E Gsineering: Synthesis of Schematic Descriptions in Mechanical Design

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Res Eng Des (1989) 1:3-18 Research in

Egsineering
lgn
© 1989 Springer-Vedag New York Inc.

Synthesis of Schematic Descriptions in Mechanical Design


K a r l T. U l r i c h , * W a r r e n P. S e e r i n g
Massachusetts Institute of Technology, Artificial Intelligence Laboratory, Cambridge, Massachusetts, U.S.A.

Abstract. This article describes a schematic synthesis planation of our choice of problem domain, a
problem and one of its solution techniques. The problem description of our synthesis technique, and a dis-
domain consists of devices that can be described as net- cussion of the project results. Our research is aimed
works of lumped-parameter, idealized elements in the at a more fundamental understanding of how de-
translational-mechanical, rotational-mechanical, fluid- signs can be generated from a specification of their
mechanical, and electrical media. Such devices include behavior. This understanding is a necessary founda-
speedometers, accelerometers, pneumatic cylinders, and
tion for better computer tools for design and for
pressure gauges. Design problems in this domain are
specified by an input quantity, an output quantity, and the e n h a n c e d design teaching and practice.
desired relationship between the input and output. The
solution technique is based on three steps: 1) generate a I.l What is Schematic Synthesis?
candidate design, 2) derive and classify the behavior of
the candidate, 3) based on the derived behavior and do- We define schematic descriptions as graphs of func-
main knowledge, modify the candidate (if possible) to tional elements. A design described schematically
bring it in line with the specification. The key idea behind consists of a specification of its constituent func-
this technique is that an abstract characterization of the tional elements and their interconnections. The key
essential properties of the candidate design expedites the distinction between schematic descriptions and
analysis and modification. The results of this work are other types o f design descriptions is that schematic
aimed at computer tools for preliminary mechanical de-
descriptions generally contain no information about
sign.
the design's geometrical and material properties.
Generating a schematic description ~ in response to
a specification of desired device behavior is sche-
matic synthesis. Researchers and practitioners have
1 Introduction
developed schematic languages for several engi-
neering domains. Here are two examples of sche-
This article describes a class o f schematic synthesis
matic descriptions and associated synthesis prob-
problems and one solution technique in the domain
lems. Figure ! is a schematic description of a
of single-input single-output (SISO) dynamic sys-
section of a process plant. The functional elements
tems. The c o m p u t e r program implementation o f the
are idealized pumps, heat exchangers, mixers, con-
ideas in this article generates a schematic descrip-
densers, etc. A synthesis problem in this domain
tion of a device in response to a specification o f the
might be: given a library of equipment elements,
desired relationship between an input quantity and
generate a schematic description of a heat pump
an output quantity. An example of this kind of syn-
fractionation column for a certain set o f specified
thesis is the generation of a schematic description o f
input and output streams. Figure 2 is a schematic
a s p e e d o m e t e r - - i n terms of idealized elements like
description of an analog circuit. The description
inertias, compliances, and r e s i s t a n c e s - - i n response
consists o f idealized e l e m e n t s - - r e s i s t o r s , diodes,
to the specification that the angular velocity of an
operational amplifiers--connected together in a
input and the deflection of an output should be pro-
graph. A synthesis problem in this domain might be:
portionally related. This article presents an intro-
generate a circuit that will take a voltage signal
duction to the schematic synthesis problem, an ex-
input and produce an output voltage corresponding
to the absolute value of the input voltage.

* Reprint requests: Massachusetts Institute of Technology,


Artificial Intelligence Laboratory, 545 Technology Square, Cam- i We use the terms schematic description, schematic, and
bridge, MA 02139, USA description synonymously in this article.
4 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

Overhead
product

[ Trim condensor
Compressor

Fractionation - ~ -
column ---

Feed - - ~
Alkylation unit
reactor [ Reboiler

Fig. 1. Schematic description of a chemical


Bottoms product process.

1.2 Schematic Synthesis o f Single-Input schematic synthesis problem in our domain is:
Single-Output Systems given a description of the input and o f the output of
a system, what is the schematic description of a
This article presents a schematic synthesis problem
system that provides the desired relationship be-
and one of its solution techniques in the domain of
tween input and output.
single-input single-output (SISO) dynamic systems.
Figure 3 shows several schematic synthesis prob-

~o"
lems in the dynamic systems domain along with a R

sketch version of example solutions. The figure is


intended to give a rough sense of what this article is R
about and of the kinds of problems our solution i~P~R~R/2
technique addresses. The first example problem is
utput
the design of a pressure gauge. The solution is to
connect the pressure source and output spring with
a piston-cylinder. The second problem is to design a g
device to produce a voltage on a resistor that is
m
proportional to an input velocity. The solution is to
use a rack-and-pinion connected to a generator. The
third problem is to produce a pressure in a fluid
capacitance that is proportional to an input angular
velocity. The solution is to connect the input
through a rotary damper to a rack-and-pinion. The
rack is then attached to a piston-cylinder connected
to the capacitance. The final problem is to convert a
voltage to an angular deflection. The solution is to
connect the voltage source with a series resistor to a
motor that is attached to the output. The output is
also connected to ground with a torsion spring. The Fig. 2. Schematic description o f an analog circuit.
Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design 5

PROBLEM SOLUTION

....
I
:<(t) ~ xCt)

pressure gage

u(t) I
?
I
t v(t)
~ v(t)

i
[. . . . --J

velocity to volts

wCt)

w(t) r ~
? P(t)

angular velocity to pressure ~ , ~

i
I e(t)
[-- ~ -- --
Fig. 3. Example problems and solutions in
volts to angle the dynamic systems domain

