| This dissertation is devoted to the study of cluster analysis in relation to the configuration of telecommunication switching cabinets. The objective is to design and configure switching cabinet models to match the pattern of customer demand such that the total number of insertions required to manufacture a given set of orders is minimized. These models are then used as the starting points in the actual manufacturing operation.; The problem of designing switching cabinet models is referred to as the q-model configuration problem. In this dissertation, the q-model problem is addressed for q = 1, 2, and 3. It turns out the that 1-model problem is relatively easy to solve. It is conjectured that the q-model problem for q ≥ 2 is np-hard due to its relation to several well-known np-hard clustering problems. Thus, for the q-model problem with q ≥ 2 the focus is on the development of appropriate heuristic procedures and the use of statistical techniques to recognize identifiable patterns in the past orders. Specifically, two constructive techniques, a local search algorithm, and a variant of local search called variable depth search, are developed as solution strategies for the q-model problem.; Problems are generated for which the optimal solution can be obtained and compared to the solution of the heuristics. For problems that the optimal solution cannot be obtained, the heuristics are compared with each other to determine which procedure (if any) performs better than the others. A lower bound on the total number of insertions for a given order set is also presented and used within the computational study to evaluate the heuristics on an empirical basis. |
