A new low-turbulence wind tunnel for animal and small vehicle flight experiments

Our understanding of animal flight benefits greatly from specialized wind tunnels designed for flying animals. Existing facilities can simulate laminar flow during straight, ascending and descending flight, as well as at different altitudes. However, the atmosphere in which animals fly is even more complex. Flow can be laminar and quiet at high altitudes but highly turbulent near the ground, and gusts can rapidly change wind speed. To study flight in both laminar and turbulent environments, a multi-purpose wind tunnel for studying animal and small vehicle flight was built at Stanford University. The tunnel is closed-circuit and can produce airspeeds up to 50 m s−1 in a rectangular test section that is 1.0 m wide, 0.82 m tall and 1.73 m long. Seamless honeycomb and screens in the airline together with a carefully designed contraction reduce centreline turbulence intensities to less than or equal to 0.030% at all operating speeds. A large diameter fan and specialized acoustic treatment allow the tunnel to operate at low noise levels of 76.4 dB at 20 m s−1. To simulate high turbulence, an active turbulence grid can increase turbulence intensities up to 45%. Finally, an open jet configuration enables stereo high-speed fluoroscopy for studying musculoskeletal control in turbulent flow.


Introduction
Wind tunnels are essential tools for studying both vehicle and animal flight in laminar and turbulent flows. The key advantage of wind tunnels for biological research is that airspeed can be finely controlled over an animal that remains stationary in the laboratory frame. This fixed frame allows direct measurements of forces, kinematics and flow fields that would otherwise be unattainable. Inspired by these advantages, several laboratories have built open-circuit wind tunnels designed for studying bird flight in the past 50 years [1][2][3][4][5][6][7][8][9]. Open-circuit tunnels typically produce moderate turbulent intensities (1-3%) 2017 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original author and source are credited. Table 1. Recent closed-circuit wind tunnels produce low turbulence intensities. Modern closed-circuit tunnels use a combination of honeycombs, screens and contractions to create extremely low turbulence intensities. Axial turbulence intensity is the root mean square fluctuation in streamwise velocity compared to the bulk airspeed. Measurements have been reported with different methods and positions within wind tunnels, so identical conditions for comparison are not available in every case. The grid over which measurements were made also varies; the grid at KTH [22], for example, includes points closer to the wall than those reported here. Where possible, tunnels were compared at airspeeds relevant for animal flight (approx. 10 m s −1 ). n.a., no data available. U. Western Ontario [25,26]  and therefore provide valuable information about animal flight in moderate turbulence. Flight environments, however, span a wide range of turbulence, from the extremely laminar flows found in the upper atmosphere (less than 0.1% [10,11]) to the extremely turbulent flows found in forests and street canyons (10-40% [12][13][14]). Turbulence significantly affects flight performance, because it promotes boundary layer transition and affects lift/drag measurements, especially if lift-generating surfaces are operating near stall [15]. As a result, turbulence has been shown to decrease the flight speed of bees [16,17] and moths [18] and increase flight costs in hummingbirds [19].
To study flight in extremely laminar environments, some laboratories have built specialized closedcircuit wind tunnels with very low turbulence levels. While modern open-circuit tunnels can produce turbulence intensities as low as 0.1% [20], they cannot achieve the extremely low turbulence and temperature control possible in closed-circuit tunnels. The Royal Institute of Technology (KTH) [21,22] and Texas A&M [23,24], for example, have closed-circuit tunnels with axial turbulence intensities of less than 0.040% and less than 0.047%, respectively (measured at 10 m s −1 ). A look at the properties of current state-of-the-art animal flight tunnels shows the trade-offs associated with low-turbulence tunnel design (table 1). The tunnel at the University of Western Ontario produces higher turbulence intensities (less than 0.3%) but is able to simulate flight at altitudes up to 7 km [25,26]. The tunnel at Lund University produces low turbulence (less than 0.06%) and tilts to simulate ascending (up to 8°) and descending (up to 6°) flight [27][28][29]. None of these tunnels, however, can produce both highly laminar and highly turbulent flows.
To advance the capabilities of animal flight wind tunnels for biomechanics and aerodynamics studies, we built a multi-purpose tunnel that produces low streamwise turbulence (less than or equal to 0.028% at the centreline over the full operating range), but also high turbulence using an active grid of spinning vanes (up to approx. 45% at the centreline). The low turbulence levels are made possible by six seamless screens, a seamless honeycomb and a 7 : 1 contraction ratio. The tunnel is also barely audible at low airspeeds. Measured noise levels in the test section reach 76.4 dB at 20 m s −1 , allowing quiet natural flight conditions for animals and minimal acoustic disturbances to the flow. The low noise is a result of acoustic treatment throughout the tunnel and an acoustic wall that separates the large diameter fan from the testing area. The tunnel test section (1.0 m wide, 0.82 m tall and 1.73 m long) can be removed to operate the tunnel with an open jet of air in the testing area, allowing the placement of large diagnostic equipment, such as biplanar fluoroscopes for high-speed musculoskeletal recordings. A water-chilled heat exchanger keeps the test section temperature steady (1σ = 0.007°C at 10 m s −1 ) over a range of 10-30°C. The tunnel is washable, and all sections are accessible via hatches, making the tunnel easy to keep sanitary for animals. A system of redundant screens and netting keeps animals safely within the testing area. The high-quality flow and the special features of the multi-purpose tunnel make it the first of its kind. Here, we give a complete description of its layout and its aeroacoustic performance. To define the performance metrics for the wind tunnel, we used the KTH MTL wind tunnel as our primary reference, because it performs well and is relatively well documented [21,22].

