SRKB subgroup: knowledge sharing for engineering modeling
Tom Gruber <Gruber@Sumex-AIM.Stanford.edu>
Message-id: <2922600431-1915582@KSL-Mac-69>
Date: Wed, 12 Aug 92 02:27:11 PDT
From: Tom Gruber <Gruber@Sumex-AIM.Stanford.edu>
To: srkb@ISI.EDU
Subject: SRKB subgroup: knowledge sharing for engineering modeling
As a working group of the DARPA Knowledge Sharing effort, we
are looking for ways to facilitate the sharing and reuse of
knowledge used in engineering modeling. Our long term goal is
to enable libraries of reusable bodies of formally represented
knowledge. In particular, we are studying the use of a set of
common terms, or representation primitives, that would allow us
to build tools that share domain models and testbed examples,
or exchange knowledge at run time. Such a set of terms is
called a common ontology. For example, the PDES organization
has formalized an ontology for describing the geometry of
parts; this allows CAD vendors to build software that can
accept geometric models in an openly-specified format.
We are looking to collect and analyze ontologies for other
sorts of engineering models, such as those for modeling the
behavior and function of electromechanical and thermodynamic
systems.
The representations in which engineering models are formulated
can help us leverage off an investment in modeling by
supporting multiple tasks; for example, formal behavior models
of physical systems can be used as inputs to programs
performing a variety of tasks, from design to diagnosis.
And since the physics underlying engineering is rather
well-established, it should be possible to share device models
across tools performing similar tasks but developed
independently.
However, the reuse of these device models is currently hampered because
the modeling formalisms are idiosyncratic. Some of the differences are
purely syntactic, but many are due to different ways in which physical
behavior can be modeled: as discrete events, causal state transitions,
differential equations, etc. To enable the sharing and reuse of
engineering models, the conceptual framework upon which each modeling
approach is based should be made explicit. We are studying the role of
declarative knowledge representations as mechanisms for sharing
engineering models by analyzing and developing ontologies for various
modeling approaches. Such ontologies are intended to play the role of
"open" specifications for software reuse and interoperability.
Two projects are currently underway in this area, and they are
described below. This message is an invitation for you to participate
in the activities of this working group. Possible contributions
include:
- identifying existing ontologies or similar formal accounts, of
engineering modeling approaches and technologies (e.g., math and
simulation tools).
- helping to formalize established engineering modeling approaches
- helping to develop modeling ontologies and make existing ontologies
publicly accessible.
If you are interested in participating, please contact
gruber@sumex-aim.stanford.edu. Thank you.
--Tom Gruber
CURRENT PROJECTS IN KNOWLEDGE SHARING FOR ENGINEERING MODELING
Two projects are currently underway that are associated with the
special interest subgroup on knowledge sharing for engineering modeling.
One is an effort by representatives of projects in qualitative
physics to specify a common language for model fragments.
Model fragments are conceptual building blocks for programs
that formulate and assemble engineering models of device
behavior, using techniques such as Compositional Modeling
[Falkenhainer and Forbus, AIJ, 1990]. For example, idealized
components such as resistors and physical processes such as
liquid flow are represented by model fragments, which are
composed to produce simulation models of complete systems. The
language under development is a unification of model
formulation and simulation systems such as QPE, DME, and QPC,
and should enable a community library of model fragments that
can be directly executed by these systems. The semantics of the
language forms and the ontological commitments assumed by these
model-formulation systems will be formalized as an ontology of KIF
definitions and axioms. This work is in progress, but it appears that it
may be possible to share a library of model fragments (via a common core
language) and still allow for extension, research, and experimentation.
A second project is concentrating on characterizing the space of engineering
models, trying to identify points in common among approaches. This work grew
out of the Summer Ontology Project, held at Stanford in 1990. In a series of
meetings, computer scientists and engineers from Stanford and local research
laboratories examined the assumptions underlying a variety of approaches used
to model four electromechanical motion-control devices. It appeared that
several modeling approaches, from digital circuit modeling to rigid body
dynamics, seemed to be based on a commitment to a class of behavior models
called lumped-element models. In a lumped-element model, the behavior of a
device is described in terms of values of state variables (unary functions)
that map a single independent variable (e.g., time, but not space) to
physical quantities (of position, force, etc.). Work was started on an
ontology to formalize notions like lumped parameter, state variable,
quantity, dimension, etc., and how they can be reflected in a modeling
representation.
Recent work is developing a family of ontologies, factoring the lumped
element ontology into modular pieces. It is formalizing the classes
of algebras used in constraints (e.g., with or without differential
equations; qualitative operators), the assumptions underlying
hierarchical component/connection topologies, the relationships
between continuous and discrete state spaces, and the various styles
of dynamics analysis (e.g., Newtonian, LaGrangian, Kane's method).
This work is complementary to the composition modeling effort; any of
these styles of modeling can be formulated using the model fragment
language.
For more information contact
Brian Falkenhainer <falkenhainer@parc.xerox.com> or
Tom Gruber <gruber@sumex-aim.stanford.edu>.