04
février
Par

“Experience capitalization to support decision making ininventive problem solving”

par Pei Zhang, ICube équipe CSIP

le lundi 18 février à 10h

Amphithéâtre de Vinci, INSA Strasbourg

Composition du jury :

Denis CAVALLUCCI, Professeur, INSA Strasbourg

Cecilia ZANNI-MERK, Professeur, INSA Rouen Normandie

Yulia HICKS, Associate Professor, Cardiff University

Mauricio CAMARGO, Professeur, Université de Lorraine

Jean-Marc Le LannProfesseur, Université de Toulouse

Jean-François BOUJUT, Professeur, Institut polytechnique de Grenoble

Résumé de la thèse

Experience plays a crucial role in the resolution of problems. When in inventive problem solving activities, experience is composed of two parts: one is the specific know-how knowledge acquired in the practice of solving previous problems, the other is the additional knowledge from other domains where the problem solver is previously acquired and is used for problem solving. This thesis is motivated by the theory of inventive problem solving (TRIZ) and the knowledge engineering technologies, where the use of classical TRIZ tools to solve a specific problem requires a large amount of experience from the problem solver. Hence, this thesis aims at proposing a new way to solve new inventive problems by capitalizing experience obtained from past problem solving activities, thus improves the efficiency and the effectiveness of the current inventive design activity. In order to facilitate the capitalization of experience, the proposed approach adopts case-based reasoning (CBR) technique. The first contribution of our approach is based on the use of the case-based reasoning for collecting and rapidly accessing the experiences. In this way, the user with less expertise can solve new problems by reusing or revising the old solutions of other people without going through the classical TRIZ process. The second contribution consists in proposing a new way to classify the physical effects basé sur Wikipedia. In the proposed approach, each effect is first represented as a vector based on the use of Wikipedia. Each category associated with this effect that is retrieved from the Wikipedia database corresponds to an axis. Such that the values along the axes for each effect correspond to the distance between this category and the effect on the shortest path linking them. These vectors are then classified using the kNN classifier. Finally, the proposed method is applied to train four classifiers of the physical effects. The aim is to allow the direct use of the available physical effects knowledge base without having pre-acquired knowledge of other domains. 

To implement the proposed approach in this thesis, a web-based application called CBRID (Case-based reasoning for Inventive Design) is developed. A particular case of ”cloth hanger” is studied to illustrate the problem solving process based on the proposed approach. In addition to that, we conducted a set of experiments to evaluate our approach in terms of effectiveness and efficiency. We considered a number of new problems whose resolution is performed in three different ways: using the classic brainstorming approach, using an existing rule-based approach and using our proposed approach. The evaluation demonstrates that our approach improves the efficiency and effectiveness of the inventive design process.

Image : Shutterstock- NicoElNino

Laisser un commentaire

Votre adresse de messagerie ne sera pas publiée. Les champs obligatoires sont indiqués par un *