See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/46432582 Challenges in Capitalizing Knowledge in Innovative Product Design Process Article · January 2009 Source: RePEc CITATION READS 1 26 3 authors: I. Saad Michel Grundstein 66 PUBLICATIONS 163 CITATIONS 72 PUBLICATIONS 362 CITATIONS Université de Picardie Jules Verne SEE PROFILE Paris Dauphine University SEE PROFILE Camille Rosenthal-Sabroux Paris Dauphine University 76 PUBLICATIONS 330 CITATIONS SEE PROFILE All content following this page was uploaded by Michel Grundstein on 28 November 2016. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and read them immediately. 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.