Continual Graph Learning for Thermal Analysis of Composite Materials under Interface Variations
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Graph Neural Network, Continual Graph Learning, Thermal Analysis
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: Utilize physics-contrained GNN to conduct efficient thermal prediction on composite materials. Propose a continual graph learning method to address thermal transfer blockage due to interface variations.
Abstract: Thermal analysis is an important topic in many fields, such as building, machinery, and microelectronics. As the types of materials in a system are increasingly diverse, conventional numerical methods or machine learning-based surrogate models face tremendous challenges in computation cost and accuracy. Furthermore, a realistic system usually suffers from random fabrication variations that induce significant errors in model prediction. To overcome these issues, we propose Graph Neural Networks (GNN) as a framework for thermal analysis of composite materials with diverse thermal conductivity and thermal interface variations. Using chiplets in microelectronics as the study case, we first partition the system into sub-blocks based on their material property. Then we develop a physics-constrained GNN as the aggregator to integrate local models of each sub-block into a system, with the edge to represent the thermal interaction. In the presence of interface variations, we introduce continual adaptation of the GNN model, using a minimum number of training samples. Compared with previous solutions, our GNN model is robust for various material and interface conditions, and efficient in the prediction of hot-spot.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8528
Loading