GOALI: Stochastic Optimization for the Scheduling of Tests in the Development of New Chemical Products

要在市场上推出一种新化学品，研发机构必须进行测试。这些主要是法规要求，如环境和安全测试。如果一个潜在的产品没有通过任何测试，它就不能进入市场。为开发新产品安排测试具有重要的经济意义。最根本的取舍是在更大的销售和更长的时间表带来的收入之间。调度的复杂性之一是其组合性，特别是当潜在产品的数量很大时。此外，任务的成本，任务成功的概率，产品的收益都存在不确定性。此外，还存在测试的优先约束、执行测试的资源约束以及制造设施的能力和投资资本约束。本课题主要研究新产品开发过程中测试任务最优调度的随机优化模型和方法。一个有效的优化模型将被开发出来，用于测试任务的最优调度，该调度最大化新现值，并且可以捕获这个问题中涉及的主要权衡和决策，以及相关的约束。初步的工作表明，可以将该问题建模为一个大规模的混合整数线性规划，该规划使用连续时间表示和随机参数的离散分布。然而，对于少量的产品和测试，这种模型在计算上可能会变得昂贵。因此，将研究依赖于切割平面、分解和逻辑的计算策略。测试优化模型与制造公司过程网络模型的集成，以评估工厂产能限制的影响，以及确定不修改设计或建造新设施的投资机会，也将进行探索。另一个关心的问题将是确定对现有设施进行必要的修改以适应新产品的生产。虽然大部分理论和计算工作将在卡内基梅隆大学进行，但将与DowElanco进行广泛的互动。最初，它将包括信息收集，以充分了解这个问题所涉及的问题。在稍后阶段，将针对农用化学品进行实际的个案研究，以检验拟议的方法。

Abstract - Grossman - 9520153 To introduce a new chemical in the market there are tests that have to be performed by and R&D organization. These are mostly regulatory requirements such as environmental and safety tests. If a potential product fails any test it cannot reach the market. Scheduling the tests for the development of new products has important economic implications. The fundamental trade-off is between greater sales and income that result from longer schedules. One of the complications in the scheduling is its combinatorial nature, especially when the number of potential products is large. In addition, there are uncertainties in the costs of the tasks, probabilities of success of the tasks, in the income of the products. Furthermore, there are precedence constraints for the tests, constrains on resources for performing the tests, and on the capacity and investment capital for the manufacturing facilities. This project is concerned with the study of stochastic optimization models and methods fmr the optimal scheduling of testing tasks for the development of new products. An effective optimization model will be developed for the optimal scheduling of testing tasks that maximizes New Present Value and that can capture the major trade-off and decisions involved in this problem, as well as the relevant constraints. Preliminary work has indicated that one can model the problem as a large-scale mixed -integer linear program that uses continuous time representation and discrete distributions for the stochastic parameters. However, this model can become computationally expensive for modest number of products and tests. Therefore, computational strategies will be investigated that rely on the use of cutting planes, decomposition and logic. The integration of the optimization model for testing with the model of the process network of the manufacturing company in order to assess the effect of plant capacity constraints as well as to identify investment opportunities for mo difying designs or building the new facilities will also be explored. Another problem of interest will be to determine the required modifications in existing facilities to accommodate the production of new produats. While most of the theoretical and computational work will be performed at Carnegie Mellon, there will be extensive interaction with DowElanco. Initially it will consist of information gathering to fully understand the issues involved in this problem. At a later stage, realistic case studies aimed at agricultural chemicals will be developed to test the proposed methods.