Challenges in Capitalizing Knowledge in Innovative Product Design Process
Inès SAAD
MIS, University of Picardie Jules Vernes
Amiens School of Management, 18 Place Saint-Michel,
80000 Amiens, France
Michel GRUNDSTEIN
and
Camille ROSENTHAL-SABROUX
LAMSADE, University of Paris-Dauphine, Place du Maréchal de Lattre de Tassigny
75775 PARIS Cedex 16
Abstract. Capitalizing on company’s knowledge is
increasingly being recognized in a private organizations
environment since managing knowledge productivity is
considered a source of competitive advantage. In this
paper we present a generalization of GAMETH
framework, that play an important role in identifying crucial
knowledge used and created in innovative product design
process. Thus, we have developed a method based on three
phases. In the first phase, we have used GAMETH to identify
the set of “reference knowledge”. During the second phase,
decision rules are inferred, through rough sets theory, from
decision assignments provided by the decision maker(s). In the
third phase, a multicriteria classification of “potential crucial
knowledge” is performed on the basis of the decision rules that
have been collectively identified by the decision maker(s).
Keywords: Knowledge Capitalizing, crucial knowledge, Multicriteria classification, Decision rules, Dominance Rough set
approach.
1. INTRODUCTION
The knowledge created in innovative product design
process has some characteristics. First, this knowledge is
specific to the innovative product. It is mainly based on
tacit knowledge [11] of the project experts gained from
previous projects and they do not necessarily apply to the
innovative product even if such experience is still
important to search new concepts. Second, generally the
lifetime of most knowledge used to develop the
innovation product is very short because one part of
knowledge is not validated in the innovative product
project development or because the company’s objectives
change rapidly. In the automotive sector, capitalizing on
the knowledge used in design process, that is, locating,
preserving, enhancing value and maintaining this
knowledge is very complex [18]. It involves more and
more heavy investments in order to convert unstructured
tacit knowledge into explicit knowledge to be integrated
in corporate memory defined as “Explicit, disembodied,
persistent representation of knowledge and information
in an organization” [2].
As resources of the company are limited, the automotive
company must define accurately the knowledge to be
integrated in the design process’s corporate memory. In
our case study, the goal is to propose a method to identify
crucial knowledge in order to justify a situation where
knowledge capitalization, specifically in the context of
decision-making, is advisable.
The rest of the paper is organized as follows. Section 2
synthesizes the related research studies. Section 3
presents experimentations. Section 4 presents the
methodology. In Section 5 we present the application of
the methodology in the automotive French Company.
Section 6 concludes the paper and presents our current
and future work.
2. RESEARCH STUDIES
In literature, there are only few works that are interested
in the identification of the knowledge on which
preservation operation need to be conducted. Several
authors, including [3] [5] [8] [9] [12] [25] consider
crucial knowledge delimitation process as a hard
operation.
The need for pertinent and crucial knowledge in any
knowledge capitalizing operation has been proved by
several authors (e.g. [1] [2] [5] [6] [20]). Only few
theoretical and empirical works are available in literature.
Concerning knowledge collection, we think that the
method proposed by [6] enables to study the area and to
clarify the needs in knowledge required to deal with
pertinent problems through the modeling and analysis of
sensitive processes in the company. This approach
involves all the actors participating in the area of the
study. Finally, the method proposed by [4] is evenly
based on both a series of interviews with the leaders and,
the study of strategic documents. These two last
approaches suppose that the leaders are able to identify
the knowledge to evaluate.
Our analysis of these approaches at the level of criteria
construction and knowledge evaluation permits us to
remark that the methods proposed by [6] construct
criteria intuitively. In turn, Tseng and Huang propose to
compute the average score of each attribute of the
knowledge as a function of the evaluations provided by
each analyst. Then, the analyst evaluates the important of
each knowledge in respect to each problem. Finally, the
average global is computed for each analyst. One
limitation of this method is that the scales used are
quantitative. However, due to the imprecise nature of the
knowledge, qualitative scales are preferred.
3. EXPERIMENTATION
We carried experiments in order to show whether the
decision rules resulting from the identification phase of
crucial knowledge are effective. We considered a set of
forty “Potential crucial knowledge” items and classified
them in two classes: (1) “not crucial knowledge” (Cl1)
and (2) “crucial knowledge” (Cl2).
