[
CogSci Summaries home |
UP |
email
]
http://www.jimdavies.org/summaries/
Goel, A., Bhatta, S. & Stroulia, E. (1997) Kritik: An Early Case-Based
Design System. In Maher, M. and Pu, P. (Eds.)Issues and
Applications of Case-Based Reasoning in Design, Mahwah, NJ:
Erlbaum, pages 87--132.
@InBook{,
ALTauthor = {Ashok Goel, Sambasiva Bhatta, and Eleni Stroulia},
ALTeditor = {M. Maher and P. Pu},
title = {Issues and Applications of Case-Based Reasoning in Design},
chapter = {Kritik: An Early Case-Based Design System},
publisher = {Erlbaum},
year = {1997},
OPTaddress = {Mahwah, NJ},
OPTpages = {87--132},
}
Author of the summary: Jim Davies, 2002, jim@jimdavies.org
Cite this paper for:
- Variant design vs. Adaptive design [2]
- SYSTEM: Kritik
- SYSTEM: Kritik2
- Combining of CBR and MBR (combining of cased-based reasoning
and model-based reasoning)
- SBF provides a vocabulary for indexing design cases [4]
- SBF allows for the learning of new indices for better future indexing.
- Cases are indexed by function
- Structural relations: part-of, containment (of a substance), inclusion
(of a component), connection (seriel (output of one becomes input of
the other and parallel (share input and output))
- types of differences [30]
- transformations
Variant design: new designs differ from old designs because of
different values of parameters [2]
Adaptive design: new designs differ in the elements they have
The early Kritik experiment "showed that CBR provides a good process
account of the variant and adaptive aspects of preliminary design." A
key insight was the combination of model-based comprehension and
case-based reasoning (CBR).[3]
Kritik's domain: Simple electric circuits and heat exchangers
Kritik2's domain: that and electomagnetic devices and circuits with
amplifiers.[4]
Kritik2's innovations over Kritik: enlarged domain, hierarchical memory.
SBF provides a vocabulary for indexing design cases [4]. In Kritik2
cases are indexed by function (because problems are specified in terms
of function [22]). SBF allows
for the learning of new indices for better future indexing.
Structure: components and substances and their interactions.[5]
Function: input and output behavioral states, pointer to behaviors
that achieve these changes.
Slots in an SBF schema can be filled with an element from the set of
primitive entities, another schema or list of schemas.[12]
Structural relations: part-of, containment (of a substance), inclusion
(of a component), connection (seriel (output of one becomes input of
the other and parallel (share input and output))[12]
Behavioral state slots:
- previous (state)
- next (state)
- enabled-by (preceding state-transition)
- enabling (succeeding state-transition)
And then one of the following:
- substance-state-schema:
- location
- main-substance
- is-a (pointer to it's type)
- contained-substances: (optional, their description at this
state)
- component-state-schema:
- component
- mode (mode of the component's operation at this
state e.g. closed switch)
Functional Specification (from Sembugamoorthy and Chandrasekaran
1986): [15]
- makes (output behavioral state)
- given (optional. input behavioral state)
- by (causal behavioral state sequence)
- stimulus (optional. trigger for process)
- provided (optional. external conditions necessary)
Functions are a subset of observable behaviors that are intended by
the designer.
State transitions get you from one behavioral state to another: [16]
- previous-state
- next-state
- by-behavior (optional. pointer to a more detailed behavior)
- using-function (optional. list of Components and their
functions.)
- as-per-domain-principle (optional. e.g. physics laws)
- parameter-relations (optional. describes changes in parameter
values.)
- condition (optional. necessary conditions)
Kritik2's knowledge of cases is specific, but the knowledge of
components and substances is generic. [18]
Components are black-boxed: [19]
- is-a
- structural-relations (between itself and other components)
- parameters (slots and values)
- modes (modes of operation)
- functions (list of functions)
- connecting-points (places where other components can connect)
Retrieval: A probe is formed based on the desired function (which is
in the problem specification.) Partially matching cases are retrieved
and ordered according to ease of adaptation. [8] It looks at
differences between the function of the considered case in memory and
the problem specs and chooses only those that it can reduce the
differences of. Kritik2 knows how to reduce substance and
substance-property-value differences. [23]
Adaptation and evaulation: "diagnoses" and "repairs" the retrieved
case.[9] The device is adapted and then qualitatively simulated to see
if it gets the desired effect. Then it's stored.
Difference types: [30]
- substance
- substance-property-value
- substance-location
- component
- component-modality
- component-parameter
Modifications (transformations): [31]
- substance substitution (generalization and specification)
- component modification (replacement, modality change,
component-modality change, component-parameter adjustment)
- relation modification (series-to-parallel and parallel-to-series)
- substructure deletion (e.g. component deletion)
- substructure insertion (e.g. substructure replication)
Summary author's notes:
- page numbers are from my personal printout
Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:
JimDavies
(jim@jimdavies.org)
Last modified: Thu Mar 14 15:07:50 EST 2002