Sensing is one of the most fundamental tasks for the monitoring, forecasting, characterization and control of complex, spatio-temporal systems.
Abstract
Sensing is one of the most fundamental tasks for the monitoring, forecasting, characterizing and controingl of complex, spatio-temporal systems. Sensors are typically limited in measuring partial differential equations-based systems and-or spatio-temporal fields. In these dynamic systems, the measurement time history, which is typically ignored in most sensor reconstruction strategies, encodes a significant amount of information that can be extracted for critical tasks. Most model-free sensing paradigms aim to map current sparse sensor measurements to the high-dimensional state space, ignoring the time-history all together. However, using modern deep learning architectures, we show that a sequence-to-vector model, such as an LSTM(long, short-term memory) network, in combination with a decoder network, dynamic trajectory information can be mapped to full state-space estimates, thus allowing one to leverage critical information embedded in the sensor time-history. Indeed, we demonstrate that by leveraging sensor trajectories with Shallow REcurrent decoder (SHRED) networks, we can train the network (i) to accurately reconstruct the full state space using arbitrary dynamical trajectories of the sensors, (ii) to construct a latent space that is conjectured to be diffeomorphic to the true dynamics, and (iii) to build reduced order models for rapid generalization (parameterization of dynamics) and (iv) to learn physics models for interpretability.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 06, 2025
- Source ID
- FA95502410141
Entities
People
- J. Nathan Kutz
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Washington