; p=roots(p); k=0;
for i=1:6; %eliminate infeasible (neg & complex) roots
if imag(p(i))==0 & real(p(i))>0; k=k+1; p(k)=p(i);end
end
[wm i]=min(p(1:k));
if moTorque>maxT; wm=min([p(1:i-1) p(i+1:k)]);end
if length(wm)~=1; disp([moTorque maxT i]);disp(p);disp(wm);error('moTorque,maxT,i,p,wm
(not scalar)');end
%if x(1)==0; wm=0; end
moTorque=moTorque/12
%SECANT METHOD will have trouble getting over torque peak
%d1=pTofW(1:6);d1=d1.*[6:-1:1];%slope of polynomial
%y=polyval(pTofW,w); yp=polyval(d1,w); dw=(y-moTorque)/yp;
%while abs(dw)>0.0001;
% w=w-dw;
% y=polyval(pTofW,w); yp=polyval(d1,w); dw=(y-moTorque)/yp;
%end
%w=w-dw
%f(1)=(x(3)/C.emag - rho*L^5*x(1)^2*ndTorque/10000)/(I.pump+I.motor);
f(1)=(moTorque - rho*L^5*x(1)^2*ndTorque/10000)/(I.pump+I.motor);
f(2)=(rho*(L*x(1))^2*ndPressure - loadPressure)/I.fluid;if f(2)<0; f(2)=0;end
f(3)=wm-x(1); %wm is fcn of x(3)
98
Armstrong 25/29
Appendix F. Simulation script and functions for induction motor, Model 3
Model that assumes a linear magnetic system (linear flux-current relation)
% punIND1.m runs ODE45 simulation of induction motor
[p pINDname] = pind3Qhp;% set motor parameters
%plot(w,Tss);xlabel('Speed (rad/s)');ylabel('Torque (N-m)')
%pause
options = odeset('reltol',.0001,'abstol',1e-6*[1 1 1 1 1 1]);
[tout,y] = ode45('pind1',[0,.5],[0 0 0 0 0 0],options,p);
plot(tout,y(:,5));
xlabel('Time (s)');
ylabel('Rotor Speed (rad/s)');
title([date ' punIND1.m,pIND1.m, ' pINDname]);
T = (3/2)*p.PP*(p.Y*p.Xm/p.we)*(y(:,1).*y(:,4) - y(:,3).*y(:,2));
figure(2);
plot(y(:,5),T);hold;
plot(w,Tss,'r');
xy=axis;xy(2)=p.we;axis(xy);
xlabel('Rotor Speed (rad/s)');
ylabel('Torque (N-m)');
[m,m2]=pconvIND(tout,y,
function [ydot] = pind1(t,y,flag,p)
% [ydot] = pind1(t,y)
% evaluate derivatives for standard model (Krause 2002, eqns 4.5-39,40
% and 4.6-6; see also Shaw & Leeb 1999) of a balanced, three phase
% induction motor with linear magnetic system. The 5 state variables are
% D and Q stator and rotor flux linkage rates, psi*, and rotor speed wr.
% Displacement (th = y(6)) is not really part of the state vector but is
% included for numerical consistency (same integration scheme). TEST NEED FOR THIS!
% w is angular velocity of the arbitrary reference frame. w=we=synchronous
% speed, e.g. 60*2pi, is most convenient for simulating the induction motor.
%p.PP rs rr Xm Xls Xlr we J Bl vds vqs vqr vdr Tl Xss Xrr Y
%Xss = p.Xls + p.Xm; Xrr = p.Xlr + p.Xm; Y = 1/(Xss*Xrr+Xm^2)
psiqs = y(1);
psids = y(2);
psiqr = y(3);
psidr = y(4);
wr = y(5);
%th = y(6);
ydot = [0 0 0 0 0 0]';
ydot(1) = p.we*(p.vqs - p.rs*p.Y*(p.Xrr*psiqs - p.Xm*psiqr)) - p.w*psids;
ydot(2) = p.we*(p.vds - p.rs*p.Y*(p.Xrr*psids - p.Xm*psidr)) + p.w*psiqs;
ydot(3) = p.we*(p.vqr - p.rr*p.Y*(p.Xss*psiqr - p.Xm*psiqs)) - (p.w-wr)*psidr;
ydot(4) = p.we*(p.vdr - p.rr*p.Y*(p.Xss*psidr - p.Xm*psids)) + (p.w-wr)*psiqr;
% PP=P/2 is number of pole pairs
T = (3/2)*p.PP*(p.Y*p.Xm/p.we)*(psiqs*psidr - psiqr*psids);
ydot(5) = p.PP*(T - p.Tl)/p.J;
ydot(6) = p.w;
;
99
Armstrong 26/29
Appendix E. Simulation script and derivatives function (X
m
solver in progress) for
Model 3 with thermal and saturation effects
% punIND1.m runs ODE45 simulation of induction motor; thermal states added 20021213
[p pINDname] = pind50hp;% set motor parameters
options = odeset('reltol',.0001,'abstol',1e-6*[1 1 1 1 1 1 1 1]);
[tout,y] = ode45('pind1sat',[0,1.4],[0 0 0 0 0 20 20 0],options,p);
figure(1); plot(tout,y(:,5));
xlabel('Time (s)'); ylabel('Rotor Speed (rad/s)');
title([date ' punIND1.m,pIND1.m, ' pINDname]);
T = (3/2)*p.PP*(p.YY*p.Xm/p.we)*(y(:,1).*y(:,4) - y(:,3).*y(:,2));
figure(2); plot(y(:,5),T);hold; plot(w,Tss,'r');
xy=axis;xy(2)=1.04*p.we;axis(xy);
xlabel('Rotor Speed (rad/s)'); ylabel('Torque (N-m)');
[m,m2]=pconvIND(tout,y,p);
function [p,pINDname]=pIND50hp
% pIND50hp.m loads motor parameters into struct p
% Call pINDparm, then pIND1 via ODE45 to simulate motor start transient
% These are the parameters for a typical 50 Hp, 480V (L-L) 3-phase motor
% Krause (2002) p.165 after JJ Cathay et al IEEE PAS 92(1399-1406) 1973.
pINDname='typ 50hp ind motor'
p.PP = 2; % Number of pole pairs =P/2
p.vnmp = 460; % nameplate voltage (line-to-line rms)
p.hpnmp = 50; % nameplate horsepower
p.rpmn = 1705; % nameplate rpm
%p.Vb=
%p.Pb=
%p.Tb=
%p.Ib=
%ORIGINAL VALUES
p.rs = 0.087; % Stator resistance
p.rr = 0.228; % Rotor resistance referred to stator (ractual*(Ns/Nr)^2)
p.Xm = 13.08; % Magnetizing impedance, in Ohms on a 60 Hz base
p.Xls = 0.302; % Stator leakage impedance, in Ohms on a 60 Hz base
p.Xlr = 0.302; % Rotor leakage impedance
p.we = 120*pi; % Base electrical frequency, rads per second (376.99 @60 Hz)
p.J = 1.662; % Rotor inertia (kg-m-s2=N-m3?)
p.Bl = 0; % Mechanical load damping coefficient
%TEST VALUES
p.rs = 0.087;
p.rr = 0.057;
p.J = 1.662;
p.Bl = 0;
p.vds = 391.9; % D axis stator voltage
p.vqs = 0.0; % Q axis stator voltage
p.vdr = 0.0; % D axis rotor voltage (used for starting large machines)
p.vqr = 0.0; % Q axis rotor voltage
p.Tl = 0.0; % Mechanical load torque
p.w = p.we; % Angular velocity of the arbitrary reference frame
p.Xss = p.Xls + p.Xm;
p.Xrr = p.Xlr + p.Xm;
p.YY = 1.0/(p.Xss*p.Xrr - p.Xm*p.Xm);
%THERMAL MODEL PARMS (C=thermal capacitance, J/K or Btu/R)
p.Cscslot= 0; % stator conductor shielded from air stream, J/K
p.Cscair = 130; % stator conductor exposed to air stream
p.Csi = 0; % stator iron (distributed RC and air coupling)
p.Crcslot= 0; % rotor conductor shielded from air stream
p.Crcair = 130; % rotor conductor exposed to air stream
p.Cri = 0; % rotor iron (distributed RC and air coupling)
p.Tamb = 20; % assume rs and rr given at Tamb
p.salpha =.00393;% stator R coeff per degree temperature rise (copper:0.00393/K)
p.ralpha =.00393;% rotor R coeff per degree temperature rise
p.uscair = 30;%123; %W/K
p.urcair = 30;%123;
100
Armstrong 27/29
function [ydot] = pind1sat(t,y,flag,p)
% [ydot] = pind1sat(t,y)
% evaluate derivatives for standard model (Krause 2002, eqns 4.14-1:17
% induction motor with linear magnetic system. The 7 state variables are
% D and Q stator and rotor flux linkage rates, psi*, rotor speed wr, and
% stator and rotor winding temperatures.
%p.PP rs rr Xm Xls Xlr we J Bl vds vqs vqr vdr Tl Xss Xrr YY
%Xss = p.Xls + p.Xm; Xrr = p.Xlr + p.Xm; YY = 1/(Xss*Xrr+Xm^2)
psiqs = y(1);
psids = y(2);
psiqr = y(3);
psidr = y(4);
wr = y(5);
Ts = y(6);
Tr = y(7);
%th= y(8);
rs = p.rs*(1 + p.salpha*(Ts-p.Tamb));
rr = p.rr*(1 + p.ralpha*(Tr-p.Tamb));
%HERE need to solve for variable Xm when core becomes saturated
Xpl = 1 / (1/p.Xm + 1/p.Xls + 1/p.Xlr);
psimq = Xpl *(psiqs/p.Xls + psiqr/p.Xlr);
psimd = Xpl *(psids/p.Xls + psidr/p.Xlr);
ydot = [0 0 0 0 0 0 0]';
ydot(1) = p.we*(p.vqs + rs*(psimq - psiqs)/p.Xls) - p.w*psids;
ydot(2) = p.we*(p.vds + rs*(psimd - psids)/p.Xls) + p.w*psiqs;
ydot(3) = p.we*(p.vqr + rr*(psimq - psiqr)/p.Xlr) - (p.w-wr)*psidr;
ydot(4) = p.we*(p.vdr + rr*(psimd - psidr)/p.Xlr) + (p.w-wr)*psiqr;
iqs = (psiqs - psimq)/p.Xls;
ids = (psids - psimd)/p.Xls;
iqr = (psiqr - psimq)/p.Xlr;
idr = (psidr - psimd)/p.Xlr;
%T = (3/2)*p.PP*(p.YY*p.Xm/p.we)*(psiqs*psidr - psiqr*psids);
T = (3/2)*(p.PP/p.we)*(psiqr*idr - psidr*iqr);
ydot(5) = p.PP*(T - p.Tl - p.Bl*wr)/p.J;
ydot(6) = ((iqs^2+ids^2)*rs-(Ts-p.Tamb)*p.uscair)/p.Cscair;
ydot(7) = ((iqr^2+idr^2)*rr-(Tr-p.Tamb)*p.urcair)/p.Crcair;
ydot(8) = p.w;
101
Armstrong 28/29
Appendix F. Simulation script and derivatives function for combined models 1 & 3
% punINDpump2.m runs ODE45 simulation of induction motor,pump,load 20021207
%MOTOR PARMS
[p pINDname] = pind3Qhp;% set motor parameters
%plot(w,Tss);xlabel('Speed (rad/s)');ylabel('Torque (N-m)')
%pause
%LOAD PARMS (I,C,...,curveParms)
pump.rho = 1.94;%density of pump fluid (water) [lbm-s^2/ft]
pump.L = 1.0; %impeller diameter [ft]
pump.Imotor=0.042;pump.Ipump=0.008;pump.Ifluid=8310;% ft-lb/s2; lb-s2/ft5
pump.Ccoupl=0.001;%1/10 is compliant, 1/100 is stiff [ft-lb/radian]
pump.Tp0=.01*polyval(pndPofndQ,.01)/polyval(pEffofndQ,.01)%else Q/eff=undefined at speed=0
y1=ones(1,9); y0=0*y1;
opts = odeset('reltol',.0001,'abstol',1e-6*y1);
[tout,y] = ode45('pindpump2',[0,1.6],y0,opts,p,pump,pPofQ,pndPofndQ,pEffofndQ);
plot(tout,y(:,5));
xlabel('Time (s)');
ylabel('Rotor Speed (rad/s)');
title([date ' punINDpump2.m,pINDpump2.m, ' pINDname]);
T = (3/2)*p.PP*(p.YY*p.Xm/p.we)*(y(:,1).*y(:,4) - y(:,3).*y(:,2));
figure(2);
plot(y(:,5),T);hold;
plot(w,Tss,'r');
xy=axis;xy(2)=p.we;axis(xy);
xlabel('Rotor Speed (rad/s)');
ylabel('Torque (N-m)');
plpump2(tout,y,pump,pPofQ,pndPofndQ,pEffofndQ)
y(:,6)=y(:,9);
[m,m2]=pconvIND(tout,y,p);
function [ydot] = pindpmp2(t,y,flag,p,pump,pPofQ,pndPofndQ,pEffofndQ)
% [ydot] = pindpmp2(t,yflag,p,pump,p3,p4,p5,p6)
% evaluate derivatives for standard model (Krause 2002, eqns 4.5-39,40
% induction motor with linear magnetic system. The 5 motor state variables
% are D and Q stator and rotor flux linkage rates, psi*, and rotor speed wr.
% The 3 load state variables are coupling deflection, pump speed and flow.
%p.PP rs rr Xm Xls Xlr we J Bl vds vqs vqr vdr Tl Xss Xrr YY
%Xss = p.Xls + p.Xm; Xrr = p.Xlr + p.Xm; Y = 1/(Xss*Xrr+Xm^2)
%pump.rho L Ccoupl Ipump Ifluid
%pTofW, pPofQ, pndPofndQ, pEffofndQ are polynomial curve fit coeffs
psiqs = y(1);
psids = y(2);
psiqr = y(3);
psidr = y(4);
wr = y(5);
delta = y(6);
wp = y(7);
flow = y(8);
%th = y(9);
ydot = [0 0 0 0 0 0 0 0 0]';
102
Armstrong 29/29
%INDUCTION MOTOR (SI units)
ydot(1) = p.we*(p.vqs - p.rs*p.YY*(p.Xrr*psiqs - p.Xm*psiqr)) - p.w*psids;
ydot(2) = p.we*(p.vds - p.rs*p.YY*(p.Xrr*psids - p.Xm*psidr)) + p.w*psiqs;
ydot(3) = p.we*(p.vqr - p.rr*p.YY*(p.Xss*psiqr - p.Xm*psiqs)) - (p.w-wr)*psidr;
ydot(4) = p.we*(p.vdr - p.rr*p.YY*(p.Xss*psidr - p.Xm*psids)) + (p.w-wr)*psiqr;
moTorque= (3/2)*p.PP*(p.YY*p.Xm/p.we)*(psiqs*psidr - psiqr*psids); %N-m
%ydot(5) = p.PP*(T - p.Tl)/p.J;
ydot(9) = p.w;
%PUMP & HYDRAULIC LOAD (change to Engrg units @motor-coupling interface)
if wp>0; flowCoeff = flow/wp/pump.L^3; else; flowCoeff=0; end
loadPressure = polyval(pPofQ,flow)*144; %convert from psi to psf
ndPressure = polyval(pndPofndQ,flowCoeff*10000);%g'
pumpEfficiency= polyval(pEffofndQ,flowCoeff*10000);%eta
if flowCoeff~=0;%>.01;
ndTorque = flowCoeff*ndPressure/pumpEfficiency;%10000f'
else;ndTorque=pump.Tp0; end
ydot(5) = p.PP*(moTorque - 1.3564*delta/pump.Ccoupl)/p.J; %convert ft-lbf to N-m
ydot(6) = wr - wp; if wp<0; ydot(6)=0;end
ydot(7) = (delta/pump.Ccoupl - pump.rho*pump.L^5*wp^2*ndTorque/10000)/pump.Ipump;
%f(1)=(moTorque - rho*L^5*x(1)^2*ndTorque/10000)/pump.Ipump;
ydot(8) = (pump.rho*(pump.L*wp)^2*ndPressure - loadPressure)/pump.Ifluid;
if ydot(8)<0; ydot(8)=0;end
103
Parameterization, Analysis & Simulation of a Heat Gun
Submitted by
Thomas A. Bowers
December 10, 2002
2.141: Modeling and Simulation of Dynamic Systems
Fall 2002
104
1 Introduction
This paper discusses the dynamic analysis and simulation of a heat gun. The system
consists of an electric heating coil and a universal AC electric motor that drives a
centrifugal fan in order to produce airflow. A diagram of the system is shown in Figure 1.
Although the system looks relatively simple there are complex interactions between
electrical, mechanical, thermal, and fluid domains.
+
-
Heating Coil
Fan
AC Motor Centrifugal
Figure 1: Schematic of Heat Gun
2 System Model
Because the system operates in four domains there are several couplers required to
convert energy from one domain into energy in another domain.
2.1 Electro-mechanical Coupling
The coupling between electrical and mechanical domains is the universal AC motor. A
diagram of the universal motor is shown in Figure 2. Because the windings on the rotor
are connected in series with windings on the two poles of the stator, this motor is able to
Schematic and graph removed due to copyright considerations.
See reference [1].
Figure 2: Universal Motor Wiring Diagram and t-N Curve
operate with an AC or DC power supply [1]. This allows the motor to be treated similarly
to a simple DC motor, which is modeled as a linear gyrator. The motor constant, K
m
, can
be determined experimentally by measuring the input voltage and current when driving
the motor at a known speed. It is evident from part b of Figure 2 that AC operation is
1
105
even more linear than DC operation for this type of motor adding validity to the use of a
linear gyrator.
2.2 Electro-thermal Coupling
The interaction between the electrical domain and the fluid domain is a thermal coupling.
To transfer energy to the fluid, the material of the electrical resistor must first heat up.
The coupling is modeled as a non-conservative two-port resistor. This is due to the fact
that electrical energy is converted to thermal energy, but thermal energy does not create
electrical energy. The power dissipated in the resistor is equal to e
2
/R. This power is
converted to thermal energy through the generation of entropy. Thermal energy is stored
in the resistor, which acts as a thermal capacitor, and transferred to the air by convection.
2.3 Thermo-Fluid Coupling
The thermal energy that is transferred through convection can be modeled using the HRS
macro element that is presented in Brown [2]. This bond is used to model heat
exchangers and allows the heater temperature to be much larger than the temperature of
the fluid at the inlet or outlet port.
2.4 Mechanical-Fluid Coupling
The centrifugal fan provides the coupling between mechanical and fluid domains. The
geometry of the fan used in this heat gun is a forward-curved blade centrifugal blower,
which is also known as a sirocco fan. As with the coupling between electrical and fluid
domains, the mechanical to fluid transmission requires a non-conservative coupler
modeled as a two-port resistor. Losses in the fan are due to vorticity, friction, and
turbulence. An efficiency coefficient, q, can be used to account for these losses in the fan.
2.5 System Bond Graph
After determining the necessary transmission elements of the system, it is possible to
create the bond graph of the system, which is shown below in Figure 3.
p e
line
e
back t
motor
GY 0 1 1 I : J S
e
e i
motor
i
coil
t
fan
RS : R
coil
R : R
motor
R
f
T
Q AP
S
R
P
amb
0 C 1 S
e
S
C
P
loss
S
loss
P,h
2
P,h
1
P
amb
HRS R 0 u S
e
Figure 3: Bond Graph Representation of Heat Gun
2
106
2.6 System Equations
Because the heat gun has a parallel circuit, it is possible to analyze much of the system as
two separate networks. The output of the motor-fan system, which is the airflow through
the heat gun, can be determined independent of the thermal characteristics of the system.
The thermal response of the system is dependent on the transient in the flow rate;
however, the flow rate transient is much faster than the heat coil transient.
The equations for the fan are the most difficult to derive, but they can be determined
based on conservation of momentum [3]. The rate of change of fluid angular momentum
is equal to the torque applied on it:
) ( r ) t
fan
=
dH
0
=
d
(V dm r = V m
u 2
r
2
V
u 1 1
dt dt
For the system blade geometry the velocity diagrams shown in Figure 4 are used to
determine V
u2
and V
u1
.
|
2
V
2
W
2
U
2
e W
1
V
1
= V
r1
U
1
V
r2
V
u2
Figure 4: Velocity Diagrams for Inlet and Outlet Flow of Centrifugal Fan
From Figure 4 it is evident that the V
u2
is equal to:
Qcot |
2
V
u 2
=U
2
+V
r 2
cot | = r
2
e +
2
b r
2
2t
2
where 2t b r
2
is the area of the fan outlet and Q is the volumetric flowrate.
2
The flow into the fan is assumed to be purely radial, which gives V
u1
= 0. Substituting
these velocities back into the torque equation results in the following expression:
2 2
t
fan
= r m
2
|
r e +
Qcot | |
|
| = Qr
2
|
r e +
Qcot | |
|
|
2
b r
2 . \
2
2t
2 \
2t
2
b r
2 .
From this is clear that the fan torque is dependent both on volumetric flowrate and
angular velocity. This is consistent with non-conservative two-port resistors, which
supports the use of this type of coupling in the system bond graph. However, the torque
equation does not account for losses in the fan. The forward-curved blade geometry is not
3
107
as efficient as a backward-curved blade or an airfoil [4]. Therefore, the fan torque needs
to be divided by the fan efficiency in order to model its non-conservative nature.
t
fan
Qr
|
r e +
Qcot | |
2 2
=
b r
2 .
q
\
2
2t
2
|
|
The pressure rise in the fan is determined from the conservation of energy (with the
efficiency loss taken into account).
2
t
fan
e = APQ AP =
er
2
|
r e +
Qcot | |
|
|
b r
2 .
q
\
2
2t
2
The pressure rise in the fan is equal to the pressure lost as the air travels through the heat
gun. An adjustable orifice varies the inlet area allowing for control of the flowrate.
Additional restrictions in the flow path include the heating coil and wall friction along the
length of the flow path [5].
2
|
2
2 2 n
|
| Q
|
|
|
| Q
|
|
|
|
+
_
f
i
L
i
| Q
AP =
V
2
V
1
+ h
l
=
2
\
A
2 . \
A
1
( )
.
|
2
.
|
i =2
2 u 2 D
i \
A ( )
.
|
|
u
i
2 n 2
| A
2
A
1
2
L
i
1 |
=
Q
AP =

Q
2
\
A
2
2
A
1
2
+
_
f
i
2
i =2
D
i
A
i
2
.
|
|
2 A
e
2
where, f
i
, L
i
, and D
i
are the friction factor, length, and diameter of the i
th
restriction
section, respectively, and A
e
is the effective area of the entire system.
The ratio L/D, also known as the effective length, is provided for various pipe geometries,
valves, and other types of restrictions.
The preceding expression for AP is in terms of Q only; however, AP found from the
energy balance equation also included the angular velocity of the fan, e. Therefore, Q
can be solved in terms of the angular velocity, e:
2
er
2
| Qcot |
|
|
|
=
Q
q
|
|
|
Q
2
cot |
2
Qer
2
= 0
2
2
2
2t
2
A r
e
2
.
2t
2
q

\
r e +
b r
2 .
2
A
e \
2e
2
b r
2
cot |
2
| cot |
2
|
2
2q
b r
2
+
b r
2 .
|
| +
A
2
2t
2 \
2t
2
Q =
e
q
A r
2
e
2 e
The angular velocity is related to the state variable, p, by the following equation:
4
108
e =
p
J
The remaining equations needed to solve for the system dynamics are provided by the
motor equations and the state equation for angular momentum, p:
e
back
= K e
m
e
line
e
back
=
R
i
motor
motor
m
t
motor
= i K
motor
p =t t
fan motor
The thermal component of the heat gun behavior is determined from the equations for the
HRS macro element and the conversion of electrical energy into thermal energy. The
thermal equations are as follows:
S
C
S
0
mc
T
coil
= e T
0
2 2
P =
e
line

e
line
T =
coil
S
R
S
R
=
R
coil
T
coil
R
coil
S
loss
1T
amb
/ T
coil
=
1T
amb
/ T
coil
=
1/ H +1/ c
p
m 1/ H +1/ c
p
Q

S
C
= S
R
S
loss
3 System Parameterization
In order to simulate the heat gun, the system parameters were determined through
measurement and experimentation. The following subsections discuss the methods used
to determine the system parameters.
3.1 Fan Parameters
Most of the parameters associated with the fan involved its geometry, as indicated by the
velocity diagrams in Figure 4. The dimensions of the fan and its blades were measured,
and the blade angle at the outlet estimated by assuming that the blade formed the arc of a
circle. According to Logan [6], the maximum efficiency of centrifugal fans varies from
70 to 90 percent. Logan also provides a table relating centrifugal fan efficiency to
volumetric flow rate. The logarithmic regression from his table provides the following
relationship between flow rate and efficiency:
q = ln( 0722 . 0 Q) + 5638 . 0
5
109
For this application the flow rate is probably on the order of 0.5 L/s, which leads to an
efficiency of about 50%. Therefore, the fan parameters are as follows:
%Fan Parameters
r1=.03066; %Inlet Blade Raidus, m
B1=59.27*pi/180; %Inlet Blade Angle, rad
b1=.01; %Inlet Height, m
r2=.0375; %Outlet Blade Radius, m
B2=10*pi/180; %Outlet Blade Angle, rad
b2=.01; %Outlet Height, m
eta=.50; %Fan Efficiency
3.2 Flow Parameters
The flow parameters were the most difficult to determine by measurement. While the
geometries of the inlet and exit were easy to measure, the internal restrictions in the heat
gun posed a problem. The inlet and outlet were treated as lossless elements in the flow
path, only contributing to the calculation of velocity into and out of the system for
Bernoullis equation. To determine the pressure losses due to the flow within the heat gun,
two major sources of loss were considered. The first was the loss associated with the flow
directional change due to the fan volute. Fox and McDonald [5] provide a graphical
estimation for losses associated with 90 pipe bends that is based on the ratio of the radius
of curvature to the diameter of the pipe, r/D. For this system the ratio is ~3, which results
in an effective length, L
e
/D, of about 13. This value is multiplied by 3 to account for 270
of volute bend. The volute pressure loss is also proportional to the loss coefficient, which
is determined by the Reynolds number and the Blasius correlation for turbulent flow in
smooth pipes:
316 . 0
f =
Re
25 . 0
The second, and more significant loss of pressure within the heat gun, is the restriction
due to the heating coil. The coil significantly reduces the diameter of the pipe for a length
of 10cm just before the outlet. In addition to this significant reduction in diameter, the
coil also creates fully turbulent flow in this section of the heat gun. Both of these effects
result in huge losses, which are difficult to determine analytically. Therefore, this
parameter, which was identified as the effective length of the coil, was left as a variable
to be optimized in the simulation. This parameter also accounts for other losses in the fan
including the losses at the inlet and outlet that were neglected. The complete set of flow
parameters used in the simulation are as follows:
%Flow Parameters
Douter=.0670; %Intake Outer Diameter, m
Dinner=.0320; %Intake Inner Diameter, m
Ain=pi*(Douter^2-Dinner^2)/4%Maximum Intake Area, m^2
A1=cos(6*theta)*Ain/2; %Restricted Intake Area, m^2
D2=.029; %Outlet Diameter, m
A2=pi*D2^2/4; %Outlet Area, m^2
rho=1.19; %Assume Constant Density, kg/m^3
mu=2e-5; %Dynamic Viscosity of Air, Ns/m^2
Dduct=.0145; %Volute Diameter, m
Aduct=pi*Dduct^2/4; %Duct Area, m^2
Re=rho*.01/Aduct*Dduct/mu; %Reynolds Number in Duct (Assume 1 L/s flow)
6
110
fduct=.316/Re^.25; %Duct Friction Factor (Blasius)
fcoil=.015; %Coil Friction Factor
Lduct=13*3; %Effective Length of Duct, L/D
Lcoil=550; %Effective Length of Coil, L/D
Acoil=A2/2; %Effective Coil Area, m^2
A=(A2^2-A1^2)/(A1^2*A2^2)+fduct*Lduct/Aduct^2+fcoil*Lcoil/Acoil^2;
3.3 Motor Parameters
The motor parameters were fairly easy to determine, though measurements were made
under several different operating conditions to fully characterize the system. The easiest
parameter to measure was the resistance of the motor windings, though this varied from
70 to 75 depending on the angle of the motor shaft. Because the system incorporated a
universal motor, its parameterization was simplified by utilizing a DC power supply. By
applying a known voltage and recording steady-state current draw and shaft speed, it was
possible to determine the motor constant, K
m
, and therewith the motor torque. The motor
speed was measured with a timing gun. K
m
is plotted versus applied voltage in Figure 5
below for two operating conditions: no restriction (cover removed), and minimum
restriction (A
1
=A
in
).
M
o
t
o
r

C
o
n
s
t
a
n
t
,

K
m

0.6
0.5
0.4
0.3
0.2
0.1
0
No Restriction (Cover Removed)
Minimum Restriction (A1=Ain)
0 10 20 30 40 50 60 70 80 90
DC Voltage (V)
Figure 5: Motor Constant, K
m
, versus DC Voltage
For increasing voltage the value of K
m
decays to about 0.9 Nm/A. However, it was shown
in Figure 2b that AC operation yields slightly different values for K
m
. Therefore, the
steady-state current and speed were also measured for 120VAC. This resulted in K
m
equal
to 0.114 Nm/A, which is about 25% larger than the motor constant corresponding to DC
operation.
In reality there is a thermal transient in the motor as the rotor and stator heat up due to
electrical losses. This results in a slight decay in the current draw that was observed to
stabilize completely after about a minute. The decrease in current during this transient
was not included in the simulation because it was very small (~10mA). Additionally, the
7
111
motor resistance was measured immediately after running in order to capture the true
steady-state characteristics of the system.
The DC measurements were also used in characterizing the low-speed characteristics of
the motor. The steady-state operating torque was found for each of the points on Figure 5
and plotted versus the motor speed. This resulted in the torque-speed curve of the load,
which is shown below in Figure 6.
0
)
i
i i in)
0.02
0.04
0.06
0.08
0.1
0.12
T
o
r
q
u
e

