Augmented Reality Guided Aerodynamic Sampling
Julian Humml
∗
ETH Zürich, Zürich, ZH, 8092
California Institute of Technology, Pasadena, CA, 91125
Victor Cohen
†
and Fernando Perez-Cruz
‡
ETH Zürich, Zürich, ZH, 8092
Morteza Gharib
§
California Institute of Technology, Pasadena, CA, 91125
Thomas Rösgen
¶
ETH Zürich, Zürich, ZH, 8092
Probe-based flow measurement systems are relevant in a wide range of applications. A
prominent application area of such systems are wind tunnel measurement campaigns. Without
prior knowledge about the flow under investigation, real-time flow feedback delivered to the
human probe operator is critical for arriving at a versatile yet accurate and time-economical
measurement. To this end, we exploit in the present work the potential of augmented reality (AR).
Here, flow feedback is provided by overlaying a real-time-updatable flow map and active learning
data models on top of the physical objects in the measurement domain. More specifically,
we apply advanced visualization techniques and present a fully functional system involving
a multi-hole pressure probe combined with motion-capture and real-time AR feedback. We
demonstrate the capabilities of the system in a typical experimental setting. Our demonstrations
document excellent qualitative and good quantitative agreement while delivering a significant
speedup in system setup and measurement time.
Nomenclature
퐴
= Gram matrix
훼
= angle probe acceptance yaw
훽
= angle probe acceptance pitch
휖
= error / uncertainty metric
퐾
= GP kernel
푝
= pressure
푞
푞
푞
= spatial points
푢
푢
푢
=
(
푢,푣,푤
)
푇
= flow vector
푟
푟
푟
=
(
휑,휓,휃
)
푇
= orientation coordinates
푅
= rotation matrix
푆
power spectral density =
푡
푡
푡
= translation vector
푤
푤
푤
= weight vector
푥
푥
푥
=
(
푥, 푦, 푧
)
푇
= position coordinates
푥
푖
, 푦
푖
= feature and target value
휎
= variance
∗
Postdoctoral Scholar, GALCIT, California Blvd. 1200 MC105-50, 91125, Pasadena, AIAA Member, hummlj@caltech.edu
†
Senior Data Scientist, Swiss Data Science Center, Turnerstrasse 1., Zürich ZH, 8092
‡
Professor, Swiss Data Science Center & ETH Zürich, Turnerstrasse 1., Zürich ZH, 8092
§
Professor, Graduate Aerospace Laboratories, AIAA Senior Member
¶
Professor em., Institute of Fluid Dynamics, ETH Zürich, Sonneggstrasse 3, 8092, Zürich.
1
Downloaded by 131.215.248.203 on June 27, 2024 | http://arc.aiaa.org | DOI: 10.2514/6.2024-1382
AIAA SCITECH 2024 Forum
8-12 January 2024, Orlando, FL
10.2514/6.2024-1382
Copyright © 2024 by Caltech and ETH Zürich, J.
Humml, C.Cohen, F. Perez-Cruz, M.Gharib, T.Rösgen. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
AIAA SciTech Forum
I. Introduction
Aerodynamic optimization is a crucial aspect of designing modern and future transportation systems, including
cars, airplanes, and trains. The main goal is to develop efficient designs that minimize energy consumption, reduce
environmental impact, and enhance customer comfort. Achieving these objectives requires a combination of numerical
simulations and experimental tests, each with its own limitations, such as cost and turnover time. Engineers rely heavily
on visualizing the flow field around the body to optimize its shape for the desired aerodynamic properties. Two branches
of fluid dynamics, Computational Fluid Dynamics (CFD) and experimental investigations are used to gain insights into
these flow structures. While CFD simulations generate rich datasets that provide valuable insights into the fluid-surface
interactions, they require attention to numerical difficulties and parameter calibration, which can be challenging in the
early stages of development. On the other hand, quantitative flow visualizations of large-scale flows around objects are
often limited by the number of measured quantities, sparsity of data, and setup turnaround time.