1.3 Importance of Schematic Synthesis substitution of physical elements for functional ele-
ments, and then to subsequently look for opportuni-
Schematic synthesis is an important part of the
ties to simplify this design by using particular physi-
larger task of generating a physical description of a
cal features of the design to perform more than one
device from a description of its behavior. This
function.
larger task can be broken into two steps: 1) generate
a schematic description of the design in terms of
functional elements; then 2) from the schematic de- 1.3.1 Reducing complexity. A schematic de-
scription, generate a physical description of a de- scription is an abstraction of a design, excluding all
vice that approximates the ideal properties of the physical information except for topology. Working
schematic description. The generation of a sche- on a design problem first in this abstract problem
matic description as a first step in design is justified space is a less complex problem than trying to de-
because the strategy reduces the complexity of the sign directly in the design space described by the
problem solving task, and because the strategy de- physical representation language of the domain. In
couples functional and physical design issues. The general, there will be many fewer functional ele-
second step, generating a physical description from ments in a schematic description language than
the schematic description, is also quite important in there will be physical elements in a physical de-
mechanical design. Our approach to this problem, scription language. Therefore, the space of possible
discussed thoroughly in [Ulrich88a], is to first cre- schematic descriptions is usually considerably
ate a very modular physical description by direct smaller than the space of possible physical descrip-
6 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

tions. Given these factors, an approach to reducing the elements are idealized dampers, springs, and
the complexity of the larger design problem is to masses. In the rotational-mechanical medium the
work first in the smaller space of schematic descrip- elements are idealized rotary dampers, torsional
tions, and then, with the resulting schematic de- springs, and rotary inertias. And in the fluid-me-
scription as a starting point, work in the physical chanical medium the elements are idealized fluid
description space. In building computer tools for resistances, fluid capacitances, and fluid inertances.
design synthesis this schematic-physical decompo- In addition to these elements there are elements
sition is one way of coping with combinatorial com- that can transform quantities in one medium to
plexity. quantities in another medium--like idealized mo-
tors, piston-cylinders, or racks-and-pinions. This
1.3.2 Decoupling functional and physical is- domain can be thought of as a kind of generalized
sues. Focusing the initial problem solving effort analog circuit domain in that the elements impose
on schematic descriptions decouples the functional constraints on generalized effort and flow variables
and structural aspects of the design problem. Gen- (corresponding to voltage and current) and their in-
erating a schematic description, without concern terconnections obey generalized Kirchoff's laws.
for a possible physical implementation, forces the The devices described in Fig. 3 are instances of
design system to get the ideal device behavior right schematic descriptions that fall into this category.
first. This strategy is sound since only devices with An important characteristic of this domain is that it
correct schematic descriptions will meet the design does not deal with any information about the geo-
specifications. Once the schematic description is metrical or material properties of these devices.
right, the synthesis of an efficient physical imple-
mentation can proceed. This decomposition helps
2.2 Representing Schematic Descriptions
to focus both human and computer problem solving.
The descriptions in Fig. 3 are shown with icons that
suggest real physical components like motors and
2 Domain Description racks-and-pinions. These icons are purely mne-
monic. We represent these devices as networks of
This section describes our choice of the lumped- idealized elements 2. Associated with each element
parameter dynamic systems domain, explains the is an effort variable and a flow variable whose prod-
schematic description representation, and defines uct is the power associated with an element. Figure
problem specifications. 4 shows each of the elements in each of the four
media we deal with. Figure 5 shows a set of trans-
forming and gyrating elements that convert quanti-
2.1 Single-Input Single-Output Dynamic Systems
ties in one medium to quantities in another medium.
The domain in which we have applied the schematic Transformers convert efforts to efforts and flows to
synthesis concepts consists of devices that can be flows. Gyrators convert efforts to flows and flows to
described as networks of lumped-parameter ideal- efforts. All of these elements are connected to-
ized elements, and whose behavior can be specified gether in series or in parallel to form networks. The
by a relationship between a single input quantity networks obey a generalized version of Kirchoff's
and a single output quantity. We call this domain current and voltage laws. That is, series elements
SISO dynamic systems or just dynamic systems. share the same flow quantity and parallel elements
Examples of devices in this domain include pres- share the same effort quantity.
sure gauges, speedometers, pneumatic cylinders,
and accelerometers. We chose this domain because
2.3 Classifying the Behavior of a Schematic
there is a well-defined set of primitive elements with
Description
which to build device descriptions, and because the
domain is of engineering interest and importance. In order to evaluate a schematic description with
The lumped-parameter elements used in the dy- respect to a design specification, a system for per-
namic systems domain are idealized generalized re- forming design must be able to derive behavior from
sistances, capacitances, and inertances, as well as
transformers and gyrators. These elements have in- 2 In performingthis research and implementingthe results as
stances in the fluid-mechanical, translational-me- computer programs we use Paynter's bondgraph language to de-
chanical, rotational-mechanical and electrical me- scribe these devices. For clarity of presentation, we have
avoided introducing the bondgraph notation here. For those
dia. For example, in the electrical medium the readers interested in a more detailed descriptionof the represen-
elements are idealized resistors, capacitors, and in- tation of this problemdomain, see [Ulrich88b]and the appendix
ductors. In the translational-mechanical medium to this paper.
Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design 7

Trans. Mechanical Rot. Mechanical Fluid Electirical Effort-Flow Relation

Inertance

MASS
+
ROT. INERTIA F. INERTANCE INDUCTOR
E =ITdF

Capacitance

dE
F = C"~-
SPRING ROT, SPRING F. CAPACITANCE CAPACITOR

Resistance

E=FR
DAMPER ROT. DAMPER F. RESISTANCE RESISTOR
Source
~_) w(t) Efforts and Flows
F(t) i(t) + v(t) specified by these
T(t) sources,

Effort Quantity
Force Torque Pressure Voltage

Flow Quantity Velocity Angular Velocity Volume Flowrate Cttrrent Fig. 4. Elements used to repre-
sent schematic descriptions

schematic descriptions. In this domain, behavior output. A simple proportionality between an effort
consists of a general characterization of the equa- or flow on the input and an effort or flow on the
tions of motion for the schematic. Classifying this output corresponds to a type number of 0. For each
behavior involves first deriving the equations of mo- derivative of the input (or integral of the output) the
tion, and second, characterizing the equations in type number is increased by 1. So, if the first deriva-
terms of their global properties. tive of the input is proportional to the output, then
The derivation of equations from the schematic the system is of type 1. If the input is proportional
description is a straightforward application of sys- to the first derivative of the output then the system
tems theory, and consists of three steps. First, im- is of type - 1 , and so on. Four canonical examples
pose the constraints from the generalized Kir- illustrate this concept. An accelerometer is a sys-
choff's laws by writing a set of network equations tem with a type number of 2, because the deflection
relating the effort and flow variables at each junc- of the indicator (the integral of the velocity of the
tion point between elements. Second, impose the output) is proportional to the input acceleration (the
constraints from the element constitutive laws. For derivative of the velocity of the input). A speedom-
each element (capacitance, resistance, inertance, eter is a system with a type number of 1, since the
transformer, gyrator, or source), establish the ap- deflection of the output (the integral of the output
propriate constraint between effort and flow vari- velocity) is proportional to the input velocity. A
ables on the element. Third, given the equations piston-cylinder is a system with a type number of 0,
generated in steps one and two, eliminate all of the
effort and flow variables except for the variables
corresponding to the desired input variable and the
desired output variables. The result of this proce-
dure is a differential equation relating the input vari-
able and its derivatives to the output variable and its
derivatives.
PUMPfrURBINE PISTON and CYLINDER
The next step in deriving the behavior of the
schematic description is to characterize its equa-
tions of motion. We have developed a characteriza-
t--
tion technique aimed at determining the nominal re-
lationship between the input quantity and output
quantity. The system is characterized according to
its type number, an idea borrowed from control the-
ory [Ogata70]. The intuition behind this character- MOTOR/GENERATOR
ization is that the type number of the system speci-
RACK AND PINION
fies which integral or derivative of the input is
proportional to which integral or derivative of the Fig. 5. Example transforming and gyrating elements
8 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

