題 目： Heat Transfer Studies: from “Feature-based” to “Rule-based”
報告人： 楊力 上海交通大學副教授
時 間： 2021 年 7 月 11 日（周日） 上午 10:00~11:00
地 點： 1號巨構5078
邀請人： 馬挺 教授
報告簡介：Geometry definition, heat transfer evaluation and design optimization are the main processes of heat transfer structure research. At present, the bottlenecks of heat transfer performance in many applications mainly come from these three aspects because of the constraints on design caused by "feature thinking". This report will introduce the limitations of feature-based traditional research methods including parametric modeling, data regression and genetic optimization, and the advantages of rule-based research methods including self-organizing geometry, operator neural networks and reinforcement learning. Taking the typical convective heat transfer problem as an example, it will be explained the significance of "rule thinking" for the study of heat transfer structure. The report shall include three parts. In the first part, in view of the limitations of parametric modeling, the presenters explored the mechanism of geometry and its programmable approach, and analyze the compatibility of geometry with flow field and scalar field through the geometric shape design method based on self-organization theory. In the second part, the presenters shall talk about the limitations of regression deep learning which is based on image processing for temperature field reconstruction. Besides, a rule-based operator deep learning method shall be proposed for multi-size, multi-working conditions, and variable topological geometry heat transfer information learning method. In the third part, the feasibility of using the intelligent body for spatial structure design is discussed in view of the limitations of parameter optimization. At the moment, the versatility of the artificial body for optimization problems is analyzed. Reinforcement learning methods are applied to significantly accelerate the optimization speed of the structure, and to a certain extent avoid the dimensional explosion.