| year | 2007 |
|---|---|
| author | Hyoung-Gon Lee |
| Keyword | MRP, Part explosion, APS, Real-time processing, Computational grid, Distributed databases, Feedforward CRP, Hierarchical structures, Main-memory virtual DB |
| Abstract | The material requirement planning (MRP) process is a core function to generate an exact and efficient plan for manufacturing enterprises so that appropriate quantities of raw materials and subassemblies are provided at the right time. Providing an executable production plan at high speed has remained a goal for MRP since the process was first developed in the 1970s. However, due to the time consuming characteristic of the process with its intensive database (DB) interaction, real time processing remained unattainable, whereas the unrealized ideal of MRP, which was prematurely heralded as an add-on module called advanced planning system (APS) in the 1990s, remains as yet unachieved. APS differs from conventional MRP in its simultaneous consideration of the aterial and resource requirements. Established studies and commercial solutions have attempted to make MRP more practical through its integration with capacity planning in a closedloop planning system. However, despite enormous research efforts the problem of processing time has remained largely unsolved, as the solution remains unattainable within a sufficiently short time for practical application. On the other hand, many studies for improving the performance of DB systems have focused on the modification of the data structure, data partitioning, and materialization strategies. In addition, grid computing technology, which aims to utilize unused computing power capacity via internet networking, is being broadly introduced in business. However, to the best of our knowledge, no attempt has yet been made to apply these advanced computing technologies to the MRP part explosion process. This paper proposes a grid enabled APS considering finite-capacitated MRP part explosion process in a distributed DB environment and demonstrates its substantially improved performance. The proposed system comprises four MRP maintenance steps: master data synchronization, job distribution, part explosion, and writing back steps. In addition to being the first attempt to apply a computational grid to the MRP domain, our proposed system features two further distinctive points. First, the entire process terminates in one step, termed feedforward capacity requirement planning (CRP), even though it contains capacity planning corrections. Second, all the part explosion processes which are performed by multiple resources are accelerated by utilizing its hierarchical structure with processing queries in the virtual DB of the main memory. In particular, the MRP part explosion process was found to be much faster using grid resources, with an almost n-fold increase in acceleration with n nodes, assuming that the master data are synchronized among the resources and that the writing back step is performed afterwards. Considering the marginal effect obtained from the main memory operation and feedforward CRP, a speedup ratio of 13n was attained for finitecapacitated MRP with deep binary-tree type bill of material (BOM) more than 10 levels. Several experimental results are presented to illustrate the performance of the proposed approach. |
| c | PhD |