(DEPSCOR-RC FY21) TOPOLOGY-AWARE LEARNING AND MODELING OF HIGH-RATE DYNAMIC SYSTEMS
Abstract
High-rate or sub-second systems are defined as systems experiencing high-rate (< 100 ms) and/or high-amplitude (acceleration > 100 gn) events. This topic is of great interest to DoD, in particular concerning hypersonic vehicles and weapons. Being able to model the dynamics of these systems accurately would yield real-time decision-making capabilities to improve operations, prevent further damage and complete failure, ensure users’ safety, and reduce economic losses. This DEPSCoR project will create a real-time modeling methodology that leverages algebraic topology to produce high-performance models in unknown, high-rate dynamic environments. Our method consists of an ensemble of topology-aware autoencoders capable of real-time learning and prediction of high-rate dynamics. Each autoencoder is a recurrent neural network encoder-decoder model tasked with learning a latent representation of an input space with a unique embedding dimension and time lag (i.e., delay vector). A particular strength of our proposed method is in the integration of topological data analysis with ensemble learning to yield topology-aware learning capabilities for high-rate dynamics. Our proposed research will improve capabilities in modeling high-rate dynamic systems and other nonstationary dynamics by developing topology-aware learning methods. By quantifying algorithmic performance using well-defined metrics, we expect to gain useful knowledge on the trade-off between modeling accuracy and computation time. We will work closely with the DEVCOM Army Research Laboratory (ARL) and Air Force Research Laboratory Munitions Directorate (AFRL/RW) to produce discoveries using realistic datasets and expert feedback. The expected outcomes are (1) concepts, methods, and mathematical frameworks for topology-aware learning of high-rate systems, (2) metrics for evaluation of modeling performance, (3) realistic, DoD-relevant datasets from high-rate systems of interest to the ARL, AFRL/RW, and other DoD organizations, and (4) results and findings of performance evaluation.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 20, 2023
- Source ID
- FA95502210303
Entities
People
- Chao Hu
Organizations
- Air Force Office of Scientific Research
- Office of the Secretary of Defense
- University of Connecticut