Abstract
Moving heat source problems are commonly seen in many manufacturing applications, while numerical modeling often takes time to analyze. This article presents a convolution neural network (CNN)-based framework for rapid prediction of temperature distribution and two methods to improve the overall efficiency and accuracy when the framework is scaled to large 3D geometries. The first method, referred to as geometric subsection training, reduces the amount of spatial data needed by over 90% for the specific 3D geometry used in this framework. The second method, referred to as the boundary-focused training method, allows for further scalability of the framework to large and/or complicated geometries using a clustering approach to classify the spatial data. Then, a tandem learning approach is adopted to train a series of CNNs for each respective cluster. These methods are implemented on a complex 3D geometry and a random sequential moving heat source as proof of concept. Results show a high level of agreement with the ground truth generated by finite element analysis. The scalability and limitations of this approach are also discussed in this article.
