Acoustic sources identified using a microphone phased array during a rocket engine test

Abstract

A new phased array of microphones, suitable for the harsh environment of a rocket launch, was built and tested during a static firing of an RS-25 engine. It uses 70 piezo-resistive, dynamic pressure sensors, optimally distributed on a 10.5-ft diameter open frame dome structure and has a 200-ft long cable bundle to carry the signals to a weather-protected cabinet containing the data systems. The test stand was imaged using an infra-red camera and a visible wavelength camera, and the beamformed noise maps were superimposed on the photographs. The first-time data from a full-scale engine test stand showed that the plume deflector at the bottom of the engine was the primary noise source. The openings of the test stand around the nozzle exit were also found to be noise sources particularly at higher frequencies. The final goal is to utilize the array during NASA’s Artemis-II launch at Kennedy Space Center.

Keywords

Phased array rocket noise beamform engine test stand launch acoustics

Introduction

Every part of a launch vehicle, launch pad, and ground operation equipment is subject to high acoustic load generated during lift-off.¹ The acoustic load is a major contributor to the vibro-acoustics environment to which every payload, vehicle structure, propellant storage and handling devices, and electronics and navigational component must be designed, tested and certified. Even a single decibel reduction of the acoustic levels translates into a sizable reduction of acoustic loadings, certification costs, operation costs and even vehicle weight. The same is true for every payload that the vehicle carries. Therefore, lowering of the acoustic level via various mitigation schemes is an important aspect of a launch pad design. The first step is the identification of the sources responsible for noise generation. Typically, single microphones are placed at different locations on a launch pad and on the vehicle to measure acoustic fluctuations. Such microphones, however, are unable to determine the locations of the noise sources. Single microphones provide a measure of the absolute level, leaving the cause of noise generation to speculation. In contrast, a phased array of microphones directly identifies the locations and relative strengths of the noise sources and is therefore capable of providing significant insights into the underlying causal effects. Effectively a phased array acts as a “sound camera” capable of finding the highest sound generating regions.

Another advantage of a phased array is its ability to “see” through the smoke, water vapor and dust cloud generated during a rocket launch. Such a smoke cloud might restrict the view of the launch pad, but sound waves will still be detectable. This allows for the identification of the noise sources at all phases of the launch from hold-down to elevation above the launch tower. A third advantage of a phased array is its ability to identify regions of the pad affected by the splashing of the rocket plumes as the vehicle ascends from the hold-down state. The zones affected by the plume impingement are typically locations of louder noise sources and are easily identifiable by a phased array.^2,3 The plume impingement zones can create thermal and structural damage to the launch platform. To mitigate such thermal and structural damage and to reduce the acoustic environment many launch pads for heavy-lift rocket vehicles employ extensive water suppression systems. Knowledge of the distribution of the acoustic sources at every stage of the launch provides actionable information to optimize the water suppression system. An important goal of the present effort is to determine the plume splashing regions on the launch pad via acoustic source identification.

The benefits of phased array technology were demonstrated earlier in static-burn tests,² static engine tests, as well as the first launch of the Northrop Grumman’s Antares vehicle.³ More recently, a phased array proved to be a valuable tool in the redesign of launch pads for Ariane⁴ and Vega⁵ vehicles. There exist many limitations of the current understanding of the acoustic sources during a rocket launch (Lubert et al.).⁶ A phased array can be a very useful tool in filling these voids.

The present array is a new design based on various lessons learnt from past applications.^2,3 The new design utilizes an open frame design to better sustain the wind and blast loads, as well as piezo-resistive microphones which are better suited for the salty air of a typical launch site. The array was designed, fabricated, and tested (using speaker sources) at NASA Ames Research Center.⁷ It was then transported to NASA Stennis Space Center (SSC) to participate in the hot fire test of an RS-25 engine. The goals were to: (a) determine the beamforming ability of the array in realistic environment; (b) determine tolerance to the weather elements, and (c) gain experience with the logistics of operation, such as lifting and mounting, disassembly, and reassembly of hardware, and optimizing procedure for data collection. Data collected during the hot-fire test provided insights into the noise sources around a full-scale engine test stand. This test is a part of a planned series of Validation and Verification (V&V) tests working towards the final deployment during the launch of Artemis-II.