Today, there are numerous advanced techniques available for measuring and visualizing large-scale flow fields [
1
] [
2
].
Among these, Particle Image Velocimetry (PIV), Particle Tracking Velocimetry (PTV), Laser Doppler Anemometry
(LDA), and classical traversing of pressure probes through a domain of interest are the most prominent. Optical
techniques rely on the seeding of the flow field in the region of interest with tracers like Helium Filled Soap Bubbles
(HFSB), oil droplets, or smoke, enabling non-intrusive, local measurements of the velocity components. However, strict
requirements for domain illumination can result in longer setup times and access control, especially when using laser
devices. Commercial measurement systems capable of investigating time-resolved flow fields are now readily available
using these optical measurement techniques. It remains important, however, to consider operational constraints, as well
as the reproducibility of the measurements, such as seeding density in so-called dead water zones.
[3] developed a new approach for manual pressure probe traversing by taking a well-known multi-hole pressure probe
and incorporating its data stream into a processing chain, allowing real-time visualization of the reconstructed flow field
for an operator guiding the probe. Although an intrusive measurement technique, the manual guidance of the probe by
the test engineer to the reconstructed coherent flow features reduces the measurement time drastically. The output of
this technique is similar to a three-dimensional steady-state CFD simulation but derived from a measurement of the
actual flow field. This enables the same post-processing pipelines for further analysis and investigation of the data.
For the aerodynamicist to interpret the measured flow field and introduce adequate steps to improve the design of the
object under investigation, powerful and intuitive visualizations are essential. Projecting the reconstructed flow field
in the field of view of an operator and overlaying with a physical world object has been explored for fluid dynamic
applications, e.g., by [
4
] [
5
] [
6
] [
7
] [
8
] [
9
] [
10
], and general engineering, e.g., [
11
] [
12
] or medical applications.
However, only the 3rd wave of computing produces devices capable of delivering the performance needed to incorporate
such systems in sophisticated real-time processes and interactive procedures.
We present a novel advancement in human-machine collaboration that features an improved visualization technique and
the ability to reconstruct global flow fields. The system has short setup times and employs advanced visualization tools,
leading to faster turnaround times and enhanced comprehension of flow field characteristics. The system is built around
an Augmented Reality (AR) headset, resulting in a significant improvement in usability. It incorporates algorithms
based on narrow artificial intelligence (AI) to provide manual guidance of the measurement probe to optimal locations,
akin to [
13
] or [
14
]. We have named this system SmartAIR, which stands for Self-guided Machine Learning Algorithm
for Real-time Assimilation, Interpolation, and Rendering of Flow Data.
II. System Setup and Measurement Chain
The SmartAIR system is composed of several individual components that deliver data streams to be combined,
analyzed and used for comprehensive visualization of the measured flow field in real-time. The subsystems are depicted
in Figure 1. The operator is guiding a multi-hole pressure probe through a flow field surrounding a test object under
investigation. The multi-hole pressure probe is equipped with retroreflective markers that allow a motion-capture system
to identify the probe’s rigid body and determine its position and orientation in space. Two high-frequency data streams
from the motion-capture system and the multi-hole pressure probe’s sensors are combined into individual time-tagged
data packages containing the information of the sampled flow field quantities. This data stream is then further evaluated
on a computer and sent to a Head-Mounted Display (HMD) worn by the operator. If an AR device is used (the system
could also be built with a Virtual Reality (VR) display), the measured flow field can directly and in real-time be
overlayed over the physical scene allowing to display the otherwise invisible properties of the measured air stream.
Statistical criteria regarding the current global measurement allow identifying the next optimal measurement location.
The spatial coordinates of this location are displayed to the operator to move the probe to the exact location establishing
2
Downloaded by 131.215.248.203 on June 27, 2024 | http://arc.aiaa.org | DOI: 10.2514/6.2024-1382