(
N
m
No Restr ction (Cover
Removed)
Minimum Restr ct on (A1=A
0 1000 2000 3000 4000 5000 6000
Motor Speed (rpm)
Figure 6: Load Torque versus Motor Speed
This figure is very informative because of the information it contains about motor friction.
At low speed operation, one would expect the contribution of torque due to flow rate to
be negligible. Therefore, at low speeds the load torque is dominated by motor friction.
Judging from the figure, the load torque at 500 to 1500 rpm is entirely due to kinetic
friction; beyond this range, the air flow begins to add to the motor load at a quadratic rate.
This provides the parameter for motor friction, which is assumed to provide a constant
resistive torque (independent of motor speed). This value is not shown on the bond graph,
but it would be represented by a constant effort source applied at the e 1-junction for
t
motor
>t
friction
. The motor parameters are as follows:
%Motor Parameters
eline=120; %Line Voltage, Vrms
Rmotor=72.5; %Motor Coil Resistance, Ohms
Km=.114; %Motor Constant, Nm/Amp
J=.0001; %Rotatioinal Inertia, kg/m^2
tau0=.03; %Motor Friction, Nm
3.4 Heat Coil Parameters
The heat coil parameters were very easy to measure. An ohmmeter was used to find the
resistance of the coil, which was then weighed to determine its mass. The coil was
assumed to be Nichrome (80%-Ni, 20%-Cr) and the specific heat was found in Incropera
8
112
and DeWitt [7]. The value for the initial entropy of the coil does not affect the result. The
only value dependent on the coil entropy is the coil temperature, and since temperature is
dependent on the difference between the instantaneous entropy and the initial entropy the
initial value is arbitrary.
%Heat Coil Parameters
Rcoil=6;
m=.05;
c=385;
S0=0;
%Heat Coil Resistance, Ohms
%Heat Coil Mass, kg
%NiChrome Specific Heat, J/kg_K
%Initial Entropy Condition
3.5 Fluid Convection Parameters
The inlet air was assumed to be at standard temperature and pressure. Therefore, the only
variable that was needed for the simulation was the convection coefficient. In order to
determine this constant, the thermal behavior of the heat gun was measured using a
thermocouple. The following two plots, Figure 7 and Figure 8, show the temperature of
the exhaust air for two operating conditions: maximum inlet area and minimum inlet area.
0
50
o
C
)
Trial 1 Trial 3
100
150
200
250
300
350
400
450
T
e
m
p
e
r
a
t
u
r
e

(

Trial 2 Average
0 5 10 15 20 25 30 35 40
Time (s)
Figure 7: Exhaust Air Temperature with Maximum Inlet Area
9
113
T
e
m
p
e
r
a
t
u
r
e

(
o
C
)
600
500
400
300
200
100
0
0 5 10 15 20 25 30 35 40
Trial 1 Trial 2 Trial 3 Average
Time (s)
Figure 8: Exhaust Air Temperature with Minimum Inlet Area
From the preceding figures it is clear that reducing the inlet area increases the output
temperature. This is due to the fact that convection is dependent on the mass flow of air
and the flow rate depends on the inlet area. These two figures can be used to adjust the
value of H in the simulation.
%Fluid Convection Parameters
Tamb=297; %Ambient Temperature, K
Pamb=101325; %Ambient Pressure, Pa
cp=1004; %Air Specific Heat @ Constant Pressure, J/kg_K
H=4.5; %Bulk Convection Constant, J/K
4 System Simulation
Using the parameters defined in Section 3 and the constitutive equations from Section 2.6,
the following MATLAB input was used to simulate the dynamic behavior of the heat gun:
%2.141 Term Project
%Simualtion of a Heat Gun
clear all
global r2 eline Rmotor Rcoil Km Tamb S0 m c H cp rho J Vt2 eta A omegaout tauout tau0
%j=1;
for j=1:2
theta=14*(j-1)*pi/180;
%theta=input('What is the Restriction Plate Angle (0-14)?')*pi/180;
Q1(j)=fzero(@solveQ,[1e-20 .02]);
10
114
omega1(j)=omegaout;
tau1(j)=tauout;
tau=[Km*eline/Rmotor 0];
omega=[0 eline/Km*30/pi];
%ODE Solver
t=0:.1:40;
[T,X]=ode45('heatgun_dot',t,[1e-10 S0]);
%Shaft Angular Speed
Omega(:,j)=X(:,1)/J;
for i=1:length(X)
%Volumetric Flow Rate
Q(j,i)=(r2*Vt2+sqrt((r2*Vt2)^2+2*eta*A*r2^2))/(eta*A*J/X(i,1));
%Fan Load
tau_fan(j,i)=rho*Q(j,i)*(r2*(r2*Omega(i,j)+Q(j,i)*Vt2))/eta;
%Exhaust Air Temperature
Tcoil(j,i)=Tamb*exp((X(i,2)-S0)/(m*c));
Sloss_dot(i)=(1-Tamb/Tcoil(j,i))/(1/H+1/(cp*rho*Q(j,i)));
Qdot(i)=Tcoil(j,i)*Sloss_dot(i);
Tair2(j,i)=Tamb+Qdot(i)/(cp*rho*Q(j,i))-273.15;
%Motor Current
eback(i)=Km*Omega(i,j);
imotor(j,i)=(eline-eback(i))/Rmotor;
end
close all
if j==2
figure
plot(omega1(1)*30/pi,tau1(1),'bo',omega1(2)*30/pi,tau1(2),'ro')
hold on
plot(omega,tau,'k--')
xlabel('Shaft Speed, (rpm)')
ylabel('Torque, (Nm)')
title('Torque-Speed Curve for Motor Indicating Steady-State Operation Points')
legend('\theta = 0 deg (max area)','\theta = 14 deg (min area)')
figure
plot(Omega(:,1)*30/pi,tau_fan(1,:),'b',Omega(:,2)*30/pi,tau_fan(2,:),'r',omega,tau-
tau0,'k--')
xlabel('Fan Speed, (rpm)')
ylabel('Load Torque, Nm')
title('Load Torque versus Speed for Minimum and Maximum Inlet Area')
figure
plot(T,Omega(:,1)*30/pi,'b',T,Omega(:,2)*30/pi,'r')
xlabel('Time, (s)')
ylabel('Shaft Speed, (rpm)')
title('Motor Speed versus Time')
figure
plot(T,Q(1,:)*100,'b',T,Q(2,:)*100,'r')
xlabel('Time, (s)')
ylabel('Volumetric Flowrate, (L/s)')
title('Volumetric Flowrate, Q, versus Time')
figure
plot(T,Tcoil(1,:)-273.15,'b',T,Tcoil(2,:)-273.15,'r')
xlabel('Time, (s)')
ylabel('Heat Coil Temperature, (decC)')
11
115
title('Heat Coil Temperature for Minimum and Maximum Inlet Area')
figure
plot(T,Tair2(1,:),'b',T,Tair2(2,:),'r')
xlabel('Time, (s)')
ylabel('Exhaust Air Temperature, (degC)')
title('Exhaust Air Temperature for Minimum and Maximum Inlet Area')
figure
plot(T,imotor(1,:),'b',T,imotor(2,:),'r')
xlabel('Time, (s)')
ylabel('Motor Current, (A)')
title('Motor Current for Minimum and Maximum Inlet Area')
end
end
The first function called on by the preceding file was an early solution to the problem that
ignored the inertia of the motor and fan. It was, therefore, a zero order system providing
the steady-state behavior of the motor and fan. The function solveQ is shown below:
function solveQ=f(Q)
global r2 eline Rmotor Km rho Vt2 eta A omegaout tauout tau0
tauout=(r2^2*eline/Km+Q*r2*Vt2)/(eta/(rho*Q)+r2^2*Rmotor/Km^2)+tau0;
omegaout=eline/Km-Rmotor*tauout/Km^2;
P=rho*omegaout*(r2*(r2*omegaout+Q*Vt2))/eta;
solveQ=sqrt((1/A)*2*P/rho)-Q;
The second function that is solved by the main function is the set of differential equations
for the system:
function heatgun_dot=f(t,x)
global r2 eline Rmotor Rcoil Km Tamb S0 m c H cp rho J Vt2 eta A tau0
heatgun_dot=[0 0]';
p=x(1);
Scoil=x(2);
w=p/J;
Q=(r2*Vt2+sqrt((r2*Vt2)^2+2*eta*A*r2^2))/(eta*A/w);
tau_fan=rho*Q*(r2*(r2*w+Q*Vt2))/eta;
tau_motor=Km*(eline-Km*w)/Rmotor;
p_dot=tau_motor-tau_fan-tau0;
mdot=Q*rho;
Tcoil=Tamb*exp((Scoil-S0)/(m*c));
Sloss_dot=(1-Tamb/Tcoil)/(1/H+1/(cp*mdot));
Sr_dot=eline^2/(Rcoil*Tcoil);
Scoil_dot=Sr_dot-Sloss_dot;
heatgun_dot(1)=p_dot;
heatgun_dot(2)=Scoil_dot;
4.1 Simulation Results
The simulation was run foru equal to 0 and 14, which correspond to the maximum and
minimum inlet areas achievable by adjustment of the restriction plate, respectively. The
12
116
following results show the outstanding correlation between the mathematical simulation
and the recorded data. The first simulation output, Figure 9 below, shows the motor
torque versus speed. This line can be drawn without simulation using the parameter K
m
,
which was found through experimentation.
Torque-Speed Curve for Motor Indicating Steady-State Operation Points
0.2
T
o
r
q
u
e
,

(
N
m
)

0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0 2000 4000 6000 8000 10000 12000
u
u i )
= 0 deg (max area)
= 14 deg (mn area
Shaft Speed, (rpm)
Figure 9: Motor Torque-Speed Curve (Linear K
m
) Including Steady-State Speeds for Minimum and
Maximum Inlet Area
The blue and red circles on the plot are also determined from the zero dynamics of the
system. Because the system is first order, with only a very small inertance due to the
motor rotor and fan, the majority of the system operation is at steady-state. The blue
circle on the plot, which indicates the steady-state motor speed and torque for the
maximum inlet area configuration, is at 6669 rpm. This is less than 0.3% larger than the
measured steady-state speed of 6650 rpm. The red circle corresponds to a smaller inlet
area and therefore a lower flow rate. Because less momentum is transferred from motor
to fluid due to the reduced flow, the resulting torque is lower and the speed is higher. An
experimental value of this motor speed could not be measured for this geometry because
the restriction plate completely covered the fan preventing the use of the timing gun.
However, the current drawn by the motor was measured for both of these conditions (it
was used to determine the motor constant for the first configuration) and can be used to
calculate the steady-state torque and speed.
For the high speed operation with the restriction fully closed, the current draw was
530mA. This results in 0.06043 Nm of torque at a speed of 6832 rpm. The MATLAB
output for this configuration was 6970 rpm, which is only 2% larger than the value
calculated by the current measurement.
13
117
The following two plots, Figure 10 and Figure 11, show the transients in the motor speed
and current. Because of the low motor and fan inertia the system is able to reach steady-
state speed in just a few seconds.
0 5 10 15 20 25 30 35 40
0
f
)
i
u
u i )
1000
2000
3000
4000
5000
6000
7000
8000
9000
S
h
a
t

S
p
e
e
d
,

(
r
p
m
Motor Speed versus Tme
= 0 deg (max area)
= 14 deg (mn area
Time, (s)
Figure 10: Motor Speed for Dynamic Simulation of Heat Gun
Motor Current for Minimum and Maximum Inlet Area
1
2
)
u
u i )
0.4
0.6
0.8
1.2
1.4
1.6
1.8
M
o
t
o
r

C
u
r
r
e
n
t
,

(
A
= 0 deg (max area)
= 14 deg (mn area
0 5 10 15 20 25 30 35 40
Time, (s)
Figure 11: Motor Current for Dynamic Simulation of Heat Gun
Figure 12 shows the volumetric flow rate of the heat gun in Liters per seconds for both
operating regimes. Substituting the steady-state flow values back into the equation for
14
118
centrifugal fan efficiency from Section 3.1 gives efficiency values of 52% for the fully
restricted fan and 53% for the fully opened fan. These values are only a fraction higher
than the value of 50%, which was used in the simulation.
Volumetric Flowrate, Q, versus Time
0
V
o
l
i
l
(
)

u
u (mi
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
u
m
e
t
r
c

F
o
w
r
a
t
e
,

L
/
s
= 0 deg (max area)
= 14 deg n area)
0 5 10 15 20 25 30 35 40
Time, (s)
Figure 12: Volumetric Flow Rate for Dynamic Simulation of Heat Gun
While an accurate measurement of flow rate was not made, the experimental load versus
speed curve shown previously in Figure 6 illustrated how flow rate contributes
quadratically to the load. A similar result is seen in Figure 13, which shows both the shaft
load and the motor torque versus speed, which is directly proportional to flow rate.
Load Torque versus Speed for Minimum and Maximum Inlet Area
0
u
u (mi
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
L
o
a
d

T
o
r
q
u
e
,

N
m

= 0 deg (max area)
= 14 deg n area)
0 2000 4000 6000 8000 10000 12000
Fan Speed, (rpm)
Figure 13: Load Curve and Motor Torque-Speed Curve
15
119
Finally, the thermal aspects of the heat gun were addressed. Figure 14 shows the
simulated exhaust air temperature for each of the system configurations considered. The
time constant of the thermal behavior was modified by adjusting the convection
coefficient. The temperature could be scaled by changing the effective length of the flow
restrictions in order to decrease volumetric flow rate. Of course this affects the steady-
state current and speed, which were known values. The following plot represents the
optimized model of the system, which was constrained to meet the known speed and
current parameters while also attempting to simulate the thermal data shown in Figure 15.
0
50
i
i i i l
u
u i )
100
150
200
250
300
350
400
450
500
E
x
h
a
u
s
t

A
r

T
e
m
p
e
r
a
t
u
r
e
,

(
d
e
g
C
)

Exhaust A r Temperature for Mnimum and Max mum In et Area
= 0 deg (max area)
= 14 deg (mn area
0 5 10 15 20 25 30 35 40
Time, (s)
Figure 14: Exhaust Air Temperature for Minimum and Maximum Inlet Area
500
450
400
350
o
T
e
m
p
e
r
a
t
u
r
e

(
C
)
300
250
200
150
100
50
0
0 5 10 15 20 25 30 35 40
Time (s)
Figure 15: Experimental Thermal Data for Minimum and Maximum Inlet Area
16
120
It is apparent from Figure 14 that the magnitude of the simulated temperature was only
about 25-30C less than the recorded temperaturean error of 5-7%. The simulation
demonstrates a remarkable correlation with the data. There are several possible sources
for the error in the simulation with the most likely source being temperature rise in the
fluid due to flow friction. This explanation seems very feasible especially when
examining the thermocouple data. In the data presented in Figure 7, which provided the
thermal data for the fully opened configuration, it was seen that the initial temperature of
the air was about 50C. Each of these tests was run with the motor already at steady state,
which suggests that the temperature rise of the fluid as it flowed through the heat gun was
about 25C before it reached the heating element. This would provide the 25C bias that
is seen in the simulation versus the actual data. This hypothesis is also supported by the
use of dissipative elements in the model of the fluid circuit without also including their
entropy contribution to the system. With a 5-7% effect on the result of the simulation, it
appears that these losses are important to the thermal dynamics of the system.
The final variable considered in order to validate the model was the temperature of the
coil. Although no data was taken for the coil temperature, the color of the wire can be
used to qualitatively approximate the steady-state coil temperature. It was observed
during thermal testing that the coil glowed in the medium to light orange color range. The
following table provided by Process Associates of America indicates the possible range
of temperatures corresponding to this color range [8].
Table 1: Metal Color versus Temperature
Table removed due to copyright considerations.
See reference [8].
From this table it is evident that the temperature of the coil was probably in the range of
890-940C.
The temperature of the coil as predicted by the mathematical model is shown in Figure 16.
With a bulk heat transfer coefficient of 4.5 W/K the simulation predicts coil temperatures
of 850 and 910C for the fully opened and fully restricted flows respectively. According
to Table 1, these temperatures would result in a coil color in the range of salmon to
medium orange, which is indeed the case.
17
121
Heat Coil Temperature for Minimum and Maximum Inlet Area
1100
1000
900
800
H
e
a
t

C
o
i
l

T
e
m
p
e
r
a
t
u
r
e
,

(
d
e
c
C
)

700
600
500
400
300
200
100
0
0 5 10 15 20 25 30 35 40
u
u i )
= 0 deg (max area)
= 14 deg (mn area
Time, (s)
Figure 16: Temperature of Heating Coil for Minimum and Maximum Inlet Area
5 Conclusion
Although the heat gun initially seemed very complicated due to the coupling between
multiple domains, it was found that the lack of energy storage elements in the system
resulted in a deficiency of system dynamics. From the constitutive equations for the heat
gun it is apparent that the fluid flow exhibits first order behavior and that the thermal
behavior depends on both the time constant of the flow rate and on the thermal time
constant of the heat coil. However, the time constant of the flow rate is so small
compared to the thermal time constant that its contribution is not even apparent in the
thermal results of the system model. Because of this, the mechanical, electrical, and fluid
flow elements of the system can essentially be treated as a zero order system with
complicated static coupling between domains.
While the dynamics of the system were relatively uninteresting, the coupling between
domains led to a rigorous analysis of domain interactions and provided tremendous
insight into non-conservative couplers. In addition, the inability of the model to fully
predict the exhaust gas temperatures led to further insight about system dynamics that
were not included in the model. A more complete model would attempt to alleviate this
problem by including the entropy generated by flow restrictions in the overall
temperature increase in the system.
Overall, the dynamic model was very successful at demonstrating the behavior of this
device in all four domains. Unfortunately, the system could not be characterized entirely
through measurement and analysis and some values, such as the effective length of the
18
122
flow restriction and the convection coefficient, were determined simply through trial and
error. However, experimental results agreed completely with the final simulated results,
with the only major discrepancy being the exhaust air temperature as discussed
previously. With this major discrepancy between the results accounted for, the model
appears to provide a complete picture of the system.
References
[1] Universal Motors. University of Michigan, Mechanical Engineering, ME 350 Webpage. Accessed
12/04/2002. http://www.engin.umich.edu/labs/csdl/ME350/motors/ac/universal/index.html.
[2] Brown, Forbes T. Engineering System Dynamics: A Unified Graph-Centered Approach. New York:
Marcel Dekker Inc. 2001.
[3] Wright, Terry. Fluid Machinery: Performance, Analysis, and Design. New York: CRC Press. 1999.
[4] Centrifugal Wheel Designs. Lau Industries Inc., Barry Blower Central Webpage. Accessed
12/08/2002. http://www.barryblower.com/centrifugal.htm.
[5] Fox, R.W. and A.T. McDonald. Introduction to Fluid Mechanics: Fifth Edition. New York: John Wiley
and Sons, Inc. 1997.
[6] Logan, Earl Jr. Turbomachinery: Basic Theory and Applications, Second Edition. New York: Marcel
Dekker, Inc. 1993.
[7] Incropera, F.P and D.P. DeWitt. Fundamentals of Heat and Mass Transfer: Fourth Edition. New York:
John Wiley & Sons. 1996.
[8] Metal Temperature by Color, Process Associates of America. 1995-2002. Website accessed 12/09/02.
http://www.processassociates.com/process/heat/metcolor.htm
19
123
Stirling Engine
Marten Byl
12/12/02
1
124
x
Te
Th Tc
=0
R
Figure 1: Schematic of Stirling Engine with key variables noted.
Introduction
In the undergraduate class 2.670 at M.I.T., the students explore basic manufacturing tech-
niques by building a stirling engine. The class is concluded by all of the students running
their engines at the same time. As the students discover, the stirling engine is very sensi-
tive to manufacturing tolerance, specically the t of the components determines both the
friction in the engine and air leakage out of the engine. The purpose of this project was
to develop a model of the stirling engine that accurately predicts the eects of leakage and
friction on engine performance.
1 Stirling Engines
Figure 1 shows a simple schematic of a stirling engine with key parameters noted. The
concept of a stirling engine is fairly simple. The engine consist of heat source, in our case
an alcohol ame, and a heat sink, ambient air, an enclosed cylinder, a heat piston, a
power piston, and a ywheel connected to the two pistons by a set of linkages. The
concept is that the heat owing through the air in the enclosed cylinder is modulate by the
position of the heat piston. When the heat piston is located directly over the ame the
heat ow into the engine is minimized while the heat ow out of the cylinder to the heat
sink is maximized. Similarly, when heat ow in is maximize, heat ow out is minimized.
While the heat piston is moving, the power piston is also moving thus converting the
thermal energy being captured by the air into mechanical motion. The ywheel then stores
this mechanical energy, thus allowing the mechanical power to ow both in and out of the
engine. The geometry of the linkages determines the relationship between the motion of the
power piston and the heat piston.
Figure 2 shows an animation of the stirling engine in operation. Frame A shows the engine
in the starting angular position. In the starting position, we see that the heat cylinder
125
S S
S S
S
S
A
B
C
D
E
F
Figure 2: Animation of stirling engine in operation.
is positioned to maximize the heat in-ow while at the same time the power piston is
positioned to maximize output power. In frame B, we see the engine has rotated such that
output power is minimized while the heat input area is reducing. In frame C, we see that
heat outow is nearing maximum while mechanical power may actually be owing back into
the engine. Frames D and E, show the transition back to heat in ow and mechanical power
outow. Frame F shows the engine moving back into the maximum thermal power in and
mechanical power out position. I would like to thank Katherine Lilienkamp (gonzo@mit.edu)
for allowing me to use her matlab code to generate these animations.
From the 2.670 class notes by Prof. David Hart [1], the stirling engine built in the class
operates with a hot temperature, T
h
, of 600 K and a cold temperature, T
c
, of 300 K. The
typical engine will produce 1 W at 400 Rpm. The typical engine will operate at between
400-600 Rpm, with exceptional engines running at speeds up to 1200 Rpm.
126
C 1 TF
1
MTF
1
R
I
0
0
R
Se:Pa
R
Se:Th
R
Se:Ta
R*cos
x Ap
V
Sa
Na
Te
Sh
Sc
Ah( )
Ac( )
Figure 3: Bond graph model of stirling engine.
2 Stirling Engine Model
Figure 3 shows the bond graph model developed for the stirling engine. The heat source is
modelled a constant temperature eort source, T
h
, which transfers entropy to the air in the
cylinder, modelled as a multi-port capacitor, through a variable resistor. Similarly, the heat
sink is modelled as a constant temperature eort source, T
c
, which also transfers entropy to
the air in the cylinder through a dierent variable resistor. As mentioned earlier, the air in
the cylinder is modelled as a multi-port capacitor. Since leakage from the cylinder is impor-
tant, one port on the multi-port capacitor tracks the mass loss through a resitor to ambient
conditions, modelled as a constant pressure eort source. A second port on the capacitor
tracks the entropy ows too and from the heat sources and the entropy loss due to mass
ow. The nal port on the capacitor is associated with the volume change. The pressure in
the cylinder acts upon the power piston which is modelled as a constant transformer. The
piston then acts upon the linkage to the ywheel, modelled as a modulated transformer.
Finally, the ywheel is modelled as an inertia, while all of the friction losses in the system
are modelled as a resistor with damping b. The major modelling assumption used in this
bond graph are:
No power transfer through the heat piston.
Mass-less pistons.
Uniform temperature for air in engine.
Lumped friction element to govern engine speed.
Uniform constant temperature sources.
All leakage from engine through power cylinder.
Motion of heat piston is sinusoid 90 degrees ahead of power piston.
There are four state variable in this system.
127
the angular position of the ywheel
the angular velocity of the ywheel

S
e
the total entropy of the air in the cylinder
N
e
the number of mols of air contained in the cylinder
The model results in the following formulation equations:
x = R
e
(1 + sin )
A
h
= A
sc
(1 + cos )
A
c
= A
sc
(1 cos ) + P
ps
x
S
h
=
A
h
(T
h
T
e
)
T
e
S
c
=
A
c
(T
e
T
c
)
T
e
N
e
= A
l
2
e
(P
e
P
a
) or A
l
2
a
(P
a
P
e
)
S
a
=
S
e
N
e
N
a
S
e
=

S
h

S
c
+

S
a
V
e
= V
c
+ A
p
x
v
e
=
V
e
mN
e
T
e
= T
o
v
e
v
o
R
Cv
exp
s
e
s
o
C
v
P
e
= P
o
v
e
v
o
(
R
Cv
+1
)
exp
s
e
s
o
C
v
F
e
= (P
e
P
a
)A
p
e
= F
e
R
e
cos
I
=
e
b
=

e
b
I
Where:
128
R
e
= 1.25 cm = Radius of linkage pivot on ywheel
A
sc
= 40 cm
2
= Heat transfer surface area of cylinder
P
pc
= 4.9 cm = Perimeter of power piston
A
h
= variable = Hot heat transfer area
A
c
= variable = Cold heat transfer area
T
e
= variable = temp of air in engine
= 100000 W/m
2
= heat transfer constant of cylinder, this was calculated as
the thermal conductance of steel with the cylinder wall thickness
A
l
= nominally 0.06 mm
2
= area of leak
A
p
= 1.9 cm
2
= area of power piston
V
c
= 40 cm
3
= volume of air cylinder
m = 29 kg/kmol = molar mass of air
R = 287 J/kg = mass gas constant air
s
o
= 2800 J/K*kg = specic entropy of air at T=300 K
T
o
= 300 K = starting temp of air in cylinder
P
o
= P
a
= 1e5 Pa = ambient pressure
C
v
= 717 J/kg*K = constant volume specic heat
b = 0.7e-3 N/rad/s = damping constant
I = 4 kg cm
2
= inertia of ywheel
Values for thermal conductance from [2]. Values for thermal dynamic properties from
[3].
Initial Conditions
The model stirling engine was initialize at ambient thermal and pressure conditions. To
start the stirling engine in motion, a short duration (0.1 s) and small magnitude external
torque (0.1 N*m = 0.6 lbf*in) was applied to the ywheel. This is sucient to accelerate the
ywheel to a nominal velocity of 500 Rpm. Further exploration showed that a nominal motor
would start with an initial velocity of 60 Rpm but the rise time to steady state operation was
on the order of 15 seconds. This long rise time resulted in impractically long computational
runs, thus for convenience the large initial velocity is used. Both large and small initial
velocities resulted in the same steady state velocity.
3 Results
For this project, I evaluated the two variables that the 2.670 students have the most control
over when building their motors. First is to evaluate the t of the power piston into the
power cylinder (Note: in my model I assumed all of the leakage occurred at this interface
but leakage occurs at other spots). I ran the simulation with no leakage, a leakage area of
0.06 mm
2
(the equivalent to having to having a 2.54 m gap between the piston and cylinder
wall), and a leakage area of 0.1 mm
2
(5 m gap). While these gaps are smaller than I would
expect in actual motors, my model does not take into account the ow resistance caused by
129
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0
100
200
300
400
500
600
700
800
900
Rotational Speed vs Time
Time (s)
S
p
e
e
d

(
R
p
m
)
No leakage
Leakage Area = 0.06 mm
2
2
Figure 4: Rotational Speed vs Time for various leakage areas.
long interface between the piston and the cylinder (length of interface 1000x gap). My
modelling assumptions thus result in very small gaps allowing large ow.
The results for these three dierent leakage areas are shown in gures 4, 5, and 6. In
gure 4, we see rotational speed of the motor in Rpm vs time. As we expected, the steady
state velocity of the motor drops as leakage increases. In fact, a very small leakage results
in the motor not running. The no leak case has a terminal speed of 800 rpm with an output
power of 3.5 W. While the nominal motor, leakage area of 0.06 mm
2
, has a steady state
speed of 400 rpm with a corresponding power output of 1 W. Figure 5 shows the plot of
temperature vs time for all three leakage cases. All of the motors operate between 550 K
and 300 K. The size of this oscillation is most likely greater than that in an actual motor
because the actual motor has quite a bit of additional thermal storage that is not included in
my model. Figure 6 shows the pressure vs time relationship for the three leakage cases. As
expected, the motor with no leakage operates at much higher pressures than the two motors
with leakage. Also as expected, the motor with leakage operate around ambient pressure.
The second variable that I evaluated was to change the friction in the engine by changing
the viscous damping on the ywheel. Figure 7 shows the rotation speed vs time plots for
the nominal leakage motor for nominal friction, 50% friction, and 200% friction. As the plot
shows the terminal speed for the 50% friction motor is 950 rpm with a corresponding power
output of 2.4 W. While the 200% friction motor does not spin.
130
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
200
400
600
Temperature vs Time for various leakage areas
No Leakage
T
e
m
p
e
r
a
t
u
r
e

(
K
)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
200
400
600
2
T
e
m
p
e
r
a
t
u
r
e

(
K
)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
200
400
600
2
Time (s)
T
e
m
p
e
r
a
t
u
r
e

(
K
)
Figure 5: Temperature vs Time for various leakage areas.
Conclusions
In general, I am pleased with the results of this term project. The simulated system responds
appropriately to changes in the leakage and system friction. The model would likely by
improved with the incorporation of additional thermal storage elements and a better model
of leakage (thus allowing more realistic air gaps). Specically, I would try to determine how
to add the thermal storage of the air in the heat cylinder. Given the diculties that I had
in setting up the simulated system with both the correct constitutive equations and system
parameters (I still do not know what was wrong with the .m le that failed with no heat
input), I am extremely pleased with these results.
References
[1] Hart, D. P. Stirling Engine Analysis 2.670 Course notes, Department of Mechanical
Engineering, Massachusetts Institute of Technology, Cambridge, MA, 1999 and 1997.
[2] White, Frank. Heat and Mass Transfer Addison-Wesley Publications, 1988.
[3] Van Wylen, G., Sonntag, R. Fundamentals of Classical Thermodynamics 3rd John
Wiley & Sons, Inc, New York, 1986.
131
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.5
1
1.5
2
x 10
5
Pressure vs Time for various leakage areas
No Leakage
P
r
e
s
s
u
r
e

(
P
a
)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.5
1
1.5
2
x 10
5
Leakage area = 0.06 mm
2
P
r
e
s
s
u
r
e

(
P
a
)
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.5
1
1.5
2
x 10
5
Leakage area = 0.1 mm
2
Time (s)
P
r
e
s
s
u
r
e