Cl1
Cl2
DM1
DM2
DM3
DM4
0,46
0,75
0,58
0,81
0,3
0,77
1
1
AVERAG
E
0,58
0,83
Figure 1. Approximation quality of crucial knowledge
4. METHODOLOGY
The methodology for crucial knowledge identification
and evaluation is composed of three phases (Figure 2). A
detailed description of it is available in [16].
Phase 1: Determining “Reference Knowledge”
The first phase is relative to constructive learning
devoted to infer the preference model of the decision
makers. Constructive learning, as opposite to descriptive
learning, suppose that the preference model is not preexisting but is interactively constructed by explicitly
implying the decision maker. Practically, it consists in
inferring, through the DRSA (Dominance-based Rough
Set Approach) [4] method which is an extension of rough
set theory [10] and which is devoted to multi-criteria
sorting problems of a set of decision rules from some
holistic information in terms of assignment examples
provided by the decision makers. This set of rules may be
used in the same project or in other similar or new
projects. However, for similar or new projects an
adaptation of the set of decision rules to the project under
consideration often required. This phase includes also the
identification, using GAMETH (Global Analysis
METHodology) framework, of a set of “Reference
crucial knowledge”.
Table 1. Quality of approximation
The evaluation of each knowledge in this test set is
carried with the help of the decision maker. Table1
reports the quality of approximation with respect to four
individual decision rules corresponding to four decision
makers (DM). The average 0,83 of Figure 1 shows that
we have various results depending on the decision
maker's preferences. In addition, the average of
approximation quality of crucial knowledge (Cl2)
determine with GAMETH framework is 0, 83.
Figure 2. The methodology for crucial knowledge
identification and evaluation
Phase 2: Constructing Preference model
The second phase includes the construct of preference
model and the evaluation of knowledge with the respect
to a convenient set of criteria [13]. Inspiring from the
systemic approach of [7] and by using the bottom-up
approach [15] [16], three sub-families of criteria where
constructed: (i) knowledge vulnerability family that are
devoted to measure the risk of knowledge lost and the
cost of its (re)creation; (ii) knowledge role family that are
used to measure the contribution of the knowledge in the
project objectives and (iii) use duration family that is
devoted to measure the use duration of the knowledge
basing on the company average and long term objectives.
The criteria used to evaluate the “knowledge of
reference” were constructed through a combination of the
top-down and bottom-up approaches. The top-down
approach was used to identify the indicators from which
the criteria g1,…, g15 are constructed. These indicators
were defined basing on the theoretical research in
knowledge engineering, strategic management and
artificial intelligence domains and on the empirical
studies conducted in the French car company see [23] for
details.
To make the evaluation phase easier, we should analyze
the “knowledge of reference”, i.e. identify the process
where the knowledge is used, the person gathers it, the
tacit level, production time and see if it is validate or not.
To evaluate each knowledge Ki in respect to the each
objective Oj, we have developed the computing model
[21] [22]. The evaluation of knowledge in respecter to
criteria of families (i) and (iii) are normally provided by
the decision maker. However, in practice the decision
makers may show some difficulty in directly evaluating
knowledge in respect to some complex criteria. To
overcome this problem, complex criteria are decomposed
into several more simple indicators. The decision makers
can easily evaluate these indictors.
Once all knowledge items are evaluated with respect to
all criteria, the next step is an iterative procedure
permitting to conjointly infer the decision rules. Two
decision classes have been defined Cl1: “non crucial
knowledge” and Cl2: “crucial knowledge”.
Phase 3: Classifying potential crucial knowledge
In the third phase, the decision maker use the preference
models (decision rules) of the different stakeholders
defined in the first phases to assign the new knowledge,
called “potential crucial knowledge”, to the classes Cl1
or Cl2. More specifically, a multi-criteria classification of
“potential crucial knowledge” is performed on the basis
of the decision rules that have been collectively identified
by the decision maker(s) in the first phase. The term of
“potential crucial knowledge” should be mapped to the
concept of “potential action” as defined in the multi-
criteria decision-aid theory, that is, “real or virtual actions
considered by at least one stakeholder as a temporally
realistic one” [14]. “Potential crucial knowledge” is the
knowledge that has been temporary, identified as crucial
by at least one stakeholder. The generated “potential
crucial knowledge” are analyzed and then evaluated
against the criteria identified in the first phase. Then, they
are assigned in one of two decision classes Cl1 or Cl2.
In fact, one “potential crucial knowledge” is regarded as
effectively crucial if there exists at least one decision rule
within the rules base, whose premises are paired with the
evaluation of this knowledge on the set of criteria.