Tunnel layout
The new wind tunnel is a closed-circuit tunnel with a footprint of 15.1 × 5.2 m and a maximum height of 4 m. The components of the tunnel are designed to work together to create flow with uniform speed, uniform temperature, low turbulence and a low noise profile (figure 1).

Fan
The airflow is driven by a single stage, 18 blade, axial fan. The fan was selected based on its low acoustic signature, which helped to minimize acoustic levels in the test section. A 1.5 m diameter fan was chosen to maintain the compact size of the wind tunnel while meeting the wind tunnel's low noise requirements. To reduce vibration transmission, the fan's steel housing was mounted on its own massive concrete foundation, separate from the building foundation and connected to the tunnel through expansion joints. The fan is powered by an 80 kW induction motor (891 r.p.m. max speed; ±0.1 r.p.m. resolution), which is housed in an enclosed nacelle. A tail-cone is provided to further improve fan efficiency as well as to provide good airflow performance downstream. The motor is water-cooled and controlled by a variable frequency drive which uses a 100 : 1 turndown ratio to regulate fan speed based on user input. The drive-motor assembly is equipped with braking in the event of an emergency stop.

Flow conditioning
A series of flow conditioning elements are used to straighten and laminarize the flow before it enters the test section. Turning vanes were designed to straighten the flow while minimizing induced turbulence and circuit pressure loss. A honeycomb in the stilling chamber (cell length/diameter = 16) straightens the flow before it enters the flow conditioning screens. To maximize flow alignment, the honeycomb was made seamless and uniform by welding together precisely crimped stainless steel strips. Downstream of the honeycomb, five seamless, fine mesh, stainless steel screens are used to reduce turbulence in the stilling chamber. The mesh screens are spaced 350 mm apart have wire diameters of 0.16 mm, nominal apertures of 0.5 × 0.5 mm and open areas of 57.4%. A sixth identical screen is located in the crossleg between corners 3 and 4 for additional turbulence reduction. Each of the screens is pre-tensioned to reduce sag and lateral displacement. To prevent flow perturbations in the stilling chamber, fairing plates were used to conceal the screen and honeycomb attachment points. Following the honeycomb and flow conditioning screens in the stilling chamber, the flow passes through a 7 : 1 fibreglass contraction,  The resulting 'open-jet' configuration allows easy access to an exposed jet of air in the testing area.
which reduces turbulence intensity by accelerating the flow. The contraction length and curvature were designed with computational fluid dynamics (CFD) simulations to reduce velocity gradients at the contraction exit, thereby minimizing the anisotropy of turbulence in the test section.

