QRM: Using Visual Information to Quantify Microstructure-Processing-Property Relationships

为了推进基础设施和制造业的技术，必须开发具有可调性能（例如强度或导电性）和可预测性能的新型高性能材料。用于预测材料性能的计算工具通常依赖于材料工程师检查材料微观子结构（称为微观结构）的图片，以确定微观结构特征与材料性能之间的关系。显微结构图像的分析既昂贵又耗时，而且可能产生不一致的结果。该奖项支持基础研究，以建立工具，实现微观结构图像分析的自动化方法，目的是更好地理解控制材料特性和性能的微观结构特征。在这个项目中，研究人员将收集大量的微观结构图像，并使用人工智能（AI）工具，包括计算机视觉和机器学习来分析它们。人工智能系统将自动识别微观结构的不同特征，测量它们，并将它们与材料的制造（加工）和性能（性能）联系起来。该系统的优点是快速、客观、通用，并且可以感知到人类不容易看到的信息。为了测试和验证这种方法，它将被应用于理解先进制造中应用的科学挑战问题，包括预测材料的强度，并确定微观结构信息的极限。该项目将为材料科学与工程专业的学生提供人工智能原理方面的额外培训。此外，该项目的方法和结果将向公众公布。微观结构的定量表示是微观结构科学的基础工具，传统上是由人先验地决定测量什么和如何测量。然而，数据科学的最新进展，包括计算机视觉（CV）和机器学习（ML），提供了从微观结构图像中提取信息的新方法。在这个项目中，pi将获取、整理和发布各种微观结构图像数据集，包括组成、处理和属性元数据，并将构建一套CV/ML工具，用于自主微观结构图像表示和量化。CV/ML方法将应用于寻找定量成分-微观结构-加工-性能关系，目标是发现材料。案例研究将侧重于实现对增材制造工艺的科学理解；加强对变形机理的认识；推进微观结构科学，包括人类无法察觉的视觉信号；以及通过黑盒机器学习方法评估科学知识的程度。该项目开发的方法和结果将通过开放获取代码和数据存储库传播，并将通过向材料科学与工程专业的学生传授计算机视觉和机器学习原理，为劳动力发展做出贡献。该奖项反映了美国国家科学基金会的法定使命，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。

To advance technology for infrastructure and manufacturing, new, high-performance materials must be developed, with tunable properties (for example, strength or electrical conductivity) and predictable performance. The computational tools used to predict material performance have often relied on Materials Engineers examining pictures of a material's microscopic substructure, termed the microstructure, to determine the relationships between microstructural features and material properties. The analysis of microstructural images can be expensive and time consuming, and can produce inconsistent results. This award supports fundamental research to build the tools to enable an automated approach to microstructure image analysis, with the aim of better understanding the microstructural features that control material properties and performance. In this project, researchers will collect a large set of microstructural images and use artificial intelligence (AI) tools including computer vision and machine learning to analyze them. The AI system will autonomously identify different features of the microstructure, measure them, and link them to how the materials was made (processing) and how it performs (properties). The advantages of this system are that it is fast, objective, general, and may perceive information that is not readily visible to humans. To test and validate this approach, it will be applied to understanding scientifically challenging problems with application in advanced manufacturing, including predicting the strength of materials, and determining the limits of microstructural information. This project will additional train Materials Science and Engineering students in the principles of AI. Furthermore, the methods and results of this project will be made publicly available.The quantitative representation of microstructure is the foundational tool of microstructural science and traditionally involves a human deciding a priori what to measure and how to measure it. However, recent advances in data science, including computer vision (CV) and machine learning (ML) offer new approaches to extracting information from microstructural images. In this project, the PIs will acquire, curate, and publish a diverse collection of microstructural image data sets, including composition, processing, and property metadata, and will build a suite of CV/ML tools for autonomous microstructural image representation and quantification. The CV/ML approach will be applied to finding quantitative composition-microstructure-processing-property relationships, with the goal of materials discovery. Case studies will focus on achieving scientific understanding of additive manufacturing processes; enhancing knowledge of deformation mechanisms; advancing microstructural science to include visual signals that are not perceptible by humans; and assessing the degree of scientific knowledge learned by a black-box ML method. The methods and results developed in this project will be disseminated via open access codes and data repositories, and will contribute to workforce development by educating Materials Science and Engineering students in the principles of computer vision and machine learning.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.