The general form of a decision rule is:
If gj (k) ≥ rgj ; j {א1,…,m} then k אCl2 where
- g1,… , gm is a family of m criteria,
- gj (k) is the performance of the knowledge k on criterion
gj
- (rg1, … , rgm) אVg1 x …x Vgm is the minimum
performance of a knowledge k on the set of criteria.
5. CASE STUDY
The proposed methodology was conceived and validated
in the French Car Company. More specifically, we have
focalized on the depollution systems. The objective of the
French car company is to transfer the knowledge
developed in the depollution system for use with:
-
Other types of vehicles
Projects concerned with definition of the new
depollution systems.
Phase 1: Determining “Reference crucial Knowledge”
To identify the “knowledge of reference”, we have
applied GAMETH framework. This framework is
composed of four steps. The first step is composed of
four substeps. The first substep permits to define the
organizational model of the depollution system project
under study, i.e., define the study area, construct the
organization chart and formalize the objectives in
hierarchical form to help the decision makers identify
sensitive processes. In the second substep we identify,
with the help of the project responsible, the sensitive
processes. Two sensitive processes are: “Choice of filter
support” and “Design and methodology of supervisor
calibration”. The third substep concerns the modeling
and analysis of these processes as well as the study of
“critical activities” associated with each process. In the
last step we identify the sources of knowledge and their
localization.
Phase 2: Constructing Preference model
Since our objective is to identify crucial knowledge, we
have analyzed and characterize those knowledge that are
mobilized in the different critical activities related to each
sensitive process. We have often called to model the
creation process of each of these knowledge. Table 2
illustrates the result of the in-depth analysis of the
knowledge relative to “the choice of material”. To assure
good choice of material, the filtration system needs to be
efficient whatever the rolling system. The choice of the
material includes the constraints relative of the engine
working, implementation and storage of residue.
Table 2. Analysis of the knowledge relative to “the choice of
material”
Three sub-families of criteria where constructed : (i)
knowledge vulnerability family including the eight
criteria g1,…,g8 that are devoted to measure the risk of
knowledge lost and the cost of its (re)creation; (ii)
knowledge role family including the criteria g9,…,g14
that are used to measure the contribution of the
knowledge in the project objectives. The criteria g9,…,g14
are specific to the depollution system project and should
be replaced by other ones for other projects. These
criteria correspond to the objectives in the contribution
degree computing model and (iii) it use duration family
including the criterion g15 that is devoted to measure the
use duration of the knowledge basing on the company
average and long term objectives.
Once criteria family is constructed, we need to evaluate
each knowledge of reference in respect to all criteria. We
have distinguished three family of criteria which permit
to measure the vulnerability of the knowledge and
implies criteria {g1 : complexity, g2 : accessibility, g3 :
substituability , g4 : validation type , g5 : transferability,
g6 : rarety, g7 : acquisition cost and g8 :acquisition time} ;
the role of each knowledge in each objective and implies
criteria g9, g10, g11, g12, g13 and g14; and use duration of
each knowledge which implies criterion {g15 : use
duration}.
As mentioned earlier, the evaluations of “knowledge of
reference” in respect to criteria g1, g2,…,g8 are provided
by the decision makers. For example, in respect to
criterion complexity, the knowledge “relative to different
characteristics that exist between depollution system
command law and the other CMM command laws" is
considered as “very complex” since this knowledge
depends on several other knowledge related to the law of
EGR (Exhaust Gaz Recirculation) command, the law of
CAN (Controller Area Network) command, the law of
gearbox command, to the injection system and to the law
of depollution system command.
To infer rules, we have constructed four decision tables
containing the evaluations of 34 "knowledge of
reference" in respect to 15 and to the assignment
examples provided by four decision makers.
We present in Table 3 an extract from the decision table
concerning the assignment of three knowledge of
reference”.
Table 3. An extraction from the decision table for one decision
maker
First, each decision maker selects the decision rules. We
have applied the DOMLEM algorithm, proposed in
DRSA [10] method to infer rules permitting to
characterize knowledge assigned to classes Cl1 and Cl2.