f------~v(t) nal relationship between the voltage and velocity


I
will be a first derivative (i.e., velocity will corre-
SPECIFICATION spond to change in voltage, so displacement will
correspond to voltage). If we want a displace-
.......... .!
V(t) ment proportional to voltage rate-of-change,
then we specify that the output velocity will be
SOLUTION nominally related to the second derivative of the
input voltage. This specification is represented
as a type number. For example, the first deriva-
tive case is type 1, and the second derivative
case is type 2. The problem specification can
Fig. 6. E x a m p l e specification and solution
only include nominal, steady-state behavior of
the device. That is, the dynamic response of the
since the output force is proportional to the input device is not part of the specification. The as-
pressure (i.e., no derivatives or integrals). A mass sumption is that in preliminary design, engineers
with a force applied to it, in which the mass velocity first want to get the nominal behavior right, and
is the output, is a system with a type number o f - I, will then optimize parameters to get a particular
because the acceleration o f the mass (the derivative frequency response.
of the output velocity) is related to the force on the 3. The third part of the specification is the identifi-
mass. The formal procedure for determining the cation of the dynamic model of the input and
type number is in the appendix. output. If, for example, we want to design a volt-
The complete equations for the system are im- meter that measures the voltage of a source
portant in refining the particular details of a design, whose voltage drops with increasing current, we
and when choosing design parameters, but only this will specify the input as consisting of an ideal-
type number specification is important in getting the ized voltage source (one that can provide an ar-
gross relation between input and output quantities bitrary amount of power) in series with a resis-
right. tance. And if our voltmeter must actuate a
spring-loaded indicator, we would specify the
2.4 Specifying a Problem output flow as acting on a translational-mechani-
cal capacitance (a spring). This part of the specifi-
A schematic synthesis problem in the dynamic sys-
cation is provided by a lumped-parameter model
tems domain is specified by a relationship between
of the input and output environment of the de-
an input quantity and an output quantity. This spec-
vice. The intuition behind this is that all dynamic
ification includes three parts: 1) specification of the
systems interact in some way with their inputs
variables of interest, 2) specification o f a relation-
and outputs. These interactions partially deter-
ship between variables, and 3) specification of sche-
mine the behavior of the system. In order to cor-
matic descriptions corresponding to a dynamic
rectly synthesize a system description, one must
model of the input and output environment of the
know what the properties of the input and output
device.
interactions are.
1. The specification of the variables of interest sim-
In summary, a specification consists of an effort or
ply involves identifying either an effort or flow
flow variable in a selected medium, an identification
variable in the desired medium. For example, if a
of the derivative or integral relation of interest (indi-
voltmeter is desired, the input is a voltage (or
cated with a type number), and a specification of a
effort in the electrical medium) and the output is
lumped-parameter model, of the input and output.
perhaps a velocity (flow in the translational-me-
chanical medium). Note that the input and out-
2.4.1 Example specification. Consider the fol-
put variables must be efforts or flows. Deriva-
lowing complete specification. In designing an air-
tives and integrals of these efforts or flows are
craft instrument that must actuate a spring-loaded
specified when the relationship between vari-
valve with a displacement proportional to rate-of-
ables is specified.
change in air pressure, we specify the following
2. Specifying a relationship between variables re-
(shown in Fig. 6):
quires the identification o f the appropriate deriv-
ative or integral of the input and output vari- 1. The input variable is fluid pressure (effort in the
ables. For example, if we want a displacement in fluid medium). The output variable is transla-
response to a voltage, we specify that the nomi- tional velocity (flow in the translational mechani-
Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design 9

cal medium). These variables correspond to the two parts. This connection can either be a direct
air pressure and the velocity of the valve. junction between the input and output parts of the
2. The relationship between output velocity and in- schematic, or it can be a connection formed by a
put pressure should be one of second derivative, sequence of transforming or gyrating elements. Any
yielding the desired relationship between dis- device that meets the design specifications must
placement and rate-of-pressure, and so would be have this connection and the power spine repre-
specified as a type 2 system. sents the minimal possible connection for a given
3. The input schematic description is a pressure- set of intermediate media. Therefore any schematic
source attached in series with a fluid resistance. description that will meet the specifications must be
The output schematic description is a grounded derivable from augmentations to this connection be-
spring. tween the input and output.
The synthesis problem is to now find a completed
3.1.2 Connecting input to output. The synthesis
schematic description that will behave as specified.
strategy for generating candidate descriptions is to
A satisfactory solution would be the description
generate a set of schematic descriptions consisting
shown at the bottom of Fig. 6.
of the input and output descriptions connected by a
power spine. These candidate descriptions are gen-
3 Solution Technique erated exhaustively, subject to the constraint that
the number of intermediate medium transforma-
Our solution to the schematic synthesis problem is tions be small (<3). The size limitation is a con-
based on the following strategy: straint imposed by engineering practice--because
of cost and reliability, it rarely is a wise design strat-
1. Generate a set of candidate schematic descrip-
egy to use more than three media in a SISO dy-
tions from which all possible solutions can be
namic system. The most important property of the
obtained through augmentations to one of these
candidate descriptions generated by constructing
descriptions.
the power spine is that they are in a certain sense
2. Derive and classify the behavior of these candi-
minimal. Minimality in this context means that any
date descriptions. schematic containing fewer than three inter-me-
3. Based on the behavior of the candidates, the de-
dium transformations and meeting the design speci-
sired behavior, and domain knowledge, modify
fication, must be derivable through the addition of
the candidates to meet the specifications.
elements to one of the candidates.
The following three subsections correspond to
these three problem solving steps.
3.2 Classijfing Behavior
Given a minimal schematic description that will
3.1 Generating Candidate Descriptions
provide some relationship between input and out-
Part of the specification of the schematic synthesis put, the next step is to classify this initial descrip-
problem is a description of the input and output tion. It is possible that the candidates meet the
environment of the desired device. The procedure specifications; it is also possible that there is a mis-
for generating a set of candidate descriptions is match between the specified behavior and the actual
based on the observation that any complete sche- behavior of the candidate descriptions, requiring
matic description that performs a transformation some modifications. The degree of agreement be-
from a quantity in the input description to a quantity tween the behavior of the candidate descriptions
in the output description must contain either a di- and the specifications is determined by first deriving
rect connection between elements in the input and the equations of motion, and then categorizing
output descriptions, or it must contain a connection these equations according to type number. If the
between input and output consisting of a sequence type number of the candidate matches the type
of transforming or gyrating elements. We call this number of the specification, then coincidentally the
connection a power spine. candidate meets the design specifications, and the
synthesis problem is solved. More likely, a modifi-
3.1.1 Concept of a power spine. A schematic de- cation to the candidate is required.
scription with a non-trivial relationship between a At this point the synthesis problem can also be
quantity associated with the input part of the de- assessed as being impossible. Since increasing the
scription and the output part of the description must type number of a system requires the addition of
contain some sort of power connection between the elements, and decreasing the type number of a sys-
10 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

