CAREER: Embedded Data Assimilation for Complex Turbulent Reacting Flows

应对气候变化的挑战需要先进、高效、低排放的燃烧技术以及受过良好教育的劳动力来理解和解决这些挑战。一个关键的限制因素是，目前的模拟技术无法准确预测低排放燃烧器和可持续燃料设计所需的湍流燃烧状态。由于计算资源的实际限制，用于工程设计的计算模拟依赖于湍流和化学物理某些方面的简化数学表达式；这些模型几乎总是彼此断开连接，因此无法捕捉到关键的物理相互作用。最近，利用机器学习和约束优化技术开发了有效的数值方法来校准流动模拟过程中的复杂模型。虽然这些模型对简单的非反应湍流是成功的，但尚未应用于高度非线性的湍流反应流动。该项目的主要目标是开发有效的方法来校准复杂湍流反应流动模拟中缺失的物理模型，包括工程几何中的流动，这将提高实际计算的预测准确性。由此产生的方法将在科学和工程的许多领域有用，并将以开源软件包的形式公开提供。该项目将通过举办一个关于湍流燃烧数据和建模的年度夏季研讨会，促进跨学科伙伴关系和跨越传统边界的学生教育。该项目还将支持一所资源不足的高中的教育和研究计划的发展，这将鼓励对能源科学的广泛理解，并参与解决国家和全球能源挑战的方案。该项目将通过开发湍流闭包和优化方法来满足对精确、高效的湍流燃烧模型的需求，这些方法适用于典型和复杂的湍流反应流模拟。一种基于伴随的优化方法可以有效地优化典型流（如湍流射流火焰和楔形火焰）的Navier-Stokes方程上的闭合模型。应用基于伴随优化的主要挑战是需要侵入式访问代码的数据结构，这对于通用计算流体动力学（CFD）求解器来说实际上是不可能实现的。为了解决这个问题，将开发一种新的协同优化框架，以利用基于伴随的规范流优化和基于集成卡尔曼（无伴随）的几何复杂流和实验数据优化。这种结合的方法将训练规范和复杂物理的模型，同时减轻当前通用CFD代码嵌入式优化的局限性。更广泛地说，科学界对开发利用大型数据集的方法感兴趣；因此，这个项目的方法有可能被跨学科广泛采用。结果数据、优化框架和训练模型将作为开源软件分发，以方便学术界和工业界的研究人员复制、重用和扩展。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。

Addressing the challenges of climate change requires advanced, efficient, low-emission combustion technologies as well as an educated workforce to understand and solve these challenges. A key limiting factor is the inability of current simulation techniques to accurately predict turbulent combustion in the regimes needed for the design of low-emission combustors and sustainable fuels. Due to practical limits on computing resources, the computational simulations used for engineering design rely on simplified mathematical expressions for some aspects of turbulence and chemical physics; these models are almost always disconnected from each other and so do not capture key physical interactions. Recently, efficient numerical methods to calibrate complex models during flow simulations have been developed using techniques from machine learning and constrained optimization. While successful for simple nonreacting turbulent flows, these models have not been applied to highly nonlinear turbulent reacting flows. The principal objective of this project is to develop efficient methods to calibrate models for the missing physics in simulations of complex turbulent reacting flows, including flows in engineering geometries, which will enhance the predictive accuracy of practical calculations. The resulting methods will be useful across many areas of science and engineering and will be made publicly available in an open-source software package. The project will facilitate interdisciplinary partnerships and student education across traditional borders by developing an annual summer symposium on data and modeling for turbulent combustion. The project will also support the development of an education and research program for an underresourced high school, which will encourage broad understanding of energy science and participation in solutions to national and global energy challenges.This project will address the need for accurate, efficient turbulent combustion models by developing turbulence closures and optimization methods for both canonical and complex simulations of turbulent reacting flows. An adjoint-based optimization method will enable efficient optimization of closure models over the Navier–Stokes equations for canonical flows such as turbulent jet flames and wedge-shaped flameholders. The primary challenge in applying adjoint-based optimization is the need for intrusive access to a code’s data structures, which is practically impossible to achieve for general-purpose computational fluid dynamics (CFD) solvers. To address this, a novel co-optimization framework will be developed to leverage both adjoint-based optimization over canonical flows and ensemble Kalman-based (adjoint-free) optimization over geometrically complex flows and experimental data. This combined approach will train models for both the canonical and complex physics while alleviating the current limitations of embedded optimization for general-purpose CFD codes. More broadly, the scientific community is interested in developing methods to leverage large datasets; therefore, this project’s methods have potential to be adopted widely across disciplines. The resulting data, optimization framework, and trained models will be distributed as open-source software to facilitate replication, reuse, and extension by researchers in academia and industry.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.