The set of decision rules identified by decision maker r
permit to establish Table 4. The result obtained are
traduced in the form of approximation quality, and
permitted us to verify the presence of inconsistences in
the decision rules. These rules are deduced from the
comparison of information related to the assignment
examples intuitively provided by each decision maker,
and the assignment generated by the algorithm. To
illustrate the incoherence, we consider the assignment of
a given decision maker r. Initially, decision maker r
assigns K11, K14, K15, K16 and K21 simultaneity to Cl1 and
Cl2. Thus, we have called this decision maker to
carefully reconsider the evaluation of each of these
knowledge. Concerning knowledge K11 and K15, the
decision maker mentioned that hesitated when he
assigned these knowledge. For knowledge K14, K16 and
K21, there is no remark and we do not modify his/her
assignment. We have reviewed with all the decision
makers that have provided inconsistent decision rules and
that are ready to modify his/her assignment examples.
Once each decision makers chooses the decision rules
relatives to different assignment examples, we determine,
jointly with the decision makers, a subset of decision
rules that permit to evaluate the crucial knowledge. Three
examples of jointly selected decision rules follows
(expressed in mathematical form):
Table 4. Approximation qualitative decision maker r
Rule 1: If g3 (k) ≥ 3 רg6 (k) ≥ 2 רg9(k) ≥ 5 רg15(k) ≥2
Then x ≥ א2Cl
Rule 2: If g3 (k) ≥ 2 רg6(k) ≥ 2 רg12(k) ≥ 4 רg15(k) ≥2
Then x ≥ א2Cl
Rule 3: If g1(k) ≥ 3 רg3(k) ≥ 2 רg8(k) ≥ 4 רg15(k) ≥ 2
Then x אCl ≥2
In the system, Rule 2 is traduced as follows:
IF Ki. Substitutable –Level is “at least weak”
and
Ki. Rarety-Level is “at least rare”
and
Ki. Competitivity is “at least high”
and
Ki.use-duration is at least “average”
THEN Ki is at least in Cl2
This rule means that a piece of knowledge Ki is
considered crucial (i.e. Ki belongs to the class of at least
crucial Cl2), if it is difficult to replace it, it is scares, have
an important impact on commercial position of the
company and also has convenient use duration.
Phase 3: Classifying potential crucial knowledge
In this phase, the system use decision rules defined in the
first step to assign new “potential crucial knowledge” to
either Cl1 or Cl2. Those assigned to Cl2 are the crucial
ones that need to be capitalized on.
1) Step1. Definition of a “potential crucial knowledge”
set: First, we have identified, with the help of the
stakeholder, the decision makers implied in this second
phase. There are 6 implied decision makers. These are
the ones that have participated to phase one plus the
responsible on the cooperation with another automobile
constructor company. With all these decision makers, we
have first retained all the knowledge that are supposed
potentially crucial and then we have combined some ones
(that they find very detailed) and removed/added some
another ones. The final list is obtained after individuals
discussion with the different decision makers and
validated through emails with all of them. The choice of
the set is facilitated by the analysis of process and
activities performed during the definition of knowledge
of reference process.
2) Step2. In-depth analysis of “potential crucial
knowledge”: we have applied for each “potential crucial
knowledge” the same process as applied in step 2 of
phase 1.
3) Step 3.Evaluation of “potential crucial knowledge”:
We have evaluated all potential crucial knowledge in
respect to all criteria constructed in step 3 of phase 1. The
obtained performance table contains the evaluation of
each “potential crucial knowledge” in respect to criteria
related to:
- The vulnerability of knowledge (i.e g1, g2, g3, g4, g5,
g6, g7, g8);
- The role of knowledge for each objective (i.e. g9, g10,
g11, g12, g13, g14) ; and
- Use duration (i.e g15)
4) Step 4. Application of decision rules:
We have used the performance table containing the
evaluation of different knowledge of reference as input in
this phase. Thus, it will be required only one rule (that
characterize knowledge required a capitalizing operation)
is verified to conclude that the knowledge is crucial.
6. CONCLUSION
In this paper we have presented a generalized method to
make GAMETH usable for any complex project. We
have developed a novel methodology that constructs the
set of “crucial knowledge”. This methodology consists of
three phases. During the first phase, decision rules are
inferred, through rough sets theory, from decision
assignments provided by the decision maker(s). It
includes the identification of a set of “reference
knowledge” and its evaluation with respect to a
convenient set of criteria. In the second phase, a
multicriteria classification of “potential crucial
knowledge” is performed on the basis of the decision
rules that have been collectively identified by the
decision maker(s).
Several points related to the methodology itself need to
be investigated. The contribution degrees model should
take into account evolution of different industrial projects
concerned by the capitalization operation. For example,
during our experiences at automobile company, some
data relative to the use of a chemical substance in the
DEPOLLUTION
system were qualified as very
important by the actors, and hence the corresponding
knowledge were computed as important by the model.
