4 présentations

• 
  15h30 - 15h55

  Integrating Assortment Planning and Shelf Space Allocation Problem with Machine Learning Approaches

  • Fatemeh Safaei, prés., Department of Mechanical, Industrial, and Aerospace Engineering, Concordia University
  • Masoumeh Kazemi Zanjani, Department of Mechanical, Industrial, and Aerospace Engineering, Concordia University
  • Satyaveer Chauhan, Department of Supply Chain and Business Technology Management, John Molson School of Business , Concordia University

  This study proposes a mixed integer nonlinear programing model for integrated assortment planning and shelf space allocation problem. The model considers asymmetric assortment affinity and allocation affinity in two-dimensional shelves by considering a deterministic demand. The objective is maximizing the total profit by determining the best combination of products and shelf assignment. However, solving this model for a large number of products becomes computationally complex. In addition, capturing the demand uncertainty for assortment planning within the same model further increases its computational complexity. Therefore, we propose to decompose the problem by first solving the assortment problem by considering demand uncertainty. This can be done with the aid of machine learning algorithms with the goal of balancing profitability and diversity. Specifically, we apply graph neural network (GNN) and Cleora that uses embedding vectors to find out the relations between products. This helps in identifying substitutes and complementary items, enhancing the profitability of the selected assortment. Moreover, we apply X-Gboost algorithm to determine which products features have the most impact on the profit. The extracted features are used for clustering the products, whereas the results of Cleora is used to improve the product selection. After finding the best assortment, we use the results as an input to solve the shelf space allocation problem.

• 
  15h55 - 16h20

  Robust Tactical Planning in hybrid Multi-Echelon Manufacturing Systems

  • Anthony Montanaro, prés., Concordia University
  • Masoumeh Kazemi, Concordia University

  With Industry 4.0 technologies, manufacturers have increased capabilities for making customized products. In this context, more manufacturers are adopting a hybrid production mode, producing both standard and customized items. Nonetheless, only a handful of production planning models in the literature can be adapted to such hybrid environments, specially when manufacturing modular-structured products in a multi-echelon job-shop environment. This study strives to close this gap by developing a robust tactical planning model in the above setting while emphasizing the uncertainty in terms of order size and resource utilization for customized items. The goal is maximizing profit under production capacity constraints at different echelons. The decisions involve machine activation, quantities of procurement and production, along with inventory levels for standard items, and lost sale for customized ones. A cardinality-constraint approach is adopted to model the uncertain capacity constraints when considering uncertain extended production time for highly customized items. By designing an experimental setting inspired by data from the literature, the performance of the robust model is evaluated under different uncertainty budgets. The results indicate that the robust model schedules more machines than the deterministic model to protect against uncertainty. Moreover, Monte-Carlo simulation experiments are conducted to assess the performance of machine assignment decisions set by the deterministic and robust models in a more realistic manner. The robust model consistently outperforms the deterministic approach, providing up to 11% (expected) profit gap.

• 
  16h20 - 16h45

  A Hybrid Scenario Decomposition Approach for Stochastic Production Planning in Smart Factories

  • Benedikt Kern, prés.,
  • Masoumeh Kazemi, Concordia University

  In this study, we present a two-stage stochastic programming model to address operational-level production planning challenges in the context of smart factories. We focus on a modular production environment within a flexible final assembly facility designed to manufacture highly customizable, modular-structured products. We formulate a stochastic mixed-integer programming (SMIP) model with the objective of minimizing total costs while optimizing machine activations and production planning. Two key sources of uncertainty are considered: stochastic machine degradation, which reduces machine reliability and affects available capacity, and variability in item processing times, which adds complexity to scheduling and planning. To hedge against the latter, we incorporate the price of robustness, enabling robust decisions. We further introduce recourse actions such as activating additional machines and scheduling maintenance to preserve production capacity. To address the computational complexity of solving the SMIP, we develop a hybrid scenario decomposition approach integrated with Benders decomposition and local branching. The proposed method is validated through extensive computational experiments, demonstrating its effectiveness in comparison to a commercial solver.
  Keywords – Operational production planning, smart factory, stochastic mixed-integer programming, scenario decomposition, Benders decomposition, price of robustness

• 
  16h45 - 17h10

  Collaborative Production Planning in a Network of Multi-Echelon Supply Chains

  • Mohammadreza Hassani, Concordia University
  • Masoumeh Kazemi Zanjani, prés., Concordia University

  The firms in multi-echelon supply chains (MESCs) often operate independently, leading to inefficiencies in procurement, production, and distribution. This issue is particularly pronounced in the manufacturing of customizable modular products (e.g., electronics and photonics devices). In such industries, component interchangeability (e.g., lenses) provides opportunities for a horizontal collaboration among the firms operating at the same echelon. The goal is to enhance cost efficiency, mitigate demand uncertainty, and optimize resource utilization. However, the development of collaborative decision-making tools and efficient cost-sharing mechanisms remain challenging. This research addresses these issues by proposing a cooperative game that facilitates horizontal collaboration in a network of MESCs. More specifically, by allowing resource sharing for the processing of compatible items in different entities, a mixed-integer programming (MIP) model and benefit-sharing framework is developed to evaluate the benefits of various coalition structures in the network. The MIP model, representing the characteristic function of the game, determines the optimal level of procurement, production, transportation, and storage in different entities to maximize the total profit of the network while respecting the processing capacities and resource-sharing limitations among the entities under each coalition structure. Shapley value and nucleolus methods are employed to quantify firms' contributions to the coalition value and fairly allocate the benefits among the participants. The proposed SC collaborative game is sub-additive; hence, the grand coalition is the most profitable coalition structure. The numerical experiments, conducted on a synthetic case study, highlight the high synergy of grand coalition. We also demonstrate the coalition stability under the two benefit-sharing mechanisms and compare their fairness level by measuring the Gini index and plotting Lorenz curves.