ments that do not influence the type number of the


system. The procedure for performing this transfor-
mation is as follows:
• The first step converts a description in terms of
particular media (electrical, fluid, etc.) into a de-
scription containing only generalized resistances,
capacitances, inertances, and sources. We do not
give the details of this transformation here, al-
<3" though it is a straightforward syntactic operation.
The transformation basically involves removing
all transforming and gyrating elements, and
changing specific resistances, capacitances, inert-
REDUNDANCY
ances, and sources into their generalized equiva-
ELIMINATION V(t)> lents. The only subtlety in this process involves
HIGHER-ORDER BEHAVIOR
the removal of gyrators. Since a gyrator converts
ELIMINATION flows to efforts and efforts to flows, the removal
of a gyrator from a schematic requires that the
Fig. 7. Example simplification rules region of the schematic associated with the gyra-
tor be modified to its complementary form. That
is, elements connected in parallel are switched to
tem requires the removal of elements, it is not pos-
being connected in series and vice versa; and ca-
sible to decrease the type number of a candidate
pacitive elements are replaced with inertial ele-
description. This is because no elements may be
ments and vice versa. Converting a particular de-
removed from the candidate. The input and output
scription to its generalized equivalent compresses
parts of the description can not be modified since
the space of possible schematic descriptions about
they are part of the problem specification; and no
which the design system must reason.
elements can be removed from the power spine
The second step is to use rewrite rules to elimi-
since the spine is a minimal connection between the
nate elements that do not influence the type num-
input and output (assuming a given set of inter-me-
ber of the description. Two example rules are
dium transformations). Therefore, if the specifica-
shown in Fig. 7 (shown with translational-me-
tion of the design calls for a type number that is less
chanical elements for clarity). There are eight
than that of the candidate, the synthesis problem
such rules. Using the rewrite rules to eliminate
has no solution.
these non-essential elements completes the trans-
formation of the schematic description into its
3.3 Modifying Candidate Schematic Descriptions most compact form.
The final step is to modify a candidate that does not The schematic description now consists of a net-
exhibit the specified behavior. This process in- work of idealized inertances, capacitances, resis-
volves three steps: 1) transform the candidate to a tances, and sources. This is a minimal characteriza-
compact description; 2)based on explicit domain tion of the candidate schematic description. A
knowledge, derive and perform a set of modifica- system described this way represents an infinite set
tions; 3) reverse the compacting transformation. of graphs that have the same type number.
The rest of this subsection is divided into parts cor-
responding to these steps. 3.3.2 Based on domain knowledge, generate mod-
ifications. The power of the compacting transfor-
3.3.1 Transform the candidate design to a com- mation in this domain is that once the candidate
pact description. The key concept that allows for description is reduced to its compact form the do-
correct modification of faulty schematics is the use main knowledge can be expressed concisely. There
of a compact representation that highlights only the is one special case and a set of general principles for
features of the description that contribute directly modifying the design to meet the problem specifica-
to its nominal behavior. The transformation of a tions. Here we explain the intuition behind the mod-
schematic description to its compact representation ification procedure, although the formal explanation
consists of first converting it to a generalized equiv- is left for the appendix.
alent description and then applying some simplify- The special case is when the schematic is of type
ing rewrite rules. The rewrite rules eliminate ele- - 1, and one desires a type 0 system. There are only
Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design 11

Non-Isolated
capacitive ~ ~ V(t)
element ~ [ "
F(t)

Type 1 s y s t e m ~ Isolated Capacitive Type 1 system ~ ' ~ Isolated Capacitive


Element Element
~ Isotated capacitive element

Type 2 system ~ " Isolated Capacitive Fig. 8. Illustration of the concept of an isolated
Element element

two cases in which a schematic can be of type - 1 . add derivatives to the equations of motion; in other
The first is if an inertial element is connected di- words, to add capacitive or inertial elements. How-
rectly to an effort source and the variables of inter- ever, there are only certain specific conditions un-
est are the input effort and the output flow. In this der which a capacitive or inertial element can influ-
case, the flow will increase linearly in time for a ence the behavior of the system. For example (we
constant input effort (thus a system of type - 1 ) . In present a translational-mechanical case for clarity,
order to change the type o f this system to 0, a resis- although the procedure would be carried out on the
tive element must be added between the inertial ele- minimal description of the system in terms of gener-
ment and ground. An example of this case is a mass alized elements), imagine a force source connected
connected to a force source. The mass accelerates to ground by a spring (Fig. 8). If the force is applied
at a constant rate, and therefore the first derivative to the spring, it deflects to a position that is deter-
o f the mass velocity is directly related to the force mined by the stiffness o f the spring and the magni-
(type - 1 ) . If a translational damper is added be- tude of the applied force. Since the integral of the
tween the mass and ground, the steady-state rela- spring velocity is directly related to the spring
tionship between velocity and force is now linear force, the system is of type 1. Now, imagine that we
(type 0). The second instance is if a capacitive ele- add another spring to the system in series with the
ment is connected directly to a flow source and the first spring (Fig. 8). Although we have added a ca-
variables of interest are the input flow and the out- pacitive element to the design, we have not altered
put effort. In this case the effort will increase lin- the type number o f the system. This new spring
early in time. In order to change the type of this does not influence the type number of the system
system to 0, a resistive element must be added be- because it is not isolated. If, however, we added a
tween the flow source and the capacitive element. spring that is separated from the original spring by a
In all other cases the type number of the system series damper (a resistive element), then the type
is exactly the number of isolated capacitive or iner- number of the system increases to 2 (Fig. 8). In this
tial elements in the schematic. The intuition behind second case each of the springs is isolated.
an isolated capacitive or inertial element is as fol- This concept of isolation is defined formally in
lows. Capacitances and inertances are the only the appendix to this article, but a simplified expla-
lumped-parameter d e m e n t s that have constitutive nation of the rule for isolation is that the capacitive
relations containing derivatives. That is, the rela- and inertial elements must be connected into the
tionship between the effort and the flow on a capaci- description in such a way that they are distinct from
tance or an inertance is an equation between one existing capacitive and inertial elements. In certain
quantity and the derivative or integral o f the other cases, this can be facilitated by the addition of resis-
quantity. This is in contrast to resistances and tive elements that serve an isolating function. Given
sources whose constitutive relations contain no de- this statement of the domain knowledge, the deriva-
rivatives or integrals. The only way to change the tion of modification operators is straightforward.
type n u m b e r o f a system is to add to it elements that E x c e p t in the case of systems of type - 1, the differ-
12 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