Testing area
The testing area can be operated in closed jet, with the test section installed, or open jet, with the test section removed and collector flaps installed (figure 1c). The closed-jet configuration allows maximum airspeeds with minimal turbulence, flow angularity and pressure variation. When operating in closed-jet configuration, a black anodized aluminium test section (airline dimensions 1.0 m wide, 0.82 m tall and 1.73 m long) connects the contraction to corner 1. A breather gap downstream of the test section assures that testing pressures are near atmospheric, allowing access for probes and supports without the need for airtight access ports. Streamwise gradients in pressure are minimized by diverging test section walls that account for boundary layer growth. The floor and ceiling divergence angle is fixed (0.5°) while the sidewall divergence angle is variable to adjust for different blockage levels (see the electronic supplementary material, SM.1).
The test section is designed for efficient high-fidelity diagnostics on animal and small vehicle flight. All components in and around the test section are matt black to facilitate high-speed videography and particle tracking. In addition, each wall of the test section contains a panel that can be removed for repair, replacement, modification or for quick access to the test section interior. Two sets of side panels were built: translucent acrylic, designed for durability during training flights; and thin optical glass, designed for the high transparency required for laser-based flow measurements. The area downstream of the test section is equipped with side and floor hatches (with windows) for easy access while working with animals (electronic supplementary material, figure SF.10). The open-jet configuration allows even easier access to live animals during training sessions in the airflow. It also allows large diagnostic machinery, such as biplanar fluoroscopy emitters and receivers, to be positioned around the flow. To switch into open-jet mode, the test section (approx. 270 kg) is removed with a hydraulic cart, and matt black aluminium collector flaps are attached to corner 1. The angle of the flaps can be adjusted to maximize pressure recovery and minimize flow divergence.

Airspeed control
Airspeeds up to 50 m s −1 in the test section are determined by measuring static pressure readings in the stilling chamber and the contraction. For both readings, pressure is mechanically averaged over two sets of four permanently installed static pressure ports. One set of ports from each section connects to a differential pressure sensor (Rosemount 3051S) tuned to measure airspeeds from 0 to 15 m s −1 . The other set connects to a differential pressure sensor tuned to measure airspeeds from 15-50 m s −1 . This multisensor approach was chosen to ensure high-precision airspeed control at both low and high speeds. A third pressure sensor calculates gauge pressure in the contraction by comparing the static pressure to atmospheric pressure. During commissioning, a Pitot-static probe (Kanomax 1 m) connected to a high-accuracy pressure transducer (MKS 690A/698A) was placed at the centreline in the test section. A second-order polynomial curve fit was generated between the facility pressure port readings and the readings from the Pitot-static probe. This curve fit is used to correct the facility pressure port readings to the true dynamic and static pressures in the test section. Airspeed is calculated from these corrected pressures by using the isentropic flow equations for an ideal gas, including correction factors for temperature and humidity (electronic supplementary material, SM.2).
The airspeed is controlled in open loop by setting the fan speed (r.p.m.), which is determined from a user-defined airspeed via a lookup

Temperature control
The temperature in the tunnel is controlled by a Proportional-Integral controller that regulates input to a water-cooled heat exchanger. The controller compares a user-defined set point temperature to the average reading of four resistance thermometers in corner 4. Based on this comparison, the controller operates two control valves that regulate the exchange of chilled water (approx. 7°C) with water pumped at constant rate through the heat exchanger. This use of a small chilled water loop to regulate the primary coolant loop was chosen to simplify the controller and thus reduce time to reach thermal equilibrium. The controller gains are automatically adjusted based on airspeed to ensure thermal equilibrium at all speeds. Because the tunnel is in a climate-controlled laboratory, a separate heater was not necessary; ambient heat and friction losses in the fan are sufficient to warm the tunnel to the airspeed-dependent temperature range (electronic supplementary material, figure SF.11).
The heat exchanger was designed to maximize water-air heat transfer, maximize temperature uniformity and minimize turbulence generation. The exchanger consists of two sets of vertical copper tubes connected to thin continuous horizontal aluminium fins, all of which are housed in an insulated galvanized steel frame to eliminate oxidation at the edges. To increase temperature uniformity, water is pumped at a constant flow rate and the flow direction reverses between the two sets of tubes. The exchanger is positioned just downstream of the diffuser following the fan. This position was chosen for three reasons: (i) the resulting back pressure helps to keep the flow attached in the diffuser, (ii) any resulting turbulence has time to decay before reaching the test section and (iii) the resulting noise and pressure drop are minimized by placing the exchanger in a section with wide area and thus low airspeed (predicted maximum of 7.1 m s −1 ).

