Shaking Outside the Box to Advance Flight Research

Shaking Outside the Box to Advance Flight Research

Lagrangian Particle Tracking (LPT) is a popular flow visualization technique that lets scientists track tracer particles in a three-dimensional volume using highspeed cameras. This method plays an important role in many fluid mechanics studies, particularly those that investigate turbulent flows.

Pushing the boundaries of this technique, some researchers are beginning to use multiple high-speed cameras driven by sophisticated 3D measurement software which uses temporal information contained in time-resolved data sets to track particles in densely seeded flows. This newer technique, called Shake the Box (STB), predicts the positions of already tracked particles and then corrects errors using image matching. The algorithms triangulate the positions of new particles in the measurement domain, enabling scientists to look at higher-particle densities with greater positional accuracy compared to traditional LPT.

STB in Action at the German Aerospace Center

A group of researchers from the German Aerospace Center (DLR) recently used the STB method to observe Collar’s triangle of forces — aerodynamic, elastic and inertial — acting on a flexible wing as it flapped inside a water tunnel. The DLR team, made up of Drs. Andreas Schroeder and Daniel Schanz, seeded the tunnel with particle trackers, flowing around the wing, to calculate volumetric pressure fields and determine the aerodynamic loads. At the same time, they used tracked markers, painted on the wing’s surface, to determine the elastic and inertial forces.

The dynamic testing rig that was used in this evaluation.

Data was captured with Phantom high-speed cameras and an integrated imaging system from LaVision, which is known for its expertise in flow and fluid dynamics, offering systems built for advanced optical measurement techniques including Time Resolved Particle Image Velocimetry (TR-PIV) and Particle Tracking Velocimetry (PTV).

According to the DLR team, these measurements are useful for the development of micro-aerial vehicles. “We were inspired by the flapping wing patterns of birds,” Schroeder said. “Understanding how wing configurations affect the flow phenomena and aerodynamic forces around the wing can help us design vehicles that generate lift and thrust more efficiently, or that can handle wind gusts with more agility.”

The DLR team carried out its experiments using a water tunnel that was open at the top and had windows at the sides and bottom. Around the tunnel, they placed eight 4-megapixel (Mpx) high-speed cameras: six Phantom v2640s and one Phantom v1840, along with one Phantom T1340. Three high-powered LED arrays provided illumination. This experimental setup, also included a motorized wing mechanism, which included three NACA 0012 wings with different flexibilities. The wing motion was inspired by nature, as the team based each wing’s form and movement parameters according to prior measurements of owls in free flight.

The eight Phantom cameras, operating at 2 kilohertz (kHz), each contained 72 GB of RAM, allowing the researchers to capture 12,597 consecutive images at 4-Mpx pixel resolution. According to Schroeder, in a typical LPT experiment, the entire measurement volume fits inside the camera’s field of view. “However, the presence of the wing in the tunnel always blocked the view for a subset of cameras as the wing moved,” he explains. To overcome this issue, Schroeder and his colleagues lined up the cameras in two groups of four, with one group facing each side of the tunnel and each unit linking back to a single recording workstation via 10 GB Ethernet. “With this setup, we always had three or four cameras recording each illuminated region in the volume, enabling us to capture the particle distribution around the entire wing.”

To determine the elastic and inertial forces acting on the wing, the researchers seeded the flow with 60-micrometer (μm) spherical polyamide particles. They also applied a random, white dotted pattern to the wing’s surface, allowing them to take time-resolved measurements of the wing’s motion and shape within the water volume. Over the course of the experiments, the team captured 12,597 consecutive images per 6.3 second run, for a total of six cycles.

How Shake the Box Works

The full view of the wing markers and particles: a) First is the raw image; b) second is the minimum over six images, extracted at the same phase position of the wing movement; c) and third is the minimum image subtracted from the camera (a-b).

Central to the success of these experiments was STB Lagrangian particle tracking which, according to the researchers, has two main ingredients. The first ingredient is called Iterative Particle Reconstruction (IPR). The DLR team’s special tracking software detected the particle peaks on the captured images and then triangulated the particle positions using images from the multiple cameras.

“In particle reconstructions, the particle’s 3D position is often slightly displaced in relation to the true position,” Schanz explains. “This can be the result of noise, image overlap or the presence of ghost particles.” To overcome this issue, the position and intensity of each reconstructed particle were optimized by fitting the particle’s 3D position and intensity to the camera images. This position optimization step, based on image matching, is integral to IPR. According to Schanz, “We can optimize a particle’s position by comparing the original image of the particle to an image we create synthetically. As a result, we can see the mismatch between the reprojected image and the original image.” This process of moving the particles around in space to optimize their position is what is meant by “shaking the box.”