v(t) ine that an engineer wanted to design a device that

l' , ?
,F-" would take as an input a current from an electrical
circuit and actuate a mass with a displacement pro-
portional to the current. Perhaps the device is to
i(O
serve as a kind of governor on an electric motor
-I-: _, circuit--an increase in motor current will move a
mass, causing an increase in the rotary inel~ia of the
Fig. 9. Specificationof current meter problem system. The problem can be stated in the dynamic
systems representation as follows: synthesize a sys-
tem that will connect the electrical system input
ence between the type number of the candidate de- description with the mechanical system output de-
sign and the type number of the specification is scription (Fig. 9) such that the integral of the flow
equal to the number of new isolated capacitive or on the output is proportion to the flow on the input
inertial elements that must be created in the de- (i.e., a system with a type number of 1). The solu-
scription. If one isolated element must be created tion technique proceeds according to the steps de-
than an isolated inertial or capacitive element could veloped in the preceding section.
be added, or a nonisolated inertial or capacitive ele-
1. Construct the power spine. In this case the
ment already in the description also could be iso-
power spine is a motor-like element (a gyrator)
lated by adding a resistive element in an appropriate
and a rack-and-pinion-like element (a trans-
location. The only constraint that must be enforced
former). The system elements that form the
when creating new isolated groups is that new ele-
power spine are between the two vertical dotted
ments be added to the power spine between the
lines. Just one candidate schematic description
compact descriptions of the input and output envi-
is shown (Fig. 10) although there will in general
ronment. This is because the addition of elements to
be several possible power spines that connect
the input and output regions of the description
the input graph chunk to the output graph chunk
would require that the input and/or output environ-
and that consist of fewer than three gyrating or
ment of the device be modified.
transforming elements.
2. Derive the behavior o f the candidate schematic
3.3.3 Reverse compacting transformation. The
description. Deriving the equations of motion
final synthesis step is to reverse the compacting
that relate the output flow to the input flow re-
transformation on the modified schematic in order
veals that the candidate schematic description is
to arrive at instantiations in specific media. The re-
a type 0 system--the output flow (a velocity) is
verse transformation is straightforward:
proportional to the input flow (a current). Since
1. Add the elements removed by rewrite opera- the candidate schematic description does not
tions. meet the specification, the transformation and
2. Change the generalized system back to specific modification steps must be carried out. Fortu-
media by re-introducing the transformers and gy- nately, since the type number of the system must
rators. be increased (from 0 to 1), the modification is
possible.
The only decision that must be made in this process
3. Transform the candidate schematic description
is where, with respect to the newly added elements,
to its compact representation. The first step is to
the gyrators and transformers should be reintro-
remove all of the gyrating and transforming ele-
duced. The approach we have taken to this decision
is to create a version of the design for each possible
location of the elements with respect to the gyrators I
and transformers. There is a range of possible posi- (I If tl II II II II II II I f I ~ M ]
tions in the various media associated with the sys- ' U -~ ,t I
tem for locations of the newly-added elements.

4 A Complete Example

This section is aimed at clarifying the synthesis


technique by presenting a concrete example. The
example is illustrated by figures 9 through 15. Imag- Fig. 10. Example of one of the possible power spines
Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design 13

Fig. 11. Description converted to generalized elements Fig. 12. Removal of redundant capacitive element

ments, replacing particular instances of inert- that occurred in step 3; that is, the generalized
ances, capacitances, resistances and sources capacitance is converted to a mass, and the gen-
with generalized inertances, capacitances, resis- eralized inertance is converted to a spring.
tances, and sources. Note that removing a gyra-
Note that there are several possible locations for
tor requires that all of the downstream 3 capaci-
the newly added element in the schematic descrip-
tive elements be changed to inertial elements,
tion. It could be a torsional spring on the motor
and that all of the downstream inertial elements
output, a.translational spring on the rack, or it could
be changed to capacitive elements; and that ele-
be an electrical inductance in the electrical portion
ments that are in parallel be changed to elements
of the system. Two of these possibilities are shown
that are in series and vice versa (Fig. I1). The
in Fig. 15.
final compacting step is to remove any superflu-
ous elements (execute any rewrite rules that may
apply). In this example one of the capacitive ele-
5 Discussion
ments closest to the source can be removed
without altering the type number of the system
This article has presented a solution to a synthesis
(Fig. I2).
problem that people have traditionally solved in an
4. Derive appropriate modifications. The design is
ad hoc, heuristic way. This section analyzes the
specified to be a type 1 system--a displacement
importance, completeness, extensibility, and com-
(integral of output flow) proportional to current
puter implementation of the synthesis technique.
(the input flow). Since the candidate schematic
description is of type 0, an isolated capacitive or
inertial element will have to be added to the 5.1 Importance and Utility of the Technique
power spine. Observing the compacted sche-
When we began thinking about the SISO dynamic
matic description reveals that an inertial element
systems synthesis problem, we gave several engi-
can be added along the power spine and will be
isolated. Adding a capacitive element, however,
will require adding a resistive element as well,
since it would otherwise not be isolated from the
existing capacitive element. These two new de-
scriptions are shown in Fig. 13.
5. Reverse the compacting transformation. The
next step is to reverse the steps carried out dur-
ing the compacting transformation (only one de-
scription is shown in Fig. 14). First the capaci-
tive element is added back to the schematic
description. Next the gyrators and transformers o 1 I
are reintroduced. Note that adding the gyrator
back into the description reverses the swapping , I
I
3 D o w n s t r e a m in this context is an arbitrary choice of direc- o I
tion within the schematic description. When a gyrator is re- ' I
moved from the design, one chooses a direction within the de-
scription as the downstream direction and then converts the Fig. 13. Two possible additions of isolated groups to minimal
elements as explained in this paragraph. description
14 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