(
P
a
)
Figure 6: Pressure vs Time for various leakage areas.
A Matlab le for model parameters
%
%Stirling Engine Simulation
%Th=Hot temp, Tc=ambient temp, Tg=gas temp, Ae=end area, As=cylinder area T
%Asc=area power cylinder, Al=leakage area, rf=radius of pivot, mu=heat transfer
clear all; close all; global Th Tc Ae As Al rf mu Apc Veo R Cv Pa
so Pp If b M Ts qa qo vo
rf=0.5*2.5/100; %radius of power linkage
Ae=(1.25*2.5)^2*pi/(4*100^2); %area of heat cylinder end plate
As=(1.25*2.5*pi*1.5*2.5/100^2); %area of oscillating portion
Apc=(0.625*2.5)^2*pi/(4*100^2); %power cylinder area
Pp=(0.625*2.5*pi/100); %power cylinder perimeter
mu=100000; %heat transfer coefficient W/K*m^2
Veo=pi*((3.69*2.5*(1.225*2.5/2)^2-2.5*2*(1.060*2.5/2)^2))/100^3;
%base volume for engine
R=287; Cv=716; Pa=1e5; so=2800;
Al=pi*((2.5*.625)^2-(2.5*.6249)^2)/(4*100^2); %leak area
If=4/(100^2);%kg*m^2 inertia of flywheel
Th=600; %hot temp
132
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0
100
200
300
400
500
600
700
800
900
1000
Rotation Speed vs Time for three different frictional losses
2
Time(s)
R
o
t
a
t
i
o
n
a
l

S
p
e
e
d

(
R
p
m
)
Nominal Friction
Friction 1/2 nominal
Friction twice nominal
Figure 7: Rotational Speed vs Time for various motor frictions.
Tc=300; %cold temp
Ts=300;
b=560e-6; %rotational damping N*m/rad/s
M=29; ni=Pa*(Veo+Apc*12.5e-3)/(Ts*R*M); Veo=(Veo+Apc*12.5e-3);
vo=Veo/(ni*M); si=ni*M*so; qa=Pa/(R*M*Tc); yi=[0 si ni 0];
options=odeset(MaxStep,0.001);
[t,y]=ode15s(@stirling,[0 5],yi,options);
B Matlab function for Stirling Engine simulation
function dy=stirling(t,y);
global Th Tc Ae As Asc Al rf mu Apc Veo R Cv Pa so Pp If b M Ts qa
qo vo
%y(1)=angle y(2)=engine entropy y(3)=moles gas y(4)=flywheel speed
dy=zeros(4,1);
x=12.5e-3*(1+sin(y(1)));
ve=Veo+x*Apc;
133
Ah=As*(1+cos(y(1)));
Ac=As*(1-cos(y(1)))+Pp*x+Ae;
ve=ve/(y(3)*M);
se=y(2)/(y(3)*M);
Te=Ts*(ve/vo)^(-R/Cv)*exp((se-so)/Cv);
Pe=Pa*(ve/vo)^(-(R/Cv+1))*exp((se-so)/Cv); dsh=mu*Ah*(Th-Te)/Te;
dsc=mu*Ac*(Te-Tc)/Te; if Pe>Pa
dn=-Al*sqrt(2*(Pe-Pa)/ve);
else
dn=Al*sqrt(2*(Pa-Pe)*qa);
end if (t>0)&(t<0.1)
tau=0.1;
else
tau=0;
end
dsa=(y(2)/(M*y(3)))*dn;
dy(2)=dsh-dsc+dsa;
dy(3)=dn;
dy(4)=((Pe-Pa)*Apc*rf*cos(y(1))-b*y(4)+tau)/If;
dy(1)=y(4);
134
Massachusetts Institute of Technology Department of Mechanical Engineering
Some useful definitions
- System: That which is to be described, analyzed & controlledanything of
interest which is to be described in detail. Often defined as a collection of objects
enclosed by a boundary, but this is not essential and the boundary may be
conceptual rather than tangible.
- Environment: All that is external to the system
1
. Everything else of interest, but
which will not be described in detail. Commonly conceived as external to the
system, but again, this is not essential.
- Open, closed: The behavior of an open system may depend upon its environment;
i.e., the two interact. A closed system does not interact with its environment.
- System Variable: A quantity, used to describe the system, which may change
with time (or space).
- System Input: A quantity that is prescribed or imposed on the system by the
environment; i.e. an independent variable.
- System Output: Any system variable of interest.
- State Determined Systems (SDS): A class of systems fully determined by a
finite set of state variables ( )
n
x x x , ,
2 1
.
- State: A minimal, complete and independent set of state variables ( )
n
x x x , ,
2 1
that uniquely describe the system.
- State Equations: To describe a state-determined systems behavior uniquely for
all t>t
0
it is sufficient to have:
(i) Values of a finite set of variables ( )
n
x x x , ,
2 1
at t
0
,
(ii) Values of a finite set system inputs ( )
r
u u u , ,
2 1
for all t>t
0
, and
(iii) A set of state equations:
( )
( )
( ) t u u u x x x f dt dx
t u u u x x x f dt dx
t u u u x x x f dt dx
r n n n
r n
r n
, , , , , ,
, , , , , ,
, , , , , ,
2 1 2 1
2 1 2 1 2 2
2 1 2 1 1 1

=
=
=
- Output equations: Any output variables ( )
m
y y y , ,
2 1
of a state-determined
system may be expressed as functions of its state and input variables:
( )
( )
( ) t u u u x x x g y
t u u u x x x g y
t u u u x x x g y
r n m m
r n
r n
, , , , , ,
, , , , , ,
, , , , , ,
2 1 2 1
2 1 2 1 2 2
2 1 2 1 1 1

=
=
=
1
It should be clear that the distinction between "system" and "environment" is not a
property of the real world, but a matter of descriptive convenience. Any given object may
be described as part of a system in one situation, and part of the environment in another.
135
Massachusetts Institute of Technology Department of Mechanical Engineering
Vector notation
A more compact notation is as follows.
- State Space: An abstract n-dimensional space defined by the state variables.
- State Vector: A point in state space defined by a complete set of state variables
( )
t
n
x x x , ,
2 1
= x .
- Input Space: An abstract r-dimensional space defined by the input variables.
- Input Vector: A point in input space defined by a complete set of input variables
( )
t
r
u u u , ,
2 1
= u .
- Output Space: An abstract m-dimensional space defined by the output variables.
- Output Vector: A point in output space defined by a complete set of output
variables ( )
t
m
y y y , ,
2 1
= y .
- State Equations: ( ) t dt d , , u x f x =
- Output Equations: ( ) t , , u x g y =
136
Integrated Modeling of Physical System Dynamics
Neville Hogan 1994 page 1
Bond Graph notation for physical system models
One of our first concerns in developing a modelling formalism is notation. To allow a concise
representation of our models, we will use pictorial diagrams, similar to the electrical network
diagram used above. Several well-established sets of pictorial symbols already exist for
depicting electrical systems, mechanical systems, and so on, but none of these notations are
adequate to be applied to all energetic systems. To emphasize the generality of energetic
considerations we will use a notation introduced by Paynter in 1959 and developed by him and
his students in the sixties bond graphs.
The net power flow between two interacting systems results in an interdependence between the
energetic states of the two systems: it bonds the two systems together into one. Consequently,
the basic symbol of the bond graph notation is a line called a bond (somewhat reminiscent of the
way chemical bonds are represented). It depicts the exchange of power between the two systems
or subsystems or elements at each end of the bond.
Figure 3.6: A bond denoting energetic interaction between two systems.
Other elements of the bond graph notation are depicted by letters and/or numbers placed at the
ends of the bond. We will introduce these symbols as we encounter the corresponding modelling
elements.
The power flow between the two systems will usually be represented as a product of two real
variables, an effort and a flow. As needed, the corresponding symbols are written adjacent to the
bond as shown below. For clarity and efficiency, we omit the box or outline representing the
boundary of each system (which is purely conceptual, anyway) and represent the interaction as
follows.
e
f
Figure 3.7: A basic bond graph.
Bond Graphs and Block Diagrams
The most important feature of the bond graph notation is that a bond explicitly represents power
flow or energy transport and distinguishes it from signal flow, the transfer of information.
Generally, the behavior of an element or system will be described mathematically as an
operation on an input variable to produce a corresponding output variable. The operator may be
dynamic, acting on one time function to produce another, or it may be static (algebraic), simply
mapping one number onto another. Those mathematical operations may be represented as the
flow of signals (e.g. the input and output) and the transformation of one into the other.
137
To represent signal flow, a familiar and versatile graphical notation is available: block diagrams.
Indeed, we have used block diagrams freely up to this point and will continue to use them.
However, before proceeding further we must recognize that block diagrams do not provide a
suitable notation for depicting physical system models because not all block diagrams represent
physical processes.
One of the most important consequences of energetic coupling between elements or systems is
that energy exchange implies interaction; a bilateral, two-way influence of each system on the
other. In contrast, block diagrams fundamentally depict a unilateral influence of one system on
another. If we wish to describe energetic interaction of two systems or elements in terms of
signal flow, then the output of one must be the input to the other and vice versa.
Usually the input to an element will be one of the power dual variables, effort or flow (though, to
reiterate, that is not essential). Consequently, because the bond represents power flow which is
determined by both of the power dual variables, the output must be the dual or conjugate of the
input; if effort is input, flow must be output and vice versa. Then, when two systems interact
energetically, we must have the situation represented by the block diagram shown in figure 3.8
(or its converse, obtained by switching A and B).
A
B
effort
flow
Figure 3.8: Block diagram of energetic interaction.
In contrast, the block diagrams shown in figure 3.9 might represent possible operations on
signals or information, but neither represents any possible energetic interaction between two
physical systems.
A
B
effort
flow
A B
effort flow flow
Figure 3.9: Two block diagrams with no physical counterparts.
Causality
Because energetic interaction is a function of two variables, when we come to describe a system
in terms of mathematical operations on numbers (i.e. signals), there are two possible choices for
138
the input and output of each element (or subsystem). In making these choices we are assigning
one variable to the role of cause (or input) and the other to the role of effect (or output), so this
choice is referred to as causality assignment. To represent this choice on a bond graph we add a
causal stroke at one end of the bond
1
as shown in figure 3.10.
Figure 3.10: Representation of causal assignment.
This graphical symbol means that the system nearest the causal stroke has effort impressed on it
as input and produces flow as output. Of necessity, the system at the other end of the bond has
flow imposed on it as input and produces effort as output. In terms of signal flow, the bond
graph of figure 3.10 is equivalent to the block diagram of figure 3.8.
We refer to the two ways of describing an element's behavior (e.g. effort in, flow out vs. flow in,
effort out) as different causal forms. Note that the two alternative causal forms may, in general,
require quite different mathematical operations. As we will see later, the causal form we use, i.e.
which variable we select as input and which we select as output, can make a lot of difference.
For example, the required mathematical operations may be well-defined in one causal form, but
not defined at all in the other.
Sign Convention
One of the important details of formulating any model is establishing a sign convention. While
this may appear to be a trivial detail, and there are no associated conceptual problems, in practice
sign conventions require considerable attention. We know that when we depress the accelerator
pedal, the speed of a car should change, and it usually does. But that information is insufficient
for almost all purposes, and especially if we wish to design a control system. It is important to
know the sign of the change: whether the car speeds up or slows down. In complex systems,
determining the sign of the effect of a given cause can require careful attention. The toil can be
minimized by working with a consistent sign convention for all energy domains, a sign
convention based on power flow. In a bond graph, the direction of positive power flow is
indicated by a half-arrow at the appropriate end of the bond.
1
If you need a mnemonic to help you to remember the convention, try this: with a little
imagination the bond with the causal stroke looks like a nail. Obviously, you would push
exert effort with the flat end of the nail.
139
Figure 3.11: Representation of power sign convention
This symbol means that the direction of positive power flow is out of system A and into system
B.
Active vs. Passive Systems
The key to our modelling formalism is to keep track of the flow of energy. We will find it useful
to distinguish between active and passive systems or elements. The precise mathematical
definition of passivity is quite subtle, but the basic idea is that a passive system (or subsystem) is
one which cannot supply an infinite amount of energy to its environment. In contrast, an active
system may supply energy to its environment indefinitely. Of course, you should realize from
this definition that an active system is an idealization, a fiction; but, as we will see, it is a very
useful one.
The sign convention we will use is that power is positive into a passive element. Much needless
confusion can be avoided by adhering to this convention. Referring to figure 3.11, by this
convention, A cannot represent a passive system, whereas B may.
The sign convention we will use for active elements is more flexible. Usually, power will be
positive out of an active element, as that makes the most physical sense; the power into the
passive elements has to come from somewhere, usually out of an active element. However, on
occasion we may wish to model an active element which is a sink, not a source, an element
which removes energy from the system independent of the system's internal state. In those cases,
we may represent power as positive into an active element.
Augmented Bond Graphs
Because the basic bond graph (figure 3.7) depicts energetic interaction independent of choice of
sign convention or assignment of causality, a graph with half-arrows (for power sign convention)
and/or strokes (for causal assignment) is sometimes referred to as an augmented bond graph.
Note that the choices power sign convention and causal assignment are quite independent
of each other.
140
Ideal Dissipative Elements
Energy cannot be destroyed, but it can be lost or dissipated, irreversibly removed from a system.
Loss or dissipation of energy may be characterized by an ideal dissipative element, ideal in the
sense that it does not store, supply or transmit energy, but simply removes it from the system.
The bond graph symbol is:
Figure 3.12: Bond graph symbol for an ideal dissipative element.
Any phenomenon characterized by an algebraic relation (possibly nonlinear) between effort and
flow will exhibit this behavior, provided the relation exists only in the first and third quadrants,
as indicated in the following diagram.
flow, f
Figure 3.13 Sketch of a possible ideally dissipative characteristic.
The restriction to the first and third quadrants ensures that the product of effort and flow is never
negative
1
, and that, by our sign convention, ensures that power always flows into the element
and never out of it; this is a passive element. We may augment the bond graph with a half-arrow
denoting the power sign convention.
Figure 3.14: Bond graph symbol for an ideal dissipator with sign convention.
1
Here we are assuming the usual convention that positive values are upwards on the vertical
axis and to the right on the horizontal axis.
141
Dissipative elements come in either of two causal forms. If the dissipator accepts flow as input,
it has resistance causality. An ideal resistor is defined as an element for which effort is a single-
valued algebraic function of flow.
e = R(f) (3.13)
Adding the causal stroke to the bond graph:
Figure 3.15: Bond graph for an ideal dissipator with sign convention and resistance causality.
The algebraic function R() is the constitutive equation for this element. The term refers to the
fact that this equation describes and is defined by the physical constitution of the device or
phenomenon being modeled. Using the constitutive equation, the power into this element may
be written as a function of the input flow alone.
P(f) = R(f) f (3.14)
Thus the power dissipated by this element is a function of its input, which is determined by the
rest of the system.
An ideal resistor may have a nonlinear constitutive relation. An ideal linear resistor is defined to
have a linear constitutive relation.
e = R f (3.15)
The power flow into an ideal linear resistor is a quadratic function of the input flow.
P(f) = R f
2
(3.16)
The parameter R is the resistance of this ideal linear dissipative element. In bond graph notation,
parameter values may be written next to the corresponding symbol as shown in the next figure
2
,
but this is optional.
:R
Figure 3.16: Bond graph convention for denoting parameter values.
2
This apparently redundant practice will make more sense when we need to distinguish between
similar types of element, e.g. resistors, with different parameter values, e.g. R
1
, R
2
, etc.
142
An example of an ideal linear resistor is the familiar (idealized) electrical resistor, characterized
by Ohm's law, which states that the voltage drop, e, across the resistor (an effort) is proportional
to the current, i, through the resistor (a flow).
e = R i (3.17)
Because the ideal linear electrical resistor is the commonest example of a dissipative element, an
ideal (nonlinear) dissipator is sometimes called a generalized resistor.
If the dissipator has the dual causal form and accepts effort as input it has conductance causality.
This is an element which has a constitutive equation defining flow as a single-valued algebraic
function of effort.
f = G(e) (3.18)
Adding the causal stroke to the bond graph:
Figure 3.17: Bond graph for an ideal dissipator with sign convention and conductance
causality.
The power flow into this element may be written as a function of the input effort alone.
P(e) = e G(e) (3.19)
An ideal linear conductance has a linear constitutive relation.
f = G e = e/R (3.20)
The power flow into this element is a quadratic function of the input effort.
P(e) = G e
2
= e
2
/R (3.21)
The parameter G = 1/R is the conductance of this ideal linear dissipative element. Again, it may
optionally be written next to the symbol for the element as shown in figure 3.18.
:G
Figure 3.18: Dissipator in conductance causality with associated parameter.
Causal Constraints
A word of caution is appropriate at this point. If we are dealing with a linear element, we are
assured that its constitutive equation can be inverted, (with the possible exception of the
degenerate cases R = 0 or G = 0) and so the element can assume either causal form. A linear
143
dissipator can be modelled equally well as a resistance or as a conductance; the choice of which
may be determined by considering other elements in the system.
For a nonlinear element the matter may not be so simple. Frequently, a nonlinear constitutive
equation will have a well-defined inverse (as in the sketch in figure 3.13), but this will not
always be the case. Consider, for example, Coulomb's friction law, commonly used to represent
dry friction.
F
friction
= B sgn(v) (3.22)
where B is a constant and sgn(
.
) is the signum function.
sgn(v)
A
1 if v > 0
0 if v = 0
-1 if v < 0
(3.23)
Figure 3.19: Dissipative characteristic of a Coulomb friction element.
Equating force with effort and speed with flow as in Table 3.1, we can see that this object is an
ideal dissipator which has resistance causality. An appropriate bond graph would be figure 3.15
and its constitutive equation is depicted in the figure 3.19.
However, this constitutive equation cannot be inverted. Whereas there is an unique force
corresponding to every speed, there is only one value of force, F = 0, which determines an
unique speed; two values of force, F = +B and F = -B correspond to infinite sets of speeds ( v > 0
and v < 0 respectively); and all other values of force do not correspond to any defined speed at
all. This model cannot be represented in conductance causality.
If the nonlinear constitutive equation of a dissipator is not invertible, the element is causally
constrained. It must be represented in whichever causal form will result in a well-defined
function which operates on the input to produce an unique output.
144
Ideal Power Sources
The ideal dissipative element defined above should be clearly distinguished from another type of
system element which can also model the removal of energy from a system. The power
dissipated in an ideal generalized resistor is a function of the state of the system; it is a passive
element which responds to changes within the system. It is also useful to define active elements
which may remove energy from a system but which are independent of its state. Being
independent of the system state, these elements may also supply energy to the system, and thus
we are led to define ideal power sources. As we might expect from our discussion of conjugate
power variables, there are two types.
An ideal effort source is an element which produces an effort independent of the flow from the
element. The bond graph symbol is:
S
e
Figure 3.20: Bond graph symbol for an ideal effort source.
An ideal flow source is an element which produces a flow independent of the effort on the
element. The bond graph symbol is:
S
f
Figure 3.21: Bond graph symbol for an ideal flow source.
Causal Constraints
By definition, an ideal effort source always imposes effort on the rest of the system.
Consequently, the causal assignment for an effort source must be as shown in figure 3.22. The
dual causal assignment would be logically incompatible with the definition of the element.
S
e
Figure 3.22: Causal assignment for an ideal effort source.
Similarly, an ideal flow source always imposes flow on the rest of the system and its causal
assignment must be as shown in figure 3.23. The dual causal assignment would be logically
incompatible with the definition of the element.
S
f
Figure 3.23: Causal assignment for an ideal flow source.
145
Note that in both cases the half-arrow depicting the direction of power flow is not determined by
the definition of the element. This is because these elements may be used to model sinks
(positive power in) as well as sources (positive power out).
Modulated Power Sources
To control something it is necessary to act on it. For a physical system this requires the addition
or removal of energy, but that is not sufficient; it is also important to be able to change one's
actions to respond to changing events. For a physical system this requires that the energy
supplied or removed be modulated, usually as a function of feedback information.
Real devices which perform this function will be examined in depth in a subsequent chapter. For
the present we will confine ourselves to defining idealized elements which would serve this
purpose: modulated power sources. An ideal modulated power source is defined as an ideal
power source with an output variable which may be modulated as a function of an input or set of
inputs.
Internal Modulation is Prohibited
This definition may seem straightforward enough, but an important restriction is that, when used
in a physical system model, the input variable(s) which modulate the power source output must
not depend directly on the internal state of the system.
To understand the reason for this restriction it is important to recognize that an ideal power
source, by virtue of its definition, does not come under the jurisdiction the first law of
thermodynamics. If we were to include a power source within the boundary of a system, then
the total energy within that boundary need not be constant. Ideal power sources are boundary
elements; they describe the flow of power into and out of a system, due to external influences,
and belong on the system boundary, not within it.
If an ideal power source were to be modulated by a variable internal to the system, then it would
be possible to describe all phenomena as modulated power sources. For example, a generalized
resistor could be described as a modulated effort source, with output, e = R(f), modulated by an
input flow, f. But that description would not retain the special properties of a dissipator. In
particular, a modulated power source is not necessarily passive, whereas a dissipator must be.
Of course, it would be possible to add those special properties as restrictions on the modulating
function for that particular source, but that would defeat the purpose of defining idealized
elements in the first place. We would, in effect, deny ourselves access to a considerable body of
knowledge which has been amassed about the behavior of physical systems. For example,
without knowing the details of their constitutive equations, we may conclude that the total
energy in a system composed only of passive elements may never increase
1
. If those passive
1
The proof of this statement requires a deeper look at the definition of passivity. This is
postponed until chapter X.
146
elements were described instead as modulated sources, we could not draw that conclusion a
priori; we would have to examine the modulating functions of each of the elements.
This prohibition of internal modulation of power sources does not preclude the modulation of a
power source via feedback based on a measurement of the internal state of a system. Indeed, we
will frequently encounter that situation in later chapters. However, the feedback modulating
functions will not, in general, be subject to the kinds of restrictions which apply to the internal
elements of a system. For example, a feedback-modulated power source will usually be an
active system, theoretically capable of injecting infinite energy into a system. That is one of the
reasons why feedback control systems are intrinsically prone to instability. But we are getting
ahead of ourselves; that is the topic of a later chapter.
The main point to be taken from this discussion is that modulated sources should be used only to
describe external influences on a system.
Power Bonds vs. Signal Flow Paths
As emphasized above, an ideal power source is a boundary element. An ideal modulated power
source also serves as a "boundary" between the description of power flow or energetic
interaction (described by bond graph notation) and signal flow or informational interaction
(which may be described by block diagram notation). To emphasize the distinction between the
two, we reserve the thick line for power flow (and use a half-arrow to denote power sign
convention), and use a thin line with a full arrow for signal flow (the usual block diagram
convention).
Note that the full arrow on a block diagram connection does not specify any sign convention, but
rather the input to a block (and the output from the connected block); it corresponds to the causal
stroke on a bond graph. A modulated power source is thus represented graphically as shown in
figure 3.24. The labels near the power bond and the signal flow path are optional; they need
only be included as necessary for clarification.
u(t)
S
e
e(t)
u(t)
S
f
f(t)
(a) (b)
Figure 3.24: Bond graph symbols for modulated power sources with signal flow and power
flow explicitly represented. (a): Modulated effort source. (b): Modulated flow source.
Explicit representation of the signal flow path is not essential, especially if the source of the
input signal is not included. In that case, the modulating input, u(t), may be written next to the
symbol for the ideal power source as shown in figure 3.25.
e(t)
u(t): S
e
u(t): S
f
f(t)
(a) (b)
147
Figure 3.25: Bond graph symbols for modulated power sources without signal flow explicitly
represented. (a): Modulated effort source. (b): Modulated flow source.
148
Ideal Symmetric Junction Elements
Having defined elements to model where energy can come from (sources) and where it can go to
(dissipators) we next need to describe the ways these pieces can be assembled. To this end we
will define ideal junction elements. Junction elements will describe the distribution of energy
between other elements, so, in keeping with the philosophy of lumped-parameter modelling, we
will assume that they don't do anything else. They don't supply, store or dissipate energy; they
are isenergic
1
.
The elements we have defined so far have a single interaction port, and one conjugate pair of
variables was sufficient to describe their energetic transactions with their environments. The
junction elements will be used to describe the interchange of energy between sets of elements,
and therefore will have many ports (obviously, a one-port junction element would be pretty
useless; it couldn't join anything to anything else). These multi-port elements obey the principle
of conservation of energy which we will apply in differential form as a power balance condition,
equation 3.4. For a junction element, stored energy and dissipated power are both zero, so
summing across all ports results in the following condition.
_
i=1
n
e
i
f
i
= 0 (3.24)
where n is the number of ports. Now, if we add a symmetry
2
condition and require that each of
the ports of a junction element be the same as each of the other ports, then it can be shown (see
appendix) that there are only two possible symmetric junction elements, and both are linear.
They are as follows.
Common Effort Junction: Type Zero
A common-effort or type zero junction has an unique effort associated with all connected bonds.
A multi-port element with n ports is characterized by n constitutive equations. The n constitutive
equations for an n-port zero junction are:
e
i
= e for i = 1 to n (3.25)
where e is the unique effort associated with the junction and the subscripts refer to labels on the
n bonds or ports. The bond graph symbol for a four-port zero junction is shown in figure 3.26.
1
Sometimes junction elements are called non-energic, but that is an odd and misleading term for
one of the most important primitives of an energy-based modelling formalism. Junction
elements are defined by the requirement that energy be constant; hence the term iso-energic or
isenergic.
2
The symmetry referred to is an invariance of properties under the operation of exchanging or
permuting the interaction ports. See Appendix.
149
Identifying the unique (common) effort next to the zero symbol as in figure 3.26 is optional, but
highly recommended.
0
Figure 3.26: Bond graph symbol for a common effort or type zero junction.
Applying the power balance condition, equation 3.24, the net power flow into the junction is
identically zero,
e
_
i=1
n
f
i
+ 0 (3.26)
where the common effort, e, has been brought outside the summation. But this identity must
hold for all values of the common effort, therefore this junction gives rise to a flow continuity
equation.
_
i=1
n
f
i
= 0 (3.27)
This flow continuity equation is a generalization of Kirchhoff's current law, encountered in
electrical circuit theory. In the electrical domain, a zero junction corresponds to a parallel
connection.
Sign Convention
Equation 3.27 is implicitly based on an assumption that the half arrows on all bonds point
inwards as shown below.
0
Figure 3.27: Sign convention assumed in flow continuity equation 3.27.
That is strictly in accordance with our sign convention for passive elements, but it is not always
convenient. Because the junction element will be used to model the transmission and
distribution of power between elements, it is more useful to allow the half arrow on any bond to
point in either direction as circumstances dictate. Whatever the orientation of the half arrows,
150
the constitutive equations for the junction require the efforts on all bonds to be identical, and that
includes sign. Therefore, for a zero-junction, the half arrow denoting power sign convention
also determines the sign of the flow. For any bond with an outward-pointing half arrow the
corresponding flow in the continuity equation must have a negative sign, as in the following
example.
f
1
- f
2
+ f
3
- f
4
= 0 (3.28)
0
1
2
3
4
Figure 3.28: Sign convention assumed in flow continuity equation 3.28.
Causal Constraints
As the junction element has multiple interaction ports, it has multiple inputs and multiple
outputs. However, the choice of what can be input and what can be output is subject to a strict
constraint. By definition, the effort of the zero junction is common to all bonds, therefore one
and only one bond may impress an effort on the junction element as shown in the following
example.
0
Figure 3.29: Example of a permissible causal assignment for a zero junction.
Any bond may impress the input effort, but only one may do so. Note that as soon as a single
bond has been chosen to determine the effort on the junction, no further choice is available; all
other bonds must provide an input flow. Conversely, that one bond must output a flow; and all
others must output the common effort. This is known as a strong causal assignment.
On the other hand, provided no bond has been chosen to impress an effort, as many as n-1 bonds
may be chosen to impose a flow on the junction. But if n-1 flows are imposed on the junction,
the n
th
flow is determined through the flow continuity equation. Therefore the n
th
bond must
determine the effort on the junction. This is known as a weak casual assignment.
Common Flow Junction: Type One
If you understand the common effort junction, then you also understand the common flow
junction, which is simply its dual. A common-flow or type one junction element has an unique
151
flow associated with all connected bonds. The n constitutive equations for an n-port one
junction are:
f
i
= f for i = 1 to n (3.29)
where f is the unique flow associated with the junction. The bond graph symbol for a four-port
one junction is shown in figure 3.30.
1
Figure 3.30: Bond graph symbol for a common flow or type zero junction.
The power balance condition for this junction element is as follows.
f
_
i=1
n
e
i
+ 0 (3.30)
where the common flow, f, has been brought outside the summation. Arguing as before, this
identity must hold for all values of the common flow, therefore this junction gives rise to an
effort compatibility equation.
_
i=1
n
e
i
= 0 (3.31)
This effort compatibility equation is a generalization of Kirchhoff's voltage law, encountered in
electrical circuit theory. In the electrical domain, a one junction corresponds to a series
connection.
Sign Convention
Equation 3.31 is implicitly based on an assumption that the half arrows on all bonds point
inwards. As with the zero junction, it is more useful to let the half arrow on any bond point in
either direction as circumstances dictate. Arguing as before, whatever the orientation of the half
arrows, the constitutive equations for the junction require the flows on all bonds to be identical
including sign. Therefore, for a one-junction, the half arrow denoting power sign convention
also determines the sign of the effort. For any bond with an outward-pointing half arrow the
corresponding effort in the compatibility equation must have a negative sign, as in the following
example.
- e
1
- e
2
+ e
3
+ e
4
= 0 (3.32)
152
1
1
2
3
4
Figure 3.31: Sign convention assumed in effort compatibility equation 3.32.
Causal Constraints
As with the zero junction, the choice of what can be input to and output from a one junction is
subject to a strict constraint. By definition, the flow is common to all bonds, therefore one and
only one bond may impose a flow on the junction element as shown in the following example.
1
Figure 3.32: Example of a permissible causal assignment for a one junction.
The choice of any single bond to determine the flow on the junction is a strong causal
assignment, and no further choice is available; all other bonds must provide an input effort.
Conversely, that one bond must output a effort; and all others must output the common flow.
If no bond is selected to impose a flow, as many as n-1 bonds may be chosen to impress an effort
on the junction. A weak causal assignment may be made by choosing n-1 efforts to be
impressed on the junction. Because the n
th
effort is determined through the effort compatibility
equation, the n
th
bond must determine the flow on the junction.
153
Generalized Energy Variables
Energetic interactions are mediated by the flow of power. Power flow through an interaction
port may be expressed as the product of two real-valued variables, an effort and a flow, and all
instantaneous interactions between systems or elements may be described in terms of these
conjugate power variables.
However, to define the energy stored in a system (i.e. its instantaneous energetic state) it is
necessary to define energy variables. Just as we may define two power variables, we may define
two dual or conjugate energy variables, obtained by integrating the power variables with respect
to time.
The first of these is generalized momentum
1
, p.
p A_
_
]
(
t
o
t
e(t)dt + p(t
o
) (4.1)
It will be associated with kinetic energy storage. This relation may be differentiated.
dp = e dt (4.2)
The conjugate variable is generalized displacement, q.
q A_
_
]
(
t
o
t
f(t)dt + q(t
o
) (4.3)
It will be associated with potential energy storage. This relation, too, may be differentiated.
dq = f dt (4.4)
The following tables provide a partial list of energy variables and notation we will use for
several energetic media.
1
Sometimes known as impulse.
154
Table 4.1
Generalized Momenta and Notation
ENERGY MEDIUM EFFORT SYMBOL MOMENTUM SYMBOL
General e p
Mechanical
translation
force F momentum or
impulse
p
Fixed-axis
mechanical rotation
torque or moment
t (or )
angular
momentum
q
Electrical voltage or
potential
difference
e (or v)
flux linkage
2

