CAREER: Integrating Graph Theory based Networks with Machine Learning for Enhanced Process Synthesis and Design

流程综合涉及到根据预定义的指标，如成本、能源消耗和环境影响，为给定的化学制造操作找到最优的工艺步骤序列。图论是一种强大的数学工具，它可以使用特定的连通性规则列举制造步骤之间的所有可能路径，并可以根据特定情况的标准生成可行路径的排序列表。然而，排名靠前的路径仍可能导致对整体流程弹性的担忧-例如，由于基础设施老化或极端天气事件，废水处理设施面临故障风险。因此，弹性被定义为承受故障并从故障中恢复的能力，被纳入为额外的衡量标准。机器学习算法可以根据过去的操作数据预测风险和失败概率，这使得能够对每条可行的路径进行弹性评估。这项研究计划将整合图论和机器学习，从根本上增强过程合成算法，以生成经济高效、环境友好和健壮的解决方案，这些解决方案可以很容易地适用于其他化学制造操作，如塑料回收或制药制造中的溶剂回收。除了对研究生和本科生进行培训外，该项目还将有助于开发以环境和社会正义为主题的开放式教育模块和程序系统案例研究。将为来自新泽西州南部缺乏服务的社区的学生和教师开展K-12 STEM推广和竞赛的互动活动。设计化学工艺网络时的一个关键限制是，最终设计仅与初始结构枚举(上层建筑)一样好，结构列举(上层建筑)是一个大型、复杂的化学加工路径选择网络，目前必须由设计工程师定义。现有的超结构综合方法大多基于启发式方法，由于不能保证在初始枚举中捕获了所有可行路径，因此有时会导致次最优。在本项目中，将使用图论通过定义明确的公理和最大结构生成算法来生成整个过程网络(在本项目中为污水处理厂网络)的详尽列举，其中可能的过程路径通过材料和工艺技术节点与代表水流的弧线的连接来表示。这种详尽的列举之后是可行的路径分析，以通过包括组合、线性和非线性模型方程和解算器的新型两层过程综合算法，基于成本、能源使用、环境影响和弹性度量来生成解决方案结构的分级列表。基于现有污水处理厂数据和资产库存的回归、分类和特征重要性建模的机器学习的集成，将允许及时预测由于老化和极端事件而导致的风险和故障概率。这种改进的流程综合方法将产生非直观、创新、成本效益高、可持续和可靠的解决方案，使市政当局和自来水公司等利益攸关方能够在设计新设施或改造现有设施时做出明智的决定。作为该项目的一部分开发的材料将包括计算算法、代码、用户手册、教程、技术论文、会议演示、教育材料、情况说明书和K-12扩展计划。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Process synthesis involves finding the optimal sequence of processing steps for a given chemical manufacturing operation based on predefined metrics such as cost, energy consumption, and environmental impact. Graph theory is a powerful mathematical tool that can enumerate all possible pathways between manufacturing steps using specific rules of connectivity and can produce a ranked list of feasible pathways given case-specific criteria. However, the top-ranked pathways may still lead to concerns regarding overall process resilience - for example, a wastewater treatment facility faces failure risks due to infrastructure aging or extreme weather events. Thus, resilience, defined as the capability to withstand failures and recover from them, is incorporated as an additional metric. Machine learning algorithms can predict risk and failure probabilities from past operational data, which enables resilience evaluation of each feasible pathway. This research program will integrate graph theory and machine learning to fundamentally enhance process synthesis algorithms to generate cost-effective, environmentally friendly, and robust solutions that can be easily adapted to other chemical manufacturing operations, such as plastics recycling or solvent recovery in pharmaceuticals manufacturing. In addition to training graduate and undergraduate students, this project will also contribute to the development of open-access educational modules and process systems case studies that feature environmental and social justice themes. Interactive activities for K-12 STEM outreach and competitions will be conducted for students and teachers from underserved communities in the southern New Jersey region.A key limitation when designing chemical process networks is that the final design is only as good as the initial structural enumeration (superstructure), a large, complex network of chemical processing path options that must currently be defined by the design engineer. Existing superstructure synthesis methods are largely based on heuristics which can sometimes lead to suboptimality since there is no guarantee that all feasible paths have been captured in the initial enumeration. In this project, graph theory will be employed to generate an exhaustive enumeration of the overall process network (in this project a wastewater treatment plant network) through well-defined axioms and a maximal structure generation algorithm, wherein the possible process pathways are represented via the connection of materials and process technology nodes with arcs that represent stream flows. This exhaustive enumeration is followed by feasible pathway analysis to generate a ranked list of solution structures based on cost, energy use, environmental impact, and resilience metrics through a novel two-layer process synthesis algorithm that includes combinatorial, linear, and nonlinear model equations and solvers. The integration of machine learning for regression, classification, and feature importance modeling based on available wastewater treatment plant data and asset inventory will allow for timely prediction of risk and failure probabilities due to aging and extreme events. This enhanced process synthesis methodology will lead to non-intuitive, innovative, cost-effective, sustainable, and robust solutions that will enable stakeholders, such as municipalities and water utility companies, to make informed decisions when designing new facilities or retrofitting existing facilities. Materials developed as part of this project will include computational algorithms, codes, user manuals, tutorials, technical papers, conference presentations, educational materials, factsheets, and K-12 outreach plans.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.