perience suggest that design in general will benefit


from formalizing some domains and from finding
synthesis techniques in those domains. In many
ways human engineers have been successful be-
cause they have been able to find some solution to a
problem. Synthesis techniques for even fairly sim-
ple engineering tasks can help engineers find a much
greater number of feasible solutions, many of which
Fig. 14. Reintroduction of redundant capacitive element (only
one description shown) will be solutions the designer has not thought of.
This work is also important because of its contri-
bution to the larger task of synthesizing a physical
neers some sample problems to see how they per- description of a device from a specification of its
formed the task; most of them found ways to avoid behavior. We propose that performing this larger
doing the exercise. One or two engineers actually synthesis task in two steps--first generating a sche-
found a solution. No one found more than one solu- matic description, and from it a physical descrip-
tion. These informal observations suggest that the t i o n - r e d u c e s the problem solving complexity and
synthesis problem is not simple for people and that focuses the problem solving effort. In this article we
synthesis techniques are not widely known. The have presented a solution to the first half of the
technique we have developed provides all possible larger design problem. We present one solution to
solutions (containing fewer than three inter-medium the second half in [Ulrich88a] that is based on the
transformations) to a specified problem. This idea of function sharing--the use of a single part in
results in an improvement in the state of the art in a design to implement more than one functional ele-
this area of synthesis. These techniques are useful ment in the device schematic description.
not only for computer programs, but also for pencil
and paper approaches. The results of this single ex-
5.2 Completeness of the Technique
Our schematic synthesis technique is in a trivial
sense incomplete--it will not generate all possible
schematic descriptions that meet the problem speci-
fications. The incompleteness results from two fac-
tors. First, there is an infinity of schematic descrip-
tions that meet the specifications. A pair of
inter-medium transforming or gyrating elements can
always be added to a schematic description to cre-
ate a new schematic description: a back-to-back
} i(O motor-generator pair added to an electrical device,
for example. Second, elements that do not change
the nominal behavior of a schematic description can
always be added to a schematic description to cre-
ate a new schematic description. An example of the
second case is the addition of a redundant capaci-
tance tO a region of the schematic description that
already has a capacitive element. Completeness
should, therefore, be defined with respect to the
compact representation of the schematic descrip-
tion, after all of the extraneous elements have been
removed, and the schematic description has been
converted to a generalized form without transform-
ers and gyrators. Under these conditions the tech-
nique is complete. The completeness rests on the
fact that the type number of a system is determined
by the number of isolated capacitive and inertial
elements within the power spine of the description.
This can be proven by the following argument. A
Fig. 15. Two possible instantiations of the modified description positive type number indicates the number of times
Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design t5

the input quantity has been differentiated. There are erai classes of devices--those with feedback, those
only two possible ways a quantity can be differenti- with hidden sources, those without any signal
ated in this domain. Either a capacitance differenti- flows.
ates an effort to produce a flow, or an inertance If the problem were extended to address multi-
differentiates a flow to produce an effort. Resis- ple-input multiple-output systems, the synthesis
tances serve to convert efforts to flows and vice procedure would become more complex. The
versa. The only way to differentiate an effort source power spine concept no longer is as strong as in the
(there is a similar argument for flow sources) is to SISO case. There are now many possible ways for
connect it to a capacitance (an instance of his case the multiple inputs and outputs to be connected to-
is a force source attached to a spring), or to connect gether other than a single sequence of idealized
it to a series resistance (to convert the effort to a transforming and gyrating elements. Prabhu and
flow) and then to a parallel inertance (to differenti- Taylor have explored some of the issues in the mul-
ate the flow into an effort). These two cases corre- tiple input and output problem [Prabhu88].
spond to single isolated capacitances and inert- The problem addressed in this article deals only
ances. Since the power spine of a device is a with the nominal behavior of a dynamic s y s t e m - -
sequence of elements, this argument applies to an we have not considered dynamic issues like
arbitrary number of differentiations. For each addi- bandwidth. To some extent these issues can be ad-
tional differentiation (increase in the type number) dressed by optimizing parameter values numeri-
an isolated capacitance or inertance must be added. cally. In other cases, filtering elements or damping
Because the synthesis technique produces all possi- must be added to the schematic descriptions to
ble ways of modifying the candidate schematic de- achieve the specific numerical specifications de-
scription in order to increase the number of isolated sired. The synthesis procedure in this article, how-
capacitances and inertances in the system, it pro- ever, is a necessary first step towards meeting more
duces a complete set of compact schematic descrip- detailed numerical design specifications. Any sche-
tions. matic description that meets a detailed numerical
specification must also meet the specification of the
nominal behavior of the device. The work in this
5.3 Extensibility
article is directly applicable to this more specific
The ideas in this article have been directed at a very synthesis problem.
specific synthesis problem. This subsection deals
with the problem of extending the ideas within the 5.3.2 Extension to other domains. The three ma-
dynamic systems domain and to other domains. jor segments of the solution to the dynamic systems
synthesis problem are: development/selection of a
5.3.1 Extension within dynamic systems do- schematic language, a procedure for generating can-
main. The devices that can be synthesized with didate schematic descriptions, and a procedure for
the technique described in this article are limited in categorizing and altering schematic descriptions
several ways: they are single-input, single-output; falling into different classes of behavior.
there is no feedback or amplification; and they can The particular schematic language we use in the
not be specified to have particular dynamic re- dynamic systems domain can not be extended to
sponse characteristics. Extension of the solution any other domain, although the properties of the
technique to erase these limitations requires either lumped-parameter systems language that make it a
an extension of the representation language, an ex- good choice for the dynamic systems problem can
tension of the synthesis procedures, or both. be used as guidelines for selecting schematic lan-
To deal with feedback or amplification, we would guages in other domains.
have to introduce the use of active elements--ele- The power spine idea in the dynamic systems
ments with external power sources. Along with the domain has incarnations in other domains as well.
active elements, signal flows would be required. In the domain of material handling systems (i.e.,
This extended representation would also require ex- design of conveyers) there must be a connection
tensions to the synthesis procedures. Specifically, between input and output. In the domain of engi-
the connection property of the domain that leads to neering structures (beams, brackets, supports,
the concept of a power spine is still valid, although struts) there must be a connection between input
the kinds of connections that are allowed are in- and output. In fact, in the structures case, perhaps
creased to include signal-only connections. Under the candidate schematic descriptions could be max-
these conditions the synthesis procedure may re- imal connections--a structure that occupies all of
quire the consideration as special cases each of sev- the possible space between the input and output
16 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