Although STB unlocks particle tracking at higher seeding densities, ghost particles remain a challenge. This is where the second main STB ingredient comes into play: exploitation of the temporal domain. Using available temporal information, the tracking software extended known particle tracks to the next time-step and then corrected prediction errors by the shaking process.

“The value of STB is its ability to use the available temporal information to reduce the reconstruction complexity of each time-step,” Schanz said. “Because we’re using multiple cameras, we have a full time-series available to us. STB extracts the additional temporal information we have from the multiple images of the same particle in the time-series.”

According to the scientist, the STB algorithms initially concentrated on the first four time-steps (T1-T4) in the series. Reconstructing the particles for each of the four time-steps enabled the research team to discriminate between the good particles and the bad. “Ghost particles pop up randomly at different locations in the time-steps, whereas true particles follow a trajectory,” Schanz explains. “In light of this fact, we can discern the good particles from the bad ones. We can also begin extending the known particle tracks to the next step in the time-series.”

At the fifth time-step (T5), the STB algorithms used the information for the particles tracked so far to predict their position. “While we can do this with a high degree of accuracy, there are still inevitable errors due to things like particle acceleration or noise,” Schanz said. “To correct these deviations, we shake the box — subtracting the projected image from the original image.”

Four Phantom cameras, installed on each side of the water tunnel, formed a common camera system.

The STB algorithm continued in this way, predicting, correcting and shaking more particles, until it worked its way through the entire time-series.

The Right Cameras for the Job

While the quantity of high-speed cameras was a requirement to see both sides of the wing, the specific camera models were chosen based on factors like frame rate, sensor resolution and ease of useability. “For these experiments, we knew we needed a repetition rate of at least 2 kHz, or roughly 2,000 frames per second — maybe higher,” Schantz said. “We also needed a sensor resolution of at least 4 megapixels to image enough particles to resolve the spatial structures of the flow.”

Meeting these technical requirements, the Phantom v2640 and v1840 are capable of at least 4,500 fps at 2048 x 1952 resolution. The third high-speed camera model used for these experiments was the newer Phantom T1340. It provided the researchers with the performance required in a much smaller form factor, ideal for environments with limited space. Each of the Phantom models feature a low noise rating (8.7 e- or below) which is a tremendous benefit that enables the capture of measurable detail in dark and traditionally difficult-to-capture regions of the image. All three camera types met the requirement for sensor resolution and exceeded the requirement for frame rate. “The Phantom cameras gave us the highly detailed images we needed to image the particles for our STB analysis,” Schanz said.

Phantom cameras work with LaVision DaVis, a complete software package for intelligent imaging applications in the field of flow measurement technology. While the researchers used their own software for the particle tracking and STB analysis, they did use LaVision software for the image capture and save process. The DLR team used to have to set up one PC per camera — hardly a flexible solution for field experiments that require multiple cameras. “For these wing experiments, we simply connected all eight Phantom cameras to a single PC configured by LaVision, which streamlined our setup significantly,” Schroeder said.

The lightning-fast download speed, enabled by the Phantom cameras’ 10 Gb Ethernet connection, was an important feature for the researchers, who previously had to wait 30 minutes or longer to download their images. “Standard Ethernet was not a viable option whenever we went out into the field and had to set up and take down our equipment quickly,” Schroeder said. “The slow download speed wasted too much time.”

Alternatively, for these wing experiments, the team could run all eight cameras and download all images simultaneously in a matter of minutes using 10 Gb Ethernet and the integrated LaVision system. Optimizing the quality of the images even further was “quiet mode.” This camera option switches off the cooling fans during the recording process. “This option eliminated a source of vibration that could have impacted the accuracy of our measurements,” Schroeder said.

A Promising Look to the Future

While the DLR research team has yet to evaluate the data in terms of its implications for flow physics — the delay being an unfortunate side-effect of the COVID-19 pandemic — the experiments incorporated many milestones, particularly regarding the potential of STB to accurately track densely seeded flows. The trials demonstrated that STB can be successfully performed using eight high-speed cameras, versus the three or four cameras typically used as part of this technique. The experiments revealed that variables, such as the lines of sight of multiple cameras, can be successfully resolved with obstacles occupying the same measurement volume as the tracer particles.

“These experiments feature the highest fluid seeding densities I’ve seen so far. The amount of cameras, the high sensor resolution and the quality of the setup have yielded such high quality data. It all points to very promising results — and future research endeavors,” said Alex Nila, the LaVision Applications Consultant who worked with the DLR research team.

“Our evaluation is ongoing,” Schanz adds. “Our next task will be to fit a 3D finite element model of the wing to the captured marker cloud, enabling us to fully describe the flexing of the wing.”

This article was written by Toni Lucatorto, Product Manager, Vision Research. For more information, visit here .