Robust Scheduling and Diagnostics Using Simulation-Based Optimization

该合同为有限容量调度方法的开发提供资金，该方法使用基于仿真的优化技术和现代制造执行系统（MES）中可用的实时信息。我们的方法是将强大的仿真优化技术与快速制造模拟器结合起来，开发一种包含可变性的调度方法，计算“最佳”交货时间，并提供灵敏度分析。我们将在研究中解决的具体问题包括：(1)将“过程清单”（即材料清单，所涉及的路线和过程的组合）与一组需求一起转换为快速“样本路径”仿真模型；(2)开发有用的目标函数，使用MES中的信息，并反映生产计划人员的愿望；(3)研究了各种梯度估计和快速优化方法；(4)开发敏感性分析方法，为生产计划人员提供有用的信息，帮助他们将不可行的生产计划变为可行（例如，增加产能，推迟交货期）；(5)使用佐治亚理工学院的虚拟工厂实验室和我们的工业合作伙伴的工厂测试这些方法。如果这项研究成功，它将在许多方面改善生产调度。首先，它应该通过计算考虑当前和预测拥堵水平的计划交货时间来改进现有的MRP II系统。因此，大型的现有软件系统只需要改造而不是替换。其次，系统将明确考虑可变性和随机性。通过考虑这些因素，可以设定适当的前置时间缓冲，而不需要额外的“填充”，从而在快速反应和履行承诺的能力方面保持竞争力。第三，所开发的方法应该能够为时间表提供有用的诊断，指出哪些需求难以达到，以及如何放宽这些需求。这应该提供一种能力，知道如何使不可行的计划以最具成本效益的方式变得可行。

9713878 Spearman This award provides funding for the development of an approach to finite capacity scheduling that uses simulation-based optimization techniques and real-time information now available in modern manufacturing execution systems (MES). Our approach will be to incorporate powerful simulation optimization techniques with a fast manufacturing simulator to develop a scheduling methodology that incorporates variability, computes "optimal"' lead times, and provides sensitivity analysis. The specific issues we will address in our research include: (1) converting a "bill of process"' (i.e., the combination of a bill of material, the involved routings, and the processes) along with a set of demands into a fast "sample path" simulation model; (2) developing useful objective functions that use information found in an MES and that reflect the desires of production planners; (3) examine various gradient estimation and fast optimization methods; (4) develop methods for sensitivity analysis that provide information useful to a production planner to assist in making an infeasible production schedule feasible (e.g., adding capacity, pushing out due dates); (5) test the methods using the Virtual Factory Laboratory at Georgia Tech and in the plants of our industrial partners. If this research is successful it will improve manufacturing scheduling in a number of ways. First, it should improve existing MRP II systems by computing planned lead times that consider current and predicted congestion levels. As such, large existing software systems need only be retrofitted rather than replaced. Second, the system will explicitly consider variability and randomness. By considering these factors, appropriate lead time buffers can be set without extra "padding" thereby remaining competitive in terms of both quick response and the ability to meet commitments. Third, the developed methodology should be able to provide useful diagnostics for a schedule indicating which requirements are difficult to attain and how they might be relaxed. This should provide the ability to know how to make an infeasible schedule feasible in the most cost effective way.