A learning-based joint flow and kinematic state estimation for bodies in highly disturbed flows
Abstract
Accurate estimation of unsteady aerodynamic flows from sparse measurements remains a fundamental challenge, particularly in the presence of strong disturbances, unknown body kinematics, and incomplete observations. This study presents a data-driven sequential estimation framework for joint reconstruction of unsteady flow fields, aerodynamic loads, and airfoil kinematics from sparse measurements. The approach combines a kinematics-aware nonlinear flow autoencoder with online filtering of streaming measurements. In the resulting reduced-order representation, the forecast and observation operators are learned directly from data, with each forecast-assimilation cycle requiring only a few milliseconds. The framework is evaluated on two-dimensional incompressible flow over an airfoil subjected to random vortical gusts while undergoing arbitrary pitch-up motions. Results demonstrate accurate reconstruction from surface pressure sensors, vertical lines of vorticity sensors, and synthetic velocimetry data with representative shadow regions. The transient informativeness of pressure sensors is quantified through their time-varying contributions to dominant observation modes. While lift is reconstructed from the leading modes, drag, pitch angle, and angular velocity require higher modes. Incorporating limited off-body measurements alongside surface pressure improves observability and reduces estimation error and uncertainty.
Related
- Probing Baryons with the Kinematic Sunyae Zel'dovich Effect and Machine Learning Derived Peculiar Velocities using DESI DR2 and ACT DR6
- Stable tracking of turbulence-disturbed laser communication spots using a time-series state fusion network
- Exoskeleton-based user's pose estimation: a hybrid geometric and machine learning approach
- Higgs signal strength estimation with machine learning under systematic uncertainties
- Quantum-enhanced mechanical sensing with machine-learning estimation
BibTeX
@misc{mousavi2026learningbased,
title = {{A learning-based joint flow and kinematic state estimation for bodies in highly disturbed flows}},
author = {Hanieh Mousavi and Anya Jones and Jeff Eldredge},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.13560},
eprint = {2609.13560},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.13560},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - A learning-based joint flow and kinematic state estimation for bodies in highly disturbed flows
AU - Hanieh Mousavi
AU - Anya Jones
AU - Jeff Eldredge
PY - 2026
DA - 2026/09/11
PB - arXiv
DO - 10.48550/arxiv.2609.13560
UR - https://arxiv.org/abs/2609.13560
AB - Accurate estimation of unsteady aerodynamic flows from sparse measurements remains a fundamental challenge, particularly in the presence of strong disturbances, unknown body kinematics, and incomplete observations. This study presents a data-driven sequential estimation framework for joint reconstruction of unsteady flow fields, aerodynamic loads, and airfoil kinematics from sparse measurements. The approach combines a kinematics-aware nonlinear flow autoencoder with online filtering of streaming measurements. In the resulting reduced-order representation, the forecast and observation operators are learned directly from data, with each forecast-assimilation cycle requiring only a few milliseconds. The framework is evaluated on two-dimensional incompressible flow over an airfoil subjected to random vortical gusts while undergoing arbitrary pitch-up motions. Results demonstrate accurate reconstruction from surface pressure sensors, vertical lines of vorticity sensors, and synthetic velocimetry data with representative shadow regions. The transient informativeness of pressure sensors is quantified through their time-varying contributions to dominant observation modes. While lift is reconstructed from the leading modes, drag, pitch angle, and angular velocity require higher modes. Incorporating limited off-body measurements alongside surface pressure improves observability and reduces estimation error and uncertainty.
ER -