Active turbulence grid
While the tunnel is designed to produce extremely laminar flow, it is also capable of simulating the highly turbulent conditions that animals and vehicles encounter in the atmospheric boundary layer. High turbulence is introduced with the use of an active grid of spinning diamond-shaped vanes, a  [32]. Our grid in particular was modelled after a similar system developed by Cekli et al. [33,34]. The grid is made up of 7 rows and 8 columns of vanes, which are actuated independently by 150 W DC motors (Maxon RE40). For low turbulence conditions, the grid of vanes can be replaced by an empty filler section (figure 1b).
The active grid of vanes can also be used to simulate non-uniform flows in the testing area. For example, by adjusting the fixed position of the vanes, a jet or wake profile can be created downstream of the grid. This technique could be used to investigate flight in turbulent wakes or jets and compare the results with low turbulence conditions. The positions of the vanes can be controlled automatically based on high-speed marker tracking. This enables the grid to react in real time to the kinematics of animals or vehicles flying in the test area.

Additional features
Expanding/contracting corners. To limit the overall circuit length while maintaining flow quality, corners 1 and 2 expand with expansion ratios of 1.17 and 1.28, respectively. The corresponding diffusion halfangles are 1.5°and 2.6°. The heat exchanger pressure drop immediately downstream of the backleg diffuser was included in the design to prevent flow separation at the expansion points. The values for the expansion ratio were chosen to minimize the tunnel footprint while maintaining flow uniformity and minimizing fan power consumption. Similar expansion ratios have been used in previous successful designs by Jacobs Engineering and were verified via CFD for the present wind tunnel.
Acoustic treatment. Acoustic material is used throughout the tunnel to create a low-noise environment in the test section. The internal side walls of the corners and crosslegs are lined with mineral wool (100 mm thickness) encased in perforated sheet metal, designed to attenuate sound in the mid-to-high frequency range. The turning vanes in corners 1-3 are filled with mineral wool as well. The vanes in corner 1 are also covered with acoustically transparent carbon fibre cloth that redirects the flow smoothly while maintaining the acoustic absorption properties of the vane. For maximum noise reduction, synthetic washable fur covers can be placed over the leading edges of the corner 1 turning vanes.
Cleaning and human safety. To maintain sanitary conditions for animals, the tunnel was designed such that it can be washed efficiently. All surfaces-both external and internal-can be cleaned with a lowpressure wash of diluted bleach, which drains through ports in the tunnel floor. The purpose of the wash is to remove flow seeding residue, animal faecal matter and/or feathers. A pulley system slides out the turbulence screens for easy cleaning access. Ceiling panels are designed to support approximately 100 kg for easy access to upper exterior sections and installation of equipment, and all internal sections are accessible through hatches on the sides or bottom of the tunnel. An interlock safety system prevents the tunnel from running when hatches near the fan are open. Finally, the entire wind tunnel laboratory is lead-shielded to meet X-ray health and safety standards for high-speed biplanar fluoroscopy.
Animal safety. A series of protective screens, laser-curtains and nets keeps animals confined to the testing area. First, an array of vertical steel strings (0.5 mm diameter, 10 mm spacing) prevents animals from flying upstream of the test section. Tensioned music wire (high-carbon steel) was chosen because of its high tensile strength and stiffness. These properties (i) minimize flow disturbance by ensuring that the strings are as straight and thin as possible and (ii) minimize wire damage due to beaks and/or claws. Vertical wires were chosen to prevent animals from perching on the strings. The strings are matt black to facilitate high-speed particle tracking and avoid unwanted visual cues for animals flying in the test section. After birds are trained to fly reliably in the test section without entering the pitch-black settling chamber, the array of wires can be removed for experiments where minimal turbulence is required. Likewise, engineering aerodynamic research will be performed without the array of wires. Therefore, the turbulence intensities reported here were measured with the array removed. Flexible netting was installed downstream of the test section and around the breather gap. Together, the wire array and the netting keep animals safely confined to the testing area. In addition, a curtained antechamber around the tunnel and a permanent metal grid between corners 1 and 2 provide failsafe layers of secondary protection for animals. Finally, the laboratory can be hosed down, the air volume is refreshed 15 times per hour, and all lights are flicker-free to meet all modern animal research standards.
Humidity measurement. While relative humidity is not actively controlled, it is measured by a humidity sensor downstream of corner 4 and was found to stay relatively constant (±0.08%) over normal operating times. The humidity measurement is factored into the air density calculation to increase accuracy when recording airspeed (electronic supplementary material, SM.