Magnetic magnetomotive
force
F not defined
Incompressible fluid
flow
pressure
difference
P pressure
momentum
I
Compressible fluid
flow
enthalpy h not defined
Thermal temperature
u (or T)
not defined
Note that generalized momentum is not defined in some media. The reason is because there is
no known kinetic energy storage phenomenon in those media. We will return to this point in a
subsequently.
2
Strictly speaking, this variable should be associated with displacement in the magnetic
medium.
155
Table 4.2
Generalized Displacements and Notation
ENERGY
MEDIUM
FLOW SYMBOL DISPLACEMENT SYMBOL
General f q
Mechanical
translation
speed or
velocity
v position or deflection x
Fixed-axis
mechanical
rotation
angular speed
or velocity
e (or O)
angle
u
Electrical current i charge q
Magnetic flux rate

flux

Incompressible
fluid flow
volumetric
flow rate
Q volume V
Compressible fluid
flow
mass flow rate m mass m
Thermal entropy flow
rate
s entropy s,
Ideal Energy-Storage Elements
We are now in a position to define ideal energy-storage elements. (Ideal in the sense of not
being contaminated by dissipation or any other non-storage phenomenon). The energy in a
system may be determined from the power flux across its boundaries
3
.
E =
]
(
t
o
t
Pdt + E(t
o
) (4.5)
Using equations 4.2 and 4.4 and the definition of effort and flow, this may be rewritten in the
following ways.
3
Once again we assume the convention that power is positive inwards.
156
E - E(t
o
) =
]
(
t
o
t
e f dt =
]
(
t
o
t
e dq =
]
(
t
o
t
f dp (4.6)
Now, if we encounter a phenomenon characterized by a relation which permits the integral in
either of the latter two forms to be evaluated so that it is not an explicit function of time, that
phenomenon may be regarded as energy storage. As you might expect, there are two
possibilities.
Generalized Capacitor
A ideal generalized capacitor is defined as any phenomenon characterized by an algebraic
relation (possibly nonlinear) for which effort is an integrable (single-valued) function of
displacement.
e = u(q) (4.7)
The algebraic function u() is the constitutive equation for this element. Note that although we
will use energy storage elements to describe dynamic behavior, this constitutive equation is a
static or memory-less function. The constitutive equation permits us to evaluate the generalized
potential energy, E
p
E
p
A_
_
]
( e dq =
]
( u(q) dq = E
p
(q) (4.8)
For this element, potential energy is a function of displacement alone. It is a generalized
potential energy storage element. The displacement, q, plays the same role as the specific
entropy and specific volume do for a pure thermodynamic substance: it is sufficient to define the
energy in the system. By convention we will define E
p
= 0 at q = 0 as shown in figure 4.1.
It will turn out to be important to distinguish potential energy from a related quantity,
(generalized) potential co-energy, E
*
p
, which is a function of effort.
E
*
p
A_
_
]
( q de = E
*
p
(e) (4.9)
Energy and co-energy are related by a Legendre transformation:
E
*
p
(e) = e q - E
p
(q) (4.10)
157
displacement, q
*
E
p
E
p
Figure 4.1: Sketch of a possible potential energy storage constitutive equation.
The relation between the two quantities is illustrated in figure 4.1. We will postpone further
discussion of co-energy until later.
Taken together, the definitions of generalized displacement and the constitutive equation for a
generalized capacitor specify a set of relations between flow, displacement and effort. These are
represented by the symbol shown in figure 4.2. Note that as this is a passive element, power
flow has been depicted as positive into it.
Figure 4.2: Bond graph symbol for an ideal capacitor.
An ideal capacitor may have a nonlinear constitutive relation. We may also define an ideal
linear capacitor, one with a linear constitutive relation.
e = q/C (4.11)
The parameter C is termed the capacitance of this ideal linear element. The potential energy
stored in an ideal linear capacitor is a quadratic function of displacement.
E
p
= q
2
/2C (4.12)
Aside: The reason for writing equation 4.11 with the proportionality constant, C, dividing the
argument is largely historical. The generalized capacitor is based on an electrical capacitor,
usually described by a linear relation between charge, q, (displacement) and voltage, e, (effort).
q = C e (4.13)
However, in the nonlinear case it may not always be possible to express charge as a function of
voltage; the fundamental definition is the one which permits the energy integral to be evaluated:
voltage as a function of charge as in equation 4.7. To be consistent the fundamental definition
yet acknowledge historical precedent, the parameter, C, has been written as in equation 4.11.
158
Another example of an ideal linear capacitor is the common model of a mechanical spring. If its
deflections are small and occur at modest rates of change, it may be well described by Hooke's
law:
F = k Ax (4.14)
where F is force, k is stiffness and Ax is deflection. Equating deflection, Ax, with displacement,
q, and force, F, with effort, e, this model provides a mechanical example of an ideal linear
potential energy storage element with capacitance 1/k. A bond graph symbol with the parameter
included is shown in figure 4.3.
Figure 4.3: Bond graph symbol for an ideal linear potential energy storage element with
capacitance 1/k.
For large length changes, the force-deflection relation for typical mechanical spring departs from
linear and the device provides a mechanical example of an ideal capacitor. In either case, the
important point is that the stored potential energy is a function only of the displacement, q.
Generalized Inertia
In an exactly dual manner, an ideal generalized inertia is defined as any phenomenon
characterized by an algebraic relation (possibly nonlinear) for which flow is an integrable
(single-valued) function of momentum.
f = +(p) (4.15)
The algebraic function +() is the constitutive equation for this element. Again, it is a static or
memory-less function. This constitutive equation permits us to evaluate the generalized kinetic
energy, E
k
.
E
k
A_
_
]
( f dp =
]
( +(p) dp = E
k
(p) (4.16)
For this element the potential energy is a function of the momentum alone. It is a generalized
kinetic energy storage element. By convention we will define E
k
= 0 at p = 0 as shown in figure
4.5.
159
momentum, p
*
E
k
E
k
Figure 4.4: Sketch of a possible kinetic energy storage constitutive equation.
It will turn out to be important to distinguish kinetic energy from a related quantity,
(generalized) kinetic co-energy, E
*
k
, which is a function of flow.
E
*
k
A_
_
]
( p df = E
*
k
(f) (4.17)
Once again, energy and co-energy are related by a Legendre transformation:
E
*
k
(f) = f p - E
k
(p) (4.18)
The relation between the two quantities is illustrated in figure 4.4.
Taken together, the definitions of generalized momentum and the constitutive equation for a
generalized inertia specify a set of relations between effort, momentum and flow. These may be
represented by the symbol shown in figure 4.5. Again, as this is a passive element, power flow
has been depicted as positive into it.
Figure 4.5: Bond graph symbol for an ideal inertia.
An ideal inertia may have a nonlinear constitutive relation. We may also define an ideal linear
inertia, one with a linear constitutive relation.
f = p/I (4.19)
The parameter I is termed the inertance of this ideal linear element. The kinetic energy stored in
an ideal linear inertia is a quadratic function of momentum.
E
k
= p
2
/2I (4.20)
160
A common mechanical example of an ideal linear kinetic energy storage element is a body in
motion. If the deflections of the body are small enough that it may be regarded as rigid, it may
be characterized by a linear relation between velocity and momentum.
v = p/m (4.21)
where v is velocity (flow), p is momentum and m is mass (inertance). A bond graph symbol with
the parameter included is shown in figure 4.6.
Figure 4.6: Bond graph symbol for an ideal linear kinetic energy storage element with
inertance m.
Once again, the reason for writing equation 4.19 with the proportionality constant, I, dividing the
argument is to be consistent with historical precedent. Equation 4.21 is frequently written with
momentum as function of velocity as follows.
p = m v (4.22)
However, in a general, nonlinear case it may not be possible to express generalized momentum
as a function of effort and the fundamental definition is the one which permits the energy
integral to be evaluated: flow as a function of momentum as in equation 4.15.
Another practice sustained by historical precedent is to equate kinetic energy with the integral of
equation 4.22 with respect to velocity. This is regrettable as this quantity is properly termed
kinetic co-energy.
E
*
k
= m v
2
/2 (4.23)
At this point we do no more than note the point of confusion and proceed. The importance of the
distinction between energy and co-energy will be discussed in depth later.
Modulated Energy Storage is Prohibited
Previously we encountered the use of modulated power sources to describe how a control system
might influence the energy supplied to or removed from a system. When we consider energy-
storage elements, an important restriction must be emphasized: modulation of energy storage
elements is prohibited.
The reason for this restriction is that a modulated energy-storage element would mean that the
total energy in a system would be a function of the modulating input or set of inputs.
Consequently, the total energy in the system would not be equal to the net power flow in across
the system boundaries.. The system equations would not be guaranteed to satisfy the first law of
thermodynamics; but that was to be the objective of our energy-based formalism.
161
The basic idea behind lumped-parameter modelling is to identify different phenomena with
different elements. To be consistent with the energy-based formalism, the function of adding or
removing energy from a system must be represented only by elements defined for that purpose:
power sources and dissipators. The point to remember is: power sources and dissipators may be
modulated; energy-storage elements may not.
162
Basic Bond Graph Notation
power bond energetic interaction e.g. A B
between (sub)systems
S
e
effort source boundary condition inpedendent
variable
S
f
flow source boundary condition inpedendent
variable
R (generalized) dissipator irreversible energy removal
C (generalized) capacitor (generalized) potential
energy storage
I (generalized) inertia (generalized) kinetic energy
storage
0 zero junction (generalized) continuity
| equation
1 one junction (generalized) compatibility
| equation
Fundamental quantities and relations
P power
e effort
f flow
P = ef
E energy
E = (edq = (fdp
] ]
p (generalized) momentum
q (generalized) displacement
163
Examples: First-Order Systems
Energy storage elements provide the basis of the state equations we will derive to describe the
dynamic processes occurring in a system. Of course, an energy storage element does not by
itself define a dynamic process it needs an input. That input will arise from the interaction
with other system components as we will see in the following examples.
A simple fluid system
The sketch in figure 4.7 depicts an open-topped cylindrical container of water with a section of
circular pipe connected at the bottom through which may leave. We wish to predict the time-
course of the volume of fluid as the container empties.
circular pipe
diameter d, length L
depth, h
water
cylindrical
container
area A
c
Figure 4.7: A simple fluid system.
At any cross section of interest, the instantaneous power transmitted is the product of the flow
velocity and the force exerted on the cross section
1
. The force, F, exerted on the cross section is
the product of pressure
2
, P
g
, and area, A
F = P
g
A (4.24)
The volumetric flow rate, Q, is the product of flow velocity, v, and area.
Q = v A (4.25)
Thus the power transmitted is the product of pressure and volumetric flow rate.
1
Strictly speaking, we may only refer to an unique flow velocity, force, pressure, etc. if the
cross-section has infinitesimal area. If the area is finite, each of the quantities represents an
average over the cross section.
2
For this example, it is assumed that we are dealing with gage pressures, i.e. that atmospheric
pressure is taken to be zero, hence the subscript g.
164
P = F v = P
g
Q (4.26)
Within the cylindrical container, the pressure at any depth, h, (e.g. the depth of the entry to the
pipe) is due to the weight, W, of the water above.
P
c
= W/A
c
= ghA
c
/A
c
= gh (4.27)
where subscript c denotes the container, is the density of water, g is gravitational acceleration
and A
c
is the area of the container. We may rewrite this relation in terms of the volume of
water, V
c
, above a given depth.
P
c
=
g
A
c
V
c
(4.28)
Consequently, we may describe the container as an ideal linear capacitor, characterized by a
linear relation between pressure (effort) and volume (displacement). The capacitance, C, is
determined by the geometric and material properties of the container, the fluid, etc.
C =
A
c
g
(4.29)
The energy stored in the capacitor is determined by the displacement variable, V
c
, and the
system parameters.
E
c
=
g
2A
c
V
2
c
(4.30)
The flow rate through the pipe is determined by the pressure difference between its ends. If we
assume the pipe is horizontal with a constant circular cross-section and the flow is laminar, we
may describe the pressure/flow-rate relation using the Hagen-Poiseuille law.
P
p
=
128L
ad
4
Q
p
(4.31)
where P is the gage pressure at the end of the pipe next to the container, subscript p denotes the
pipe, L is the length of the pipe, d is its diameter and is the absolute viscosity of water. This is
the equation of an ideal linear resistor with resistance, R, determined by the geometric and
material properties of the pipe, the fluid, etc.
R =
128L
ad
4
(4.32)
The remaining idealized element in this system is the junction between the pipe and the
container. If we assume the pressure at the entry to the pipe is the same as the pressure in the
container at that depth, that common pressure defines a zero junction and a bond graph is as
shown in figure 4.8.
165
0
R
C
g
A
c
128L
d
4
P = P
c
p
: :
Figure 4.8: Bond graph of the simple fluid system.
Given this graph it is straightforward to determine state equations. We are interested in the rate
of the change of the volume in the container which (from the definition of a generalized
displacement) is due to the flow rate into it.
dV
c
/dt = Q
c
(4.33)
That flow rate is determined from the continuity equation of the zero junction. Reading signs
from the half arrows on the graph:
Q
c
= - Q
p
(4.34)
The flow rate in the pipe is determined by inverting equation 4.30.
Q
p
= P
p
/R (4.35)
From the zero junction definition, the pipe pressure is the same as the container pressure.
P
p
= P
c
(4.36)
The container pressure is determined from equation 4.28. Successively substituting (4.28 into
4.35 into 4.34 into 4.33 into 4.32) yields a first-order linear state equation.
dV
c
/dt = -V
c
/RC (4.37)
Note that this simple system has one energy-storage element and is characterized by a first-order
state equation. The state variable, V
c
, is directly related to the stored energy. This simple state
equation may readily be integrated.
]
(
t
o
t
dV
c
/V
c
=
]
(
t
o
t
-dt/RC (4.38)
ln{V
c
(t)/V
c
(t
o
)} = (t
o
- t)/RC (4.39)
V
c
(t) = V
c
(t
o
)e
(t
o
- t)/RC
(4.40)
Note that to predict the behavior of this first-order system we require one initial condition which
is related to the energy stored at time t
o
.
166
A simple electrical system
Figure 4.9 shows a diagram of a simple electrical circuit consisting of a capacitor connected to a
resistor.
C R
e
0
R
C
e
Bond graph
Electrical circuit diagram
Figure 4.9: A simple electrical system and a corresponding bond graph.
Assuming a common voltage drop across the resistor and capacitor, a corresponding bond graph
is also shown, which is the same as the bond graph of the previous example. This is no accident:
the two systems exhibit analogous energetic behavior.
Assuming an ideal linear capacitor, its charge, q, is proportional to the voltage difference, e,
across it.
e = q/C (4.41)
where the capacitance, C, is a physical property of the device. The rate of change of charge is
the current, i
C
, flowing into the capacitor.
dq/dt = i
C
(4.42)
Assuming an ideal linear electrical resistor, the current through it, i
R
, is proportional to the
voltage difference across it, e, as determined by Ohm's law
i
R
= e/R (4.43)
where the resistance, R, is a physical property of the device.
A zero junction describes the interaction between these elements. Its associated flow continuity
equation requires that the current into the capacitor is the negative of the current into the resistor.
i
C
= -i
R
(4.44)
One reasonable choice of state variable is the charge on the capacitor. A first-order state
equation is obtained by substitution (4.41 into 4.43 into 4.44 into 4.42).
dq/dt = -e/RC (4.45)
167
Assuming the obvious analogy between like quantities (voltage drop and pressure difference are
both effort variables, charge and volume are both displacement variables, etc.) equations 4.45
and 4.37 are clearly similar; the time response for discharging the capacitor is of the same form
as the time response for emptying the fluid container and will be described by an equation
similar to equation 4.40.
The analogous dynamic behavior of these two systems is represented by their similar bond
graphs. In this regard, bond graphs provide a unified notation for depicting different physical
systems.
A nonlinear fluid system
In this example consider a system similar to that of figure 4.7 but instead of a container with a
constant cross-sectional area, assume the radius, r, of the circular cross-section varies with height
as follows.
r = ah
n
n > 0 (4.46)
where n is a positive constant and a is a scaling factor. The volume varies with depth as follows.
V
c
=
]
(
o
h
aa
2
h
2n
dh =
aa
2
h
2n+1
2n+1
(4.47)
Pressure varies with depth as before (equation 4.27) and the relation between pressure and
volume may be obtained by substitution.
P
c
= g
(
(
2n+1
aa
2
V
c
1
2n+1
(4.48)
Whereas in the previous example the container was described as an ideal linear capacitor, in this
example it may be described as an ideal capacitor.
The same bond graph (figure 4.8 or 4.9) may be used to represent this system too, the only
difference being that the capacitor is characterized by a nonlinear constitutive equation. Indeed,
the special case n = 0 corresponds to a container with a constant cross-sectional area; substituting
n = 0, equation 4.48 reduces to equation 4.28. A state equation may be obtained following the
same procedure as above but using equation 4.48 in place of equation 4.28.
dV
c
dt
= -
g
R

(
(
2n+1
aa
2
V
c
1
2n+1
(4.49)
As before, this system contains a single (nonlinear) energy-storage element and is characterized
by a first-order (nonlinear) state equation. Once again, it is again straightforward to integrate.
The case n = 0 has been considered above. For n > 0 the result is as follows.
168
]
(
(
t
o
t
V
c
-
1
2n+1
dV
c
=
]
(
(
(
t
o
t
-
g
R

(
(
2n+1
aa
2
1
2n+1
dt (4.50)
V
c
(t)
2n
2n+1
= V
c
(t
o
)
2n
2n+1
+
2n
2n+1
g
R

(
(
2n+1
aa
2
1
2n+1
(t
o
- t) (4.51)
V
c
(t) =
V
c
(t
o
)
2n
2n+1
+
2n
2n+1
g
R

(
(
2n+1
aa
2
1
2n+1
(t
o
- t)
2n+1
2n
(4.52)
Though rather clumsy-looking, this equation, valid for V
c
_ 0, yields the simple behavior shown
in figure 4.10.
Comparing these examples, some points should be noted:
A bond graph such as that of figure 4.8 or 4.9 is an abstract representation of a family of systems.
Until the constitutive equations of the elements have been specified, state equations cannot be
determined. However, all members of the family of systems represented by the bond graph share
certain qualitative features. In the above examples, first-order state equations are sufficient to
describe energetic transactions within the system; that is a consequence of the single energy-
storage element. The behavior of the systems are qualitatively similar; all exhibit a non-
oscillating decay to the equilibrium state V
c
= 0 or q = 0.
169
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Volume
time
n=1
n=1/2
n=1/8
n=1/4
n=1/16
n=0
Volume vs. time as the
container empties for
several values of the
exponent n.
Figure 4.10: Time course of emptying for different container shapes. Fixed parameters have
been assigned arbitrary values.
It is a common misconception to regard linear and nonlinear systems as radically different. A
more enlightened perspective is to consider linear systems as special cases with certain
interesting properties. In the examples above all of the nonlinear systems make the reasonable
prediction that a finite time is required to empty the container. From equation 4.51 the time to
empty is given by
t
empty
= t
o
+
V
c
(t
o
)
2n
2n+1
2n
2n+1
g
R

(
(
2n+1
aa
2
1
2n+1
n = 0 (4.53)
In contrast, it can be seen from equation 4.39 that the time for the linear system to empty is
infinite. But as the exponent n becomes small the time to empty becomes large and the same
result could be obtained by taking n to zero in the limit in equation 4.53.
170
lim t
empty
=
n C 0
(4.54)
Another interesting property of the linear case is that the responses starting from different initial
conditions are all of the same form; that is, they are identical when multiplied by a scaling factor
inversely proportional to the initial condition. This is clearly not the case for the nonlinear
systems: The value of a response at each point in time can be regarded as the initial condition
for that response for all future time; yet for sufficiently large times the nonlinear responses are
identically zero and therefore cannot be scaled to match any non-zero response.
171
Real Power Sources: Static Characteristics
The primitive elements which have been defined so far are intended to describe aspects of
energetic behavior and they can be used in the construction of detailed models of specific
physical systems. We should also be able to use energetic considerations to draw some general
conclusions about behavior common to all physical systems; after all, that is one of the
advertised benefits of an energy-based formalism. In this section we will briefly consider some
physical limitations of real power sources (or sinks). Real power sources will exhibit dynamic
behavior, but we will confine ourselves for the present to their static characteristics.
As defined above, an ideal effort source could sustain that effort even if its output flow grew to
infinity and therefore, in principle, it could deliver infinite power; the same is true for an ideal
flow source. This is unlikely, to say the least; the amount of power which can be delivered by
any real device built to date is limited. The static behavior of any power supply may be
characterized as a relation between effort and flow. The maximum power output defines a
hyperbola on the effort-flow plane
e f = P
max
(3.33)
where P
max
is the maximum power output. If its output power is limited, the effort-flow
relation of the power source must remain within this bound, therefore the effort developed by
any real power supply must decline as the output flow becomes sufficiently large. Conversely,
the flow produced must decline as the effort required grows sufficiently large.
e f = P
flow, f
Maximum power limit
Maximum effort
max
limit
Effort-flow
Maximum flow
relation must
limit
remain in this
region.
Figure 3.33: Common limitations of real power sources.
The limitation on power output does not preclude infinite effort at zero flow, nor infinite flow at
zero effort, but no real device can generate infinite flow or withstand infinite effort. The
maximum effort limit is a horizontal line on the effort-flow plane; the maximum flow limit is a
172
vertical line. All of these limitations are illustrated in figure 3.33 for the first quadrant
1
. The
effort-flow relation of the power source must stay below and to the left of these bounds. Similar
limitations apply in the other quadrants.
Reasoning this way, we may conclude that all real power sources must exhibit some dependence
of effort on flow (or vice versa). A relation between effort and flow suggests a dissipator and if
the effort-flow relation is limited as described above, it is possible to describe this aspect of any
real power source by a combination of an ideal source and a dissipator as we did for the battery.
There are two possible forms. One is a combination of an ideal effort source and a resistor
coupled to a one junction as shown in figure 3.34. This is a generalized Thevenin equivalent
network and the figure is a bond graph version of figure 3.3.
S
e
R
Figure 3.34: Bond graph symbol for a Thevenin equivalent network.
In this bond graph, the half arrows indicate that positive power is directed out of the ideal source
element and also out of the network; a reasonable choice for a model of a power source.
Following convention, positive power is directed into the dissipator.
The other form uses an ideal flow source and an ideal resistor coupled to a zero junction as
shown in figure 3.35. This is a generalized Norton equivalent network.
0
S
f
R
Figure 3.35: Bond graph symbol for a Norton equivalent network.
These two networks provide versatile, general-purpose models for describing the static
characteristics of real power sources. Note that if the constitutive equation of the dissipator in
these models can be inverted, the two networks are duals and may be interchanged freely.
1
Note that in this figure we have assumed that positive power flow is directed out of the device.
173
Hamiltonian Systems: Ideal Oscillators
Consider a system composed of an ideal capacitor and an ideal inertia interacting through a
common-flow junction as shown in figure 4.11. A physical example would be a rigid body
connected to a spring (so that a point on the spring and a point on the rigid body move at the
same velocity) as shown schematically in figure 4.11.
rigid
body
elastic
spring
rigid support
C
Bond graph velocity
(v or f)
displacement
(x or q)
1
f
Mechanical schematic
Figure 4.11: Mechanical schematic and corresponding bond graph for an ideal oscillator.
In general notation, the constitutive equations for the capacitor and the inertia (subscripts c and i,
respectively) are
e
c
= u(q
c
) (4.55)
f
i
= +(p
i
) (4.56)
The differential form of their defining equations suggest that displacement and momentum may
be used as state variables.
dq
c
/dt = f
c
(4.57)
dp
i
/dt = e
i
(4.58)
By definition, the flows of the elements connected to the one-junction are identical.
f
c
= f
i
(4.59)
From the associated compatibility equation the efforts sum to zero.
e
i
= -e
c
(4.60)
174
By direct substitution (4.56 into 4.59 into 4.57 and 4.55 into 4.60 into 4.58) these equations may
be assembled into a pair of first-order differential equations.
dq
c
/dt = +(p
i
) (4.61)
dp
i
/dt = -u(q
c
) (4.62)
These are state equations for this system. The reason that q
c
and p
i
are a good choice for state
variables (other choices are possible see below) is that they define the energetic state of the
system. In fact, any set of numbers from which we can define the energetic state of a system will
suffice as state variables. And the reason for that is because the dynamic behavior of a physical
system is due to the movement of energy within the system.
It is informative to rewrite these equations in terms of the total energy, H(p
i
,q
c
), of the system.
H(p
i
,q
c
)
A_
_ E
k
(p
i
) + E
p
(q
c
) (4.63)
From the definitions of potential and kinetic energy we may write the constitutive equations of
the capacitor and inertia as partial derivatives of the total energy.
+(p
i
) = cH/cp
i
(4.64)
u(q
c
) = cH/cq
c
(4.65)
The state equations may then be written as follows.
dq
c
/dt = cH/cp
i
(4.66)
dp
i
/dt = -cH/cq
c
(4.67)
This is the Hamiltonian form of the state equations (named after Sir William Rowan Hamilton,
the illustrious Dublinman who first formulated it) and the generalized displacement and
momentum are known as Hamiltonian or energy state variables. Hamiltonian systems play a
fundamental role in modern physics, and have application at all scales, from celestial mechanics
to quantum mechanics.
A more compact way to write these equations, which leads to further physical insight, is to write
the displacement, q
c
, and the momentum, p
i
, as a state vector, r.
r
A_
_
(
(
(
q
c
p
i
(4.68)
The state equations then become:
175
d
dt
(
(
(
q
c
p
i
=
(
(
(
0 1
-1 0
(
(
cH/cq
c
cH/cp
i
(4.69)
Alternatively:
'
r = -J cH/cr (4.70)
where
J
A_
_
(
(
(
0 -1
1 0
(4.71)
This form of the equations is known as the symplectic form. The symplectic matrix, J, has an
interesting property:
J
-1
= J
t
= -J (4.72)
It is therefore one of the square roots of minus unity.
J J = -1 (4.73)
Experience with simple differential equations will have taught you to associate the square root of
minus one with oscillatory behavior, and this case is no exception. The oscillatory character of
Hamiltonian systems is easiest to see if we consider a linear case by replacing the pure capacitor
and inertia with an ideal capacitor and inertia, for example a Hookeian spring of stiffness k and a
Newtonian rigid body of mass m (e.g. as shown in figure 4.11). The state equations are then as
follows:
d
dt
(
(
(
q
c
p
i
=
(
(
(
0 1/m
-k 0

(
(
(
q
c
p
i
(4.74)
Double-differentiating q
c
and substituting for p
i
yields the familiar second-order differential
equation of a simple harmonic oscillator.
''
q
c
+ e
2
q
c
= 0 (4.75)
where
e
A_
_ k/m (4.76)
176
The same result can be derived directly from the first-order symplectic form, and that yields
some further insight and understanding of how the oscillatory behavior arises. To simplify
matters
1
, let us suppose that k = m = 1. The state equations become:
d
dt
(
(
(
q
c
p
i
=
(
(
(
0 1
-1 0

(
(
(
q
c
p
i
(4.77)
or
'
r = -J r (4.78)
Equation 4.77 shows that the rate of change of the state vector is always orthogonal to the state
vector their inner product is zero.
r
t '
r = -r
t
J r = 0 (4.79)
Now remember that the behavior of a state-determined system can be represented geometrically
as the motion of a point in an abstract state-space. Geometrically, the tangent to the path of the
point representing the state is always at right angles to the line joining that point to the origin, as
shown in figure 4.12. If the rate of change is non-zero, that is only possible if the state trajectory
is a circle.
A state trajectory which closes on itself represents periodic behavior. A circular state trajectory
means that the magnitude of the state vector is constant. From equation 4.77 the magnitude of
the rate vector is proportional to the magnitude of the state vector, hence the circle is traversed at
constant speed. As a result, q
c
executes a sinusoidal oscillation.
1
More fundamentally, we could reason that because the fundamental form of the system
behavior should not depend on an arbitrary choice of the units for time or space, we may choose
them so that k and m are numerically equal to unity.
177
r
p
i
q
c
'
r
state
vector
rate vector
state
trajectory
Figure 4.12: Geometric representation of the behavior of equation 4.77. The orthogonality of
the rate and state vectors means that the state trajectory must be circular.
This result can be derived formally by integrating the matrix differential equations but the value
of these manipulations is to illustrate that the oscillatory behavior comes from the symplectic
form of the equations, rather than from any particular choice of parameter values. Consequently,
we expect oscillatory behavior in a nonlinear system with this symplectic form, and this is in fact
the case. Indeed, one of the reasons why Hamiltonian systems are so ubiquitous is because they
are fundamentally oscillators, and oscillatory phenomena are found at all scales, from stellar
objects to subatomic particles.
If we return to the derivation of the equations, we can see that in this case the symplectic form is
due to the interaction between two energy storage elements of dual type through a one-junction.
As we will see shortly, interaction between two energy storage elements of the same type
through a junction structure composed of zero- and one-junctions does not lead to the symplectic
structure and therefore does not lead to oscillatory behavior.
178
Hamiltonian Systems: Ideal Oscillators
Consider a system composed of an ideal capacitor and an ideal inertia interacting through a
common-flow junction as shown in figure 4.11. A physical example would be a rigid body
connected to a spring (so that a point on the spring and a point on the rigid body move at the
same velocity) as shown schematically in figure 4.11.
rigid
body
elastic
spring
rigid support
C
Bond graph velocity
(v or f)
displacement
(x or q)
1
f
Mechanical schematic
Figure 4.11: Mechanical schematic and corresponding bond graph for an ideal oscillator.
In general notation, the constitutive equations for the capacitor and the inertia (subscripts c and i,
respectively) are
e
c
= u(q
c
) (4.55)
f
i
= +(p
i
) (4.56)
The differential form of their defining equations suggest that displacement and momentum may
be used as state variables.
dq
c
/dt = f
c
(4.57)
dp
i
/dt = e
i
(4.58)
By definition, the flows of the elements connected to the one-junction are identical.
f
c
= f
i
(4.59)
From the associated compatibility equation the efforts sum to zero.
e
i
= -e
c
(4.60)
179
By direct substitution (4.56 into 4.59 into 4.57 and 4.55 into 4.60 into 4.58) these equations may
be assembled into a pair of first-order differential equations.
dq
c
/dt = +(p
i
) (4.61)
dp
i
/dt = -u(q
c
) (4.62)
These are state equations for this system. The reason that q
c
and p
i
are a good choice for state
variables (other choices are possible see below) is that they define the energetic state of the
system. In fact, any set of numbers from which we can define the energetic state of a system will
suffice as state variables. And the reason for that is because the dynamic behavior of a physical
system is due to the movement of energy within the system.
It is informative to rewrite these equations in terms of the total energy, H(p
i
,q
c
), of the system.
H(p
i
,q
c
)
A_
_ E
k
(p
i
) + E
p
(q
c
) (4.63)
From the definitions of potential and kinetic energy we may write the constitutive equations of
the capacitor and inertia as partial derivatives of the total energy.
+(p
i
) = cH/cp
i
(4.64)
u(q
c
) = cH/cq
c
(4.65)
The state equations may then be written as follows.
dq
c
/dt = cH/cp
i
(4.66)
dp
i
/dt = -cH/cq
c
(4.67)
This is the Hamiltonian form of the state equations (named after Sir William Rowan Hamilton,
the illustrious Dublinman who first formulated it) and the generalized displacement and
momentum are known as Hamiltonian or energy state variables. Hamiltonian systems play a
fundamental role in modern physics, and have application at all scales, from celestial mechanics
to quantum mechanics.
A more compact way to write these equations, which leads to further physical insight, is to write
the displacement, q
c
, and the momentum, p
i
, as a state vector, r.
r
A_
_
(
(
(
q
c
p
i
(4.68)
The state equations then become:
180
d
dt
(
(
(
q
c
p
i
=
(
(
(
0 1
-1 0
(
(
cH/cq
c
cH/cp
i
(4.69)
Alternatively:
'
r = -J cH/cr (4.70)
where
J
A_
_
(
(
(
0 -1
1 0
(4.71)
This form of the equations is known as the symplectic form. The symplectic matrix, J, has an
interesting property:
J
-1
= J
t
= -J (4.72)
It is therefore one of the square roots of minus unity.
J J = -1 (4.73)
Experience with simple differential equations will have taught you to associate the square root of
minus one with oscillatory behavior, and this case is no exception. The oscillatory character of
Hamiltonian systems is easiest to see if we consider a linear case by replacing the pure capacitor
and inertia with an ideal capacitor and inertia, for example a Hookeian spring of stiffness k and a
Newtonian rigid body of mass m (e.g. as shown in figure 4.11). The state equations are then as
follows:
d
dt
(
(
(
q
c
p
i
=
(
(
(
0 1/m
-k 0

