A physics-consistent data-driven paradigm for designing complex material systems

复杂的材料系统，如混合材料、合金和复合材料，正在改变从航空航天工程到生物医疗设备行业的现代技术格局。应用无穷无尽：具有更高强度和更低密度的新型合金为提高可重复使用的航空航天器的可靠性提供了无限的机会。使用磁性软物质的小型软机器人正在靶向药物输送和微创手术方面建立新的前沿。目前的材料设计方法在很大程度上依赖于试错法，由于设计参数多，对材料的不确定性和多尺度结构-功能关系缺乏了解，设计效率低，效果不佳。为了填补这些知识空白，提高复杂材料系统的设计效率，我的研究团队将开发一种新的物理一致的数据驱动的材料设计范式。我们设计范例的主要新奇之处在于它将基础物理与现代数据科学相结合。它将通过多尺度、多物理计算框架对复杂材料系统的行为进行建模，并利用数据挖掘和机器学习技术进行高效的材料设计和优化。具体地说，我们将构建准确和健壮的计算技术，以数字方式生成与物理一致的数据，以补充稀缺的实验数据。我们将制定有效的均化方案，将材料的宏观性质与微观结构和机制联系起来。结合上述数值策略和机器学习技术，我们将揭示隐藏的材料多尺度结构-功能关系，否则很难发现。我们还将系统地量化由其随机微结构和制造工艺引起的材料不确定性，同时确定优化目标性能的材料设计参数。物理一致的数据驱动的材料设计范例代表了材料建模、表征和发现的一般方法，可以应用于各种复杂的材料系统，并促进材料创新。这些创新将进一步推动医疗器械、柔性电子产品和软机器人领域的技术变革，产生显著的社会效益、环境效益和经济效益。此外，我们的研究还将推进尺度桥、不确定性量化和高维约束优化的理论和技术，这些都是许多科学学科面临的基础性研究问题。这项研究位于固体力学、数值方法、材料科学和科学计算的交叉点，将为参与研究的学生传授关键技能，如理论推导和计算编程，以在未来的跨学科研究和开发版图中取得成功。

Complex material systems, such as hybrids, alloys, and composites, are transforming modern technological landscapes, ranging from aerospace engineering to the biomedical device industry. The applications are endless: novel alloys with higher strength and lower density open unlimited opportunities to enhance the reliability of reusable aerospace vehicles. Small-scale soft robots using magnetic soft matter are establishing new frontiers in targeted drug delivery and minimally invasive surgery. The current material design approach relies heavily on trial and error, which is inefficient and ineffective due to the large number of design parameters and the lack of understanding about material uncertainties and multiscale structure-function relationships. To fill these knowledge gaps and increase the design efficiency of complex material systems, my research team will develop a novel physics-consistent data-driven material design paradigm. The key novelty of our design paradigm is its integration of fundamental physics with modern data science. It will model the behaviors of complex material systems through a multiscale, multiphysics computational framework and leverage data mining and machine learning techniques for efficient material design and optimization. Specifically, we will construct accurate and robust computational techniques to numerically generate physics-consistent data complementary to scarce experimental data. We will devise effective homogenization schemes to connect material macroscopic properties with microscopic structures and mechanisms. Combining the above numerical strategies with machine learning techniques, we will reveal hidden material multiscale structure-function relationships that are otherwise difficult to discover. We will also systematically quantify material uncertainties originating from their random microstructures and fabrication processes, while identifying the material design parameters that optimize targeted performances. The physics-consistent data-driven material design paradigm represents a general methodology for material modeling, characterization, and discovery, which can be applied to various complex material systems and facilitate material innovations. These innovations will further drive technological transformations in medical devices, flexible electronics, and soft robotics, generating significant social, environmental, and economic benefits. Moreover, our research will advance the theories and techniques in scale-bridging, uncertainty quantification, and high-dimensional constrained optimization, which are fundamental research questions faced by many scientific disciplines. Located at the intersection of solid mechanics, numerical methods, materials science, and scientific computing, this research will impart participating students with key skills, such as theoretical derivation and computational programming, to succeed in the interdisciplinary research and development landscape of the future.