Tunnel performance
Following installation, tests were performed to quantify the aeroacoustic properties of the wind tunnel. The flow in the test section is stable (1σ airspeed = 0.018 m s −1 at the mid-range speed, 25 m s −1 ), uniform (1σ airspeed = 0.023 m s −1 at 25 m s −1 ), straight (1σ deviation angle = 0.14°at 25 m s −1 ), low turbulence (axial turbulence intensity = 0.021% at 25 m s −1 ) and quiet (81.8 dB at 25 m s −1 ). The datasets resulting from the aeroacoustic testing have been uploaded as part of the electronic supplementary material. Details of the testing procedures and results are given below.

Airspeed stability and uniformity
The airspeed in the test section remained stable over the full operating range (0-50 m s −1 ; figure 2a). To test stability, airspeed was recorded for 240 s at 1 Hz from the calibrated pressure ports. Seven set point airspeeds were considered over the operating range: 5, 10, 20, 25, 30, 40 and 50 m s −1 . Airspeed was more steady at the lowest speed (1σ = 0.017 m s −1 ) than the highest speed (1σ = 0.060 m s −1 ), but all airspeeds remained stable to within less than 0.2% of the mean. The highest airspeed value (50 m s −1 ) was confirmed using measurements from the Pitot-static probe at the centreline.
The airspeed was spatially uniform in the test section (figure 2b). To test uniformity, airspeed was recorded from the Pitot-static probe (60 s at 10 Hz) at each position of the 4 × 4 testing grid. Airspeed uniformity was comparable at the two airspeeds considered: 1σ = 0.020 m s −1 at 10 m s −1 and 1σ = 0.023 m s −1 at 25 m s −1 . For completeness, we also present the total pressure uniformity by using data from only the front port of the Pitot-static probe (electronic supplementary material, figure SF.3). Total pressure variation was slightly lower at 10 m s −1 (1σ = 0.344 Pa) compared with 25 m s −1 (1σ = 0.970 Pa). The uniformity in both airspeed and total pressure demonstrate that total pressure, static pressure and dynamic pressure are all uniform in the test section over the full operating range.
The direction of the flow was straight and uniform in the test section (figure 2c). A flow angle probe measured flow direction by comparing pressure readings from two ports on opposing sides of the probe.

Temperature and humidity stability and uniformity
The temperature and relative humidity remained stable over the full operating range (0-50 m s −1 ; figure 3a). To test stability, temperature and humidity were recorded by the facility probes in the stilling   controller and the constant flow heat exchanger. As relative humidity was not actively controlled, its value is primarily a function of the ambient laboratory conditions. For example, the 20 m s −1 condition was measured on a different day than the other airspeeds, resulting in the different humidity value recorded for that airspeed.
The temperature was also spatially uniform in the test section (figure 3b). To test uniformity, temperature was recorded by a resistance thermometer (60 s at 10 Hz) at each position of the 4 × 4 testing grid. The facility temperature changes slightly in the time it takes to switch between grid point trials. To separate the temporal and spatial variations in temperature, temperatures were corrected by comparing to an average facility temperature (electronic supplementary material, SM.4). Temperature was uniform at each of the two flow airspeeds considered, though more uniform at the lower of the two airspeeds:

