On Thursday, April 18, 2002, Sean Gahagan successfully presented his Ph.D. proposal to the Department of Mechanical Engineering, University of Maryland. The dissertation title is Simulation and optimization of production control for lean manufacturing transition. The proposal abstract follows below. Mr. Gahagan works at Northrop Grumman Advanced Micro Electronics Center in Linthicum, Md., as a manufacturing process engineer. The research is directed by Dr. Jeffrey W. Herrmann. For more information, see the Adaptable Simulation Models project.
Lean manufacturing is an operations management philosophy focused on reducing waste in a manufacturing system. A principle type of manufacturing waste is work in process (WIP) inventory, which lean manufacturing limits through the use of just in time (JIT) production control techniques. JIT advocates pull production control. Pull production control specifies finite WIP buffers between each process. These inventories are not processed until there is adequate space in the next downstream buffer. Thus, a workstation responds whenever a customer (or downstream workstation) removes parts from its output inventory buffer. Pull systems are typically controlled by simple, often visual, shop floor mechanisms. In contrast, many manufacturing systems use push production control, driven by a materials requirements planning (MRP) system. When a customer places an order to such a system, the required components are allocated to the order and released to the shop floor, regardless of system loading. The order is completed as quickly as possible at each process, regardless of downstream conditions. As a result, WIP piles up in front of slower processes. MRP systems are controlled by sometimes-elaborate information systems. In recent years, there has been great interest in transforming push systems to pull in order to control WIP and to reduce costly information system infrastructure. Though there has been a great deal of research about what should be done, there has been little about how best to do it.
Managers charged with lean transformation face two difficult questions 1.) What will the final system configuration be, and 2.) How shall it be transformed while maintaining throughput? Although JIT advocates pure pull production control, where every WIP buffer is controlled with a pull mechanism, as an optimum system configuration, recent literature suggests that optimum production control may make use of both push and one or more types of pull in the same system. Hybrid production control describes a system where the initial processes are controlled with pull and the final processes with push. Constant WIP (CONWIP) production control uses pull to control the overall WIP inventory in a system, but push to control production. For a given system, there are many possible combinations of these control mechanisms. The literature is full of answers to this question, but most solutions address only single product systems. Such problems are easily described and may be solved analytically. As a result, many studies in this area focus on a rigorous optimization, rather than a useful solution. The proposed research will address multiple product systems and multi-component products; systems whose complexity demands simulation based, rather than analytical, techniques. The lean transformation process is less well defined. Most of the available literature discusses transformation in anecdotal terms. Those authors who have addressed it directly have greatly simplified the transformation process by assuming that WIP in the system can be used to fill the pull buffers, failing to recognize the complexity of MRP. Systems controlled by MRP have already allocated their WIP to specific orders, but pull demands unallocated inventory in the buffers. This means that to transition from push to pull, the system must either reduce demand or increase production to fill the buffers with unallocated WIP while continuing to fulfill new customer orders. It is unclear which policy is superior or how transition duration affects the process. The proposed research will investigate this question for the first time.
A fundamental element of the proposed research is the development of easily manipulated production control simulation models. Efficient simulation programming demands the use of re-usable software objects. Discrete Event Simulation (DES) models typically make extensive use of a fundamental object called a server to control the flow of the simulation. Servers coordinate the interaction between a queue and a system resource. However, the classic server model can only emulate push production control. To model any other production control policy, servers must be linked to external control elements. For large models, such control quickly becomes unwieldy. Therefore, to efficiently model a system under pull production control, a new fundamental element is necessary. The Multi-Flow Modeling Paradigm (MFMP) was developed to meet this need. MFMP considers more than just the flow of customers through the system. It also considers the flow of demand and of system resources. It is more flexible than previous approaches because it treats all inputs identically. However, the integrated nature of MFMP requires greater structure to represent a complete system. The author has developed a three-tier hierarchical model, called the Production Control Framework (PCF), to provide a basis for MFMP modeling. The PCF describes the production control policy of a system parametrically, facilitating easy manipulation. Using this approach, a library of new software objects was created within a popular commercial simulation package. Models created with these objects can describe a wide variety of push and pull production control policies simply by changing a set of simple parameters.
The proposed research will pursue the following two-phase plan to explore production control of systems undergoing lean manufacturing transformation. The first phase will focus on the modeling and optimization of production control, building on the work already completed. First, a taxonomy of production control mechanisms will be developed to include both push and a variety of pull (e.g. Kanban, CONWIP and Hybrid) production control strategies, adding the mathematical specificity of the PCF that previous research lacks. Second, the PCF simulation object library will be expanded to include all of the production controls addressed in the taxonomy. Third, a simulation based optimization algorithm will be developed to search the production control space of a PCF model to identify the optimum steady state production control policy. This step will evaluate the performance of the algorithm by conducting experiments and comparing the results and computational effort to that of other approaches, including simple heuristics. The second phase will concentrate on the optimization of production control transition. First, a taxonomy of production control transition strategies will be developed. Again, the PCF will lend mathematical specificity. Second, a simulation based optimization algorithm, likely based on that used in phase one, will be developed to optimize production control transition strategies. The algorithm will include identification of possible transition strategies based on the current and future system configurations, decomposition of the strategy into discrete steps, parametric optimization of the system at each step using the steady state optimization algorithm, and selection of an optimum transition strategy based on a cumulative measure of system performance. Experiments performed in this step will be used to identify an optimum lean production control transition strategy for one or more significant test cases, perhaps derived from industry.
The proposed research will yield significant academic and practical benefits. Though optimization of pull production control has been attempted previously, it has never been done over such a complex production control space. Optimization of production control transition has never been attempted. The response of a manufacturing system to changes in production control mechanisms is poorly understood at this time. The proposed research will yield a greater understanding of this complex process and, for the first time, provide practitioners of lean manufacturing techniques guidance about how best to transform their systems.
Return:
The University of Maryland
Institute for Systems Research
Computer
Integrated Manufacturing Laboratory
Last updated on April 19, 2002, by Jeffrey W. Herrmann.