points of the structure. Then, any valid schematic 6 Related W o r k


description might be derivable from material re-
moval operations on this candidate schematic de- A more detailed presentation of this work along
scription. The classification and modification proce- with a procedure for generating physical descrip-
dures in the dynamic systems domain are problem tions from the schematic descriptions is in [Ul-
specific; however, the general idea is applicable in rich88b]. The key idea behind generating a physical
many other domains. Imagine for example, a syn- description from a schematic description is to first
thesis problem in the shaft and bearing system do- use a standard physical component for each func-
main. A minimal characterization of the system in tional element in the schematic description, and
terms of the types of interactions between adjacent then to subsequently look for opportunities to sim-
elements could be derived in order to compute the plify the resulting device. Doyle describes work
degrees of freedom of the device, or to determine aimed at generating explanations of device behavior
whether or not the device could be assembled. [Doyle88]. In particular, the input to the explana-
tion problem is a time history of events or device
states. The output is a network of causal connec-
5.4 Computer Implementation
tions between the events in the observation. Each
In this article, we have focused almost exclusively connection corresponds to a physical phenomenon
on the principles and the procedures for performing or mechanism. This causal network can be thought
the schematic synthesis task. We have deliberately of as a sort of schematic description of the device.
avoided a discussion of any particular computer Doyle works with a very broad class of devices,
program that implements these ideas. Our belief is including toasters, bicycle coaster brakes, and tire
that programs in this context serve two purposes: gauges. To deal with this wide variety of devices, he
first, they may be tools for solidifying and testing has developed a very rich representation for physi-
hypotheses about how to perform the synthesis cal phenomena, including many kinds of transduc-
task; second, they may be embodiments of the final tion and transport. The key contribution of this
procedures, useful as tools or demonstrations. We work is the exploration of the power of various
have written a computer program that implements a types of constraints in controlling the combinatorial
solution to the schematic synthesis problem de- complexity of the explanation problem. [Prabhu88]
scribed in this article. The program was written to derives some mathematical properties of multi-in-
fulfill the first purpose--to clarify the thinking in- put, multi-output devices characterized by power
volved in devising a solution technique--and to that flows between inputs and outputs. Prabhu uses a
extent was quite useful. The final version of the bondgraph representation for these devices, and
program does not exactly match the solution tech- concentrates on synthesizing the junction-structure
nique presented in this article. The actual program or network of transducing elements that can link the
deviates from the theory we present in one major inputs with the outputs while dividing the power
way: it relies on a set of debugging rules to modify flows according to specification. Because the bond-
the candidate schematic descriptions rather than graph is a formal mathematical representation, cer-
performing the compacting transformation and sub- tain results such as soundness and completeness
sequent modifications. An example debugging oper- can be proven. [Kannapan87] develops an instance
ator is: I f the output is a flow variable, the input is a of a schematic language for mechanical design. The
flow variable, and differentiation is desired, add a language includes primitives shafts, bearings, han-
series resistance and capacitance to ground. These dles, supports, etc. Pieces of devices described with
rules were derived by analyzing human design rea- this language are combined together to form new
soning in debugging deficient schematic descrip- designs. [Williams88] is one of the very few pieces
tions, and in fact correspond to the same operations of work that deals with the synthesis of a design
as are derived by the technique presented in this configuration. Williams explores the domain of digi-
article. The debugging rules are, however, much tal circuit design at the FET level. His program is
more complicated than the procedure using the con- based upon using qualitative analysis of the behav-
cept of a compact representation and isolated inert- ior of a preliminary circuit design to guide design
ances and capacitances. It was not until we began modifications. Williams develops a time-based de-
exploring the completeness of the debugging proce- pendency scheme for tracking the influence of one
dure that we realized that there was a more compact event in time on another event. [Roylance83] is
way of characterizing the schematic descriptions probably the first effort at synthesizing analog cir-
and that the domain knowledge could be expressed cuit configurations computationally. In this mas-
more concisely. ter's thesis, Roylance describes a design-rule-based
Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design 17

system that designs simple R L C circuits. The de- [Ogata70]


sign rules allow the system to backward chain from Ogata, K., Modern ControlEngineering, Prentice-Hall, 1970. p.
an equation specifying the desired circuit behavior. 284.
[Ressler84] is a thesis describing a computational [Paynter61]
Paynter, H.M., Analysis and Design of Engineering Systems,
procedure for designing operational amplifiers. The The MIT Press, Cambridge, MA 1961.
procedure is based upon a hierarchical circuit gram- [Prabhu88]
mar. Amplifiers are viewed as consisting of three Prabhu, D. and D.L. Taylor, "Some Issues in the Generation of
stages. Stage one may be implemented as a differen- Topology of Systems with Constant Power-Flow Input-Out-
tial pair, a cur/rent cancellation configuration or a put Requirements," Proceedings of the 1988 ASME Design
Automation Conference, Kissimmee, Florida, September
super beta circuit. A differential pair m a y be viewed 1988.
as a load and emitter coupler pair. A load can be [Ressler84]
resistive, a current mirror, a simple pair, or a Ressler, A.L., "A Circuit Grammar for Operational Amplifier
Darlington pair. There are similar expansions for Design," Massachusetts Institute of TechnologyArtificial In-
the other amplifier stages. The design procedure is telligence Laboratory Technical Report 807, January 1984.
[Rieger77]
to implement the amplifier with the simplest possi- Rieger, C. and M. Grinberg, "The Declarative Representation
ble pieces, then analyze the resulting circuit at each and Procedural Simulation of Causality in Physical Mecha-
stage in the instantiation process. If the circuit will nisms," Proceedings of the Fifth International Joint Confer-
not meet the specifications, then a more complex ence on Artificial Intelligence, VoL l, p. 250, 1977.
option is chosen. [Rieger77] is one o f the first pa- [Rosenberg83]
Rosenberg, R.C., and D.C. Karnopp, Introduction to Physical
pers attempting to describe the causal structure of a System Dynamics, McGraw-Hill, 1983.
mechanical device. This article presents a language [Roylance83]
for describing devices, and develops a model in this Roylance, G., "A Simple Model of Circuit Design," Massachu-
language for a thermostat. The functional elements setts Institute of Technology Artificial Intelligence Labora-
of the model are terms like: continuous and one- tory Technical Report 703, 1983.
[UIrich88a]
shot enablement, state coupling, state equivalence, Ulrich, K.T. and W.P. Seering, "Function Sharing in Mechani-
state antagonism, threshold, and rate confluence. cal Design," Proceedings of the Seventh National Confer-
This language was designed to allow descriptions o f ence on Artificial Intelligence (AAAI-88), St. Paul, MN, Au-
complex, highly non-linear device behavior, that gust I988.
can not be described with differential equations. lUlrich88b]
Ulrich, K.T. and W.P. Seering, "Computation and Pre-Paramet-
Rieger uses these descriptions to simulate the be- tic Design," MIT Artificial Intelligence Laboratory Technical
havior of the devices by assigning a computational Report 1043, October 1988.
behavior of each functional element in the device [Williams88]
description. Williams, B., "Principled Design Based on Topologies of Inter-
action," Massachusetts Institute of Technology Department
of Electrical Engineering and Computer Science, Ph.D. The-
Acknowledgments. This article describes work performed at the sis, 1988.
Artificial Intelligence Laboratory of the Massachusetts Institute
of Technology. Support for the laboratory's research is provided
in part by the Defense Advanced Research Projects Agency of
the United States Department of Defense under the Office of Appendix
Naval Research contract N00014-85-K-0124, and the National
Science Foundation under grant DMC 8618776. The presentation
A.1 Determining the Type Number from the
and substance of the research described in this article were im- System Equations
proved in response to helpful comments by Randall Davis, David
The procedure for determining the type number of a
Gossard, Patrick Winston, and the reviewers.
system given a differential equation relating an in-
put and output quantity is as follows:
References 1. Rearrange the equation so that the terms consist-
[Doyle88] ing o f derivatives of the input are on the right-
Doyle, R.J., "Hypothesizing Device Mechanisms: Opening Up hand side (RHS), and the terms consisting of de-
the Black Box," Massachusetts Institute of Technology De- rivatives of the output are on the lefthand side
partment of Electrical Engineering and Computer Science (LHS).
Ph.D. Thesis, 1988.
2. Perform a LaPlace transformation on the expres-
[Kannapan87]
Kannapan, S. and K.M. Marshek, "Design Synthetic Reason- sion leaving an equation between polynomials in
ing: A Program For Research," Mechanical Systems and De- the LaPlace operator, s, and the input and output
sign Technical Report 202, University of Texas at Austin, quantity. (In the case o f equations of this form,
Department of Mechanical Engineering. the LaPlace transform is performed by simply
18 Ulrich & Seering: Synthesis of Schematic Descriptions in Mechanical Design