Turbulence intensity
Turbulence intensities are low in the test section over the full operating range (0-50 m s −1 ; figure 4a). Two turbulence intensities were considered: the axial turbulence intensity, q u ≡ u /Ū, and the transverse turbulence intensity, q v ≡ v /Ū,where u and v are the root mean square fluctuations in streamwise and vertical flow speed, respectively, andŪ is the time-averaged bulk flow speed. To calculate turbulence intensities, velocity data were sampled at 10 kHz using an x-film hot wire probe (TSI Model 1241-20). The frequency response rate of the probe is airspeed dependent, but it is always above 100 kHz and thus more than 10 times our sampling rate. The AC and DC components of the signal were sampled together (DC coupled), with no gain or offset, using a low-noise anemometer (CTA Thermal Anemometer AN-1003). The maximum signal to noise ratio for the sampled signal was 96 dB based on the 16 bit analogueto-digital conversion. Notch filters were applied between 60i ± 1 Hz, where i = 1, 2, 3, . . . , to remove electrical noise from the velocity signal (electronic supplementary material, figure SF.4). The energy content of the turbulence was concentrated below 1 kHz (electronic supplementary material, figure SF.5). At all airspeeds considered (10,20,25,30,35,40, 45 and 50 m s −1 ), both axial and transverse turbulence at the centreline were found to be low compared with existing animal flight and small flight vehicle wind tunnels: q u ≤ 0.028% and q v ≤ 0.030%.
Highpass filtering has a large effect on the axial turbulence intensities reported [15]. Previous studies have used varying filters, documented with varying degrees of precision, making it difficult to compare the turbulence performance of existing tunnels. To overcome this complication, we applied five different filter types based on the entries of table 1 (see electronic supplementary material, SM.5 for detail on the filtering and analysis technique). Filtering has only a moderate effect on transverse turbulence intensities, but a significant effect on axial turbulence intensities. This is to be expected since lowfrequency streamwise travelling waves primarily affect axial flow speeds. A comparison of unfiltered data suggests that our facility has a higher travelling wave contribution than the KTH wind tunnel (average central q u = 0.144% at 10 m s -1 compared with less than 0.1% reported at KTH at 10 m s −1 ) [14].
Turbulence intensities were not only low at the centreline, but also throughout the test section (figure 4b). Axial and transverse turbulence intensities were measured at each grid point using the same technique as was used at the centreline (TSI X-Wire Model 1241-20). Both axial and transverse intensities showed little variation across the 4 × 4 grid (1σ < 0.01%) for both speeds at which the uniformity was measured.