Eight months later, this substance is not used any more in
the project. One possible solution to tackle this problem
is to use robustness analysis [14]. More precisely, this
type of uncertainty may be modeled in terms of scenarios
corresponding to the possible combinations of different
values attributed by each actor to the contribution of each
knowledge to each objective.
[17]
[18]
7. REFERENCES
[1] Dieng-Kuntz, R., D. Minier, F. Corby, M. Ruzicka, O.
Corby, L. Alamarguy, Ph. Hiep Luong (2004), Medical
Ontology and Virtual Staff for a Health Network, In
Enrico Motta and Nigel Shadboldt eds, Engineering
Knowledge in the Age of the Semantic Web: Proceedings
of the 14th International Conference, EKAW 2004,
Whittlebury Hall, UK, October 5-8, 2004, Springer, Vol
3257, pp. 187 - 202.
[2] Dieng, R., O. Corby, A. Giboin and M. Rybière (1999).
Methods and tools for corporate knowledge management.
Technical report, INRIA, ACACIA project.
[3] Greco, S., B. Matarazzo and R. Slowinski (2001). , Rough
sets theory for multicriteria decision analysis. European
Journal of Operational Research, 129, 1-47.
[4] Grzymala-Busse, J. A new version of the rule induction
system LERS. Fundamental Informaticae, pages 27-39,
1997.
[5] Grundstein, M., C. Rosenthal-Sabroux, A. Pachulski (2003).
Reinforcing Decision Aid by Capitalizing on Company’s
Knowledge. European Journal of Operational Research,
145, 256-272.
[6] Grundstein, M. (2000). From capitalizing on Company
Knowledge to Knowledge Management. In: Knowledge
Management, Classic and Contemporary Works (Morey,
D., M. Maybury and B. Thuraisingham, Ed.), Chap. 12,
pp. 261-287. The MIT Press, Massachusetts.
[7] Le Moigne (1977). La théorie du système général. Puf,
Paris.
[8] Noh, J.B., K.C. Lee, J.K. Kim, J.K. Lee and S.H. Kim
(2000). A cased-based reasoning approach to cognitive
map-driven tacit knowledge management. Expert Systems
with application, 19, 249-259.
[9] Nonaka, I. and H. Takeuchi (1995). The knowledge creating
company: How Japanese companies create the dynamics
of innovation. Oxford University Press, New York.
[10] Pawlak, Z., (1982). Rough sets. International Journal of
Information & Computer Sciences, 11, 341-356.
[11] Polanyi, M., (1966). The tacit dimension. Routledge &
Kegan Paul Ltd. London.
[12] Pomian, J., C. Roche (2002). Management des
connaissances et organisation du travail. L’harmattan,
Paris.
[13] Roy, B. (1998). A missing link in OR-DA: robustness
analysis. Foundations of Computing and Decision
Sciences.
[14] Roy, B. and D. Bouyssou (1995). Aide multicritère à la
décision, Economica, Paris.
[15] Saad, I., (2005). Une contribution méthodologique pour
l’aide à l’identification et l’évaluation des connaissances
nécessitant une opération de capitalisation. Thèse de
doctorat, Université Paris-Dauphine. 23, 141-160.
[16] Saad, I., Grundstein M. and C. Rosenthal-Sabroux (2006),
An Approach to Assess Knowledge Cruciality for
[19]
[20]
Industrial Project, 12th IFAC Symposium on Information
Control Problems in Manufacturing (INCOM 2006), 17-19
may, 2006(accepted).
Saad, I., C. Rosenthal-Sabroux and M. Grundstein (2005).
Improving the Decision Making Process in The Design
project by Capitalizing on Company’s Crucial Knowledge.
Group Decision and Negotiation, 14, 131-145.
Saad, I., M. Grundstein and C. Rosenthal-Saboux (2003).
Locating the Company’s Crucial knowledge to Specify
Corporate Memory: A Case Study in Automotive
Company. In: Knowledge management and organisational
workshop proceedings, IJCAI, pp. 75-84, Acapulco.
Stefanowski, J. and D. Vanderpooten. Rough sets, fuzzy
sets and knowledge discovery, A general two stages
approaches to rule induction from examples, pages 317–
325. Springer, 1994.
Tseng, B. and C. Huang (2005). Capitalizing on
Knowledge: A Novel Approach to Crucial Knowledge
Determination. IEEE Transactions on Systems, Man, and
Cybernetics Part A: Systems and Humans. 35, 919-931.