(
(
(
q
c
p
i
(4.74)
Double-differentiating q
c
and substituting for p
i
yields the familiar second-order differential
equation of a simple harmonic oscillator.
''
q
c
+ e
2
q
c
= 0 (4.75)
where
e
A_
_ k/m (4.76)
181
The same result can be derived directly from the first-order symplectic form, and that yields
some further insight and understanding of how the oscillatory behavior arises. To simplify
matters
1
, let us suppose that k = m = 1. The state equations become:
d
dt
(
(
(
q
c
p
i
=
(
(
(
0 1
-1 0

(
(
(
q
c
p
i
(4.77)
or
'
r = -J r (4.78)
Equation 4.77 shows that the rate of change of the state vector is always orthogonal to the state
vector their inner product is zero.
r
t '
r = -r
t
J r = 0 (4.79)
Now remember that the behavior of a state-determined system can be represented geometrically
as the motion of a point in an abstract state-space. Geometrically, the tangent to the path of the
point representing the state is always at right angles to the line joining that point to the origin, as
shown in figure 4.12. If the rate of change is non-zero, that is only possible if the state trajectory
is a circle.
A state trajectory which closes on itself represents periodic behavior. A circular state trajectory
means that the magnitude of the state vector is constant. From equation 4.77 the magnitude of
the rate vector is proportional to the magnitude of the state vector, hence the circle is traversed at
constant speed. As a result, q
c
executes a sinusoidal oscillation.
1
More fundamentally, we could reason that because the fundamental form of the system
behavior should not depend on an arbitrary choice of the units for time or space, we may choose
them so that k and m are numerically equal to unity.
182
r
p
i
q
c
'
r
state
vector
rate vector
state
trajectory
Figure 4.12: Geometric representation of the behavior of equation 4.77. The orthogonality of
the rate and state vectors means that the state trajectory must be circular.
This result can be derived formally by integrating the matrix differential equations but the value
of these manipulations is to illustrate that the oscillatory behavior comes from the symplectic
form of the equations, rather than from any particular choice of parameter values. Consequently,
we expect oscillatory behavior in a nonlinear system with this symplectic form, and this is in fact
the case. Indeed, one of the reasons why Hamiltonian systems are so ubiquitous is because they
are fundamentally oscillators, and oscillatory phenomena are found at all scales, from stellar
objects to subatomic particles.
If we return to the derivation of the equations, we can see that in this case the symplectic form is
due to the interaction between two energy storage elements of dual type through a one-junction.
As we will see shortly, interaction between two energy storage elements of the same type
through a junction structure composed of zero- and one-junctions does not lead to the symplectic
structure and therefore does not lead to oscillatory behavior.
183
Choice of State Variables
It must be kept in mind that there is no unique set of state variables (though as we will see, in
many cases, some choices are clearly superior to others).
Referring again to the system depicted in figure 4.11 and assuming an ideal (linear) capacitor
and inertia, their constitutive equations may be differentiated as follows.
F
c
= k x
c
(4.80)
dF
c
/dt = k v
c
(4.81)
v
i
= p
i
/m (4.82)
dv
i
/dt = F
i
/m (4.83)
This suggests that F
c
and v
i
would be a reasonable choice for state variables. Equations 4.81 and
4.83 may be combined using the constitutive equations of the one-junction.
v
c
= v
i
(4.84)
F
i
= -F
c
(4.85)
The following state equations result.
F
c (
(
(
F
c 0 k
(
(

(
(
d
(4.86) =
dt
(
-1/m 0 v
i
v
i
At first glance (comparing equation 4.86 with equation 4.74) it might appear that because the
system matrix in these equations has changed, we have a different model. Not so; we have
merely expressed it in different coordinates. Double differentiating the velocity and substituting
'
for F
c
again yields the equation of a simple harmonic oscillator.
''
v
i
+ e
2
v
i
= 0 (4.87)
where e = k/m as before.
In fact we can readily recover the original state equations. Denote the new state variables as r*.
(
(
(
F
c
_
r*
A
(4.88) _
v
i
Write the state equations in matrix notation.
r
'
* = A r* (4.89)
184
where A is the system matrix of equation 4.86. The new state variables, r*, are related to the
original state variables, r, by a constant, linear transformation matrix M.
F
c k 0
(
(

q
c(
(
(
(
(
(4.90) =
(
0 1/m v
i
p
i
or
r* = M r (4.91)
The transformation matrix is non-singular, so the inverse transformation exists.
F
c(
(
(
1/k 0
(
(

q
c

(
(
(4.92) =
(
0 m p
i
v
i
or
r = M
-1
r* (4.93)
Differentiating 4.93 and substituting equations 4.89 and 4.91 we recover the equation 4.74, our
original state equations.
'
r = M
-1
r
'
* = M
-1
A r* = M
-1
A M r (4.94)
1/k 0 0 k k 0
q
c
(
(
(
(
(
(
q
c
(
(

(
(
(
(
d
(4.95) =
dt
0 m -1/m 0 0 1/m p
i
p
i
0 1/m
(
(
(
q
c
(
(
q
c
(
(
d
(4.74) =
dt
-k 0 p
i
p
i
If we regard state variables as coordinates of a state space, equation 4.91 describes a (non-
singular) transformation of coordinates. But as there are an infinity of non-singular constant
transformation matrices, we can see why state variables, which uniquely characterize a system's
state, are not themselves unique the infinity of equivalent sets of state variables corresponds
to the infinity of possible choices of coordinates. Clearly the underlying physical behavior
should remain the same under a change of coordinates, though the details of the description will
be different in the different coordinate frames, hence the difference between equations 4.74 and
4.86. On the other hand, experience with choice of coordinates in mathematical analysis should
have taught you that for certain purposes, some coordinate choices are more convenient than
others.
Because effort (force) and flow (velocity) together define the power into an element, this second
choice is known as power state variables. It is a particularly convenient choice if the elements
185
are linear, and is widely used in electric circuit theory (where the power variables are voltage
and current); hence these variables are sometimes called circuit state variables.
Whichever state variables we choose, the minimum number necessary to describe energetic
transactions in this system is two. Physically, there are two independent energy storing elements
in the system, and the dynamic process arises from exchange of energy between them.
Therefore, to define the energetic state of the system at any moment we require two quantities.
Another way to think of this: each of the two energy storage elements embodies a differential
equation which is integrated to determine the solution, the state trajectory of the system. Two
initial conditions will be needed to determine the constants of integration, and this is another
way of saying that we require two numbers to determine the energetic state of the system.
186
Example: A Non-Oscillating Second-Order System
The sketch below depicts two plates which can slide relative to one another in the horizontal
direction.
Figure 4.21: Mechanical system with two movable plates
Assuming the plates are rigid and that we only care about horizontal motion, each plate may be
represented by an ideal inertia. We will assume two sources of energy dissipation. One is the
friction due to the motion of the top plate relative to the bottom; the other is the friction due to
the motion of the bottom plate relative to ground. For simplicity, we will assume linear friction
in both cases (although Coulomb friction would probably be a better model).
Identifying the Junction Structure
Having identified the energic elements in the system, it remains to establish the junction
structure. This step usually requires the most care. To describe the system we will need to know
the motion of each plate, so it is reasonable to associate a one-junction with each inertia; the flow
on each one-junction represents the velocity of the corresponding plate.
Figure 4.22: Bond graph elements representing the two plates.
We can proceed from here in two ways. Perhaps the most fundamental way is to look for efforts
or flows that are common to two or more elements, because they will define zero- (common
effort) and one- (common flow) junctions respectively. The friction between the lower plate and
ground is due to the velocity of the lower plate, so that dissipative element is attached to the one-
junction on the corresponding inertia.
187
Figure 4.23: Bond graph of the two plates with friction due to motion relative to ground.
Any force exerted on the upper plate is generated by friction between the two plates. Thus the
other dissipative element is attached to a zero junction between the two one-junctions and the
complete system bond graph is as shown below.
Figure 4.24: Bond graph of the two plates with both sources of friction.
Two of the bonds in this graph are without half arrows. How can they be chosen? Well, the
friction between the two plates acts to transmit energy between the plates. Therefore, it makes
reasonable sense to depict power as flowing from one plate to the other, for example, as follows.
Figure 4.25: Complete system bond graph with power signs included.
But we are being a little glib; the choice of power signs bears closer inspection. Suppose, for
some reason, we elected to depict the half arrows as follows:
1
0
1
R R

: b
1
m
1
b
2
m
2
:
: :
v
1
v
2
v
3
188
Figure 4.26: Complete system bond with alternative power signs.
At first glance, this would seem to be equally reasonable. However, a little care is required. The
force generated by the friction between the plates is a function of the relative velocity of the
plates. But from the signs in the bond graph above, the continuity equation associated with the
zero junction would be:
v
1
+ v
2
+ v
3
= 0 (4.103)
This would mean that the relative velocity would be expressed as the negative sum of the plate
velocities.
v
3
= -(v
1
+ v
2
) (4.104)
Can that be right? Strictly speaking, it is not wrong. However, it would require that the
direction of positive velocity for one plate be defined as opposite to that for the other plate. The
direction of positive velocity is purely a matter of convention and there is nothing physically
wrong with doing this, but it would be perverse. Worse, it would be confusing.
A much saner approach is to use the same convention for both inertias and use a junction
structure which represents the relative velocity correctly. Write down a one-junction for each of
the three distinct velocities, v
1
, v
2
, and the relative velocity, v
3
, then add a zero-junction between
them. The signs on the bonds should be chosen so that the continuity equation yields the
required definition of relative velocity.
Figure 4.27: A junction structure which identifies v
3
as a relative velocity.
Taking the signs into account, the continuity equation for the zero junction is
v
1
- v
2
- v
3
= 0 (4.105)
Now the relative velocity is the difference of the plate velocities, which makes much better
sense.
v
3
= v
1
- v
2
(4.106)
Following this line of reasoning, a clumsier but surer procedure for identifying the junction
structure is as follows. Identify all relevant velocities (including the velocity of the inertial
reference frame, v
o
) and assign a one-junction to each. Next identify all relevant relative
velocities and assign a separate one-junction to each of these. Now write an appropriately signed
zero-junction which will correctly represent each relative velocity as follows.
189
Figure 4.28: A junction structure for the two-plate system identifying all relevant velocities.
Now it easy to see that the dissipative elements are associated with the relative velocities. The
ground is always at zero velocity, (by assumption) independent of what the system does, so
attach a flow source to the corresponding one-junction. The inertias go on the remaining one-
junctions as before.
Figure 4.25: Complete system bond graph with reference velocity explicitly represented.
Next simplify this graph by eliminating superfluous elements. If the effort or flow on any bond
or element is identically zero, the associated power flow is always zero and the bond may be
removed. This eliminates the flow source, the two immediately adjacent bonds and the one-
junction between them. If any two-port one- or zero-junction has one half arrow pointing
inwards and one pointing outwards, it may be eliminated because both the efforts and the flows
on the two bonds are identical. The resulting graph is shown below. It is essentially the same as
in figure 4.25.
190
Figure 4.26: Complete system bond graph after superfluous elements are removed.
Aside: There is nothing incorrect about including superfluous elements, but it usually simplifies
matters to eliminate them; however, sometimes it is helpful to include a superfluous junction to
identify a particular effort or flow.
The important point to note here is that choosing the power signs (orienting the half arrows) is an
important part of identifying the junction structure.
Assigning Causality
Next we add causal strokes to the graph. Remember, we want as many energy storage elements
as possible to have integral causality. We begin with the inertia of the lower plate. In integral
causality, it accepts a force input an determines a velocity as its output. That velocity is the
input to the one-junction and therefore determines the causal assignment on the other two bonds
of that one-junction as shown in figure 4.27.
Figure 4.27: Partial causal assignment for the two-plate system.
This is an example of a strong causal assignment. The same velocity is input to the friction
between the lower plate and ground (b
1
) and the zero-junction. But the velocity input to the
zero-junction does not uniquely determine the outputs of the zero-junction, and the consequences
of this choice of integral causality on the lower inertia do not propagate any further.
Thus we are free to choose integral causality for the second inertia. When we do so, the zero
junction receives a second velocity input. In this situation, the consequences of this causal
choice do propagate through the zero-junction. Two of the three bonds of the zero junction input
velocities, therefore the remaining bond must input a force as shown in figure 4.28. This is
because something in this case the friction due to relative motion must determine the
unique effort on the junction. This is an example of a weak causal assignment.
191
Figure 4.28: Complete causal assignment for the two-plate system.
Deriving State Equations
To obtain state equations, we begin by choosing state variables. As the energy storage elements
give rise to the system's dynamic behavior, we choose variables associated with the independent
energy storage elements. In this case, one such choice would be the momenta of the two inertias;
they will suffice to define the energy in the system. But remember, state variables are not
unique. Another choice would be the velocities of the two inertias. As all elements in this
system are linear, we will use the latter.
Differentiating the constitutive equations for the inertias and using the differential definitions of
momentum, we obtain:
dv
1
/dt = (1/m
1
) F
1
(4.107)
dv
2
/dt = (1/m
2
) F
2
(4.108)
Expressions for the forces F
1
and F
2
are obtained by combining the junction structure equations
and the constitutive equations of the dissipative elements. There are numerous ways to write and
combine these equations, but we need expressions in causal form, as assignment statements (e.g.
equation 4.106 rather than equation 4.105). The required form, complete with correct signs, can
be found by reading the equations directly from the half arrows and causal strokes on the bond
graph.
Thus the causal strokes on the zero junction show that the force F
2
is determined solely by the
dissipator b
2
. Subscripting variables in an obvious way:
F
2
= F
R2
= b
2
v
R2
(4.109)
The velocity v
R2
is determined from the junction structure as follows.
v
R2
= (v
1
- v
2
) (4.110)
Substituting 4.110 into 4.109 into 4.108 yields a state equation.
dv
2
/dt = (1/m
2
)b
2
(v
1
- v
2
) (4.111)
The causal strokes on the one junction indicate that the force F
1
is determined by both
dissipators. Again, using obvious subscripts:
192
F
1
= -(F
R1
+ F
R2
) (4.112)
The dissipator b
2
is in resistance causal form.
F
R1
= b
1
v
R1
= b
1
v
1
(4.113)
Substituting 4.110 into 4.109, 4.113 and 4.109 into 4.112, and 4.112 into 4.108 yields another
state equation.
dv
1
/dt = -(1/m
1
)(b
1
v
1
+ b
2
(v
1
- v
2
)) (4.114)
Equations 4.114 and 4.111 may be re-written in the standard integrable form
'
x = Ax.
d
dt
(
(
(
v
1
v
2
=
(
(
(
-(b
1
+ b
2
)/m
1
b
2
/m
1
b
2
/m
2
-b
2
/m
2
(
(
(
v
1
v
2
(4.115)
As the system has two independent energy storage elements, it is second order. However, if we
examine the system matrix, A, we can see that the off diagonal elements may be unequal but
have the same sign. The antisymmetry of signs we found in the harmonic oscillator is absent.
That sign antisymmetry was what gave rise to oscillatory behavior and we might expect that this
system will not exhibit oscillatory behavior. In fact, this result can be derived formally by
integrating the matrix differential equations. In general, energy storage elements of the same
type connected by one- and zero-junctions do not give rise to oscillation.
Sign Convention for Energy Storage Elements
Note that once sign and causality have been determined on the bond graph, equations follow
directly. The basic sign convention, that power flow is positive into all passive elements, should
not be violated, as it is what allows us to write the constitutive equations of the elements in the
usual way. By this sign convention, a positive effort applied to an inertia will cause its flow to
increase. In the mechanical domain, a positive force will cause positive acceleration.
In contrast, if the half arrow were to point away from the element, this would mean that power
was positive out of the element. In the mechanical domain, positive force would cause
deceleration, not acceleration. Note the negative sign in the constitutive equation shown in
figure 4.29.
Figure 4.29: Consequence of power-positive-out convention on an energy-storage element.
If you were perverse and were to follow this sign convention consistently, everything would be
fine; but being human and therefore prone to error, there's a good chance that you might forget to
reverse the sign on the constitutive equations. Don't do it! Stay with the convention that power
is positive into a passive element.
193
Dependent Energy Storage Elements
In the foregoing examples we found that one state variable was associated with the energy stored in
each energy storage element. Will every energy storage element give rise to an unique state
variable? Not necessarily, as we will see below when we consider two energy storage elements of
the same type connected by a simple junction.
Suppose we wish to model one dimension of the motion of two space vehicles in a vacuum under
free-fall conditions (i.e. zero net gravitational effects). As we are only concerned with their overall
motion it seems reasonable to model each vehicle as an ideal Newtonian rigid body. But then the
two vehicles dock with one another such that they subsequently move with the same velocity. As
the two now move with a common velocity (flow), they interact through a one junction and a bond
graph for one-dimensional motion of the system would be as in figure 4.13.
I 1 : m
1
2
: m
v
I
Figure 4.13: Bond graph representation of common-velocity coupling between two inertias.
Because these are ideal (linear) inertias, we may differentiate their constitutive equations as before:
dv
1
/dt = F
1
/m
1
(4.96)
dv
2
/dt = F
2
/m
2
(4.97)
The equations associated with the one-junction may be written as follows:
v
1
= v
2
(4.98)
F
1
= -F
2
(4.99)
In this case, simple substitution does not lead to state equations; a little manipulation is required.
dv
1
/dt = -(m
2
dv
1
/dt)/m
1
(4.100)
Note this is an implicit equation. Rearranging:
(m
1
+ m
2
)dv
1
/dt = 0 (4.101)
Assuming m
1
and m
2
are non-zero constants, dv
1
/dt = 0. Integrating, the system state trajectory is:
v
1
(t) = constant (4.102)
This system only requires one constant of integration, and therefore only one state variable. Yet the
model had two storage elements. Why doesn't it require two state variables as in the previous
example? Because the two energy storage elements in this model are not independent. Because of
the one-junction, the velocity or momentum of one determines the velocity or momentum of the
other; given the masses of both bodies, knowing the energy of one is sufficient to determine the
energy of the other. Therefore only one state variable is needed; only one initial condition is
required to determine the single constant of integration.
194
Causality and Dependent Energy Storage Elements
In previous examples, state equations were obtained by a simple process of substitution, yet in the
simple example above, further algebraic manipulation was required. This is a typical consequence
of dependent energy storage elements and, as one might expect, in more complex systems the
algebraic manipulations can become formidable, even prohibitively so. It would be useful to know
about dependent energy-storage elements before attempting to derive equations. How may we do
so?
The inter-dependence of energy storage elements is easily discovered by considering causality. It
refers to the choice of input and output which must be made when we come to describe a system in
terms of mathematical operations
1
on numbers.
As we will see below, it is also closely related to cause and effect in the usual temporal meaning of
those words, though in this context the consequences of causality may be unfamiliar. It is generally
accepted that physical systems must be causal. If a cause produces an effect, then that effect may
not precede the cause in time. Said another way, if a system is causal, its present output cannot
depend on its future input.
Integral Causality
If we describe a system in terms of an input and an output then the equations used to model the
system can be usefully treated as a sequence of operators which act upon an input to produce an
output. Thus, in the first-order examples above, an ideal capacitor comprises two operations: an
input flow is integrated to yield an output displacement; that displacement in turn determines an
effort. This integral causal form of the capacitor equations may be represented by the operational
block diagram shown in figure 4.14.
Conversely, for an inertia, an input effort may be integrated to yield an output momentum; that
momentum in turn determines a flow. The integral causal form of the inertia equations may be
represented by the operational block diagram shown in figure 4.15.
C
Figure 4.14: Block diagram representation of the mathematical operations associated with the
integral causal form of a generalized capacitor. A corresponding bond graph is also shown.
1
For that reason it is sometimes referred to as operational causality.
195
I
Figure 4.15: Block diagram representation of the mathematical operations associated with the
integral causal form of a generalized inertia. A corresponding bond graph is also shown.
When an inertia is coupled to a capacitor through a one-junction, (as in the Hamiltonian system
above) the operations required to model the complete system may be represented by combining the
operational block diagrams for each of the pieces as shown in figure 4.16.
The defining equation for the one-junction (the identity of the flow variables) is simply a connection
in the block diagram; The associated compatibility equation is (in this simple case) a sign change
operation. This diagram shows that the complete system comprises a closed loop of operations, a
chain of causes and effects. Both the inertia and the capacitor are in integral causal form and that
means that all of the operations well-defined.
Figure 4.16: Block diagram representation of the mathematical operations associated with a
capacitor and an inertia connected through a one-junction. An equivalent bond graph is also shown.
Derivative Causality
If we attempt to represent two inertias and a one junction (as in the example of the space vehicles)
using block diagrams, we encounter a difficulty: one of the inertias has to accept flow as input and
196
produce effort as output, the opposite of what is shown in figure 4.15. Simply reversing the
orientation of the arrows on the block diagram is meaningless; one of the two inertias (it doesn't
matter which) must be represented using a different set of operations, the inverse of those depicted
in figure 4.15. The constitutive equation for the inertia must be replaced by its inverse; the time-
integration operator must be replaced with a time-derivative operator, as shown in figure 4.17. As in
this figure, an asterisk will often be used to call attention to the time-derivative operator. An energy-
storage element which is represented by a time-derivative operation is said to be in derivative causal
form.
Figure 4.17 shows that, as before, the complete system comprises a closed loop of operations, a
chain of causes and effects. However, unlike figure 4.16, the derivative causality of one of the
inertias in figure 4.17 means that all of the operations in the loop may not be well-defined. First,
the inverse of the constitutive equation for the inertia may not exist. In the example above it does, as
the constitutive equation is linear, but in general a well-defined function may not have a well-defined
inverse. A second and more profound problem stems from the time-derivative operation; it is
highly undesirable because time differentiation is a physically impossible operation.
Figure 4.17: Block diagram representation of the mathematical operations associated with two
inertias connected through a one-junction. An equivalent bond graph is also shown.
There are several ways to understand this. The most fundamental is based on the definition of a
time derivative as the limit of the difference of two values of a function divided by their time
separation as the latter goes toward zero. To obtain the time derivative of a function at any point in
time one must take the difference of two values of the function, one in the immediate past, and one
in the immediate future. This is illustrated in figure 4.18. The output of a time derivative operation
depends on future values of its input, and therefore it is non-causal in the usual physical sense.
An alternative argument: The input of any operational block may be any well-defined function of
time, including a discontinuous function such as a step function. For the integration operator,
discontinuities present no problem: the input force applied to an inertial object may change
abruptly; its velocity, momentum and stored kinetic energy will not.
197
Figure 4.18: Diagram illustrating the fundamental definition of a derivative operation.
In contrast, the derivative operator will generate infinite output in response to a discontinuous input,
and this infinity is a departure from the physically realizable. A discontinuous change in the
velocity of an inertial object would require an infinite force. Furthermore, the kinetic energy stored
in the inertial object would also change discontinuously, and that would require an infinite power
flow. Neither of these infinities is physically possible. For these reasons, a time differentiation
operation cannot be physically realized; it can at best be approximated.
How much does this matter? We know from thermodynamics that real macroscopic phenomena
are not reversible. The time sequencing of events is a fundamental aspect of physical system
behavior. Therefore causality is an important aspect of our description of macroscopic physical
phenomena. The point is that to use a time derivative operation in a model of a physical system is a
major misrepresentation of physical reality as we understand it and it should avoided.
Derivative Causality Indicates Dependent Energy Storage
However, to put this discussion in perspective, every mathematical model falls short of reality,
usually very far short; for example, the idea of a rigid body (which we used in the example with the
two space vehicles) is itself a fiction, albeit a very convenient one a typical space vehicle is
anything but rigid. It can then be argued that a time-derivative operation is intrinsically no worse
than this and other modelling approximations.
Indeed, when two inertias are coupled by a one-junction, the resulting derivative causality really
means that the two inertias are one object, one rigid body. That is the true meaning of inter-
dependence of energy storage elements: in the model they are not distinct energy storage elements,
despite appearances to the contrary. These two modelling approximations rigid-body models
and time-derivative operations are intimately related.
If we consider an object undergoing translational motion along one dimension in space, the
assumption that it is rigid is equivalent to assuming that the (infinitely) large collection of fictional
"mass points" which comprise the object all move with a common speed. In bond-graph notation, if
each mass point were represented by an inertia, all would be connected to a single one-junction
representing the common speed. But in terms of mathematical operations, only one of those
inertias can determine the common speed associated with the one-junction; only one inertia can have
integral causality. All the others must accept that speed as an input; they must be in derivative
causality. The entire collection of mass points is a single independent energy storage element; a
198
single number (the common momentum or common speed) is sufficient to determine the stored
energy.
Preferred Causality for Energy Storage Elements
A point to be taken from this discussion is that, if possible, energy-storage elements should be
independent and have integral causality. But why? Why does it matter if an energy-storage
element is dependent? If it is regarded as merely another modelling approximation, why should the
time-derivative operation be avoided?
Heed the admonition of the the fourteenth-century scholar William of Ockham
2
:
"Entia praeter necessitatem non sunt multiplicanda"
In the modelling context: Approximations must not be made needlessly; wherever possible, the
integration operation should be used rather than the derivative operation.
Every energy-storage element which can be described using an integration operator should be. It
will require one initial condition to determine its constant of integration, and therefore will give rise
to one state variable; energy storage elements which have integral causality are independent.
Conversely, any energy storage element which must be described using a derivative operation will
not require an independent initial condition and therefore will not give rise to a state variable; energy
storage elements which have derivative causality are dependent.
A final comment: Elements such as power sources are subject to causal constraints; their definition
will only permit one causal form. In contrast, in certain situations which we will discuss later,
derivative causality is unavoidable, or the steps required to eliminate it would obscure the insight
that the model is supposed to provide. The integral causal form is the preferred causal form for
energy-storage elements; it is not essential.
2
If I have my Latin correct, this translates as "Entities are not to be multiplied beyond necessity"; it
is popularly known as Occam's Razor.
199
Causality and Model Formulation
Causal considerations can be extremely helpful at the stage of formulating a model because they
can identify the consequences of particular choices. Consider the following example.
Two large, identical electrical capacitors, initially charged to different voltages, are to be
connected together by two bars of copper as shown conceptually in the figure 4.30. This system
would be a reasonable laboratory simulation of one of the problems of switching in an electrical
power system. The copper bars are large, several square centimeters in cross sectional area. We
wish to formulate the simplest possible model which is competent to describe the events that
happen as the contact is closed.
The simplest model for this system would appear to be two capacitors and to keep things simple
we may assume they are ideal. The resistance of the connector would appear to be negligible.
Ignoring surface contact phenomena, the resistance of a conductor is the product of resistivity
and length divided by cross-sectional area
R = l/A (4.116)
where R is resistance, is resistivity, l is length and A is area. Copper is an excellent conductor;
the resistivity of drawn copper is listed as 1.724 O-cm. Even if the copper bar is tens of
centimeters long, its resistance will be measured in micro-ohms.
Figure 4.30: Conceptual sketch of a capacitor and switch system.
When the switch is closed the two capacitors share a common voltage, which implies that their
connection should be modeled as a zero-junction. As a first step in thinking about this model it
is useful to assign causality. As a result, without deriving equations (admittedly a simple enough
200
task in this case) it is clear that the two capacitors are not independent; assign either capacitor
integral causality and, because of the zero-junction, the other must have derivative causality as
Does this make physical sense? Yes. Given the capacitance of both capacitors, a knowledge of
the energy stored in one is sufficient to determine its charge or voltage; the voltage is the same
for both capacitors, hence the energy stored in the second capacitor can be computed. Therefore
only one state variable is needed to characterize the energetic state of this system.
0 C C
Figure 4.31: Model of two capacitors connected.
Simple though this system is, there is plenty of room for confusion: before connection the
charges on the two capacitors could be specified independently; afterwards, our model says they
cannot. Do we need two initial conditions or one?
The confusion arises because we have two structurally different models for the system before
and after connection. That is, different mathematical operations will be used to describe the
capacitors in the two different situations. This is easily seen by comparing the causality of the
capacitors in the two situations. A model of the unconnected capacitors would be as shown in
figure 4.32. By assumption, no current may flow from the capacitors, so this boundary condition
is modeled by a flow source imposing zero flow on the capacitor. Before connection, each of the
two capacitors is independent and we are free to assign integral causality to each.
C C S
f
:
0 0
:
S
f
Figure 4.32: Model of the two unconnected capacitors.
There is nothing wrong with using two structurally different models to describe different
operating conditions for a system, though care must be taken to ensure that the models are
mutually consistent and do not contradict each other. Is there some contradiction or
inconsistency here? No. In the absence of charge leakage, the total charge in the system must be
the same before and after connection. Using obvious subscripts:
(q
1
+ q
2
)
before
= (q
1
+ q
2
)
after
(4.117)
(Ce
1
+ Ce
2
)
before
= (Ce
1
+ Ce
2
)
after
= 2Ce
after
(4.118)
Thus the two models may be made mutually consistent by deriving the single initial condition for
the model after connection from the state of the model just before connection.
e
after
=
|
\
e
1
+ e
2
2
|
|
.
before
(4.119)
However, if we attempt to concatenate the two pieces of the model valid before connection
(figure 4.32) to obtain a model for the same two capacitors after connection (figure 4.31) we
201
discover a "causal conflict". This "causal conflict" means that the pieces cannot just be
assembled, the entire model for the coupled system has to be reformulated; new equations have
to be written. In this case, the new equations require a different number of state variables than
the old. But you don't have to work out the equations to see this; it is evident from the change of
causal form for one of the capacitors from integral causality to derivative causality.
In this case we were asked for a model to describe the events that happen as the contact is closed.
But if we want to describe the events at switch closure, we have a problem: How do we reconcile
the different descriptions in the two regimes?
The key to understanding the dynamics of an energetic system is to keep track of energy. In this
linear system, energy is numerically equal to co-energy. Before contact:
*
1
E
p
before
=
2
C(e
1
2
+ e
2
2
)
before
(4.120)
After contact:
*
1
E
p
after
=
2
C(e
1
2
+ e
2
2
)
after
= C
|
\
e
1
+ e
2
2
|
|
.
2
(4.121)
before
Therefore:
* *
|
\
E
p
before
- E
p
after
= C
e
1
e
2
-
2 2
|
|
.
2
_ 0 (4.122)
Energy is lost during switch closure. To model this phenomenon we must include at least one
dissipative element
1
, however small.
The point to be taken here is that by considering causality we uncover an aspect of the model
which we may want to modify. It is important to stress that "causal conflicts" are not errors.
They simply spell out consequences of modelling assumptions which may otherwise be obscure.
Think of them as waving a banner which reads "Are you sure you really meant your model to
have this property?"
Junction Structure
Suppose we decide to include a dissipator. The simplest model we can assemble will include
two capacitors and an ideal dissipator. How should they be connected? Any current that leaves
one capacitor flows through the dissipator and enters the other capacitor; this indicates a one-
junction and suggests the bond graph shown in figure 4.33. Assigning causality we see that both
1
In practice, the dissipative phenomenon encountered on switch closure is dominated by an arc
which "jumps the gap" before metal-to-metal contact is made rather than by the resistance of the
metallic components.
202
capacitors may be given integral causality and the dissipator has conductance causality; it
determines the current as a function of the voltages on the two capacitors.
C 1 C
R
Figure 4.33: Model of the two capacitors including dissipation.
But this graph has a serious flaw. We expect the dissipator to determine current as a function of
the potential difference across its terminals. Yet the compatibility equation (Kirchhoff's voltage
law) associated with the junction in figure 4.33 would be written as follows (using obvious
subscripts and reading the signs and the required causal form from the graph):
e
R
= -(e
1
+ e
2
) (4.123)
Thus the equation for the common current would depend on the negative sum of the capacitor
voltages.
e
1
+ e
2
i
R
= -
R
(4.124)
This is the same problem we encountered previously. In fact, it is possible to define a sign
convention for the capacitor and dissipator voltages so that equation 4.124 is correct, but it is
quite counter-intuitive. A better course of action is to revise the junction structure so that the
expected potential difference is identified as shown in figure 4.34.
1 0 C 0 C
e
1
e
2
0
:
e
1
-
e
2
R
Figure 4.34: Revised junction structure which identifies the potential difference across the
dissipator.
The zero junction in the center may be eliminated, because, reading the half-arrows, its equations
merely state that the efforts and flows on the associated bonds are identical. The same is true for
the zero junction on the right, but not for that on the left. The compatibility equation for this
junction states that the flows on the two associated bonds are equal and opposite; this junction
cannot be eliminated. A simplified bond graph is shown in figure 4.35.
203
C 0 1 C
1 4 2
3
R
Figure 4.35: Simplified bond graph for the two capacitors with dissipation.
The numbers written beside the bonds in figure 4.35 are not strictly necessary. They are
included solely to facilitate the following derivation.
State Equations
At this point, deriving state equations is straightforward. As all elements in this system are
linear we might use circuit variables (capacitor voltages in this case) for state variables. But we
may equally well use energy variables (capacitor charges in this case) and, in fact, will do so.
The definition of generalized displacement yields:
dq
1
/dt = i
1
(4.125)
dq
2
/dt = i
2
(4.126)
Reading the causal strokes on the graph, i
1
is determined by i
4
which is in turn determined
by i
3
; i
2
is identical to i
3
. Reading the signs from the half arrows:
i
1
= -i
4
= -i
3
(4.127)
The dissipator determines i
3
.
1
i
3
=
R
e
3
(4.128)
e
3
is determined by e
2
and e
4
; e
4
is identical to e
1
. Reading the signs from the half arrows:
e
3
= e
4
- e
2
(4.129)
e
4
= e
1
(4.130)
The capacitors determine e
1
and e
2
.
q
1
e
1
=
C
(4.131)
q
2
e
2
=
C
(4.132)
State equations are now obtained by substitution.
204
dq
1
/dt = -
1
R
|
\
|
|
.
(4.133)
q
1
q
2
-
C C
dq
2
/dt =
1
R
|
\
|
|
.
q
1
q
2
(4.134) -
C C
These equations may be written in standard integrable matrix form x
'
= Ax.
-1/RC 1/RC
(
(

