A GPU Cluster for Differentiable Neural Computing in Manufacturing Modeling of Hypersonic Materials
Abstract
Optimizing the manufacturing process to fabricate new composites for hypersonic applications is very challenging since the state-of-the-art processing methods remain manufacturing by hand, and each process involves many inter-dependent steps and can take months. On the other hand, as the underlying physics of the manufacturing processes is very complex and not fully understood, computational models based on first principles are not available. Data-driven modeling based on recent advances in AI and machine learning opens up new possibilities to tackle this challenge and show great promise. However, there are many challenges due to massive data requirements and a lack of the capabilities to estimate the uncertainties from multiple sources. Therefore, we aim to develop a multi-fidelity physics-informed Bayesian deep learning framework to identify and reduce the uncertainties involved in the manufacturing processes of carbon-carbon composites. The overarching goal is to understand, model, and optimize the manufacturing processes of C-C composites with respect to their performance properties for hypersonic applications by integrating physics prior knowledge, numerical techniques, in-situ sensing data within a physics-informed scientific machine learning (PiSciML) framework based on differentiable neural programming.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310082
Entities
People
- Jian-Xun Wang
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Notre Dame