Ateme-Nguema, Barthélemy H. (2007). Conception optimale des cellules de fabrication flexibles basée sur l'approche par réseaux de neurones. Thèse de doctorat électronique, Montréal, École de technologie supérieure.
Prévisualisation |
PDF
Télécharger (12MB) | Prévisualisation |
Résumé
Cette thèse propose une heuristique hybride de résolution des problèmes de formation cellulaire. Notre approche en trois étapes s'amorce par la sélection du meilleur cheminement de fabrication en mettant l'accent sur la minimisation des coûts opérationnels. La seconde phase forme les ateliers de fabrication en utilisant un réseau de neurones de type Hopfield quantifié et fluctuant jumelé à une méthode d'optimisation locale représentée par « la recherche avec les tabous ». L'ultime phase de cette heuristique fut centrée sur la réduction ou l'élimination des transferts intercellulaires par la mise en place d'un équilibre entre maintenir les transferts, dédoubler les machines permettant ces transferts et recourir à la sous-traitance. Sur la base des simulations réalisées, nous obtenons des solutions réalisables 100% du temps alors que les meilleures dispositions sont déterminées 68 fois sur 100. De plus, notre approche est, en moyenne, 22 à 30 fois plus rapide qu'un réseau de Hopfield classiques dont les neurones prennent des valeurs discrètes ou continues.
Titre traduit
Optimal design of flexible manufacturing cells based on neural networks approach
Résumé traduit
Essentially, manufacturing industry always had major challenge to increase the profitability. Universalization, competitiveness, integration, innovation and technology transfer resolutely constitute the «leitmotiv» of this beginning of century. Moreover, turbulence and the change in manufacturing sector twinned with the smallness life cycle of products, with the consumers increased requirements and the continuous technological developments oblige the industrialists and the scientific community to the search, without end, of new concepts, new methods or work approaches, to innovate and push back the limits to the innovation and knowledge.
The flexible manufacturing cells concept is an application of the group technology developed in the middle of last century and returns to the physical organization of an · industrial installation by supporting production flows and by minimizing handling. Moreover, the dynamic cellular manufacturing systems (DCMS) constantly revalue the physical configuration and the composition of an industrial cell, so as to obtain an «optimal» installation according to marginal costs of handling and configuration on a
specific horizon. They make possible to have a highly dynamic production system able to answer to a turbulent environment. The design of the flexible manufacturing cells has an impact on the performance and the profitability of a producing entity seeking to reduce its costs and to offer a product centered on the customer requirements, focused on the satisfaction of the shareholders and centered on the employees responsible for the creation of the richness.
Qualified like «np-complete», the cell formation problems already hold the scientific community attention since decades and several approaches were developed. However, vis-a-vis of big sizes matrices, the algorithms, the heuristic ones and/or the mathematical approaches post major deficiencies which cause dissatisfaction and obliges us to pay more attention in order to find a manner of solving them in an effective and efficient way.
The three stages resolution algorithm which we propose starts with the selection of the routings of production in order to identify the best way to minimize the operational costs. Thereafter, hybrid heuristics of resolution, composed by a quantized and fluctuated Hopfield neural network and the Tabu search method, allows the global approach to determine the cells composition and to assign the parts, clustered in family, with the various flexible manufacturing systems formed. The elimination and/or the
reduction of the intercellular transfers found in the obtained cellular group constitute the ultimate phase of the process that we propose. In our research, the hybrid resolution model uses a quantized and fluctuate Hopfield neural network because, the neurons quantization could reduce the network size and maintaining its storage capacity whereas the fluctuation allows the Hopfield network to escape, regularly, of the local minima in which it could be trapped. Simulations and tests prove that our approach is effective and efficient. In fact, it determines feasible solutions 100% of time, «optimal» or «near optimal» 68% of time and is, approximatively, 28 to 30 times faster, on average, that a traditional network of Hopfield having neurons with continuous or discrete values.
Type de document: | Mémoire ou thèse (Thèse de doctorat électronique) |
---|---|
Renseignements supplémentaires: | "Thèse présentée à l'École de technologie supérieure comme exigence partielle à l'obtention du doctorat en génie". Bibliogr. : f. ([302]-317). |
Mots-clés libres: | Approche, Artificiel, Cellule, Conception, Evaluation, Fabrication, Flexible, Hopfield, Neurone, Optimal, Production, Reseau |
Directeur de mémoire/thèse: | Directeur de mémoire/thèse Dao, Thien-My |
Programme: | Doctorat en génie > Génie |
Date de dépôt: | 04 avr. 2011 17:44 |
Dernière modification: | 04 nov. 2016 23:46 |
URI: | https://espace.etsmtl.ca/id/eprint/548 |
Gestion Actions (Identification requise)
Dernière vérification avant le dépôt |