replacing s" for d'/dt'). Multiply both sides of candidate graph once it has been simplified to con-
the equation by s raised to some power to elimi- tain only generalized resistances, capacitances, in-
nate any s terms in the denominators of any frac- ertances, and sources. This procedure corresponds
tions. Factor out as many s's as possible from to the informal description given in section 3.3.2.
both sides o f the equation. If an s term can be
1. In the case of a bondgraph of type - t with an
factored out o f both sides of the equation, sim-
effort source as input, adding a l - - R (a new one-
plify the equation by dividing both sides by the
junction with an attached R) along the power
lower-order of the two factored-out s terms.
spine will make it a type 0 bondgraph. In the case
3. We call the order of the remaining factored out s
of a bondgraph of type - l with a flow source as
term the type number of the system. If the s term
input, adding a 0 - - R along the power spine will
is on the RHS, then the type number is positive.
make it a type 0 bondgraph.
If the s term is on the L H S , then the type number
2. In all other cases, the type number o f a graph is
is negative.
exactly the number of isolated 1 - - C or 0 - - 1
As an example consider the following equation be- groups along the power spine of the graph. An
tween an input pressure and an output velocity (cor- explanation of isolation is given below.
responding to piston-cylinder connected to a mass,
The power spine of a graph connects the input n-
a spring and a damper):
port to the output n-port. The spine can be thought
of as a string of bondgraph groups of the form
d2 d d
M~V + B ~ t V + K V = A ~ttP (x--A, y--B, z - - C ) where x, y and z are n-ports and
A, B, and C are 1-ports. The bondgraph groups
1 - - C and 0 - - I cause differentiation of the input if
Performing the LaPlace transform leaves: they are isolated. This differentiation is what deter-
mines the type number of a system. An isolated
MVs 2 + BVs + KV = APs.
1 - - C or 0 - - I group is one whose neighbors (on ei-
ther side of it in the p o w e r spine string) are isolating
Since there is one s that can be factored out of the
groups.
RHS, the system has a type number of l, meaning
that the input pressure changes with the integral of 1. An isolating group for a I - - C is a 0 - - R , a 0 - - I ,
the output velocity (position). One important prop- or a O--SE; a 1--I or 1 - - R occurring at the last
erty of this derivation procedure is that it is al- 1-port in the bondgraph sequence; or combina-
gorithmic and can be automated. tions of 1 - - R ' s or l - - l ' s and any other isolating
groups (SE indicates effort source, SF indicates
flow source). This last c a s e - - a combination of
A.2 An Explanation of Isolated Groups Using 1 - - R ' s or l - - l ' s and any isolating g r o u p - - i s
Bondgraphs 4 also an isolating group because a I - - R or a 1--1
Although we have used mnemonic icons to show next to 1 - - C does not influence the type number
the schematic descriptions in this article, the repre- of a graph.
sentation we have used in this project is bondgraphs 2. The isolating groups for a 0 - - I are either l - - R ' s ,
[Paynter61, Rosenberg83]. In the bondgraph repre- l - - C ' s , or 1 - - S F ' s ; 0 - - C ' s or 0 - - R ' s occurring
sentation, our descriptions are simply sequences of as the last 1-port in a bondgraph sequence; or
2-ports and zero- or one-junctions with attached l- combinations of 0 - - C ' s or 0 m R ' s and any other
ports that connect a source element to an n-port isolating group.
corresponding to the output quantity of interest. In Given this statement o f the domain knowledge, the
this context, a power spine is a sequence o f trans- derivation of modification operators is trivial. Ex-
formers and/or gyrators connecting a graph chunk cept in the case of systems of type - I , the differ-
that represents the input environment of the device ence between the type number of the candidate
to a graph chunk that represents the output environ- design and the type number of the specification
ment of the device. Given this representation, the is equal to the number of new isolated 1 - - C ' s or
concept of an isolated group can be defined as a 0 - - I ' s that must be created in the graph. If one
syntactic pattern within the graph sequence. In this isolated group must be created then an isolated 0 - - I
section, we given the procedure for modifying a or a 1 - - C could be added, or a non-isolated 0 - - I or
1 - - C already in the graph could be isolated by add-
4 This section presumes familiarity with bondgraph notation. ing an isolating group.

You might also like