Tractable supply chain production planning, modeling nonlinear lead time and quality of service constraints

O.m. Anli, Michael C. Caramanis, Ioannis Ch. PaschalidisJournal of Manufacturing Systems 2007 · DOI

TL;DR

This paper introduces a novel, computationally tractable algorithm for supply chain (SC) production planning that explicitly accounts for dynamic lead times and quality of service (QoS) constraints. By overcoming the limitations of traditional fixed lead time models, this approach demonstrates significant reductions in inventory and backlog costs, leading to more efficient and responsive supply chains.

What Problem Does This Paper Address?

Modern global supply chains face increasing pressure to deliver customized products quickly and reliably. Traditional planning methods, like Material Requirements Planning (MRP), often simplify lead times to constants, ignoring their dynamic nature which is influenced by factors like facility utilization, production mix, and lot sizes. This simplification leads to inefficiencies, chronic backlogs, and fluctuating inventories. Furthermore, ensuring a high "quality of service" (QoS), specifically controlling stockout probabilities, is crucial for customer satisfaction but is difficult to integrate into planning models due to its probabilistic and nonlinear nature. The core challenge is to develop a planning framework that can handle these complex, nonlinear relationships in a computationally tractable manner, moving beyond the limitations of current industry practices.

Methods

The authors propose a "time-scale-driven decentralized data communication and decision support architecture." This framework decomposes the SC planning problem into three layers:

  • Centralized Planning Coordination Layer: This layer determines tentative production targets and material release schedules. It uses a linear programming (LP) based fluid model approximation, but crucially, it incorporates nonlinear constraints derived from the other layers.
  • Centralized Quality of Service (QoS) Coordination Layer: This layer estimates hedging inventory policies (safety stock) required to meet exogenously specified QoS levels. It utilizes large deviations theory and G/G/1 queuing approximations to model probabilistic stockout constraints.
  • Decentralized Performance Evaluation and Demand Information Layer: This layer models the short-term stochastic dynamics of individual facilities. It estimates time-averaged performance measures like work-in-process (WIP) and lead times, along with their sensitivities to planning targets. This layer uses analytical probabilistic approaches (mean value analysis) and Monte Carlo simulations to capture nonlinear lead time and QoS dynamics.

The key innovation lies in an iterative algorithm that allows these layers to collaborate. The planning layer uses hyperplane approximations (tangent planes) to the nonlinear WIP and FGI constraints provided by the performance evaluation and QoS layers. This iterative process, known as "outer linearization," refines the production plan until nonlinear constraints are satisfied within a tolerance, achieving a computationally tractable solution for a complex problem.

Datasets

Not discussed in the paper. The authors use a simulated three-facility, five-part type supply chain for their numerical experiments.

Key Insights and Ideas

  • Dynamic Lead Times Matter: Ignoring the nonlinear relationship between facility utilization and lead time leads to suboptimal planning and increased costs. The paper demonstrates that modeling these dynamics is crucial for efficiency.
  • QoS as a Constraint: Explicitly modeling probabilistic quality of service (stockout probability) as a constraint, rather than an outcome, allows for more robust planning and better customer satisfaction.
  • Time-Scale Decomposition: Breaking down the problem by time scale (coarse for planning, fine for operational dynamics) is essential for managing complexity.
  • Iterative Linearization: The "outer linearization" approach effectively handles nonlinear constraints by iteratively adding tangent hyperplanes, making a complex optimization problem tractable.
  • Beyond "Vanilla" MRP: The paper shows that industry practice, often relying on fixed lead times, can be significantly improved upon by incorporating more realistic dynamic models.

Key Results

The proposed Optimal Open Loop Feedback Controller (OOLFC) approach was compared against two industry practice proxies: the Limited Information Exchange (LIX) approach and the Certainty Equivalent (CE) approach.

  • Cost Reduction: OOLFC consistently achieved lower total costs across various demand scenarios. For instance, in scenario 6, OOLFC had 42% lower costs than LIX.
  • Inventory and Backlog Reduction: The OOLFC approach led to significant reductions in WIP, FGI, and backlog costs compared to the other methods.
  • Computational Efficiency: The iterative algorithm converged quickly, typically within 16 master problem subproblem iterations, demonstrating its computational tractability.

Limitations and Future Work

  • Model Simplifications: The paper acknowledges that the models used for performance evaluation (e.g., simplified queuing network models) are approximations. More accurate models, such as those using Monte Carlo simulation or complex analytical models, could be employed.
  • Part-Type Specificity: The hedging inventory policy is aggregated for a fictitious part type. While the paper suggests methods for disaggregating this for cost accounting, it doesn't provide part-type-specific hedging quantities for actual production floor implementation, which would require more complex real-time allocation policies.
  • Future Research: The authors suggest ongoing research into improving QoS layer algorithms and enhancing fluid Monte Carlo simulation models for multi-echelon production systems. They also highlight the potential of sensor networks and RFID technologies to provide more affordable and timely data for dynamic lead time modeling.

Summary updated

BibTeX
@article{anli2007tractable,
  author = {O.M. Anli and Michael C. Caramanis and Ioannis Ch. Paschalidis},
  title = {{Tractable supply chain production planning, modeling nonlinear lead time and quality of service constraints}},
  journal = {Journal of Manufacturing Systems},
  year = {2007},
  doi = {10.1016/j.jmsy.2008.05.001},
}

Loading related content...

Figure