10:30 AM - 10:55 AM
Solving a production scheduling problem: from the model to the integration by the user
Solving a production scheduling problem is an important part of a project, but it remains incomplete if the solution is not used by the company. This presentation will focus on the resolution and integration of a scheduling problem for the machining of metal parts on CNC machines. This project is in partnership with the company APN. The objective is to find in which order to machine the products while minimizing the tardiness and taking into consideration the capacity of the available resources, the delivery dates, the transition times between tasks, and other particularities of the machines. It is necessary to consider certain functionalities wanted by the user. This problem is solved using constraint programming and a large neighborhood search. A solution is produced each day, so it is also possible to use parts of the previous day's solution as a starting point. An important part of the project is not only the solving of the problem, but also the integration with our industrial partner. The acceptability of the solution is an important part of completing an industrial project. Part of the presentation will focus on the integration with our partner.
10:55 AM - 11:20 AM
Production planning for short seasonal demand with forecast evolution
We consider the problem of planning for a short seasonal demand when decisions have to be made using uncertain and evolving forecasts. Capacity is limited and carrying inventory is expensive. Hence, it is essential to make efficient production decisions ahead of the selling season. The evolution of demand forecasts follows the martingale model of forecast evolution, which captures the property that forecast accuracy increases when approaching the selling season. We integrate forecast evolution in a production planning model with a fill-rate service-level constraint and inventory costs. The optimal production policy is determined through a dynamic programming model for the single-product case and is adapted into an iterative heuristic for the multi-product case. Through repeated rolling-horizon simulations, we show that stochastic models that do not account for forecast evolution often fail to reach the service-level targets. Explicitly integrating the uncertain forecast evolution leads to high demand satisfaction and cost reductions. We also identify the impact of product correlation and timing of uncertainty on the production policy and its performance.
11:20 AM - 11:45 AM
Ordonnancement d'une cellule robotisée de soudage 4.0
Nous considérons une cellule robotisée de soudage dotée d’un robot transporteur, d’un robot positionneur et de deux robots soudeurs. Notre projet vise à concevoir un ordonnancement de bonne qualité pour un partenaire industriel qui tient compte du processus de soudage, et de différentes contraintes d’entreposage et de transport à l’intérieur de la cellule. Deux méthodes de résolution seront présentées : une méthode exacte basée sur la programmation en nombres entiers et une heuristique utilisant la recherche à voisinage large (LNS). Des résultats numériques sur des instances fournies par le partenaire du projet seront présentées.
11:45 AM - 12:10 PM
An efficient heuristic for a job shop scheduling problem in reconfigurable manufacturing systems
A main component of Reconfigurable Manufacturing Systems (RMSs) is the Reconfigurable Machine Tool (RMT). RMTs are generally used for a specific set of operations required to satisfy customized requests and have the advantage to be converted with relatively low costs when these requests change. We propose to address a job shop scheduling problem with RMTs where the objective is to minimize the makespan while limiting the number of reconfigurations over a finite planning horizon. Two solution methods are proposed and compared: an exact method using the branch-and-cut procedure of CPLEX and a two-step heuristic method. The first step of the heuristic partitions the set of products in different disjoint groups and defines a schedule for each group using a mixed integer programming model. The second step uses a shortest path model to define the sequence in which the groups of products must be scheduled. Preliminary results will be presented to point out the merits of the proposed heuristic.