CAREER: Parallel Nonlinear Programming Techniques for Optimization in Rapid Therapeutics Manufacturing

LairdIntelligence优点：考虑到美国国内反复出现的药品短缺的潜在后果，需要能够快速更换产品的可重新配置的设施。在快速治疗药物制造(RTM)设施中，人们关心的是为快速生产给定产品而执行批量过程的放大和优化的能力。这些问题可以表示为具有不确定性的大规模动态优化问题。该项目致力于开发新的优化策略，在新兴的科学计算体系结构上使用高效的并行计算来解决这些批处理优化问题。工作有以下研究目标：1.结构化非线性规划问题在集群和多核机器上的并行求解--专注于开发块结构化非线性规划(NLP)问题的内部分解方法，使得在严格的非线性框架下能够高效地并行求解先前难以解决的问题。结构化NLP问题在新兴大规模并行体系结构上的并行解决--专注于扩展这些NLP算法以有效地利用这些负担得起的大规模并行体系结构的能力。快速治疗制造可重构系统的批量过程优化--将并行算法与模块化建模框架相集成，并开发严格的动态模型来支持可重构治疗制造设施的放大和批量优化。广泛影响：这是一个综合计划，以实现以下教育目标：1.提高本科生的科学研究意识--PI一直在追求几名有前途的本科生的职业发展，每年将有两名本科生接受研究职业培训。通过积极参与在线技术社区鼓励终身学习--为了鼓励终身学习和积极参与在线技术社区，PI将把维基和论坛软件的使用与本科数值方法课程结合起来。学生团队将被要求贡献内容，并将根据他们发布的材料的参与程度和同行评级进行评分。为研究生和专业人员有效使用和开发下一代PSE工具做好准备--PI将开发一门新的研究生水平课程，旨在展示现代计算架构在科学计算中的潜力。此外，还将开发一个短期课程，在瑞典隆德大学和德克萨斯农工大学国家治疗制造中心内授课。在代表不足的群体中促进研究职业生涯--PI将积极与德克萨斯A&A&M大学的路易斯·斯托克斯少数族裔参与联盟(LSAMP)合作，作为研究导师，鼓励少数族裔本科生和硕士学生攻读博士学位并从事研究。这项研究有可能通过其对快速制造的影响，显著提高治疗制造业的弹性。此外，新的并行算法本质上是通用的，将通过硬币或开放源码基金会提供，供其他研究人员用于解决以前难以处理的非线性优化问题。研究成果将在经评审的期刊和国家会议上传播。

0955205LairdIntellectual Merit: Given the potential consequences of recurring drug shortages within the U.S., there is a need for reconfigurable facilities that are capable of rapid product changeover. In a rapid therapeutics manufacturing (RTM) facility, one is concerned with the ability to perform scale-up and optimization of the batch process for the rapid production of a given product. These problems can be represented as large-scale dynamic optimization problems with uncertainty. This project seeks to develop novel optimization strategies that can solve these batch optimization problems using efficient parallel computation on emerging scientific computing architectures.The work has the following research objectives:1. Parallel Solution of Structured Nonlinear Programming Problems on Clusters andMulticore Machines -- focus on the development of an internal decomposition approach for block structured nonlinear programming (NLP) problems, enabling efficient parallel solution of previously intractable problems in a rigorous nonlinear framework.2. Parallel Solution of Structured NLP Problems on Emerging Massively ParallelArchitectures -- focus on extending these NLP algorithms to effectively exploit the capabilities of these affordable, massively parallel architectures.3. Batch Process Optimization in Reconfigurable Systems for Rapid TherapeuticsManufacturing -- integrate the parallel algorithms with a modular modeling framework and develop rigorous dynamic models to support scale-up and batch optimization in reconfigurable therapeutics manufacturing facilities.Broader Impacts: This is an integrated plan to meet the following education goals:1. Increasing Scientific Research Awareness Among Undergraduates -- the PI has pursued the career development of several promising undergraduates, and two undergraduates per year will be trained for a career in research.2. Encouraging Lifelong Learning Through Active Participation in Online TechnicalCommunities -- to encourage lifelong learning and active participation in an online technical community, the PI will integrate the use of wiki and forum software with the undergraduate course on numerical methods. Student teams will be required to contribute content and will be graded on the level of involvement and peer rating of their posted material.3. Preparation of Graduate Students and Professionals for Effective Use and Development of the Next Generation of PSE Tools -- the PI will develop a new graduate level course designed to demonstrate the potential of modern computing architectures for scientific computing. In addition, a short course will be developed to be taught at Lund University (Sweden) and within the National Center for Therapeutics Manufacturing at Texas A&M University.4. Promoting Research Careers Among Underrepresented Groups -- the PI will workactively with the Louis Stokes Alliance for Minority Participation (LSAMP) at Texas A&M University as a research mentor to encourage minority undergraduates and Masters students to pursue doctoral studies and a career in research.This research has the potential to significantly improve resiliency in the therapeutics manufacturing industry through its impact on rapid manufacturing. In addition, the novel parallel algorithms are general in nature and will be made available through the COIN-OR open-source foundation to be used by other researchers to tackle previously intractable nonlinear optimization problems. Research results will be disseminated in refereed journals and at national conferences.