The Intelligent Integration of Digital Twins with Physical Experiments

燃气涡轮机行业的新技术开发--从最初的研究项目到在生产发动机中的应用--通常与高成本和长时间相关联。目前，该行业的一个优先事项是减少开发过程的费用，并使采用提高性能的技术创新成为一个更加精简和具有成本效益的程序。数字双胞胎-用于高精度CFD模拟的单个燃气轮机的高分辨率，特定于单元的计算机表示-有可能在开发过程中进行冗余的某些测试和组件验证阶段，我的调查将集中在这种潜力上。通常，新技术首先在研究或学术环境中进行研究，在这种环境中，其行为可以被测量，量化和更彻底地理解。如果一项技术有潜力以某种方式提高发动机性能，则可以在公司的研究设施中进行进一步研究：可以建造专用的部件试验台，并且正在研究的部件也将被引入开发发动机并在更现实的条件下进行测试。最后，如果在这个阶段成功，该组件将被引入生产引擎，然后进行最终测试，在移交给客户之前验证机器。近年来，专门的组件测试和开发引擎测试已经变得越来越不经济，并且偶尔会出现这种测试超出客户移交的情况。用一种更便宜、更快的替代品来取代这一开发阶段，为公司工程师提供同等或更大的效用，显然会带来巨大的好处。我的项目将调查如何专用组件测试台架和开发发动机测试台架可能被替换为一个单一的“原型发动机”测试阶段，通过使用数字孪生补充，以提高效率和降低成本。而不是孤立地执行单个组件的详尽测试，组件将被安装到原型引擎中，其中存在可用于生成模拟的数字孪生副本。给定来自该改装原型发动机的试验数据，以及已知性能敏感的某些外部因素的测量结果（例如，大气条件、燃料质量、机器健康指数等），可以与“基线”原型数字孪生模拟进行公平的比较。从理论上讲，新组件的影响可以从物理测量和模拟结果之间的差异中推断出来。以这种方式取代组件测试台阶段可能会降低整体成本，并更快地集成新技术。为了能够实现这一目标，对历史数字孪生和物理实验数据的调查是必要的。特别是，必须充分了解对不同外部条件的敏感性以及实验结果的可重复性和可靠性-为了使提案有效，必须展示如何在可变条件下将数字孪生结果可靠地映射到物理结果。必须准确地量化可以作出预测和推断业绩衡量标准的确定程度。为此，我将采用统计灵敏度方法（如Sobol分析），并将为燃气涡轮机部件建立一些新的低阶模型，以便与可变条件下的数字孪生模拟和物理实验结果进行比较。

The development of new technology in the gas turbine industry - from initial research projects to application in production engines - is often associated with large costs and long timescales. It is now a priority in this industry to reduce the expense of the development process, and to make the introduction of performance-enhancing technological innovations a more streamlined and cost-effective procedure. Digital Twins - high resolution, unit-specific computer representations of individual gas turbines used for high-precision CFD simulations - have the potential to make redundant certain stages of testing and component validation in the development process, and it will be this potential about which my investigations will focus. Typically, novel technology is first investigated in a research or academic setting, in which its behaviour can be measured, quantified, and more thoroughly understood. If a technology has potential to improve engine performance in some way, it may be further investigated in company research facilities: dedicated component test rigs may be constructed, and the part being investigated will also be introduced into a development engine and tested under more realistic conditions. Finally, if successful at this stage, the component will be introduced into a production engine, before final tests are run, validating the machine before handover to a customer. In recent years, it has been the case that dedicated component testing and the development engine testing has become increasingly uneconomic, and it has occasionally been the case that such tests overrun to a point even beyond customer handover. Replacing this stage of development with a cheaper, quicker alternative that provides equal or greater utility to company engineers would clearly be of great benefit. My project will investigate how dedicated component test rigs and development engine test rigs may be replaced with a single 'prototype engine' test stage, complimented by the use of a Digital Twin, to improve efficiency and lower costs. Instead of performing exhaustive tests of a single component in isolation, the component would be instead installed into a prototype engine, of which there exists a Digital Twin copy available to generate simulations. Given test data from this adapted prototype engine, and measurements of certain externalities to which performance is known to be sensitive (e.g. atmospheric conditions, fuel quality, machine health indices etc.), a fair comparison could be made against a 'baseline' prototype digital twin simulation. In theory, the impact of the new component could be inferred from differences between the physical measurements and the simulation results. Replacing the component test rig stage in this way could potentially result in reduced overall costs and more rapid integration of new technologies. To be able to achieve this goal, an investigation into historical Digital Twin and physical experimental data is necessary. In particular, the sensitivities to different external conditions, and the reproducibility and reliability of experimental results must be well understood - for the proposal to work, it must be shown how Digital Twin results can be reliably mapped to physical results under variable conditions. The degree of certainty with which predictions can be made and measures of performance can be inferred must be accurately quantified. To do this, I will be making novel use of statistical sensitivity methods (such as Sobol analysis), and will be producing a number of new low-order models for gas turbine components, against which the results of Digital Twin simulations and physical experiments under variable conditions may be compared.