SenseDynamics

Predicting aerodynamics forces from sensor data

Started
May 1, 2020
Status
Completed
Share this project

Abstract

To advance state-of-the-art Machine Learning capabilities when creating precise surrogate models of transient nonlinear physical phenomena related to aerodynamics with known (quantified probability density function describing it) and traceable uncertainty. Create the capability to reconstruct such complex physical transient phenomenon with a minimum amount of discrete real sensorial input of finite and calibrated precision.

People

Collaborators

SDSC Team:
Christian Donner
Natasa Tagasovska
Guillaume Obozinski

PI | Partners:

EPFL, Geodetic Engineering Lab:

  • Dr. Iordan Doytchinov
  • Dr. Jan Skaloud

More info

EPFL, Unsteady Flow Diagnostics Lab:

  • Prof. Karen Mulleners
  • Dr. Guoshenge He
  • Dr. Daniel Fernex

More info

EPFL, Chair of Computational Mathematics and Simulation Science:

  • Prof. Jan Hesthaven
  • Dr. Qian Wang
  • Dr. Junming Duan

More info

Federal Institute of Metrology:

  • Dr. Henri Baumann

More info

description

Motivation

Aerodynamic navigation requires real-time capabilities to reliably infer and predict forces that act on the vehicle, based on sensor measurements. While physical simulations are complex and cannot provide fast predictions, surrogate Machine Learning models should be investigated to emulate the physics accurately.

Proposed Approach / Solution

We apply probabilistic time-series models to perform “now-casting” and “forecasting” the dynamics of a pitching airfoil subject to constant incoming wind. These models should detect model accurately the forces that act on the wing, and detect when there is danger of stall, i.e., flow disruption. Here we investigated several models, like linear regression, feed-forward neural networks, and hidden Markov models (HMM), that predict the lift force (force pushing the wing upwards) based on as few pressure sensors as possible placed on the wing. We found, that with a minimum of 4 sensors, one can predict accurately the lift force. In addition, the HMM model could even give acceptable predictions, if one of these sensors fails for some reason, without retraining the model.

Impact

As a proof-concept we showed, that probabilistic models can accurately predict the forces that act on the wing, and that they can extrapolate to unseen conditions. This is an essential step towards machine-learning-based autonomous navigation of flight vehicles.

Figure 1: An illustration of a stall development over an airfoil, as recorded in the experimental data. In the first row, each panel A – E represents one of five different snapshots of the airflow.  Every state is characterized by a significantly different behavior from the others (some exhibiting more turbulence as C and D, compared to A and B) which makes the prediction task challenging. Namely, as depicted in the second row, the machine learning model should be able to accurately predict a volatile output (compare grey and dashed red line), along with a confidence envelope around it (red band) which will faithfully capture the variability in the data.

Gallery

Annexe

Publications

  • Donner Christian, Tagasovska Natasa, He Guosheng, Mulleners Karen Shimazaki Hideaki, Obozinski Guillaume “Learning interpretable latent dynamics for a 2D airfoil system“, presented as a RobustML workshop paper at ICLR 2021

Additional resources

Bibliography

  1. Deparday, J., Mulleners, K. (2019). Modeling the interplay between the shear layer and leading edge suction during dynamic stall Physics of Fluids 31(10), 107104. https://dx.doi.org/10.1063/1.5121312
  2. Mulleners, K., Raffel, M. (2013). Dynamic stall development Experiments in Fluids 54(2), 1469 1477. https://dx.doi.org/10.1007/s00348-013-1469-7
  3. Mulleners, K., Raffel, M. (2012). The onset of dynamic stall revisited Experiments in Fluids 52(3), 779 793. https://dx.doi.org/10.1007/s00348-011-1118-y

Publications

Related Pages

More projects

ML-L3DNDT

Completed
Robust and scalable Machine Learning algorithms for Laue 3-Dimensional Neutron Diffraction Tomography
Big Science Data

BioDetect

Completed
Deep Learning for Biodiversity Detection and Classification
Energy, Climate & Environment

IRMA

In Progress
Interpretable and Robust Machine Learning for Mobility Analysis
No items found.

FLBI

In Progress
Feature Learning for Bayesian Inference
No items found.

News

Latest news

Smartair | An active learning algorithm for real-time acquisition and regression of flow field data
May 1, 2024

Smartair | An active learning algorithm for real-time acquisition and regression of flow field data

Smartair | An active learning algorithm for real-time acquisition and regression of flow field data

We’ve developed a smart solution for wind tunnel testing that learns as it works, providing accurate results faster. It provides an accurate mean flow field and turbulence field reconstruction while shortening the sampling time.
The Promise of AI in Pharmaceutical Manufacturing
April 22, 2024

The Promise of AI in Pharmaceutical Manufacturing

The Promise of AI in Pharmaceutical Manufacturing

Innovation in pharmaceutical manufacturing raises key questions: How will AI change our operations? What does this mean for the skills of our workforce? How will it reshape our collaborative efforts? And crucially, how can we fully leverage these changes?
Efficient and scalable graph generation through iterative local expansion
March 20, 2024

Efficient and scalable graph generation through iterative local expansion

Efficient and scalable graph generation through iterative local expansion

Have you ever considered the complexity of generating large-scale, intricate graphs akin to those that represent the vast relational structures of our world? Our research introduces a pioneering approach to graph generation that tackles the scalability and complexity of creating such expansive, real-world graphs.

Contact us

Let’s talk Data Science

Do you need our services or expertise?
Contact us for your next Data Science project!