MAINTENANCE DES SYSTÈMES DISTRIBUÉS : MÉTHODES D'AIDE À LA DÉCISION TEMPS-RÉEL
K. P. Adzakpa · HAL (Le Centre pour la Communication Scientifique Directe) · 2004
The technological systems have greatly evolved during the last past decades. In the same time, their installations are more and more scattered on different distributed sites. This emphasized the increasing concern about the maintenance decision making processes where cooperative approaches are required, especially in distributed systems. In this thesis, we study methods for real-time decision making in the maintenance of large distributed multisite systems with important logistic constraints (set-up times), where the available maintenance resources are shared between the different sites in the system. The aims are to plan and schedule the maintenance tasks and assign the resources in real-time to them, while minimizing the costs that are involved. The different costs occur namely when the system operates in degradation states, when the maintenance is too often undertaken, or in case of insufficient maintenance. These costs depend on different parameters namely the weighted or unweighted response time (flow-time) of the maintenance tasks, their tardiness and their earliness. The costs also introduce linear and convex combinations of these different parameters and are generally based on the availability function which reveals the degradation state of the different assets in the system. To reach the targets, different problems ought to be solved in order to guarantee a minimal required availability to the system, and assign in real-time the resources to the maintenance tasks. This must take into account the different constraints on the maintenance tasks which have unequal release dates and occur in real-time in the system. The set-up times between these tasks are important and depend on their sequence. Moreover, the tasks are linked by precedence constraints and are characterized by their priority and their emergency which are of a great importance in maintenance. The resolution methods that we propose are based on scheduling approaches, with local optimality rules (depending on the time) that we prove in this document. The decision making rules consider, in a first time, a single maintenance resource with a link to single machine scheduling problems. We developed lower bounds for the off-line use of the rules, before adapting them in algorithms for real-time decision making in order to help the maintenance manager in the cooperative maintenance planning in distributed systems. The main goal is to minimize the different costs while controlling the intervention delay of the maintenance resources. This document contains 6 main chapters. The first one is a general introduction to maintenance and presents the different practices of maintenance in the industry. In the second chapter, the main aspects of maintenance in the scientific research are analyzed. Then we propose in chapter 3 different approaches to solve the problem relative to the critical states costs of the system, which correspond to the maintenance tasks' response-time. We develop, in chapter 4, approaches to handle the different costs which are due to deadlines on the maintenance actions. In this chapter, we consider namely the combined cost of the critical states and the maintenance tasks' tardiness on the one hand, and the costs due to the tasks' earliness and tardiness on the other hand. In order to analyze the performances of the different approaches developed in chapter 3 and chapter 4, and study the robustness of these approaches in a random environment, we simulated them in chapter 5 with uncertainties. Finally we present in chapter 6 the main conclusions of the document and future extensions to this work.