Applying Simheuristics for Safety Stock and Planned Lead Time Optimization in a Rolling Horizon MRP System under Uncertainty

Material requirements planning (MRP) is one of the main production planning approaches implemented in enterprise resource planning systems, and one that is broadly applied in practice.

However, a lot of planning parameters are needed to conduct the MRP execution. In addition, MRP is an algorithmic approach to handle the production planning problem, i.e. no optimization is conducted within the planning run, which is applied in a rolling horizon manner to real production systems that face different uncertainties: stochastic processing times, machine failures, constrained worker availability, or raw material shortages. By optimizing the MRP parameter setting, the negative effects of these uncertainties can be reduced.

This paper discusses new simulation optimization methods that allow to optimize the MRP parameters safety stock and planned lead time in a multi-stage and multi-item production system with stochastic customers’ order amounts, customer required lead times, and machine setup times.

For a relatively simple production system, a discrete-event simulation model is developed in AnyLogic to conduct the rolling horizon MRP approach and mimic a shop floor exposed to the stated uncertainties.

The approach of this paper is based on simheuristics, which is a hybrid methodology that combines operational optimization based on metaheuristics and simulation as an iterative process Rabe et al. (2020).

Modeling A Stochastic MRP System

The MRP system under study has been modeled in the AnyLogic simulation software. The model implements a mutli-stage and multi-item production system. The simulation model can handle stochastic demands as well as random processing times, and provides a standard MRP logic to treat customers’ orders. The role of the simheuristic is to set safety stock and planned lead time parameters in order to minimize the overall cost, which is the sum of inventory and backorder costs.

For the simulation experiments, a simple production structure is used. Despite considering only three levels, the example demonstrates how simheuristics can be meaningfully integrated into an MRP system. Model framework also included the statechart of the AnyLogic model with the master production schedule (MPS) and standard MRP steps.

Production structure and MRP simulation framework in AnyLogic
Production structure and MRP simulation framework in AnyLogic

In this study, researchers also developed 2 extensions of basic heuristic algorithm in order to generate better solutions:

  • Ext1 - consists of an initialization phase, i.e. applying the basic heuristic, and a range reduction phase where the range of the triangular distribution is reduced;
  • Ext2 - which also includes an initialization phase according to the original heuristic, applies the parameter values of a pool of best solutions for the lower and upper bounds of the triangular distribution.

The main target of the simheuristic algorithm is to find those parameter combinations providing the lowest average overall cost with the available simulation budget.

Result

Reseachers performed 42 different simulation experiments in total, 14 for each stochastic setting (low, medium, and high), with 20 replications per simulation iteration.

Results show that the proposed approach is quite promising, specially in a research area where there is a lack of similar studies considering MRP systems under uncertainty conditions. Both heuristic extensions – Ext1 and Ext2, which change the upper and lower bounds of the scenario parameters – perform better than the basic heuristic and also much better than the Random variant with static upper and lower bounds.

相关文章