(
(
q
1
q
1
(
(
d
(4.135) =
dt
1/RC -1/RC q
2
q
2
Notice that the off-diagonal terms in the system matrix lack the sign asymmetry required in an
oscillator. In fact, all elements in the system matrix have the same magnitude and the pattern of
signs comes from the fact that i
1
= -i
2
. It is instructive to change coordinates and use the sum
and difference of the two charges as state variables. The transformation equation is as follows.
1 -1
(
(

(
(
q
d
q
1
(
(
(4.136) =
1 1 q
s
q
2
The inverse transformation also exists.
1/2 1/2
(
(

(
(
q
1
q
d
(
(
(4.137) =
-1/2 1/2 q
2
q
s
From equation 4.94, the transformed state equations are:
1 -1 -1/RC 1/RC 1/2 1/2
(
(

(
(
q
d
q
d
(
(

(
(
(
(
d
(4.138) =
dt
1 1 1/RC -1/RC -1/2 1/2 q
s
q
s
-2/RC 0
(
(

(
(
q
d
q
d
(
(
d
(4.139) =
dt
0 0 q
s
q
s
With this change of coordinates the state equations are trivial to integrate. The solutions are:
q
d
(t) = q
d
(t
o
)e
(t
o
- t)2/RC
(4.140)
q
s
(t) = q
s
(t
o
) (4.141)
Hence the second state equation merely reiterates the fact that charge is conserved in this system.
As we expect from the pattern of signs in the system matrix, this system cannot exhibit
oscillatory behavior.
205
A Switch Behaves as a Modulated Dissipator
As a final remark on this simple example, if we compare figures 4.32 and 4.35 we may note that
the causality on capacitors is same in both cases, before and after connection. This indicates that
it is possible to represent the models for both regimes as a single bond graph, shown in figure
4.36. The important addition is that the switch has been represented as a modulated dissipator.
A suitable constitutive equation would be:
u
i
R
=
R
e
R
(4.142)
where u is a modulating or control input which can assume only two real values, 1 or 0,
corresponding to connected and disconnected, respectively.
i
R
=
0 if u = 0
e
R
R
if u = 1
(4.143)
C 0 1 C
modulating or
control input, u
R
Figure 4.36: Model with modulated dissipator representing connected and unconnected
behavior.
Note that this modulated dissipator is subject to a strict causal constraint: it must have
conductance causality, determining an output flow (current) as a function of an input effort
(voltage). The inverse of equation 4.142 is not defined for u = 0.
Modulated dissipators will be encountered frequently in describing control systems where
signal-level events influence power-level behavior.
206
Examples
Ideal asymmetric two-port junction elements, gyrators and transformers, describe the
transmission or transduction of power, one of the most fundamental aspects of all machinery.
An abundant variety of devices that may be modeled using transformers or gyrators.
Rotation to Translation
Almost all mechanisms involve some relation between rotation and translation, as the following
examples illustrate.
Pivoted Lever Arm
One of the simplest examples is the pivoted lever arm shown in figure 5.7. Rotation of the shaft
results in motion of the end of the arm. Assuming the arm is rigid and the pivot constrains it to
motion about a fixed axis, a relation between displacements (the angle of the arm, u, and the
horizontal position of its end, x) may be derived from simple geometry.
x = l sin(u) (5.19)
A relation between flows (the angular speed of the arm, e, and the horizontal linear velocity of
the end, v) may be obtained by differentiating with respect to time.
v = l cos(u) e (5.20)
u
arm of length l
shaft pivots about fixed axis
x
Figure 5.7: Diagram of a simple rotation-to-translation transformer.
This is one of the equations of a transformer relating the translational and rotational domains.
The other equation relates horizontal force, F, at the end to moment, , about the axis. The
moment arm is the perpendicular distance from the end to the axis.
= l cos(u) F (5.21)
207
Equations 5.20 and 5.21 define a modulated transformer (e.g figure 5.6) with a single parameter
T = l cos (u) modulated, in this case, by the angle u. As required by the definition of a pure
transformer, power in equals power out (Fv = e). Note that this fundamentally stems from the
fact that the same geometric quantity, the perpendicular distance from the end to the axis, is the
parameter in both equations.
Scotch Yoke
When the shaft in figure 5.7 rotates, the vertical position of the end of the arm also changes, a
fact ignored in equation 5.19. It may seem that an important part of the mechanical behavior has
been ignored, but it should be recognized that there are numerous contrivances by which the
vertical position may be made irrelevant. One example is the Scotch Yoke shown in figure 5.8.
u
wheel
slider
bearing
roller
x
Figure 5.8: Another rotation-to-translation transformer: a Scotch Yoke.
208
Wheel
It should not be assumed that linearity may only be achieved for small displacements. Appropriate
design can result in devices which are linear over large ranges of displacements. Perhaps the
simplest and most ubiquitous example is the wheel. Figure 5.9 depicts a wheel which may rotate
about a fixed axis. The rim of the wheel contacts a sliding bar. Due to friction between the rim and
the slider, when the wheel rotates, the slider translates. Assuming the wheel and slider are rigid and
there is no slip between rim and slider, simple geometry provides a relation between displacements.
x = r (5.22)
where r is the radius of the wheel. As before, a relation between flows (speed of the slider, v, and
angular speed of the wheel, ) may be obtained by differentiating with respect to time.
v = r (5.23)
The relation between force, F, on the slider to the moment, , about the shaft is also linear.
= r F (5.24)
These are the equations of a linear transformer (e.g. figure 5.3) relating the translational and
rotational domains. Again, the same geometric parameter, r, shows up in both equations.
wheel
x
shaft
slider
Figure 5.9: A linear rotation-to-translation transformer composed of a wheel and slider.
Nevertheless, it is well to keep in mind that in general, a transformer may be modulated, embodying
a nonlinear relation between efforts and flows. As any practical machine designer knows, the radius
of any real wheel will not be constant. The wheel may be non-circular or may be mounted
eccentrically. In either case, its radius will be a function of its angle. If these effects are sufficiently
important, they may modeled by describing the device as a modulated transformer as depicted in
figure 5.6.
209
Parasitic Dynamics
As with other elements of a lumped-parameter model, an ideal transformer such as that of figure 5.3
or 5.6 describes a single phenomenon: power transmission. Real power transmission devices such
as those sketched in figures 5.7 to 5.9 also store and dissipate energy. Being secondary (and
somewhat athwart the primary function of power transmission) these phenomena are often referred
to as parasitic. A competent model of a real power transmission device may need to include these
"parasitic" effects. This may readily be accomplished using the energy storage and dissipation
elements defined earlier.
For example, the wheel and slider of figure 5.9 might exhibit substantial energic phenomena in both
domains. Likely candidates in the rotational domain are the inertia of the wheel and shaft and the
friction due to the bearings which support the shaft (not shown in figure 5.9). All of the rotational
phenomena are associated with a single speed, so the inertia, friction and the rotational side of the
transformer are connected by a one-junction.
friction
friction
bearing
R
R
R friction
between
and slider
between rim
slider and
frame
1
r v
rim
TF 1 0 1
v
slider
wheel
and shaft
mass of
inertia
I I slider
Rotational domain Translational domain
Figure 5.10: Bond graph of a model of the wheel and slider of figure 5.9 including plausible
energic effects in the rotational and translational domains.
Inertia and friction are also plausible in the translational domain. There are two distinct frictional
components: One is due to the motion of the slider relative to the frame. That component and the
inertia of the slider are associated with the same velocity, v
slider
, so the two elements are connected
to a one junction. The other component is due to the possibility of relative motion between the
slider and the wheel rim. The translational velocity of the wheel rim at the point of contact, v
rim
, is
the translation port of the transformer. The frictional element is connected to a zero junction
because the force on it is the same as that transmitted to the slider. Including these phenomena, a
model of the system might be as shown in figure 5.10.
Of course, it should not be assumed that figure 5.10 is necessarily a better model. For some
purposes, to include these phenomena may be to add superfluous detail and an ideal transformer
may be the best model.
210
Rotation to Fluid Flow
Another common phenomenon in machinery is the transport of fluids. Pumps are among the more
widespread machine components, as may be seen by examining a typical automobile engine.
Direction of rotation
Input
shaft
Fluid
input
port
Fluid
output
port
Fluid
flow
direction
Entrapped
volume
Direction of rotation
Figure 5.11: A sketch of a gear pump is shown on the left. A cross section is shown on the right.
Gear Pump
One of the more robust and versatile types of pump is the gear pump. As illustrated in figure 5.11,
two meshing gears rotate inside a close-fitting housing. Fluid entrapped between the gears and the
housing is drawn from the input port to the output port. A gear pump is generally used to pump
incompressible fluids and, by design, is a positive-displacement pump: aside from a small amount
of leakage between the gears and the housing, a fixed volume of fluid is pumped per unit rotation of
the drive shaft. If we neglect any leakage in the device sketched in figure 5.11, for every sixty
degrees of shaft rotation, each of the two gears pumps the volume, V
e
, entrapped between one gear-
tooth and the housing. The relation between displacements is as follows.
6 V
e
V =
= K (5.25)
where V is fluid volume, is shaft angle in radians and K is a constant. A relation between flows is
obtained by differentiating this relation between displacements with respect to time.
Q = K (5.26)
where Q is volumetric flow rate and is angular speed. This is one of the equations of a
transformer relating the fluid flow and rotational domains. The other equation relates the pressure
difference, P, between the input and output fluid ports to the moment, applied to the drive shaft. It
could be obtained by calculating the pressure-induced forces on the faces of the gear teeth and
taking moments about the shaft axis, but that requires careful thought and tedious calculation. A far
211
simpler method is to remember that power into an ideal transformer must equal power out of it (PQ
= ) thus the relation between efforts must be as follows.
= K P (5.27)
Hydraulic Motor
An important feature of all power transduction devices is that, in general, power may be transferred
in both directions. A transformer or gyrator is fundamentally a two-way device. Consequently, the
positive-displacement gear pump depicted in figure 5.11, designed to transmit power from the
rotational to the fluid domain, is also one of the basic designs for a hydraulic motor which transmits
power from the fluid to the rotational domain
1
.
Parasitic Dynamics
In addition to the power transmission it is designed for, a real pump or motor is likely to exhibit
energy storage and dissipation phenomena. In the rotational domain, friction and the inertia of the
rotating components may be significant. A single speed is associated with the inertia, friction and
the rotational side of the pump and these elements are connected by a one-junction.
In the fluid domain, there are at least three distinct sources of energy dissipation, each associated
with a different flow rate. The first is due to leakage, e.g. between the gears and the housing end-
plates. This leakage flow is not proportional to the speed of the shaft, but is related to the pressure
difference between input and output. If the leakage flow is not zero, then the total flow though the
pump is different from the component due to shaft rotation, and each of these two flows may have a
distinct dissipation phenomenon associated with it. For example, the orifice head losses at entry
and exit are related to total flow while the head loss due to flow past the circular walls of the
chamber is related to the flow component due to shaft rotation. The required junction structure is
1
However, a motor and a pump may differ in important aspects of their design details.
212
chamber leakage orifice
losses losses losses
1 TF
I
R
1
inertia of
gears and
shaft
bearing
friction
1 0
R R
0
0
R
P
Q
input
pump
total
output
P
P
Q
K
:
Rotational domain Fluid domain
Figure 5.12: Bond graph of a model of the gear pump of figure 5.11 including plausible energic
phenomena in the rotational and fluid domains.
As before, this is not necessarily a better model than the ideal transformer, merely more detailed.
The extra detail comes with a price: to be useful, the parameters of the additional elements must be
determined and that may require considerable effort.
213
Neville Hogan page 1
Magnetic electro-mechanical machines
Lorentz Force
A magnetic field exerts force on a moving charge. The Lorentz equation:
f = q(E + v B)
where
f: force exerted on charge q
E: electric field strength
v: velocity of the moving charge
B: magnetic flux density
Consider a stationary straight conductor perpendicular to a vertically-oriented magnetic field.
Magnetic flux
B
F i
Current
Force
Stationary
conductor
Motion of charges
An electric field is oriented parallel to the wire. As charges move along the wire, the magnetic
field makes them try to move sideways, exerting a force on the wire. The lateral force due to all
the charge in the wire is:
f = Al (v B)
where
: density of charge in the wire (charge per unit volume)
l: length of the wire in the magnetic field
A: its cross-sectional area
The moving charges constitute a current, i
i = Av
The lateral force on the wire is proportional to the current flowing in it.
f = l (i B)
For the orthogonal orientations shown in the figure, the vectors may be represented by their
magnitudes.
f = l B i
214
This is one of a pair of equations that describe how electromagnetic phenomena can transfer
power between mechanical and electrical systems. The same physical phenomenon also relates
velocity and voltage. Consider the same wire perpendicular to the same magnetic field, but
moving as shown
Magnetic flux
B
Moving
conductor
Electro-motive
force
Motion of charges
Velocity
v
A component of charge motion is the same as the wire motion. The magnetic field makes
charges try to move along the length of the wire from left to right. The resulting electromotive
force (emf) opposes the current and is known as back-emf.
The size of the back-emf may be deduced as follows. Voltage between two points is the work
required to move a unit charge from one to the other. If a unit charge moves along the wire from
right to left the work done against the electromagnetic force is
e = v B l
This is the other of the pair of equations that describe how electromagnetic phenomena can
transfer power between mechanical and electrical systems.
Two important points:
1. The interaction is bi-lateral (i.e., two-way). If an electrical current generates a
mechanical force mechanical velocity generates a back-emf.
2. The interaction is power-continuous. Power is transferred from one domain to the other;
no power is dissipated; no energy is stored; electrical power in equals mechanical power
out (and v.v.).
P
electrical
= e i = (v B l) i = v (B l i) = v f = P
mechanical
Power continuity is not a modeling approximation. It arises from the underlying physics. The
same physical quantity (magnetic flux density times wire length) is the parameter of the force-
current relation and the voltage-velocity relation
215
D'Arsonval Galvanometer
Many electrical instruments (ammeters, voltmeters, etc.) are variants of the D'Arsonval
galvanometer. A rectangular coil of wire pivots in a magnetic field as shown in figure 2.
Restraining
spring
Magnetic
field
Square
coil
+
-
S
N
u
Figure 2: Sketch of a D'Arsonval galvanometer.
Current flowing parallel to the axis of rotation generates a torque to rotate the coil. Current
flowing in the ends of the coil generates a force along the axis. Assuming the magnetic flux is
vertical across the length and width of the coil, the total torque about the axis is:
t = 2NBlh cos(u) i
where
t: clockwise torque about the axis
N: number of turns of wire
B: magnetic flux density
l: length of the coil
h: half its height.
This torque is counteracted by a rotational spring
As the coils rotate, a back-emf is generated.
e = 2NBlh cos(u) e
where
e: angular speed of the coil.
Note that the same parameter, 2NBlh cos(u), shows up in both equations.
216
Direct Current Permanent Magnet Electric Motor
With a different geometry the dependence on angle can be substantially reduced or eliminated
Radial flux lines
Stationary
permeable
core
Rotating
cylindrical
coil
(much of the
magnetic field
is not shown)
Stationary
permanent
magnets
N
S
Figure 3: Schematic end-view of a D'Arsonval galvanometer modified to reduce the angle-
dependence of the transduction equations.
Features:
- rotating cylindrical coil
- stationary permeable core
- shaped permanent magnets
- constant radial gap between magnets and core
- the magnetic field in the gap is oriented radially
If all turns of the coil are in the radial field the torque due to a current in the coil is independent
of angle.
0 30 60 -60 90 -30 -90
Angle
Figure 4: Sketch of half of the constant-current torque/angle relation resulting from the design of
figure 3.
From symmetry the torque/angle relation for the other half of the circle is the negative of that
shown above.
The reversal of torque can be eliminated by reversing the current when the angle passes through
90. A mechanical commutator is sketched in figure 5.
217
N
S
Split-ring
commutator
rotates with
coil
Brushes are stationary
electrical contacts
Electrical connection
to the coil is through
the commutator
+
-
Figure 5: Schematic of the mechanical commutation system used in a direct-current permanent
magnet motor.
Electrical connection is through a set of stationary conductors called brushes
1
. They contact a
split ring called a commutator that rotates with the coil. This commutator design is used in a
direct-current permanent-magnet motor (DCPMM). The same effect may be achieved
electronically. That approach is used in a brushless DCPMM.
Assuming perfect commutation the relation between torque and current for a DCPMM is
t = K
t
i
K
t
: torque constant, a parameter determined by the mechanical, magnetic and electrical
configuration of the device.
There is also a corresponding relation between voltage and rotational speed.
e = K
e
e
K
e
: back-emf or voltage constant.
Excerpt from a manufacturer's specification sheet for a direct-current permanent-magnet motor.
MOTOR CONSTANTS:
(at 25 deg C)
SYMBOL UNITS
torque constant KT oz in/amp 5.03
back emf constant KE volts/krpm 3.72
terminal resistance RT ohms 1.400
armature resistance RA ohms 1.120
average friction torque TF oz in 3.0
1
The reason for this terminology is historical the earliest successful designs used wire
brushes for this purpose.
218
viscous damping constant KD oz in/krpm 0.59
moment of inertia JM oz in sec-sec 0.0028
armature inductance L micro henry <100.0
temperature coefficient of KE C %/deg c rise -0.02
These specifications imply that the torque and back-emf constants are distinct parameters.
different symbols: KT, KE
different units: oz-in/amp, volts/krpm
But if we express both constants in mks units
K
t
= 0.0355 N-m/amp
K
e
= 0.0355 volt-sec/rad
As required by the physics of electromagnetic power transduction, the constants are in fact
identical.
Parasitic Dynamics
The Lorentz force yields two equations describing power-continuous electro-mechanical
transduction. A practical electric motor also includes energy storage and/or power dissipation in
the electrical and mechanical domains. A competent model may include these effects.
Electrical side
- inductance of the coil
- resistance of the coil
- resistance of electrical connectors ("terminal resistance")
Mechanical side
- inertia of the rotating components (coil, shaft, etc.)
- friction of brushes sliding on the commutator
- viscous drag due to entrained air
Connections:
Electrical side:
- only one distinct currentseries connection of inductor, resistor & model element for
back-emf due to velocity
Mechanical side:
- only one distinct speedcommon-velocity connection of inertia, friction & model
element for torque due to motor current
219
coil
inductance
coil &
terminal
resistance
rotor inertia
friction losses
back
emf
motor
torque
electrical domain rotational domain
Schematic diagram of DCPMM model with parasitic dynamics
Motor vs. Generator
Lorentz-force electro-mechanical transduction is bi-lateral. An electric motor uses it to convert
electrical power into rotational power. An electrical generator uses it to convert rotational power
into electrical power.
Tachometer
Lorentz-force electro-mechanical transduction is also used for sensing. A DC permanent magnet
tachometer generates voltage proportional to angular velocity as described by the velocity-
voltage equation above.
220
PHYSICAL BASIS OF ANALOGIES
Equivalences: Transforming Through Transformers and Gyrators
One useful way to enhance your understanding of a system involving multiple energy domains is to
develop an equivalent dynamic system in a single domain of your choice. To do this, any
asymmetric junction elements (transformers or gyrators) coupling the multiple domains must be
eliminated, if possible. Indeed, even if the original system involves only a single energy medium,
identifying an equivalent system without transformers and gyrators can be useful. In effect, it is
helpful to know how an element (or even an entire subsystem) would appear to behave if it were
observed through a transformer or gyrator. By loose analogy with viewing an object through a lens,
we speak of transforming elements or subsystems through transformers or gyrators.
Transforming Through a Transformer
Any element or subsystem transformed through a transformer retains its dynamic character, though
its parameter values change. That is, a zero junction still appears to behave as a zero junction, a
capacitor still appears to behave as a capacitor, a flow source still appears to behave as a flow
source, and so forth. This is true whether the elements are linear or nonlinear.
The equivalent (transformed) active source elements are related to the original source elements in
proportion to the transformer coefficient, as summarized in table 5.2.
The relation for two-port asymmetric junctions (transformers and gyrators) is equally
straightforward and is summarized in table 5.3. The other possible causal assignments are treated
similarly.
Multi-port symmetric junctions (one and zero junctions) may also be transformed as summarized in
table 5.4. Representative causal assignments are shown. The other possible causal assignments are
treated similarly.
Table 5.2
Transforming active elements through a transformer.
Initial bond graph: Equivalent bond graph:
Constitutive
equations:
Constitutive
equations:
1
TF S : e(t)
2
e
T
1
S
e
: Te(t)
e
1
= T e
2
; e
2
= e(t) e
1
= Te(t)
1
TF
2
1/T
S
f
: f(t)
1
S
f
: f(t)/T
f
1
= (1/T) f
2
; f
2
= f(t) f
1
= f(t)/T
221
Initial bond graph:
Table 5.3
Transforming asymmetric junction elements through a transformer.
Equivalent bond graph:
Constitutive
equations:
Constitutive
equations:
TF
1 2 3
T
1
TF
T
2
TF
T
1
T
2
1 3
e
1
= T
1
e
2
; e
2
= T
2
e
3
e
1
= T
1
T
2
e
3
f
2
= T
1
f
1
; f
3
= T
2
f
2
f
3
= T
1
T
2
f
1
TF
1 2 3
T G
GY GY
1
T G
3
e
1
= T e
2
; e
2
= G f
3
e
1
= TG f
3
f
2
= T f
1
; e
3
= G f
2
e
3
= TG f
1
222
Initial bond graph:
Table 5.4
Transforming symmetric junction elements through a transformer.
Constitutive
equations:
Constitutive
equations:
TF
1 2
T
1
4
3
TF
3
T
1
1
TF
5
2
4
T
e
1
=T(e
3
+ e
4
) e
1
= T e
3
+ T e
4
f
3
= T f
1
; f
4
= T f
1
f
3
= T f
1
; f
4
= T f
1
TF
1 2
T
0
4
3
TF
3
1/T
1
0
TF
5
2
4
T
e
1
= T e
4
; e
3
= e
4
e
1
= T e
4
; e
3
= (1/T)T e
4
f
4
= T f
1
- f
3
f
4
= T f
1
- T(1/T) f
3
For passive elements, the relation is quite different. If a passive element is linear, the parameter of
the equivalent (transformed) element is related to the parameter of the original element through the
square of the transformer coefficient. One way to remember this is to note that to obtain the
equivalent relation for an active element, we had to "look through" the transformer once. In
contrast, to obtain the equivalent relation for a passive element, we had to "look through" the
transformer twice, and had to multiply (or divide) by the transformer coefficient on each pass.
These relations are summarized in table 5.5
223
Initial bond graph:
Table 5.5
Transforming passive elements through a transformer.
Constitutive
equations:
Constitutive
equations:
Constitutive
equations:
Constitutive
equations:
TF C: C
1 2
T
C: C/T
2
1
de
2 1
e
1
= T e
2
;
dt
=
C
f
2
; f
2
= T f
1
de
1 T
2
dt
=
C
f
1
TF I : I
1 2
1/T
1
I : T
2
I
dt
=
I
1 df
2 1 1
f
1
=
T
f
2
; e
2
; e
2
=
T
e
1
df
1 1
dt
=
T
2
I
e
1
TF R : R
1 2
T
R : T
2
R
1
e
1
= T e
2
; e
2
= R f
2
; f
2
= T f
1 e
1
= T
2
R f
1
TF R : G
1 2
1/T
R : G/T
2
1
f
1
=
T
f
2
; f
2
= G e
2
; e
2
=
T
e
1
1 1
f
1
=
T
2
e
1
G
Transforming Through a Gyrator
Any element or subsystem transformed through a gyrator takes on the dynamic character of the
dual element, and in addition its parameter values may change. That is, a zero junction appears to
behave as a one junction, a capacitor appears to behave as an inertia, a flow source appears to behave
as an effort source, and so forth. This is true whether the elements are linear or nonlinear.
224
Initial bond graph:
Table 5.6
Transforming active elements through a gyrator.
Constitutive
equations:
Constitutive
equations:
Table 5.7
1
GY
2
G
S
f
: f(t) S
e
: Gf(t)
1
e
1
= G f
2
; f
2
= f(t) e
1
= Gf(t)
1
GY S : e(t)
2
e
1/G
1
S
f
: e(t)/G
f
1
= (1/G) e
2
; e
2
= e(t) f
1
= e(t)/G
Transforming asymmetric junction elements through a gyrator.
Constitutive
equations:
Constitutive
equations:
GY
1 2 3
G 1/T
TF GY
1
G/T
3
e
1
= G f
2
; f
2
= (1/T) f
3
e
1
= (G/T) f
3
e
2
= G f
1
; e
3
= (1/T) e
2
f
3
= (G/T) f
1
GY
1 2
GY
G
1
1/G
3
2
TF
1
G
1
/G
2
3
e
1
= G
1
f
2
; f
2
= (1/G
2
) e
3
e
1
= (G
1
/G
2
) e
3
e
2
= G
1
f
1
; f
3
= (1/G
2
) e
2
f
3
= (G
1
/G
2
) f
1
225
Initial bond graph:
Table 5.8
Transforming symmetric junction elements through a gyrator.
Constitutive
equations:
Constitutive
equations:
GY
1 2
G
1
4
3
GY
3
G
1
0
GY
5
2
4
1/G
e
1
=G f
3
; f
4
= f
3
e
1
= G f
3
; f
4
= (1/G)G f
3
e
3
= G f
1
- e
4
e
3
= G(f
1
- (1/G)e
4
)
GY
1 2
G
0
4
3
GY
3
G
1
1
GY
5
2
4
G
e
1
= G(f
3
+ f
4
) e
1
= G f
3
+ G f
4
e
3
= G f
1
; e
4
= G f
1
e
3
= G f
1
; e
4
= G f
1
The equivalent (transformed) active source elements are related to the original source elements in
proportion to the gyrator coefficient, as summarized in table 5.6.
The relation for two-port asymmetric junctions (transformers and gyrators) is summarized in table
5.7. The other possible causal assignments are treated similarly.
Multi-port symmetric junctions (one and zero junctions) may also be transformed as summarized in
table 5.8. Representative causal assignments are shown. The other possible causal assignments are
treated similarly.
As with transforming through a transformer, if a linear passive element is transformed through a
gyrator, the parameter of the equivalent element is related to the parameter of the original element
through the square of the gyrator coefficient. These relations are summarized in table 5.9
226
Table 5.9
Transforming passive elements through a gyrator.
Constitutive
equations:
Constitutive
equations:
Constitutive
equations:
Constitutive
equations:
GY I : I
1 2
G
C: I /G
2
1
df
2 1
dt
=
I
e
1
= G f
2
; e
2
; e
2
= G f
1
de
1 G
2
dt
=
I
f
1
GY C: C
1 2
1/G
1
I : G
2
C
1 de
2 1 1
f
1
=
G
e
2
;
dt
=
C
f
2
; f
2
=
G
e
1
df
1 1
dt
=
G
2
C
e
1
GY R : G
r
1 2
G
R : G
2
/G
r
1
e
2
e
1
= G f
2
; f
2
=
G
r
; e
2
= G f
1 e
1
=
G
r
f
1
G
2
GY R : R
1 2
1/G
R : G
r
/G
2
1
f
1
=
G
e
2
; e
2
= R f
2
; f
2
=
G
e
1
1 1
f
1
=
G
2
e
1
R
These tables are not intended to be memorized; they are presented primarily to demonstrate the
various equivalences. Summarizing: transforming an element through a transformer may change
its parameters but not its form. Transforming an element through a gyrator changes the element to
its dual, and may also change its parameter values. The parameters of passive linear elements are
scaled by the square of the transformer or gyrator parameter. The constitutive equations of other
linear elements are changed in proportion to the first power of the parameter.
These equivalences permit entire subsystems within a bond graph to be transformed through a
transformer or gyrator, thereby enhancing insight into the consequences of energetic interaction and
simplifying the graph and the subsequent process of deriving equations . Beware, however, that the
physical meaning of the quantities represented on the graph may become obscured.
The fact that transforming through a gyrator dualizes an element has some intriguing consequences.
A transformer is, in a sense, a superfluous element, as it could always be represented by a pair of
gyrators. By a similar argument, one of each pair of the other dual elements (0 and 1, C and I, S
e
227
and S
f
) could always be replaced by the other transformed through a gyrator. Thus any bond-
graph composed of the nine primitive elements we have defined so far (R, S
e
, S
f
, 0, 1, C, I, GY, TF)
could be constructed from a set of five primitive elements, for example: (R, S
f
, 0, C, GY). This
