Nous étudions un problème de planification de la production sur une machine multitâche dans l'industrie de la fabrication de bottes d'hiver. Nous devons planifier la production d'une machine à injection de caoutchouc ayant huit bras et pouvant utiliser deux couleurs. L'objectif consiste à trouver un plan de production (horaire de production, heures des changement de couleurs, de modèles, de pointures, nombre de quarts de travail, ...) qui permet de terminer la fabrication le plus tôt possible tout en respectant les contraintes pratiques de l'entreprise. Les méthodes et le logiciel développés ont été implantés en entreprise avec succès.

This paper is about the hybrid flow shop with nowait in process. We are given two centres in which a set of identical machines are put in parallel to process n jobs so as to minimize the makespan. A new heuristic, based on the Gilmore and Gomory algorithm, is presented to solve the above problem. An extensive experimental study is conducted to compare our solution with the algorithm of Lu et al. (2002), which is known in the literature to be the most efficient up to date.

In this talk we present a new extension of the well-known P//Cmax problem to a multi-period context. The problem is given by "t" different non-preemptive sets of tasks (batches) to be processed in a fixed sequence. Each set includes variable amounts of up to "n" different types of tasks. In addition, a non-negligible set-up cost has to be paid whenever the type of task to process on the same machine changes. The multiperiod P//Cmax problem consists of finding the schedule of tasks that minimizes the total makespan. In order to solve such a difficult problem, we propose a decomposition approach where batches are considered as periods, the final state of a period reflecting the setting of the "m" existing machines. Tasks are then scheduled one period at the time by a new heuristic that accounts for the initial state and the tasks in the current batch, but also for the type of the tasks in the upcoming period.

In their 1995 seminal report Waldspurger and Weihl introduced stride scheduling as a general technique for resource management based on the number of tickets issued to or purchased by clients competing for resources. Stride scheduling has been studied experimentally. However, we are unaware of any work done on its theory. In our presentation, we show that Stride scheduling is a parametric method of apportionment, which provides us with a number of fundamental properties of the basic stride scheduling. We also discuss two main performance metrics of stride scheduling: throughput error and response time variability. Finally, we present an algorithm for minimizing throughput error based on the solution to the product rate variation problem (PRVP) used to sequence just-in-time systems.