In this paper, we use graph-based techniques to investigate the use of geometric deep learning (GDL) in the classification and down-selection of aircraft thermal management systems (TMS). Previous work developed an enumerative graph generation procedure using a component catalog with network structure constraints to represent novel aircraft TMSs as graphs. However, as with many enumerative approaches, combinatorial explosion limits its efficacy in many real-world problems, particularly when simulations and optimization must be performed on the many (automatically-generated) physics models. Therefore, we present an approach that takes the directed graphs representing aircraft TMSs and use GDL to predict the critical characteristics of the remaining graphs. This paper's findings demonstrate that incorporating additional graph-based features enhances performance, achieving an accuracy of 97% for determining a graph's compilability and simulatability while using only 5% of the data for training. By applying iterative classification methods, we also successfully segmented the total set of graphs into more specific groups with an average inclusion of 84.7 of the top 100 highest-performing graphs, achieved by training on 45% of the data.

翻译：本文采用图技术研究几何深度学习（GDL）在飞机热管理系统（TMS）分类与优选中的应用。前期工作开发了一种基于组件目录与网络结构约束的枚举图生成程序，将新型飞机热管理系统表示为图结构。然而，与众多枚举方法类似，组合爆炸问题限制了该方法在诸多实际问题中的有效性，尤其在需要针对大量（自动生成的）物理模型进行仿真与优化时。为此，我们提出一种方法：以表示飞机热管理系统的有向图为输入，利用GDL预测其余图的关键特性。本文结果表明，引入额外图特征可提升性能：在仅使用5%数据进行训练的情况下，判断图的编译性与可仿真性的准确率达97%。通过应用迭代分类方法，我们成功将总图集划分为更具体的子群，使前100个最高性能图的平均包含率达到84.7%，而训练数据仅占45%。