Array hardware

The custom-built, one-of-a-kind, phased array uses 70 dynamic pressure (Kulite) sensors laid out in a predetermined arrangement (Figure 1). The goal of the layout is to maximize the array resolution over a wide range of frequencies, and to reduce the side lobes (fake noise sources) in the resulting sound maps. To design the layout, microphone locations were first determined on a two-dimensional base circle with a 62.7 inch radius. Subsequently, they were raised on a hemispherical surface with a radius of 88.67 inch. Inside the base circle, 24 of the microphones were positioned on logarithmic spirals while the rest were randomly scattered. The resulting pattern of microphones produces a narrow point-spread function (psf). Figure 2 shows the ideal array response at two different frequencies using two different methods of analysis, conventional and functional beamforming.

For conventional beamforming the minimum resolvable detail is due to the diffraction-limit of the array, i.e., when the first diffraction minimum of the image of one source point coincides with the maximum of another. For a circular aperture of diameter d, the Rayleigh resolution in radians θ_R is as follows

\sin θ_{R} = 1.22 \frac{λ}{d}

(1)

Conventional beamforming created large 3 dB spot sizes of 16° and 3.17° at 200 Hz and 1 kHz, respectively. The spot sizes are close to the predicted value from the Rayleigh criteria (Equation (1)). Functional beamforming processing created spot sizes smaller by a factor of ∼4: 4.19° and 0.84° for the same two frequencies. The side-lobe rejection also improved significantly. These beamforming models will be described in Section Phased array data processing. A goal of the present work was to determine if such reductions were attainable in actual tests.

Figure 3

shows a photograph of the newly built phased array with its three primary components: (a) the hemispherical dome shaped lattice structure that holds 70 dynamic pressure sensors, (b) a 200-ft long cable bundle, and (c) an air-conditioned cabinet to hold all electronic components and computers. The cylinder (also known as the canister) at the center of the dome structure holds one visible wavelength video camera, and a long infra-red wavelength (2 micron–10 micron meter) video camera (Figure 4(b)

). A glass and a Germanium window, placed on the front cap, provide clear view of the launch pad and the vehicle in the visible and infra-red wavelengths, respectively. Additionally, amplifiers for all 70 microphones are also housed inside the canister (Figure 4(a)). To monitor the temperature of the amplifier rack a resistance temperature detector (RTD) is attached to the amplifier housing. Additionally, a three-axis accelerometer is placed on the mounting rod for the amplifier rack to monitor the vibration levels expected during the launch. Finally, a ½-inch diameter tube is placed close to the amplifier housing to bring in dry air to cool the amplifiers, and to maintain a low humidity environment at the interior of the canister. The array is expected to be mounted close to a launch pad weeks before the actual launch with little access. The cooling air will limit the temperature and humidity rise inside the canister.

A more detailed description of the array hardware can be found in Ref. 7. The cable bundle passes direct current (DC) power and signals between the array dome and the data acquisition cabinet. The 200-ft long cable bundle is suitable for raising and mounting the array on nearby tall structures (such as lightening towers, water tanks or any other suitable structure) so that a clear view of the top surface of the launch platform can be visible to the array cameras. The data acquisition cabinet holds a 24-bit primary and 16-bit backup data acquisition systems to digitize the microphone and the accelerometer signals. Microphone data was collected at a rate of 20,480 samples/second and IR images were collected at 12.5 frames/second. The cabinet is made of steel for protection from the launch environment. It also has an air-conditioning unit to cool all electronics during the expected multiple weeks of continuous operation in the hot sun.

The hemispherical shape of the array dome and the triangular lattice frames are known to produce structural stiffness while minimizing the overall weight. The lattice structure was inspired by a similar dome shaped array built by Burnside and Horne⁸ for use in a wind tunnel. Acoustic phased arrays used for previous rocket launch studies^2,3 had solid cross-section areas, which caused them to catch the overpressure and the acoustic waves from rocket launches creating large vibrational responses. The present design allows for minimization of the wind and acoustic loadings and is expected to have much less vibration response. This was confirmed from multiple load analyses and calculations of the modal response of the frame.

The acoustic sensors on the phased array are piezo-resistive dynamic pressure sensors manufactured by Kulite Corporation. There are three different types of sensors on the phased array: larger 0.15 inch diameter XTEH-10L-190S-25A, and XTL-190-25A, and miniature 0.072-inch diameter XCL-072-25A. All are of absolute pressure gauges with 25 pounds per square inch absolute (psia) range and produce full-scale output of 0–5 V. The sensors were tested in a jet flow facility to ensure phase matching. The microphones were approximately mounted based on the designed location. The exact locations were established after the completion of the dome assembly and mounting. A photogrammetry-based technique using a V-STARS instrument (of Geodetic Systems) was used for this purpose. A single camera was used to take many photographs of the array, reference points, and guide poles and markers mounted on and around the array. Later on, a proprietary software was used to extract the microphone coordinates to an accuracy less than ±0.01 inch.