Acoustics
A separate mount was built for the acoustic probe (electronic supplementary material, figure SF.6). A forward-swept fibreglass strut extended through a slotted floor panel in the bottom of the test section. The strut was supported by an aluminium frame mounted to the bottom of the test section. Acoustic levels were measured at the end of the strut using a 60 kHz microphone (B&K 0.5 in 4166). To improve accuracy, the signal was sent through a signal conditioner (B&K NEXUS 4-Channel) and a pre-amplifier (B&K 2669), resulting in a hardware-induced 60 Hz highpass filter. The cooling pump was on during acoustic testing, and tape was placed over screw hole indentations in the frame, as well as gaps between movable windows and the test section, as is common practice for acoustic measurements.
Following Johansson [22], a streamlined GRAS custom pinhole nose cone was attached to the front of the microphone before measuring acoustic levels. The main effect of the nose cone is to attenuate the sound pressure level (SPL) at frequencies above 1500 Hz (figure 5a), which are presumed to be a result of small-scale turbulence and therefore not relevant to the acoustic signature of the wind tunnel [22]. The nose cone also introduced a small jump in SPL around 8 kHz, presumably due to resonance in the nose cone. This effect is negligible, because the SPL contribution at these high frequencies (approx. 10 dB) is several orders of magnitude lower than contributions in the 60-1500 Hz range (approx. 70 dB). However, at some frequencies below 1500 Hz, nose-cone resonance increased the acoustic response by up to 8 dB. This amplification is likely to cause the reported overall acoustic levels to be slightly overestimated.
To reduce noise further, pile fabric (synthetic fur) was wrapped around the leading edge of the acoustic turning vanes in corner 1 (electronic supplementary material, figure SF.7). This addition of pile fabric led to a small but measurable decrease in the test section SPL, especially at higher frequencies (figure 5b).
Frequency contributions to sound levels in the test section showed a slow decrease in magnitude with several localized tones ( figure 5a-c). Many of the tones occur in the zero airspeed condition and are therefore attributed to acoustic eigenmodes in the test section. The i-jth transverse eigenfrequency, f ij , of    [20]), 0.5-1 kHz bandpass (Saarland [7,8]), 1 Hz-10 kHz bandpass (Texas A&M [23,24]), 2-200 Hz bandpass (USC [30]), and 20 Hz highpass (KTH [21,22]) filters have been used to report intensity. Note: because our sampling rate was 10 kHz, any cutoff frequencies above 5 kHz had no effect on our reported intensities. Transverse turbulence intensities show much less dependence on cutoff frequency, presumably because of low-frequency streamwise travelling waves. Data from the curves are tabulated in electronic supplementary material, tables ST.1 and ST.2. (b) Axial ((i)(ii)) and transverse ((iii)(iv)) turbulence intensities are measured over a 4 × 4 grid in the test section (black dots indicate grid points). A 20 Hz highpass filter was applied to the data following the methods used to quantify the KTH tunnel [14,15]. Noise levels in the test section are barely audible at low speeds, then increase with airspeed; noise decreases steadily with frequency except for localized tones. (a) SPL is measured at the centreline of the test section, both with a standard nose cone (standard mic) and with a GRAS custom pinhole nose cone mounted on the front. The nose cone was used to attenuate high-frequency turbulence (above 1500 Hz), but it also caused a slight amplification of frequencies below 1500 Hz and a slight jump in SPL around 8000 Hz, presumably due to resonance in the nose-cone cavity. All further trials were performed with the nose cone attached. Overall, acoustic levels in the test section are low; at the lowest airspeeds the wind tunnel is barely audible above the room background noise. The overall sound pressure level (OASPL) rises steadily with airspeed, with unfiltered values ranging from 53.2 dB at 0 m s −1 to 76.4 dB at 20 m s −1 to 94.5 dB at 50 m s −1 (figure 5d). As with turbulence intensity, the frequency cutoff of a highpass filter has a significant effect on the reported OASPL, making it difficult to compare existing wind tunnels. For this reason, we report our results with multiple filters. The first filter is the hardware-induced 60 Hz highpass filter from the signal conditioner. A 180 Hz filter was applied with the same routine that was used for turbulence intensity (electronic supplementary material, SM.5). For a third filtering option, we demonstrate the effect of applying a transfer function inspired by Johansson [22]

Active turbulence grid
The active turbulence grid can increase turbulence intensities up to 16 or 45%, depending on the highpass filter used. To facilitate comparison with other studies using active grids, we applied a grid motion that has been studied extensively [33,34]. For this motion, the vertical vanes stay at a fixed angle to generate constant blockage, and the horizontal vanes oscillate with a given maximum velocity, v max , and amplitude, ϕ H . We studied the axial turbulence intensity, q u , over a range of motion inputs spanning the operating range of the motors driving the vanes ( 6a). The low value occurs when the vanes are fully open and motionless, passively introducing turbulence into the flow. As with the intensities in the empty test section, applying a highpass filter changes the reported intensity. With a 0.1 Hz highpass filter, the range changes to 12.4-44.6%; with a 1 Hz highpass filter, 12.0-40.1%; with a 20 Hz highpass filter, 8.1-16.1%. The significant reduction in maximum turbulence intensity with the 20 Hz highpass filter demonstrates the fact that the grid is primarily injecting turbulence at low frequencies, which do not have time to decay into broadband turbulence since the flow is not likely to be fully developed by the time it reaches the probe [36]. We chose to focus on unfiltered turbulence values for the active grid, because unlike the flow with no grid, where low-frequency perturbations represent travelling waves in the tunnel, the flow with the grid has low-frequency perturbations that have been injected purposefully.
The active grid can also be used to create non-uniform velocity profiles that include values up to ±40% of the mean airspeed. These profiles are created by adjusting the fixed position of the vanes. Changing the vane position can change the blockage and therefore the pressure drop across the active grid, which changes the mean airspeed in the tunnel at a fixed fan speed. When the airspeed changes, it can take approximately 10 s to reach a new equilibrium as the mass of air is accelerated in the tunnel. One way to avoid this delay is by switching between profiles with the same mean airspeed. We used this concept to demonstrate a switch from uniform flow to jet and wake profiles in approximately 1 s (figure 6b). The profiles were measured with an array of 16 wind sensors (Modern Device Rev P hot film; 10 Hz [37]) spaced 5 cm apart horizontally, mounted on the testing sting centred in the test section. Airspeed data for each sensor were collected at 10 Hz and averaged over 30 s.
While the 1 s delay is still longer than the timescales of wingbeats or rapid flight manoeuvres, the profiles can adapt to the changing position of an animal or vehicle. We demonstrated this possibility by adding a 4 mm diameter retroreflective marker to the base of a small quadcopter (Estes Proto X) flying downstream of the active grid. Six infrared cameras (Qualisys Oqus 7 plus; 1000 Hz; 3 MP) tracked the position of the quadcopter, and the position data were relayed to the active grid control system. All horizontal vanes were held fully open, and all vertical vanes were closed except for the one closest to the lateral position of the quadcopter. The result is a jet with constant blockage that follows the position of the quadcopter (figure 6c). While the quadcopter was moving too quickly for the jet to match its position in real time, the successful tracking shows that the active grid could send gust perturbations using a closed-loop controller based on the positions/orientations of animals/vehicles in the test section.

