ITR: Simulation and Optimization of Thermal Manufacturing of Materials Under Uncertainty: Application to Optical Fiber Drawing

先进材料构成了我们日常生活的支柱，包括支持信息技术(IT)企业的硬件。这些材料的热制造涉及在多个时间和空间尺度上发生的强耦合传输现象。过程仿真模型建立在对过程控制物理现象描述的基础上，对过程的理解和发展起着重要的指导作用。然而，由于两个主要原因，它们的全部潜力在实践中仍然没有实现。首先，模拟能力和实践之间存在根本差距，因为实际过程是在不精确和不确定的阴云中进行的，而模拟模型在处理过程变量的方式上是确定性的。其次，由于需要在多个尺度上对物理现象进行严格的解析，过程模拟在计算上通常是乏味的。当考虑过程不确定性的能力被嵌入到仿真框架中时，以及进一步地，当仿真模型被用于优化工作时，计算需求被极大地增强。研究试图开发旨在解决上述挑战的高级计算方法。将开发一个随机建模框架，以将过程不确定性的影响纳入模拟。为了解决实现快速和高效计算的挑战，将开发出在不同环境中基于代理的计算战略和基于投资组合的创新技术，用于在不确定情况下进行大规模优化。先进的计算方法将应用于光纤制造过程，这是光网络行业中的一个重要过程，代表了一般热制造过程中涉及的多尺度物理现象的复杂性。这些方法可以很容易地应用于其他材料加工应用。

Advanced materials form the backbone of our everyday life, including the hardware that supports the information technology (IT) enterprise. Thermal manufacturing of these materials involves strongly coupled transport phenomena that occur at multiple temporal and spatial scales. Process simulation models, based on description of the governing physical phenomena, play an important role in guiding process understanding and development. However, their complete potential remains unrealized in practice, for two principal reasons. First, a fundamental gap exists between simulation capabilities and practice in that whereas practical processes are carried out amidst a cloud of impreciseness and uncertainty, the simulation models are deterministic in the way they treat the process variables. Secondly, process simulations are often computationally tedious owing to the need for a rigorous resolution of the physical phenomena at multiple scales. The computational demands are tremendously intensified when the ability to account for process uncertainty is embedded in the simulation framework, and further, when the simulation models are used in an optimization endeavor.The research seeks to develop advanced computational methods aimed at addressing the foregoing challenges. A stochastic modeling framework will be developed for incorporating the effects of process uncertainties in the simulations. Towards addressing the challenge of enabling rapid and efficient computations, agent-based computing strategies in a heterogeneous environment and an innovative portfolio-based technique for large-scale optimization under uncertainty will be developed. The advanced computational methods will be applied to an optical fiber manufacturing process, which is an important process in the optical networking industry and typifies the complexities of the multiscale physical phenomena involved in general thermal manufacturing processes. The methodologies may be readily applied to other materials processing applications.