| Titre : | Towards Predictive Maintenance of Intelligent Production Systems |
| Auteurs : | Abdelwahab Hamadouche, Auteur ; khadidja yahyaoui, Directeur de thèse |
| Type de document : | texte manuscrit |
| Editeur : | Université mustapha stambouli de Mascara:Faculté des sciences exactes, 2023 |
| ISBN/ISSN/EAN : | SE02358T |
| Format : | 87P. / couv. ill. / 29cm. |
| Accompagnement : | disque optique numérique (CD-ROM) |
| Langues: | Anglais |
| Note de contenu : |
Despite the common idea about predictive maintenance that it is a costly service sold by a big tech company, like Matlab or Microsoft, that provides the factory’s technicians with a black box, In this thesis, we provide a guide to implementing that particular service with ease, starting by explaining the production system to the maintenance problem and moving on to provide a step-by-step tutorial theoretically and practically with a real-world example using the latest emerging technologies and algorithms. While doing our research we encountered a shortage of research in that particular area. For example, in some of the proposed algorithms, like CNN, there is no research that tries to adopt that particular algorithm on sequential timestamped data, a lot of researchers just use it the same way it’s used with imagery data, the same words can be said on transformers. The same shortage can be found in the preprocessing step and features engineering, there is almost no public research in this area despite its drastic importance. Another important work, to continue where we stopped here, can be a guideline to adopt reinforced learning to improve predictive maintenance pre-trained models. For the time being, we couldn’t find publicly published research on that particular problem. Also, one major part that was out of the scope of this research is the installation of sensors and their networking protocol. For a smart production system team willing to adopt predictive maintenance those were the missing parts from this thesis. Furthermore, the work done in this thesis can be optimised and taken to another level by stacking the models and training a meta-model for a better outcome. Although this was discussed theoretically, it was impossible to be applied by us due to the lack of sufficient hardware, since this can be computationally costly. Lastly, in the example presented in this research, we did not try to get the best models, as it comes to no use, the main objective was to present a guide, with that being said the presented models in this research are not optimised thus can not be used for a transfer learning technique. Transfer learning can rarely be applied to a predictive maintenance problem for production systems since, as discussed, production systems have very different kinds of data and it needs casespecific implementation. The ultimate goal is to maximize the operational efficiency of factories, eliminating any downtimes and making maintenance autonomous by the machines on their own with no human intervention. In other words, the objective is to design production systems that can function independently, completely removing the necessity for human involvement, even for maintenance. |
Exemplaires (1)
| Code-barres | Cote | Support | Localisation | Section | Disponibilité |
|---|---|---|---|---|---|
| SE02358T | INF944 | Livre audio | Bibliothèque des Sciences Exactes | 7-Mémoires Master | Consultation sur place Exclu du prêt |
Aucun avis, veuillez vous identifier pour ajouter le vôtre !
Accueil