possibility and its implications have been explored at some length in the bond-graph literature.
However, most engineers are quite comfortable with the dual concepts of inertia and elasticity,
inductance and capacitance, and so forth, and it is not clear that this further abstraction would
enhance understanding of the physical system.
228
Examples of Equivalences
Electrical Inductor
An electrical inductor is usually implemented as a coil of wire wrapped on a magnetically
permeable core, e.g. as sketched in figure 5.13. Thus it embodies a gyrational transduction
between the electrical and magnetic domains and a capacitive energy storage element in the
magnetic domain, as depicted by the bond graph of figure 5.14. It is most often used as an
electrical circuit component. Assuming magnetic linearity, the equivalent behavior in the
electrical domain is obtained by combining equations 5.33, 5.34 and 5.35:
N
2
=
R
i = L i (5.64)
where L = N
2
/R is the inductance of the coil. Thus the behavior of the electrical inductor arises
from the behavior of a capacitor in the magnetic domain transformed through a gyrator.
Differentiating results in a form of the constitutive equation which may be more familiar.
e = L di/dt (5.65)
Electrical Transformer with Parasitic Energy Storage
A model of an electrical transformer including energy storage in the magnetic field was
presented in figure 5.17, consisting of a pair of gyrators between the electrical and magnetic
domains and a capacitive energy storage element in the magnetic domain. Reflecting the
magnetic capacitor through the gyrator on the right results in the equivalent network shown
below.
e
2
e
1
S
f
N
1
TF 0 S
e
i
2
i
1
i
L2
N
2
N
2
2
I
:
R
Figure 5.30: Bond graph of an equivalent network representing the transformer model of figure
5.17.
The source elements in figure 5.30 represent an assumed current input on the the left and voltage
input on the right. After causality has been assigned, an equation for the current output on the
right may be read from the graph.
N
1
i
2
=
N
2
i
1
- i
L2
(5.66)
229
where i
L2
is the inductor current, determined from the inductor constitutive equation expressed
in integral causal form.
di
L2
N
2
2
dt
=
R
e
2
(5.67)
Differentiating equation 5.66 and substituting equation 5.67 yields equation 5.45 as before.
Alternatively, the magnetic capacitor could also be transformed through the gyrator on the left of
figure 5.17. The resulting equivalent network is shown below.
e
2
e
1
S
f
0
N
1
TF S
e
i
2
i
1
i
L1
N
2
N
1
2
I
:
R
Figure 5.31: Bond graph of an alternative equivalent network representing the transformer
model of figure 5.17.
The same network could be obtained by transforming the inductor of figure 5.30 through the
transformer. Consequently, the values of the two equivalent inductors should be related by the
square of the transformer coefficient, and they are.
L
1
=
N
1
2
R
=
2
N
2
2
R
=
2
N
1
N
1
(
(
(
(
L
2
(5.68)
N
2
N
2
230
Hydraulic Ram
Figure 5.32 shows a sketch of a hydraulic actuator of the kind commonly used in heavy earth-
moving machinery. The difference in fluid pressure acting on the two sides of the piston exerts a
force on the rod. As pressure is defined as force per unit area, the power transduction phenomenon
connecting the hydraulic and translational domains is that of a transformer. In general, the piston
area in the two chambers need not be the same
1
, and the rate of increase of the volume of one
chamber, Q
c1
, need not equal the rate of decrease of the volume of the other chamber, Q
c2
, so two
transformers may be needed to describe the piston. They are connected to a one-junction on the
translational side (representing the unique piston velocity) as shown in figure 5.33(a).
cylinder
rod
fluid lines
(a)
F
v
Q
piston
o-ring
seals
supply
line
return line
P
P P
Q
c1 c2
1
2
1
P
2
(b)
Figure 5.32: Sketch of a hydraulic actuator (a). A cross-sectional view is shown in (b).
1
For example, the sketch shows a rod protruding from each side of the piston, but single-rod
cylinders are also common; with a single rod, the area on one side of the piston exceeds that on the
other side by the area of the rod.
231
If the piston area in each chamber is the same, then the two transformers may be replaced by a
single equivalent transformer as shown in figure 5.33(b). Any increase in volume of one chamber
equals the decrease in volume of the other and the transformer is connected to a one junction in the
fluid domain (representing the common rate of volume change, Q
c
). Constitutive equations are then
as follows.
F = A (P
c1
- P
c2
) (5.69)
Q
c
= A v (5.70)
TF
1
1
TF
2
Q
c1
1
1
Q
c2
A
A
v
TF 1 1
Q
c
A v
(a) (b)
Figure 5.33: Bond graph fragments representing power transduction in the hydraulic actuator.
(a): general model for unequal piston areas. (b): special case of equal piston areas.
Parasitic Dynamics
Like other power transduction devices, a hydraulic ram also embodies energy storage and
dissipation effects which may need to be modeled. For simplicity, we will assume the hydraulic
fluid to be incompressible (though that is often one of the prominent "parasitic" dynamic
phenomena) and model only the inertial effects due to acceleration of the fluid and the power
dissipation due to its motion through the lines and orifices. On the mechanical side we will model
the inertia of the piston and rods (considered to be a single rigid body) and the friction due to the
sliding seals, described for simplicity as a single linear friction element (though in reality, a
significant component of the dissipative behavior is likely to resemble dry friction).
Friction due to flow in a pipe was considered in Chapter 4. A crude model of the inertial effects
due to acceleration of the fluid may be derived as follows. Consider an incompressible fluid
moving in a pipe of uniform cross-sectional area, A
p
. The mass, m, of the "slug" of fluid between
two cross sections separated by a distance l is determined by the dimensions of the pipe and the
density, .
m = lA
p
(5.71)
Considerable simplification is obtained by assuming that the fluid flow velocity has the same value,
v, at all points. That velocity is determined by the volumetric flow rate, Q, divided by the cross
sectional area.
v = Q/A
p
(5.72)
Given these assumptions, the kinetic co-energy of the moving fluid is:
232
* m
lA
p Q
2
l
E
k
=
2
v
2
=
2
A
p
2
=
A
p
Q
2
(5.73)
Thus, by analogy with a rigid body in translation, the energy stored in the moving fluid may be
described by an ideal inertia, characterized by a linear relation between a flow, Q, and a generalized
momentum, , termed pressure momentum.
Q = (5.74)
I
The parameter I = l/A
p
is the inertance of the fluid, determined by the geometric and material
properties of the pipe and the fluid. From the fundamental definition of generalized momentum, the
pressure difference accelerating the fluid is the time derivative of the pressure momentum.
P = d/dt (5.75)
Thus pressure difference is proportional to the rate of change of fluid flow rate.
l dQ
P =
A
p
dt
(5.76)
A bond graph of the complete system is shown in figure 5.34(a). For the purpose of this example,
it has been assumed that the supply line is connected to a power source at constant pressure, P
1
, the
return line to a sump at constant pressure, P
2
, (e.g. atmospheric pressure) and that the mechanical
load is a constant force, F.
We may analyze this system by assigning causality. After assigning the required causality to the
source elements, we are free to assign integral causality to one of the energy storage elements
(arbitrarily chosen in figure 5.34(b) to be the translational inertia). Propagating the causal
consequences of that choice through the graph completes the causality assignment.
Causality assignment reveals that two of the three inertias are dependent energy-storage elements.
The velocity of the piston determines the flow rate in the fluid lines and a single variable is
sufficient to characterize the kinetic energy stored in both the translating masses and the moving
fluid. In effect, the piston and rods and the fluid in the lines have been described as a single "rigid
body" even though the fluid in the lines would hardly be considered rigid in the usual sense of the
word.
233
P : S
e
1
TF 1 1
Q
c
A
v
1
1
Q
Q
2
1
R
I
R
I
R : b
: m
:
1
I
R :
2
R :
1
S
e
: F
P : S
e
2
I
:
I
2
Hydraulic domain Translational domain
(a)
I
S
e
TF 1 1
1
1
R
R
I
R
S
e
S
e
I
(b)
Figure 5.34: (a) Bond graph of a model of the hydraulic actuator including parasitic effects. (b)
Bond graph with causality assigned.
The parasitic fluid elements may be transformed into the translational domain. The fluid inertias are
equivalent to translational masses. Combining equations 5.69, 5.70 and 5.76 the effective mass due
to each fluid line is
m
eff
= A
2
I = A
2
l/A
p
(5.77)
The fluid resistors are equivalent to linear translational friction elements. The effective friction
coefficient due to each fluid line is
b
eff
= A
2
R (5.78)
234
I
S
e
I
I
1 TF 1 S
e
S
e
R
R
R
(a)
S
e
P
1
:
I : m + A I + A I
2 1
2 2
TF 1 1
R
S
e
S
e
F :
P
2
A
:
: b + A R + A R
1 2
2 2
(b)
Figure 5.35: Bond graph of figure 5.34 with (a) parasitic fluid elements transformed into the
translational domain and (b) inertial and dissipative elements combined.
When the parasitic fluid elements are transformed into the translational domain it becomes clear that
all of the parasitic elements are connected to a single one-junction as shown in figure 5.35(a). The
compatibility equation associated with the one-junction means that forces sum at the one-junction.
As a result, the parasitic elements may be combined as shown in figure 5.35(b) into a single
effective inertia with an apparent mass of
m
app
= m + A
2
I
1
+ A
2
I
2
(5.79)
and a single effective friction element with an apparent friction coefficient of
b
app
= b + A
2
R
1
+ A
2
R
2
(5.80)
235
Momentum-based fluid-mechanical power transduction.
A domestic vacuum cleaner is a good example of a consumer product that fundamentally
involves power transduction between the electrical, mechanical and fluid domains. How should
the transduction mechanism of the pump be modeled? The best answer is based on
understanding the basic fluid-mechanical transduction phenomenon in a turbine or centrifugal
pump. This is due to a change of fluid momentum which requires a force or torque. Thus fluid
flow is related to rotational effort; this transduction phenomenon may be modeled by a gyrator.
Suppose you didn't see this; how could you deduce it? One way is to consider what is required
to reproduce observable behavior of the vacuum cleaner. A key clue is the observation that
when the air hole of a vacuum cleaner is blocked the vacuum cleaner speeds up. Let's try to
model this phenomenon.
Motor: Assume no significant electrical dynamics; model the electric motor as a torque source;
include motor inertia, bearing friction.
Se
I
R
1 to pump
:B
m
:J
m
t
m
:
Fluid flow: Include orifice resistance, neglect hose capacitance and fluid inertia.
from
pump
R:R
orifice
First, try modeling the pump as a transformer. The blocked flow boundary condition may be
represented as a flow source with zero flow as follows.
1 Se
I
R
1 TF
R
Sf:0
:J
m
:B
m
:R
orifice
t
m
:
236
Propagating causality, it can be seen that this model would predict that blocking the air hole
would stall the motor.
Now try modeling the pump as a gyrator:
1 Se
I
R
1 GY
R
Sf:0
:J
m
:B
m
:R
orifice
t
m
:
In this case, propagating causality, it can be seen that this model would predict that blocking the
air hole will not stall the motor. If we remove all bonds with zero power flow (shown in the
following as dashed lines) we can see that the effect of blocking the flow is to eliminate the
influence of the fluid resistance on the motor; i.e. the load on the motor decreases.
1 Se
I
R
1 GY
R
Sf:0
:J
m
:B
m
:R
orifice
t
m
:
Influence of leakage flow
One might reasonably argue that this marked difference between the predictions of the two
models of the pump is an artifact of neglecting the possibility of leakage flow through the pump.
If the motor were immobilized, it would still be possible to move air through the pump, yet the
transformer model above would predict otherwise. To rectify that error we might include a
leakage flow in the fluid flow model as follows:
from
pump
0
R:R
leakage
R:R
orifice
237
Now if we model the pump as a transformer and impose a blocked-flow boundary condition we
find that the motor is not immobilized. Removing all bonds with zero flow (shown as dashed
lines) we find that blocked flow eliminates the influence on the motor of orifice resistance but
not leakage resistance.
Se
I
R
1 TF 1 0
R
Sf:0
R
:J
m
:B
m
:R
orifice
t
m
:
R
leakage
:
Modeling the pump as a gyrator yields a superficially similar result.
Se
I
R
1 GY 1 0
R
Sf:0
R
t
m
:
:J
m
:B
m
However, the fundamental difference between these two models may be seen by considering the
equivalent rotational systems. If the pump is modeled as a transformer and a blocked-flow
boundary condition is imposed, the equivalent rotational system is as follows.
Se
I
R
1 1 0
R
Sf:0
R
::
B
eq.,
leakage
eq.,
orifice
B
t
m
:
:J
m
:B
m
This may be easier to understand and visualize when represented using more conventional
graphic symbols for rotational systems. First, without the flow blocked:
238
J
m
B
m
eq.,
orifice
B
t
m
B
eq.,
leakage
Then with the flow blocked as follows:
J
m
B
m
eq.,
orifice
B
t
m
B
eq.,
leakage
blocked flow is equivalent to
immobilizing this component
From this we may see that blocking the flow is equivalent to increasing the value of B
eq., orifice
without limit, which increases the load on the motor.
If the pump is modeled as a gyrator and a blocked-flow boundary condition is imposed, the
equivalent rotational system is as follows.
Se
I
R
1 1 0
R
Se:0
R
::
B*
eq.,
leakage
eq.,
orifice
B*
t
m
:
:J
m
:B
m
To denote the fact that the equivalent rotational dissipators have different values form the
previous case, they have been denoted by B*
eq., leakage
and B*
eq., orifice
respectively. Again,
it may be easier to understand and visualize this using the more conventional graphic symbols
for rotational systems. First, without the flow blocked:
239
J
m
B
m
eq., orifice
B*
t
m
+B*
eq., leakage
Then with the flow blocked:
J
m
B
m
eq., orifice
B*
t
m
+B*
eq., leakage
blocked flow is equivalent to
disconnecting this component
Note that in contrast to the previous case, blocking the flow is equivalent to decreasing the value
of B
eq., orifice
without limit, which decreases the load on the motor.
240
STATE EQUATION DERIVATION
Summary of Basic Bond Graph Elements
A large class of physical systems may be described using the basic lumped parameter elements
ideal active and passive one-port elements connected by multiport junction elements. The
variables and primitive elements of the energy-based formalism are summarized in the following
tables.
Table 6.1 Fundamental variables
Energy and power:
E - E
o
=
]
(P dt
Conjugate power variables: P = e f
Conjugate energy variables:
p - p
o
=
]
(e dt
q - q
o
=
]
(f dt
Table 6.2 Active one-port elements (sources or boundary elements)
effort source flow source
constitutive equation e = e(t) f = f(t)
bond graph symbol
S
e
S
f
Table 6.3 Passive one-port element constitutive equations
capacitor inertia resistor
general form
e = u(q) f = +(p) e = I(f)
linear form e = q/C f = p/I e = Rf
bond graph symbol
C I R
241
Table 6.4 Symmetric multi-port junction elements
common flow junction common effort junction
constitutive equations f
i
= f
j
, j = 1,n
_
j=1
n
o
j
e
j
= 0
e
i
= e
j
, j = 1,n
_
j=1
n
o
j
f
j
= 0
bond graph symbol
1
1
2
n
3
0
1
2
n
3
where o
j
=
+1 if power positive in
1 if power positive out
Table 6.5 Asymmetric two-port junction elements
transformer gyrator
constitutive equations e
i
= T e
j
f
j
= T f
i
e
i
= G f
j
e
j
= G f
i
bond graph symbol
TF GY
Sign Convention
Half arrows denote the direction of positive power flow. For passive elements power flow is
positive inwards. For active elements there is no fixed convention, though it is common to
denote power flow as positive outwards. Multiport junction elements may serve to define effort
or flow differences (e.g. pressure difference, relative motion). In that case the following sign
convention is recommended.
242
Causality Assignment
The nine primitive elements may be used to represent energy storage and power flow within an
energetic system. The energetic interactions represented by the elements constrain (but do not
uniquely determine) the mathematical operations which may be used to describe the system
behavior. The causality assignment procedure identifies those constraints. The procedure is
repeated below (this time including the gyrator and transformer juction elements).
1. All elements in a bond graph which are constrained to have an unique causality must have
the appropriate causality assigned and denoted by a causal stroke at the appropriate end of
the associated bond.
1a. Active one-port elements (e.g. power sources) are always constrained to an unique
causality.
1b. Remember that in some cases passive one-port elements may be constrained to an
unique causality because of their constitutive equations; for example, this is often true
of resistors.
2. Examine the junction elements connected to all elements with causal assignments. Propagate
causal assignments through the junction structure until no further causal assignments can be
made.
2a. If a one-junction has its unique flow determined by one of its bonds, then all other
bonds must be assigned flow output causality.
2b. If a zero-junction has its unique effort determined by one of its bonds, then all other
bonds must be assigned effort output causality.
2c. If a transformer has a flow input determined on one of its bonds, then a flow output
must be assigned to its other bond. Equivalently, if an effort input is determined on one
of its bonds, then an effort output must be assigned to its other bond.
2d. If a gyrator has a flow input determined on one of its bonds, then an effort output must
be assigned to its other bond. Conversely, if an effort input is determined on one of its
bonds, then a flow output must be assigned to its other bond.
3. Choose one of the energy storage elements which does not already have a causal assignment
(it doesn't matter which one) and assign it integral causality (effort input to an inertia, flow
input to a capacitor).
3a. Return to step 2 and propagate this causal assignment through the junction structure
until no further causal assignments can be made.
3b. Repeat steps 3 and 3a until all energy-storage elements have a causal assignment.
243
4. Any elements remaining which have not been given a causal assignment are dissipators
which are indifferent to their causal assignment. Choose one or those dissipators (it doesn't
matter which one), assign it an arbitrary causality
4a. Return to step 2 and propagate this causal assignment through the junction structure
until no further causal assignments can be made.
4b. Repeat steps 4 and 4a until all elements have a causal assignment.
If any of the steps in this procedure cannot be completed without violating the causal constraints
imposed by the junction structure, the model is operationally inconsistent. At least one quantity
has been determined by more than one relationship. This is an error which must be corrected.
In addition to uncovering logical inconsistencies in the system model, causality assignment also
serves to identify dependent and independent energy storage elements. If, in the process, any
energy storing element is assigned derivative causality, then that is a dependent storage element.
Its stored energy is determined by the variables associated with the element from which the
causal propagation began.
Derivative causality on an energy storage element is not an error but it can have undesirable
consequences; it may lead to extensive algebra in deriving state equations; it may also lead to
numerical difficulties when simulating system behavior on a computer. If one or more energy
storage elements in a model have derivative causality, you may want to modify the model to
eliminate the dependencies, but this is not essential; indeed it is not always possible.
Causality Assignment and State Variables
Models composed of assemblies of the nine primitive elements are state-determined. To derive
state equations, state variables must be chosen. In general, state variables required to describe a
system are not unique; many equivalent sets may be identified and the best choice depends on
the problem under consideration.
For an energetic system the state variables must at least uniquely define the energy stored in the
system. The minimum number of state variables required is determined by the number of
independent energy storage elements in the system model. Those energy-storage elements which
have been assigned integral causality are independent. The energy stored in the element at a
given time may be prescribed as an independent initial condition and gives rise to an
independent state variable.
244
Substitution Methods for Deriving System Equations
A system model represented by a bond graph with a complete and unambiguous causal
assignment may be used to derive system equations by a straightforward method of successive
substitution as follows.
Models with Linear Energy Storage Elements: Power Variables
In the important special case in which all of the model elements are linear, a convenient choice
of state variables is the efforts and flows associated with independent energy storage elements
known as power state variables or circuit state variables. That choice is the basis of the
following efficient, systematic procedure for deriving state equations by successive substitution.
5. Choose power state variables. These are the output variables from the independent energy
storage elements, and the power conjugate variables are the input variables.
5a. efforts associated with independent capacitors.
5b. flows associated with independent inertias.
6. Using the constitutive equation for the corresponding energy storage element, write the rate
of change of each state variable as an explicit function of its power conjugate variable.
de f
6a. independent capacitor: =
dq 1
=
dt dt C C
df e
6b. independent inertia: =
dp 1
=
dt dt I I
7. Reading from the causal strokes on the graph, write an expression for the input variable
for each independent energy storing element as a function of the output variables from one or
more of the other elements in the system.
7a. This step proceeds by direct substitution using the constitutive equation(s) for each of
the elements as needed, expressed in the input/output form specified by the causal
strokes on the graph.
7b. Continue substitution until the input to each independent energy storing element is
expressed as a function only of one or more of the state variables or of the independent
variables associated with sources.
This seven-step causality assignment and substitution procedure will result in state equations for
the system model. With a little practice, in most cases state equations can be written by
inspection from the causally augmented bond graph of the system model.
245
Example: Electric Circuit
An electric circuit schematic is shown in figure 6.1a. This schematic is already a model of the
system and specifies the energetic elements to be used: an inertia, capacitor, resistor and effort
source. It is evident from the schematic that all elements share the same flow and thus are
connected by a single one-junction. A bond graph for the circuit, with sign convention
included, is shown in figure 6.1b.
Figure 6.1a Figure 6.1b
Assume all elements have linear constitutive equations. The effort source must impress an effort
on the rest of the system as shown in figure 6.1c. Assigning integral causality to the inertia and
propagating that causal assignment through the one-junction completes the causal assignment as
in figure 6.1d.
Figure 6.1c Figure 6.1d
From the causal graph, the capacitor and inductor are independent energy storage elements.
Power (or circuit) state variables are the capacitor voltage, e
C
, and the inductor current, i
L
.
From the constitutive equations for these elements:
d
dt
e
C
=
i
C
C
(6.1)
d
dt
i
L
=
e
L
L
(6.2)
246
Reading from the causal graph, the inductor determines the current associated with the one-
junction, which in turn determines the current for the other connected elements. Thus,
i
C
= i
L
(6.3)
Substituting, one state equation is:
d
dt
e
C
=
i
L
C
(6.4)
Again reading from the causal graph, the inductor voltage is determined by the other three one-
port elements.
e
L
= e
S
e
C
e
R
(6.5)
Note that the signs in this junction equation are read directly from the half-arrows on the graph.
In words, voltage into the inductor equals voltage into the junction from the source minus the
voltages out of the junction into the capacitor and the resistor.
From the causal graph, the appropriate form of the constitutive equation for the dissipator is
resistance causality.
e
R
= R i
R
(6.6)
Again reading from the causal graph, the resistor current is determined by the inductor current.
i
R
= i
L
(6.7)
Substituting, the second state equation is:
d 1
dt
i
L
=
L
(e
S
e
C
R i
R
) (6.8)
Note that the substitution process stops when the rate of change of a state variable has been
expressed as a function of state and input variables. The state equations may be written in the
conventional vector/matrix form as follows.
(
(
d

e
C(
dt
i
L
=
0 1/C
1/L R/L
(
(
(
(
e
C(
i
L
+
0
1/L
(
(
e
S
(6.9)
247
State Variables and Energy
Causality assignment procedures may be used to identify a minimum set of variables necessary
to define the energetic state of a system. However, it is important to recognize that everything of
interest about a system will not always be determined by its energetic state. We will frequently
need to choose state variables in addition to those associated with independent energy storage
elements. This is illustrated in the following example.
Example: A Trivial Mechanical System
rigid
body
F
applied
force
F
:
S
e
1
v
I
:
m
x
(a) (b)
Figure 6.3
Consider a rigid body subject to an applied force as depicted in figure 6.3a. A corresponding
bond graph with casual assignment is shown in figure 6.3b. There is a single independent energy
storage element in this system and its energetic state is determined by a single variable, either the
momentum of the rigid body which determines the kinetic energy
p
2
E
k
=
2m
(6.37)
or the coresponding speed which determines the kinetic co-energy
E
*
k
=
1
2
mv
2
(6.38)
A corresponding state equation is as follows.
d
dt
v =
F
m
(6.39)
However, we will often be concerned with the location of the rigid body, not just with its speed
or energy. In that case we will need an additional state variable and a suitable set of state
equations is as follows.
(
(
0 1

(
(
(
(
0
1/m
F (6.40)
d
dt
(
(
x x
v 0 0 v
248
By a similar argument, in other cases we may need other state variables (e.g. the time integral of
position, etc.). Thus the state variables identified by the causality assignment procedures are
merely a minimum set required to define the system energy.
No Fixed Rules!
To summarize: there are no fixed rules for choosing state variables. Using the substitution
methods reviewed above, the following guidelines are offered: generally, for systems with linear
energy-storage elements, power variables are appropriate; for systems with nonlinear energy-
storage elements, energy variables are recommended. For mechanical systems, Lagrangian
variables are recommended. Most important, use variables that make the most physical sense to
you.
249
Models with Nonlinear Energy Storage Elements: Energy Variables
If any of the energy storage elements in a model have nonlinear constitutive equations, then
power or circuit variables may be a poor choice for the state variables associated with those
elements. This is because differentiating a nonlinear constitutive equation (in step 6 above) will
not necessarily result in a function only of power variables and their rates of change. For
example, consider a nonlinear capacitor:
de
dt
=
?F
?q
(q)
dq
dt
=
?F
?q
(q) f (6.10)
Now it is necessary to substitute for the variable q as well as the variable f. One might try to do
so by inverting the capacitor constitutive equation.
de
dt
=
?F
?q
( ) F
-1
(e) f (6.11)
However, there are two problems with this approach: First of all, the required inverse function
may not exist. Secondly, even if it does, the required algebra may be quite tedious. A better
alternative is to choose different state variables: the displacements and momenta associated with
independent energy storage elements known as energy state variables or Hamiltonian state
variables. Steps 5 and 6 of the substitution procedure are changed as follows.
5. Choose energy state variables. These are the displacements associated with independent
capacitors and the momenta associated with independent inertias. The rate of change of each
state variable is equal to the input variable to the corresponding independent energy storage
element.
5a. independent capacitor: dq/dt = f
5b. independent inertia: dp/dt = e
6. Using its constitutive equation, write the output variable for each independent energy storage
element as a function of the corresponding state variable.
6a. independent capacitor: e = u(q)
6b. independent inertia: f = +(p)
The rest of the substitution procedure remains unchanged.
Example: Nonlinear Electric Circuit
Consider the electric circuit of figure 6.1a but assume that the capacitor and inductor have the
following nonlinear constitutive equations. For the capacitor:
250
e
C
=
1
C
q
C
q
s
q
s
2
q
C
2
(6.12)
where q
s
is the saturation charge, (a constant) the maximum charge which may be stored in the
capacitor. Note that for q
C
<< q
s
the constitutive equation reduces to that of an ideal (linear)
capacitor.
lim e
C
q
C
A0
=
q
C
C
(6.13)
For the inductor:
i
L
=
1
L
s
2

