CDS&E: GOALI: Paints/Coatings In-Silico Product Design and Real-Time Product-Quality Monitoring and Control

现代油漆/涂料（P/C）产品是复杂的化学品混合物，包括聚合物树脂，颜料分散剂和其他添加剂。为了描述P/C质量，如颜色强度，耐用性和保质期，需要大量的消费者规格。这些消费者属性对P/C成分的性质和量以及制备条件的依赖性是复杂的，知之甚少，并且目前不可能使用基于物理的数学模型来预测。然而，这与P/C成分本身相反，P/C成分的性质通常是很好理解的，并且可以通过严格的化学反应和混合模型提前预测。预计该项目将开发一种能够预测这些复杂混合物最终性质的模型。这将有助于产品设计，实时质量预测，缺陷检测和诊断，以及产品质量监测和控制。这项工作的预期经济影响是加快油漆/涂料（P/C）产品的设计和定制。 该研究计划旨在克服使用混合仿真方法预测P/C最终产品质量的挑战，该方法将机器学习方法与基于物理的建模元素相结合。其核心是，来自此次合作的工业合作伙伴的数十年制造数据将用于使用统计机器学习技术揭示制造工艺条件与人们知之甚少的P/C产品质量之间的关系。这将创建人工神经网络形式的黑盒模型，其将基于物理的成分建模元件的预测作为输入，并将预测最终的P/C质量。这项研究将产生强大的计算方法，在硅片P/C产品设计，实时P/C产品质量预测，产品缺陷检测和诊断，并将使方法来监测和控制P/C产品质量。计算方法可以直接应用或扩展到其他制造过程。该团队计划通过Drexel Co-op Program将这项研究系统地整合到本科教育中。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Modern paint/coating (P/C) products are complex mixtures of chemicals that include polymer resins, pigment dispersants, and other additives. To describe P/C qualities such as color strength, durability, and shelf life, a vast set of consumer specifications are required. The dependence of these consumer attributes on the properties and amounts of the P/C ingredients and the preparation conditions is complex, poorly understood, and currently impossible to predict using physically based mathematical models. This is in contrast to the P/C ingredients themselves, however, whose properties generally are well-understood and can be predicted in advance by rigorous chemical reaction and mixing models. This project is expected to develop a model capable of predicting final properties of these complex mixtures. This is expected to aid in product design, and real-time quality prediction, defect detection and diagnosis, and product quality monitoring and control. The expected economic impact of this work is faster design and customization of paint/coating (P/C) products. This research program aims to overcome the challenges of predicting P/C final product qualities using a hybrid simulation approach that combines machine learning methods with physically-based modeling elements. At its core, decades of manufacturing data from the industrial partner of this collaboration will be used to uncover relationships between manufacturing processing conditions and the poorly understood P/C product qualities using a statistical machine learning technique. This will create a black-box model in the form of an artificial neural network which will take as input the predictions of the physically based ingredient modeling elements and will predict final P/C qualities. This research will produce robust computational methods for in-silico P/C product design, real-time P/C product quality prediction, product defect detection and diagnosis, and will enable methods to monitor and control P/C product quality. The computational methods can be applied directly or extended to other manufacturing processes. The team plans to integrate this research systematically into undergraduate education through the Drexel Co-op Program.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.