Conclusion
The aeroacoustic performance of the new wind tunnel at Stanford is comparable to existing state-of-theart wind tunnels. We obtained streamwise turbulence levels less than or equal to 0.030% throughout the  Here, the concept is demonstrated by opening one vertical vane upstream of a small quadcopter with a marker tracked by a motion capture system. Instantaneous airspeed profiles are shown at two time steps (t 1 and t 2 ) corresponding with two frames from the video (electronic supplementary material, video SV.1; selected frames processed with Adobe Photoshop: Max Levels to 80 to improve contrast).
test section at 10 m s −1 , compared with less than 0.04% at KTH [21] using the same filtering technique, demonstrating comparable turbulence levels. While we are unable to recreate the filtering used at Lund [27][28][29], their use of 1024 samples taken at 1 kHz suggests an effective highpass cutoff frequency of approximately 1 Hz. When using a 1 Hz highpass frequency, our streamwise turbulence intensities were approximately 0.05%, comparable to the values reported at Lund. These values are also comparable to those at Texas A&M, where 1 Hz was also used as a lower cutoff frequency [23,24]. The wind tunnel at KTH used similar acoustic treatment and testing, allowing a straightforward comparison of our sound levels. Noise levels of 69 dB were reported at 35 m s −1 in the KTH tunnel, which is lower than our reported 84.4 dB [22]. The difference with KTH could be somewhat exaggerated because of the transfer function used during acoustic measurements at KTH. When we apply a transfer function inspired by theirs, our reported levels decrease by approximately 6% (electronic supplementary material, table ST.3). Even with this filter the KTH tunnel is more quiet in comparison; regardless, both tunnels are barely audible in the test section at animal flight airspeeds.
Our new wind tunnel has comparatively low turbulence, low flow angularity, high flow uniformity and high temperature uniformity. Two additional features make our animal flight wind tunnel unique: the ability to switch from open to closed jet configuration and the ability to tune turbulence intensity with an active grid. Existing tunnels are uniquely suited for studying different aspects of animal flight. With the ability to vary climb angle, the tunnel at Lund [27,29] is ideal for studying migration and ecology. With the ability to vary pressure, the tunnel at Western Ontario [25,26] is ideal for studying highaltitude flight and bioclimatology. The abilities to vary turbulence and perform open-jet experiments with biplanar fluoroscopy make the new wind tunnel at Stanford ideal for studying comparative biomechanics.
Ethics. The new wind tunnel facility complies with all state and federal regulations governing the humane treatment of research animals. The Administrative Panel on Laboratory Animal Care at Stanford endorsed the facility after commissioning, and they perform biannual inspections to ensure the facility continues to comply with all protocols. The laboratory animal care programme at Stanford is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care.