Modelling and optimization of supply chains on complex networks

Simone Göttlich, Michaël Herty, Axel KlarCommunications 2006 · DOI

TL;DR

This paper presents a continuous model for complex supply chain networks using partial differential equations (PDEs), extending previous work to handle intricate network structures. The authors then introduce and numerically investigate optimization problems within these networks, demonstrating the efficiency of their PDE approach compared to discrete event simulations.

What Problem Does This Paper Address?

Supply chains are complex systems involving multiple stages, each with its own processing dynamics. While discrete event simulations are common, they can become computationally expensive for large, intricate networks. Continuous models using PDEs offer a potentially more efficient alternative. This paper addresses the challenge of modeling and optimizing supply chains that go beyond simple linear chains, incorporating complex network topologies with branching and merging points. Specifically, it extends existing PDE-based supply chain models to handle these more realistic network structures and explores how to optimize key operational parameters within them.

Methods

The core of the paper is the development and application of a continuous model for supply chains based on partial differential equations. This model represents the flow of goods through each "processor" (e.g., a factory, a warehouse) as a continuous density that evolves according to an advection equation. The processing capacity of each processor is modeled by a flux function that is capped by its maximum processing rate.

A key innovation is the modeling of "vertices" where supply chains connect. These vertices can be of different types:

  • Degree two vertices: Simple connections between two processors, modeled with queue dynamics.
  • Dispersing vertices: Where a supply chain splits into multiple paths. Distribution rates are introduced to control the flux into each outgoing path.
  • Merging vertices: Where multiple supply chains converge into one. Queue dynamics are adapted to handle the combined inflow.

The paper also formulates and solves optimization problems. These problems aim to find optimal control parameters, such as distribution rates at branching points or processing velocities of individual units, to minimize cost functions (e.g., queue lengths) or maximize throughput. The optimization is performed using standard numerical methods, including Upwind discretization for PDEs, explicit Euler for ODEs, and the trapezoid rule for integration, all within Matlab's Optimization Toolbox.

Datasets

Not discussed in the paper. The paper uses simulated network structures and parameters for its examples and case studies.

Key Insights and Ideas

  • PDEs for Complex Networks: The paper demonstrates that PDEs can effectively model and simulate supply chains with complex, non-linear network structures (branching and merging), which are challenging for traditional discrete event simulations.
  • Vertex Modeling is Crucial: The way connections between processors (vertices) are modeled significantly impacts the overall system dynamics. The paper introduces specific mathematical formulations for different vertex types (degree two, dispersing, merging) that capture essential queueing and flow control behaviors.
  • Optimization of Network Parameters: The framework allows for the optimization of critical operational parameters like distribution rates at junctions and processing velocities. This reveals non-intuitive optimal settings that balance throughput and queue management.
  • Efficiency of PDE Approach: For larger and more complex networks, the PDE-based approach shows a significant computational advantage over discrete event simulations, especially as the number of parts or the simulation time increases.

Key Results

  • Numerical Optimization: The paper presents several numerical examples of optimization problems. For instance, in one scenario, optimal distribution rates were found to significantly reduce queue lengths and improve overall system performance compared to naive choices.
  • Processing Velocity Optimization: In a production line example, optimizing processing velocities led to a substantial reduction in queue buildup and a decrease in total outflow, highlighting a trade-off between speed and system stability.
  • CPU Time Comparison: Tables in the paper demonstrate that for a given number of parts and simulation time, the PDE model is often comparable to or faster than discrete event simulations, especially for larger problem instances. For example, computing 10,000 parts with a coarse discretization took around 3 seconds for the PDE model versus 11 seconds for DES. For 50,000 parts with a finer discretization, the PDE model took about 13.6 seconds, while DES took over 59 seconds.

Limitations and Future Work

  • Limitations:
    • The paper focuses on continuous models, which might not capture all discrete aspects of real-world supply chains (e.g., individual item tracking).
  • The modeling of queues at vertices is simplified.
  • The paper mentions that the distribution rates ($\alpha$) might depend on time or external parameters, but this more complex case is deferred to future work.
  • Future Work:
    • Developing improved optimization procedures, such as adjoint and sensitivity calculus.
  • Extending the model to include time-dependent controls.
  • Exploring connections to discrete optimization methods like Mixed-Integer Problems (MIPs).
  • Investigating more complex scenarios with time-varying distribution rates.

Summary updated

Abstract

In this paper, we extend the model for supply chains developed in The model consists of partial differential equations governing the dynamics on each processor. Furthermore, a modelling of different types of vertices is motivated and discussed. Then, optimization problems are introduced and numerically investigated. A comparison of computing times shows the efficiency of partial differential equations for solving supply chain problems.
BibTeX
@inproceedings{gottlich2006modelling,
  author = {Simone Göttlich and Michaël Herty and Axel Klar},
  title = {{Modelling and optimization of supply chains on complex networks}},
  booktitle = {Communications},
  year = {2006},
  doi = {10.4310/cms.2006.v4.n2.a3},
}

Loading related content...

Figure