3 Presentations

• 
  10:30 AM - 10:55 AM

  End to end learning models for the zone sizing problem in manual warehouses

  • Allyson Silva, presenter, Université Laval
  • Leandro C. Coelho, Université Laval
  • Maryam Darvish, Université Laval
  • Kess Jan Roodbergen, University of Groningen

  In a typical manual warehouse, storage policies are used to determine where each product should be located to minimize the expected average route length traveled by pickers to retrieve demanded products. The most popular policy is the ABC storage, which splits the storage area into three zones and assigns the products with the highest expected demands to the best zones. Arbitrary zone sizes, known to perform well in the average warehousing setting, are commonly used in practice. However, they may lead to major efficiency loss in many common settings. We train machine learning regression models -- ordinary least squares, regression tree, random forest, and multilayer perceptron -- to approximate the decisions provided by an oracle to the zone sizing problem by showing its behavior. The oracle used is a simulator implemented to generate the routes that pickers will follow for a large combination of zone sizes and several warehouse features, such as its layout, the demand characteristics, and the storage and routing rules in use. Our results indicate that our models can mimic the simulator and suggest zone sizes that improve the picking efficiency compared to the arbitrary rules commonly used in practice, being considerably faster than rerunning the simulator again.

• 
  10:55 AM - 11:20 AM

  Flexibility in a stochastic multi-level lot-sizing problem

  • Narges Sereshti, HEC Montreal
  • Jans Raf, presenter, HEC Montréal
  • Yossiri Adulyasak, HEC Montreal

  In the multi-level lot-sizing problem, we must make decisions on when and how much to produce over multiple periods, both for end products and components in the Bill-of-Material (BOM). We consider a stochastic demand and a backlog-related service level. The problem is modeled as a two-stage stochastic programming problem which is solved using a Sample Average Approximation method. The base case is the static case in which both setup decisions and production quantity decisions are made in the first stage. We analyse the value of the flexibility obtained by deciding on the production quantity decisions in the second stage for some of the products in the BOM. Different BOM structures are examined: serial, assembly and general.

• 
  11:20 AM - 11:45 AM

  Mathematical models for the short-term line balancing problem with multiple manning in the 3PL context

  • Gislaine Mara Melega, presenter, HEC/CIRRELT Montréal
  • Sanae Belgarch,
  • Raf Jans, HEC Montréal
  • Julie Paquette, HEC Montréal

  In this study, new mathematical models for the problem of designing and managing production assembly lines in the context of third-party logistic providers (3PLs) are presented. In 3PL facilities, the production lines have a short life span, from just a few hours to a few days, and lines are not fixed, but rather assembled/disassembled as needed. Multiple lines for different orders may need to be configured simultaneously and these lines use the same limited group of available employees. As tasks are mainly executed manually, each station can be occupied by more than one employee (multiple manning). Thus, in this problem two types of decisions must be made: how to group the tasks into stations and how many employees must be assigned to each station. The objectives might be associated with the minimization of the line cycle time or the minimization of the total production cost of all lines. The proposed mathematical models for the problem are based on a network flow formulation, to specifically address the special case of lines having serial tasks. We present a computational study solving real-life instances and a sensitivity analysis of key parameters aiming to better understand the trade-offs of the problem.