Production Strategies for Batch Processing with Random Yield

奥尔森9713727这项拨款为研究具有随机产量的批量生产环境提供资金。在这些环境中，产品是成批生产的，因为产品类型之间的安装时间或安装成本都很高。此外，生产的产出被认为是高度可变的。这项工作的动机来自汽车零部件行业，特别是一家生产模压和喷漆塑料保险杠的工厂。这家工厂的油漆生产线产量很难预测，而且不同油漆颜色之间的转换成本很高。这项研究将进一步发展：(A)在连续时间环境中考虑随机收益率的模型；(B)使用离散和连续近似获得最优值的可计算误差界的这些模型的求解方法；(C)这些随机收益率模型最优解的一般性质的表征，为管理决策提供洞察；以及(D)总体上离散和连续时间模型的比较，洞察每种方法的基本模型特征及其与潜在情况的关系。这项研究的主要影响将是改善随机收益率系统中的决策。它将提供具有可计算界限的方法，这些方法可以应用于从机械加工到绘画再到半导体制造的各种环境中。该方法将在真实数据上进行测试，以确保它在复杂的制造环境中可实时实施。这项研究的结果预计将对对美国竞争力至关重要的行业(如汽车零部件行业)产生重大影响，这些行业的产品收益率变化很大。调查人员将与感兴趣的公司合作，以确保结果是相关的。这项研究还有望在排队论、调度和随机规划等领域做出重大的学术贡献。

Olsen 9713727 This grant provides funding for the study of batch production environments with random yield. In these environments, products are produced in batches because either setup times or setup costs between product types are significant. Furthermore, the output from production is assumed to be highly variable. The motivation for this work is from the automotive parts industry and particularly a plant which produces molded and painted plastic bumpers. The paint line in this plant has highly unpredictable yield and changeovers between different paint colors are expensive. This research will further the development of (a) models that take into account random yield in a continuous time environment; (b) solution methods for these models that achieve computable error bounds on optimal values using both discrete and continuous approximations; (c) characterizations of general properties of optimal solutions of these random yield models that provide insight into managerial decisions; and (d) comparisons of discrete and continuous time models in general with insight into fundamental model characteristics of each approach and their relationship to underlying situations. The main impact of the research will be to improve decision making in random yield systems. It will provide methods that have computable bounds and that can be applied in a variety of settings from machining to painting to semiconductor fabrication. The methodology will be tested on real data to ensure that it is implementable in real time in complex manufacturing environments. The results of this research are expected to have a significant effect on industries critical to U.S. competitiveness (such as the automotive parts industry) where product yield is highly variable. The investigators will work with interested companies to ensure that the results are relevant. The research is also expected to have significant academic contributions in the areas of queueing theory, scheduling, and stochastic programming.