L
2
(6.14)
where
s
is the saturation flux linkage, (a constant) the maximum flux linkge which may stored
in the inductor. For
L
<<
s
the constitutive equation reduces to that of an ideal (linear)
inductor.
lim i
L
l
L
A0
=
L
L
(6.15)
If we were to choose the capacitor voltage as a state variable, differentiating the constitutive
equation would result in the following relation.
d
dt
e
C
=
\
|
.
|
| q
s
C(q
s
2
q
C
2
)
1/2
+
q
C
2
q
s
C(q
s
2
q
C
2
)
3/2
i
C
(6.16)
The nonlinear capacitor equation may be inverted as follows.
q
C
=
q
s
e
C
C
q
s
2
+ e
C
2
C
2
(6.17)
Therefore, in this case, the charge q
C
may be eliminated from equation 6.16 and the capacitor
voltage could be used as a state variable. The resulting state equation is a little intimidating:
d
dt
e
C
=
\
|
.
|
|
q
s
C
\
|
.
|
|
q
s
2

q
s
2
e
C
2
C
2
q
s
2
+ e
C
2
C
2
1/2
+
\
|
.
|
| q
s
2
e
C
2
C
2
q
s
2
+ e
C
2
C
2
q
s
C
\
|
.
|
|
q
s
2

q
s
2
e
C
2
C
2
q
s
2
+ e
C
2
C
2
3/2
i
C
(6.18)
251
By a similar argument, the inductor current could also be used as a state variable. However, it is
far simpler to use the capacitor charge and the inductor flux linkages as state variables.
d
dt
q
C
= i
C
(6.19)
d
dt

L
= e
L
(6.20)
Reading the junction equation from the causal graph (figure 6.1d) i
C
= i
L
. Using the nonlinear
inductor constitutive equation (6.14) we obtain one state equation.
dq
C
dt
=
1
L
s
2

L
2
(6.21)
As before, reading from the causal graph, the inductor voltage is determined by the other three
one-port elements
e
L
= e
S
e
C
e
R
(6.22)
Substitute for e
C
and e
R
in equation 6.22 using the constitutive equations of the nonlinear
capacitor (6.12) and the resistor (6.6).
d
dt

L
= e
S

1
C
q
C
q
s
q
s
2
q
C
2
R i
R
(6.23)
Reading the junction equation from the causal graph, the resistor current is determined by the
inductor current i
R
= i
L
. Substituting using the inductor constitutive equation, the second state
equation is:
d
dt

L
= e
S

1
C
q
C
q
s
q
s
2
q
C
2
R
s
2

L
2
(6.24)
As before, the substitution process stops when the rate of change of a state variable has been
expressed as a function of state and input variables. Note that in this nonlinear system, there is
no clear way to express the state equations in vector/matrix form.
Energy variables may also be used for systems composed exclusively of linear elements. For
this reason, energy variables have been proposed as the exclusive choice of state variables for
systems
1
represented by bond graphs. However, there are several reasons why this is not
recommended:
1
Karnopp, Dean and Rosenberg, Ronald (1975) System Dynamics: A Unified Approach, John
Wiley & Sons, New York.
252
physical system behavior is fundamentally independent of the choice of variables, so we
should expect no universal rule for choosing state variables.
energy variables may needlessly complicate the equation derivation process.
energy variables may be less familiar and less comprehensible than the corresponding power
variables.
This last point may seem trivial; in fact, it is probably the most important. One of the primary
reasons for developing models is to enhance understanding. For most of us, the current in an
inductor is more meaningful than the corresponding flux linkage; the speed of a mass is more
readily visualized than its momentum.
253
Models of Mechanical Systems: Lagrangian Variables
For (sub)system models of phenomena in the domain of mechanical translation or rotation,
neither power variables nor energy variables are necessarily the best choice. Mechanical inertias
have linear constitutive equations (except at speeds approaching light speed a rare occurence
for systems of interest to a practicing engineer). Thus it is convenient to work with flows
(speeds) of independent inertias as state variables. On the other hand, mechanical capacitors
elastic elements frequently have nonlinear constitutive equations. Even in the linear case,
rather than work with efforts, it is generally easier to identify displacements and relate their time
derivatives to the speeds of associated inertias. For mechanical systems the most convenient
state variables are the flows of independent inertias and the corresponding displacements
which may be termed Lagrangian state variables. An obvious hybrid of the above two
procedures can be used to derive state equations. It is illustrated by the following simple
example.
Example: Linear Mechanical System
rigid
body
spring
rigid
body
spring
x
x
1
2
F
applied
force
x
2
stationary frame
Figure 6.2a
A mechanical system schematic is shown in figure 6.2a. This schematic represents a model of
the system and specifies the lumped-parameter energetic elements to be used. A corresponding
bond graph is shown in figure 6.2b. It is evident that a single displacement or speed is common
to the leftmost spring and inertia which are therefore connected by a one junction. Similarly, a
single displacement or speed is common to the rightmost inertia and the source of the applied
force which are therefore also connected by a one junction. The force in the remaining spring is
applied to both inertias and a zero junction connects these three elements as shown. Note that a
sign convention is chosen so that the displacement of the rightmost spring is determined by the
difference of the displacements of the two inertias.
0
C
1 1 S
e
v
1
v
2
1/k
1
: :
F
I
: m
1
C
:
1/k
2
I
: m
2
254
Figure 6.2b
Assume all elements have linear constitutive equations. Causality is assigned following the
procedure described above with the result shown in figure 6.2c.
0 C 1 1 S
e
v
1
v
2
1/k
1
: :
F
I
: m
1
C
:
1/k
2
I
: m
2
Figure 6.2c
From the causal graph, all energy storage elements are independent. Choose as state variables
the speeds of the two inertias, v
1
and v
2
, and the corresponding displacements, x
1
and x
2
, shown
in figure 6.2a. The first two state equations are:
d
dt
x
1
= v
1
(6.25)
d
dt
x
2
= v
2
(6.26)
Subscripting variables in an obvious way, from the constitutive equations of the inertias we may
write:
d
F
1
dt
v
1
= (6.27)
m
1
d
F
2
dt
v
2
= (6.28)
m
2
Reading from the causal graph, the force on the leftmost inertia is determined by the two springs.
F
1
= F
k2
F
k1
(6.29)
The constitutive equations of the springs relate these forces to the corresponding displacements.
F
k1
= k
1
x
1
(6.30)
F
k2
= k
2
Ax
2
(6.31)
The displacement of the rightmost spring may be related to the positions of the inertias by
reading directly from the causal graph.
255
Ax
2
= x
2
x
1
(6.32)
Substituting yields a third state equation.
d 1
dt
v
1
=
m
1
[
k
2
(x
2
x
1
) k
1
x
1]
(6.33)
Reading from the causal graph, the force on the rightmost inertia is determined by the associated
spring and the applied force.
F
2
= F F
k2
(6.34)
Substituting the constitutive equation of the spring yields the fourth state equation.
d 1
dt
v
2
=
m
2
[
F k
2
(x
2
x
1
)
]
(6.35)
The state equations may be written in standard vector/matrix form as follows.
x
1
x
2
v
1
v
2
(
(
(
0 0 1 0
0 0 0 1
(k
1
+k
2
)/m
1
k
2
/m
1
0 0
k
2
/m
2
k
2
/m
2
0 0
(
(
(
x
1
x
2
v
1
v
2
0
d
dt
(
(
(
+
(
(
(
1/m
2
0
F (6.36)
0
256
December 6, 1994 page 1 Neville Hogan
Ideal asymmetric junction elements
Relax the symmetry assumption and examine the resulting junction structure. For
simplicity, consider two-port junction elements.
As before, assume instantaneous power transmission between the ports without
storage or dissipation of energy. Characterize the power flow in and out of a two-
port junction structure using four real-valued wave-scattering variables. Using
vector notation:
u =
(
( u
1
u
2
(A.1)
v =
(
( v
1
v
2
(A.2)
The input and output power flows are the square of the length of these vectors,
their inner products.
P
in
=
_
i=1
2
u
i
2
= u
t
u (A.3)
P
out
=
_
i=1
2
v
i
2
= v
t
v (A.4)
The constitutive equations of the junction structure may be written as follows.
v = f(u) (A.5)
Geometrically, the requirement that power in equal power out means that the
length of the vector v must equal the length of the vector u, i.e. their tips must lie
on the perimeter of a circle (see figure A.1).
For any two particular values of u and v, the algebraic relation f(
.
) is equivalent to
a rotation operator.
v = S(u) u (A.6)
where the square matrix S is known as a scattering matrix.
257
v
1
2
v
u
1
u
2
u
v
S need not be a constant matrix, but may in general depend on the power flux
through the junction, hence the notation S(u). However, S is subject to important
restrictions. In particular,
v
t
v = u
t
S
t
Su = u
t
u (A.7)
S is ortho-normal matrix: the vectors formed by each of its rows (or columns) are
(i) orthogonal and (ii) have unit magnitude; its transpose is its inverse.
S
t
S = 1 (A.8)
This constrains the coefficients of the scattering matrix as follows.
S =
(
( a b
c d
(A.9)
a
2
+ c
2
= 1 (A.10)
ab + cd = 0 (A.11)
b
2
+ d
2
= 1 (A.12)
258
As there are only three independent equations and four unknown quantities, we see
that this junction is characterized by a single parameter. We may also write the
orthogonality condition as
SS
t
= 1 (A.13)
which yields the following equations.
a
2
+ b
2
= 1 (A.14)
ac + bd = 0 (A.15)
c
2
+ d
2
= 1 (A.16)
There are four possible solutions to these equations. Combining A.10 and A.16,
a
2
= 1 - c
2
= d
2
. Thus a = d.
If a = d then b =c = 1 - a
2
.
One solution
Choosing the positive root yields one solution. Assuming the coefficient a to be
the undetermined parameter,
S =
(
(
(
a 1 - a
2
- 1 - a
2
a
(A.17)
Rewrite in terms of effort and flow variables.
e =
(
( e
1
e
2
(A.18)
f =
(
( f
1
f
2
(A.19)
The relation between efforts and wave-scattering variables is as follows.
e = (u - v) c = c (1 - S) u (A.20)
where c is a scaling constant. The relation between flows and wave-scattering
variables is as follows.
259
f = (u + v)/c = 1/c (1 + S) u (A.21)
If |a| = 1 then 1 + S and 1 - S are nonsingular matrices and the input wave
scattering variables u
1
and u
2
may be eliminated as follows.
e = c
2
(1 - S) (1 + S)
-1
f (A.22)
(
(
( e
1
e
2
= c
2
(
(
(
0 - (1 - a)/(1 + a)
(1 - a)/(1 + a) 0

(
(
( f
1
f
2
(A.23)
Writing G = c
2
(1 - a)/(1 + a) we obtain the equation for an ideal gyrator.
(
(
( e
1
e
2
=
(
(
( 0 -G
G 0

(
(
( f
1
f
2
(A.24)
Note that equations A.20 and A.21 imply a sign convention in effort-flow
coordinates such that power is positive inwards on both ports.
P
net inwards
= e
t
f = u
t
u - v
t
v (A.25)
To follow the more common sign convention we may simply change the sign of f
2
in equation A.24.
If a = 1, e is identically zero for all values of f. No energy is exchanged between
the ports and the junction structure behaves like a dissipator with zero resistance.
If a = -1, f is identically zero for all values of e. No energy is exchanged between
the ports and the junction structure behaves like a dissipator with infinite resistance
(zero conductance).
260
A second solution
Choosing a = d and using the negative root yields another solution. Again
assuming the coefficient a to be the undetermined parameter,
S =
(
(
(
a - 1 - a
2
1 - a
2
a
(A.26)
In this case the relation between efforts and flows is
(
(
( e
1
e
2
= c
2
(
(
(
0 (1 - a)/(1 + a)
- (1 - a)/(1 + a) 0

(
(
( f
1
f
2
(A.27)
Again we obtain the equation for an ideal gyrator.
(
(
( e
1
e
2
=
(
(
( 0 G
-G 0

(
(
( f
1
f
2
(A.28)
261
A third solution
If a = - d, b = c = 1 - a
2
. Using the positive root and assuming a to be the
undetermined parameter
S =
(
(
(
a 1 - a
2
1 - a
2
-a
(A.29)
In this case the matrices 1 + S and 1 - S are singular for all values of the parameter
a.
However, equations A.20 and A.21 may be combined as follows:
(
(
(
e
1
/c
e
2
/c
cf
1
cf
2
=
(
(
(
1 - S
-----
1 + S

(
(
( u
1
u
2
(A.30)
(
(
(
(
e
1
/c
e
2
/c
cf
1
cf
2
=
(
(
(
(
1 - a - 1 - a
2
- 1 - a
2
1 + a
1 + a 1 - a
2
1 - a
2
1 - a
(
(
( u
1
u
2
(A.31)
If |a| = 1, the 4 x 2 matrix relating efforts and flows to the input scattering variables
contains two nonsingular 2 x 2 submatrices.
(
(
( e
2
/c
cf
1
=
(
(
(
- 1 - a
2
1 + a
1 + a 1 - a
2
(
(
( u
1
u
2
(A.32)
(
(
( e
1
/c
cf
2
=
(
(
(
1 - a - 1 - a
2
1 - a
2
1 - a

(
(
( u
1
u
2
(A.33)
Solving the second of these for u and substituting into the first we obtain a relation
between efforts and flows.
262
(
(
( e
2
f
1
=
(
(
(
- (1 + a)/(1 - a) 0
0 (1 + a)/(1 - a)

(
(
( e
1
f
2
(A.34)
Writing T = (1 + a)/(1 - a) we obtain the equation for an ideal transformer.
(
(
( e
2
f
1
=
(
(
( -T 0
0 T

(
(
( e
1
f
2
(A.35)
To follow the more common sign convention we may change the sign of e
2
.
If the parameter a = 1, an argument similar to that used above shows that a
degenerate case results in which no energy is exchanged between the ports.
263
Final solution
Choosing a = d and using the negative root we obtain the fourth solution.
S =
(
(
(
a - 1 - a
2
- 1 - a
2
-a
(A.36)
Once again, the matrices 1 + S and 1 - S are singular for all values of the parameter
a, but by rearranging equations A.20 and A.21 as before the corresponding relation
between efforts and flows is
(
(
( e
2
f
1
=
(
(
(
(1 + a)/(1 - a) 0
0 - (1 + a)/(1 - a)

(
(
( e
1
f
2
(A.37)
Again we obtain the equation for an ideal transformer
(
(
( e
2
f
1
=
(
(
( T 0
0 -T

(
(
( e
1
f
2
(A.38)
264
Two-port junction elements
There are only two possible power-continuous, asymmetric two-port junction
elements, the gyrator and the transformer.
Unlike the ideal symmetric junction elements (0 and 1) the ideal asymmetric
junction elements may be nonlinear.
The relation between efforts and flows must have a multiplicative form.
The general asymmetric junction elements are a modulated gyrator (MGY) and a
modulated transformer (MTF) respectively.
265
Junction elements in network models.
Classify by number of ports and examine the possible structures that result.
Using only one-port elements, no more than two elements can be assembled.
Combining two two-ports yields another two-port.
At most two one-port elements could be connected.
Thus three-port junction elements are necessary. They are also sufficient insofar as
junctions with more than three ports may be constructed from assemblages of
three-ports.
An assembly of two three-port elements yields a four-port element.
Using two- and three-port junctions, an arbitrary number of one-port (or n-port)
elements may be assembled into a system.
Three-port junction elements
A junction structure or junction element is treated as a system with certain special
properties which are derived from energetic considerations.
266
An ideal junction structure is power-continuous:
power is transmitted instantaneously without storing or dissipating energy.
instantaneous power out equals instantaneous power in.
Power flow in and out of a three-port may be characterized by six real-valued
wave-scattering variables
u
1
, u
2
, u
3
, define the input power flow
v
1
, v
2
, v
3
, define the output power flow
Using vector notation:
u =
(
(
u
1
u
2
u
3
(A.1)
v =
(
(
v
1
v
2
v
3
(A.2)
The input and output power flows are the square of the length of these vectors,
their inner products.
P
in
=
_
i=1
3
u
i
2
= u
t
u (A.3)
P
out
=
_
i=1
3
v
i
2
= v
t
v (A.4)
267
Constitutive equations of the junction structure may be written in vector form as an
algebraic relation between input and output scattering variables.
v = f(u) (A.5)
This algebraic relation is constrained by the requirement that power in equal power
out. Geometrically, the length of the vector v must equal the length of the vector
u. The input and output vectors are both confined to the surface of a sphere.
v
1
3
v
2
v
u
1
u
2
u
3
u
v
Consequently, for any two particular values of u and v, the algebraic relation f(
.
) is
equivalent to a rotation operator and equation A.5 may be re-written as follows.
v = S(u) u (A.6)
where the square matrix S is known as a scattering matrix.
268
S need not be a constant matrix, but may in general depend on the power flux
through the junction, hence the notation S(u). However, S is subject to important
restrictions. In particular,
v
t
v = u
t
S
t
Su = u
t
u (A.7)
S is ortho-normal: the vectors formed by each of its rows (or columns) are (i)
orthogonal and (ii) have unit magnitude; its transpose is its inverse.
S
t
S = 1 (A.8)
(This is simply a mathematical statement of the conclusion that S is a rotation
operator.)
269
Symmetric three-port junction elements
To evaluate S, we assume an additional symmetry property in the mathematical
sense of invariance under the action of a group operator.
Specifically, assume the constitutive equation is invariant under the operation of
permuting or exchanging the ports of the junction structure.
In terms of the coefficients of the scattering matrix S:
S =
(
(
( a b b
b a b
b b a
(A.9)
The power output along any port depends upon the power input along all three
ports, itself and its two neighbors.
Each of the neighboring ports contribute the same fraction, b, of their input power
That fraction may be different from the portion, a, of the power input which is
"reflected" back out the same port.
Each of the three ports is the same in this regard.
Using equation A.9 in equation A.8 yields three equations for the coefficients a and
b, but only two are distinct.
a
2
+ 2b
2
= 1 (A.10)
b
2
+ 2ab = 0 (A.11)
These equations admit only two solutions,
a = 1/3; b = -2/3 (A.12)
or
a = -1/3; b = 2/3 (A.13)
so the symmetry condition constrains the scattering matrix to be a constant.
There are only two possible symmetric, power-continuous, three-port junctions.
They are both linear.
270
Transform back to effort and flow variables.
e =
(
(
e
1
e
2
e
3
(A.14)
f =
(
(
f
1
f
2
f
3
(A.15)
The relation between efforts and wave-scattering variables is as follows.
e = (u - v) c = c (1 - S) u (A.16)
where c is a scaling constant. Using equation A.12 for the value of S
S =
(
(
( 1/3 -2/3 -2/3
-2/3 1/3 -2/3
-2/3 -2/3 1/3
(A.17)
it can be seen that
e = c
(
(
( 2/3 2/3 2/3
2/3 2/3 2/3
2/3 2/3 2/3
u (A.18)
The efforts on each of the ports are equal. This is a common effort or type zero
junction.
e
1
= e
2
= e
3
(A.19)
271
The relation between flows and wave-scattering variables is as follows.
f = (u + v)/c = 1/c (1 + S) u (A.20)
Using the same value for S
f =
1
c
(
(
( 4/3 -2/3 -2/3
-2/3 4/3 -2/3
-2/3 -2/3 4/3
u (A.21)
Summing down the columns, it can be seen that the relation between flows on the
ports is
f
1
+ f
2
+ f
3
= 0 (A.22)
This is a continuity equation
1
. It is a derived property of the zero junction.
1
Note that power on all ports of the junction has been defined to be positive inwards.
272
Using the other of the two values of S (equation A.13).
S =
(
(
( -1/3 2/3 2/3
2/3 -1/3 2/3
2/3 2/3 -1/3
(A.23)
From equation A.20 we find
f =
1
c
(
(
( 2/3 2/3 2/3
2/3 2/3 2/3
2/3 2/3 2/3
u (A.24)
The flows on each of the ports of this junction are equal. This is a common flow or
type one junction.
f
1
= f
2
= f
3
(A.25)
Using equation A.16
e = c
(
(
( 4/3 -2/3 -2/3
-2/3 4/3 -2/3
-2/3 -2/3 4/3
u (A.26)
Summing down the columns, it can be seen that the relation between efforts is
e
1
+ e
2
+ e
3
= 0 (A.27)
This is a generalized compatibility equation. It is a derived property of the zero
junction.
Equating effort and voltage, flow with current, we have derived Kirchhoff's current
and voltage laws from energetic "first principles".
We have also shown that an exactly analogous pair of equations can be derived for
any of the energetic media and that they must always be linear.
273
Conservation principles and symmetric junctions
In terms of efforts and flows, power continuity constrains the constitutive
equations for a junction as follows.
_
e
i
f
i
= 0 (A.28)
The bond graph zero-junction is associated with a generalization of Kirchhoff's
node (current) law, a statement of flow (current) continuity that constrains the
constitutive equations for a zero junction as follows.
_
f
i
= 0 (A.29)
But these two constraints are not sufficient to define the zero junction .
A junction element known as a circulator (used in microwave circuit theory)
satisfies the requirements of power continuity (equation A.28) and current
continuity (equation A.29), but it is quite distinct from a zero junction.
The defining property of a zero junction is its symmetry the effort is invariant
under the operation of permuting or exchanging the ports of the junction structure.
e
i
= e
j
i, j (A.30)
Combined with this condition, either of the previous two constraints implies the
other. Thus, suppressing the subscript on the common effort, the power continuity
condition may be written as follows.
e
_
f
i
= 0 (A.31)
If this relation is to be identically true for all values of the (common) effort, the
flow continuity condition must be satisfied.
A similar argument holds for the one-junction.
Thus symmetry is the fundamental property, continuity or compatibility are not.
274
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
2.151: Advanced System Dynamics and Control Spring 2003
QUIZ 1
This quiz is closed-book. You have 1.5 hours to complete it.
275
Problem 1(30pts):
For each of the following bond-graphs, determine the systems minimum order. For the rst three
systems, determine whether the system could exhibit oscillatory (resonant) behavior.
1
I C
0
C
1
Se GY
R R
C
C
0
I
1 Sf TF
R C
1 0
I I
C
I
1 Sf
R 1 0 I
0
C I
1 Se 0
GY
TF TF 1
C
Se
276
Problem 2(30pts):
For each of the following systems, draw a corresponding bond-graph and determine the systems
minimum order. Assume all elements have linear constitutive relations. For the rst three systems,
choose appropriate state variables and write state equations.
k
b
m
k
b
m
F(t)
Assume the cylinder (with inertia I) rolls without slipping.
F(t)
R
k
I
Assume no leaking, and incompressible uid ow with density .
F(t)
Ap
d
l
Vs(t)
+
-
R R
R R
R
277
Problem 3:
Typical examples of common linear drives are described in the attached pages
taken from the Warner Electric catalog. Consider the in-line configuration
shown at the bottom of the page. A permanent-magnet DC motor is used to drive
a recirculating-ball screw (a nut and screw with ball-bearings interposed
between the nut and screw so that frictional energy losses are minimized).
The motor shaft is connected directly to the screw. The nut is prevented from
rotating but may slide as the screw rotates within it.
The screw has 5 threads per inch with a pitch diameter of 0.5 inches. The
screw may be considered to be a steel cylinder 18 inches long with diameter
equal to the screw pitch diameter. The nut may be considered to be a steel
cylinder 3 inches long, outside diameter 3 inches, with a hole though its
center equal to the pitch diameter of the screw. To keep things simple,
ignore all frictional losses due to the recirculating-ball screw and the
sliding nut.
Parameters of the motor are given in the table below.
Excerpt from specification sheet of a direct-current permanent-magnet motor
MOTOR CONSTANTS:
INTRINSIC (AT 25 DEG C)
SYMBOL UNITS
TORQUE CONSTANT KT OZ IN/AMP 5.03
BACK EMF CONSTANT KE VOLTS/KRPM 3.72
TERMINAL RESISTANCE RT OHMS 1.400
ARMATURE RESISTANCE RA OHMS 1.120
VISCOUS DAMPING CONSTANT KD OZ IN/KRPM 0.59
MOMENT OF INERTIA JM OZ IN SEC-SEC 0.0028
ARMATURE INDUCTANCE L MICRO HENRY <100.0
TEMPERATURE COEFFICIENT OF KE C %/DEG C RISE -0.02
a. First assume the DC motor is driven by a current-controlled amplifier
(i.e., the current applied to the electrical terminals of the motor may be
specified independent of the voltage required). Assuming all model elements
have linear constitutive equations, develop a model competent to describe the
linear displacement of the rod end in response to motor current input.
Express your model as a bond graph, clearly identifying the different energy
domains.
b. What is the order of this model?
c. Next assume the DC motor is driven by a voltage-controlled amplifier
(i.e., the voltage applied to the electrical terminals of the motor may be
specified independent of the current required). Assuming all model elements
have linear constitutive equations, develop a model competent to describe the
linear displacement of the rod end in response to motor voltage input.
Express your model as a bond graph, clearly identifying the different energy
domains.
d. What is the order of this model?
e. Suppose an external force is applied to the rod end. Numerically
evaluate the total apparent translational inertia opposing that force (i.e.,
the equivalent mass seen at the rod end due to everything that moves).
<Pages from Warner Electric catalog referenced above
have been removed due to copyright considerations.>
278
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
2.151: Advanced System Dynamics and Control Spring 2003
QUIZ 1
Problem 1(30pts):
Two possible bond graphs are shown below. The minimum order is two. Though this system has
a capacitance and inertance in integral causality, the gyrator does not allow oscillations.
1
I C
0
C
1
Se GY
R R
1
I C
0
C
1
Se GY
R R
Two possible bond graphs are shown below. This system has a minimum order of ve. It can
oscillate.
C
C
0
I
1 Sf TF
R C
1 0
I I C
C
0
I
1 Sf TF
R C
1 0
I I
This system has a minimum order of two. It can oscillate.
C
I
1 Sf
R 1 0 I
0
279
This system has a minimum order of two.
C I
1 Se 0
GY
TF TF 1
C
Se
C I
1 Se 0
GY
TF TF 1
C
Se
Problem 2(30pts):
k
b
m
k
b
m
F(t)
1
I2
C2 0
C1
1 Se
R
R
1
I1
We see from the bond graph the system is at least of fourth order, with states associated with
the two springs and two masses. Choosing the displacement across both springs and the momentum
of the two masses, our state equations are,
d
dt
x
1
x
2
p
1
p
2
0 0 1/m 1/m
0 0 0 1/m
k 0 b/m b/m
k k b/m 2b/m
x
1
x
2
p
1
p
2
0
0
1
0
F(t) (1)
If the force across the springs and the velocity of the masses is our state, then we have,
d
dt
F
1
F
2
x
1
x
2
0 0 k k
0 0 0 k
1/m 0 b/m b/m
1/m 1/m b/m 2b/m
F
1
F
2
x
1
x
2
0
0
1/m
0
F(t) (2)
280
F(t)
R
k
I
1
I
C 1
0 Se
TF R
II
For the bond graph, not that the velocity of the cylinder can be related to the velocity of the
rack, v and the angular velocity, , as follows,
v
m
= v r (3)
This relation indicates the need for two distinct ows and a zero junction. Choosing the spring
displacement, the cylinder velocity and angular velocity, the state equations are,
d
dt
x
c
v
m
0 1 0
k/m 0 0
0 0 0
x
c
v
0
1/m
r/I
F(t) (4)
F(t)
Ap
d
l
1
R
Se TF
I
The bond graph above is an adequate description of the system. The uid resistance can be a
sum of the resistance through the pipe and the resistance due to the exit restriction. Choosing the
volume ow rate as our state we get the following,
Q = RQ +
A
p
I
F(t) where I =
4L
d
2
and R =
32L
d
2
+ R
restriciton
(5)
281
The bond graph below indicates the system is zeroth order.
Vs(t)
+
-
R R
R R
R
1
0
1
Se R R
1
0
0
R R R
Problem 3(40pts):
a), b) A possible bond graph for the system driven with a current source is shown in the gure
below. Note that including damping in the screws motion, or resistance in the motors circuit does
not aect the minimum order, which is one. However, in order to describe the displacement of the
nut, we need to add a second variable in our system description, raising the order to two.
c), d) A possible bond graph for the system driven with a voltage source is shown below. Again,
including resistive type elements does not change the minimum order, which is now three when we
include the displacement of the nut.
e) The total apparent inertia, seen through the rods end, is a sum of the nuts and extension
tubes mass with the inertia of the screw and the motor armature as seen through the transformer.
This eective inertia is,
I
effective
= m
nut
+ m
tube
+
1
T
2
(J
screw
+ J
armature
) (6)
where T is the transformer modulus for the screw-nut interface.
T =
1in
5threads
1thread
rev
1rev
2radians
1ft
120
(7)
Then nut dimensions are given in the problem statement. To calculate the extension tube mass,
well assume a length of one foot, and radii of 3in and 2.9in.
m
nut
= (r
2
o
r
2
i
)L = (487lb/ft
3
)(3
2
.5
2
)in
2
(
ft
2
144in
2
)3in(
1in
12ft
) = 23.2lb (8)
m
tube
= (r
2
o
r
2
i
)L = (487lb/ft
3
)(3
2
2.9
2
)in
2
(
ft
2
144in
2
)1ft = 6.3lb (9)
J
screw
=
1
2
mr
2
=
1
2
(487lb/ft)18in(
1ft
12in
)(.5in)
4
(
1ft
12in
)
4
= 3.5 10
3
lbft
2
(10)
J
armature
= 0.0028ozins
2
(
1lb
16oz
)(
ft
12in
)(
32.2ft
s
3
) = 4.7 10
4
lbft
2
(11)
I
effective
= 23.2lb + 6.3lb + 497.4lb + 66.8lb = 593.6lb (12)
Note that the inertia of the screw and extension tube combined is only 5% of the total inertia
seen at the rod end.
282
1
I I e e I e
1 Sf GY
R R
TF 1
1
CTRIC
I
C IC
R T TI
I
C IC
TR S TI
I
1
I I e e I e
1 Se GY
R R
TF 1
1
CTRIC
I
C IC
R T TI
I
C IC
TR S TI
I
283

2 141 Fall 2002 Modeling and Simulation of Dynamic Systems Volume III

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

2 141 Fall 2002 Modeling and Simulation of Dynamic Systems Volume III

Caricato da

Copyright:

Formati disponibili

2.

141 Fall 2002

cos 144 384

(in terms of total entropy)

Choosing state variables, S

the angular velocity of the ywheel

Potrebbero piacerti anche