Record:   Prev Next
作者 Perry, Thomas C
書名 Product selection in semiconductor manufacturing by solving a dynamic stochastic multiple knapsack problem
國際標準書號 0496149970
book jacket
說明 172 p
附註 Source: Dissertation Abstracts International, Volume: 65-11, Section: B, page: 5973
Adviser: Joseph C. Hartman
Thesis (Ph.D.)--Lehigh University, 2004
Traditional revenue management problems consider the selection of arrivals until some specific deadline. This is often practical for the transportation industry but not in continuous manufacturing situations. Arrivals (production order) enter a manufacturing facility and stay for some period of time utilizing resources (production lines). A stochastic dynamic programming (SDP) model is presented that allocates resources among orders which arrive to and depart from the system. Defined as the dynamic stochastic multi-knapsack problem (DSMKP), this model incorporates a capacity window that represents a capacity utilization vector modeled by a series of knapsacks. A generic technique is proposed to represent state vectors as numerically equivalent numbers. This simplifies the SDP computer implementation and allows for flexibility in system changes without re-writes of the computer code. This technique extends to applications beyond this model and is valuable when performing "what-if" types of analyses. To address the "curse of dimensionality" inherent in dynamic programs, several approximation techniques are used to estimate the value function, including state aggregation, state node reduction, linear programming, integer programming and Monte-Carlo simulation. The approximations significantly reduce the solution time of the DSMKP while improving the quality of the solution by effectively increasing the solution horizon. A 6 period problem with 10 periods of approximation, having 456,976 states per period and 218 transition paths for each state was estimated to take on the order of 5000 hours to solve by traditional methods. Using the techniques in this paper, a solution was achieved in less than 3 hours. Furthermore, a far greater differentiation in state values is achieved, leading to better decisions. Although this problem is motivated by a semiconductor manufacturing firm, its general form is applicable to other settings. Although we do not address the pricing issue that is prevalent in perishable asset revenue management (PARM) problems, we believe the model can provide valuable information in these situations and also in capacity planning decisions
School code: 0105
DDC
Host Item Dissertation Abstracts International 65-11B
主題 Engineering, Industrial
0546
Alt Author Lehigh University
Record:   Prev Next