Res Eng Des (1990) 1:149-166 Research in Engineering Design
© 1990 Springer-Verlag New York Inc.
Mechanism Comparison and Classification for Design
Leo Joskowicz
IBM T. J. Watson Research Center, Yorktown Heights, New York, USA
Abstract. When designing new devices, engineers sel-
domly start their design from scratch. Instead, they rede-
sign or compose existing ones. Finding candidate devices
based on their properties is thus an essential design sub-
task. To be effective, future intelligent computer-aided
design (CAD) systems must be able to compare and eval-
uate existing devices. Due to the large number of known
devices, such systems must create and maintain a de-
vices' knowledge base, indexed by the properties of the
devices.
This paper presents a new method to compare and
classify mechanical devices according to their kinematic
properties. Mechanism comparison determines when two
mechanisms are kinematically equivalent. Mechanism
classification organizes classes of equivalent mechanisms
for efficient retrieval. Both tasks require describing
mechanisms' behaviors from various perspectives and at
multiple levels of abstraction. To produce such descrip-
tions, we develop a set of operators to simplify and a'b-
stract kinematic descriptions derived from configuration
spaces. Simplification operators ignore irrelevant infor-
mation by incorporating constraints and assumptions.
Abstraction operators ignore detail by defining multiple
levels of resolution. The resulting hierarchy supports the
design process throughout its evolution, from the concep-
tual to the detailed phase. It provides an explicit link
between behavioral specifications and part geometry.
1 Introduction
When designing new devices, engineers seldomly
start their design from scratch. They either modify
existing devices or build new devices by composing
existing ones. For example, the front-wheel drive
car was designed by modifying key elements of the
rear-wheel drive car, not by designing a new car.
Likewise, new transmissions are designed with ex-
isting components, such as ball bearings and gears.
In both cases, searching for a known device that
best matches part of the given specifications is a
Reprint requests: IBM T. J. Watson Research Center, P.O.
Box 704, Yorktown Heights, NY 10598, USA
crucial design subtask. Finding such a device re-
quires comparing and evaluating existing devices.
Engineers compare and evaluate existing design
solutions based on their knowledge and prior expe-
rience. This knowledge guides them through thou-
sands of known mechanisms--amplifiers, door
locks, carburetors--and directs them to the impor-
tant properties of each. Both the focus and level of
detail at which the devices are examined vary
widely from task to task. For example, when de-
signing a coupling to connect a clutch to a gearbox,
the detailed functioning of the clutch and gearbox
are of secondary importance. Focusing on the prop-
erties of the clutch and gearbox shafts suffices. Or
compare the design of precision gear transmissions
and car window regulators. The former requires
precise models of gear types--bevel, helicoidal,
conical--and the exact gear ratios. Such distinc-
tions are unnecessary for the later.
To be effective, future intelligent computer-aided
design (CAD) systems must be able to compare and
evaluate existing design solutions. Due to the large
number of known devices, such systems must main-
tain a devices' knowledge base, indexed by the
properties of the devices. Creating and maintaining
comprehensive engineering knowledge bases re-
quires the ability to analyze, describe, compare,
and classify devices according to their functional
and behavioral properties. This in turn requires de-
scribing devices at different levels of abstraction
and from different simplifying views.
This paper describes a new method to compare
and classify mechanical devices according to their
kinematic properties. Kinematic properties de-
scribe types of object motions and their relation-
ships. Mechanism comparison determines when
two mechanisms are kinematically equivalent.
Mechanism classification organizes classes of
equivalent mechanisms for efficient retrieval. Both
tasks require describing mechanisms' behaviors
from various perspectives and at multiple levels of
abstraction. To produce such descriptions, we de-
velop a set of operators to simplify and abstract the
behavioral descriptions produced by existing mech-