Test setup

The RS-25 engine static firing provided a good opportunity to test the array in a realistic environment. The engine was tested in the A-1 Test Stand of NASA Stennis Space Center. It is a single-position, vertical-firing facility where an engine is held above the 5th deck (Figure 5(a)) and is supplied with liquefied Hydrogen and Oxygen. The plume out of the nozzle (Figure 5(b))

passes through an “aspirator opening” on the 4th deck and impinges on a J-deflector. The flame deflector is made up of 21 stacked angular segments – or water boxes – each drilled with a pattern of holes to direct water as needed to cool the stand’s flame deflector. As much as 170,000 gallons of water per minute at 225 pounds per square inch pressure is supplied to the A-1 Test Stand during a test. The water is primarily used to cool the flame deflector and to keep it undamaged as it redirects engine exhaust.⁹ The steam and the plume mixture flow out horizontally over an open channel flume.

The array was placed on the west side, 460 ft away from the test stand (Figure 6

). The distance is comparable to the proposed separation in the Artemis application. A 50-ft tall scaffold structure was built, and the array dome was hoisted and mounted atop the scaffold. The dome was oriented to face the test stand. From the 50 ft height the cameras inside the array had unobstructed view of the entire test stand and a part of the flume. A part of the cable bundle was routed along the side of the scaffold and the data acquisition cabinet was placed at the base. The cabinet was supplied with electrical power, cooling air, and Ethernet connection for remote operation of the data acquisition computers. In addition, an engine start-stop signal was supplied for simultaneous recording with the microphone data.

Phased array data processing

Fundamentally, a phased array detects the curvature of the acoustic waves to determine the angular location of the noise sources. The information is extracted by calculating cross-spectrum of pressure fluctuations between every possible pair of microphones on the array. In addition to the desired sound signals p_m,s(t) the time-signal generated by the microphones p_m,t(t) carries two different sources of noise. For the piezo-resistive sensors used in the present array the electronic noise was high. The equivalent acoustic pressure level was around 105 dB (the dB scale uses the standard 2.9 × 10⁻⁹ psi as the reference pressure). Let p_m,e(t) be the pressure fluctuations when the electronic noise is multiplied by the sensor calibration constants. Additionally, for launch acoustics application there are host of weaker spurious sources that create background fluctuations p_m,b(t). For example, the pressure fluctuations from the strong wind blowing on the array can be 100 dB or higher. There are weaker sound sources such as valve and gas flow generated noise, and reflections of various kinds that contaminate the measured pressure fluctuations p_m(t). Let m = 1, 2, …M be indices for microphones, then the following holds for the total fluctuations measured by the mth microphone

p_{m, t} (t) = p_{m, s} (t) + p_{m, b} (t) + p_{m, e} (t)

(2)

Assuming that $p_{m, s} (t), p_{m, b} (t), a n d p_{m, e} (t)$ are mutually uncorrelated and stationary, the following relationship holds for the cross spectrum between a microphone pair m and m^′ ¹⁰

G_{m m^{'}, t} = 〈 {p_{m, t}}^{*} p_{m^{'}, t} 〉 = G_{m m^{'}, s} + G_{m m^{'}, b} + G_{m m^{'}, e}

(3)

The contribution from the electronic noise can be subtracted by taking a set of data p_m,e(t) at an indoor condition with the sound sources turned off, and then by calculating the noise-induced cross spectrum $G_{m m^{'}, e}$ for every pair of microphones.

G_{m m^{'}, e} = 〈 {p_{m, e}}^{*} p_{m^{'}, e} 〉

(4)

Direct subtraction of the noise cross-spectra from the total measured was found to be problematic for the beamformed map since the phase spectrum became contaminated by the random noise of the former spectra. Therefore, only the magnitude of the noise-induced cross-spectrum was subtracted from that of the total measured, leaving the phase part intact

G_{m m^{'}, t} (f) = | G_{m m^{'}, t} | e^{- i {f Φ}_{m m^{'}}}

G_{m m^{'}, s b} (f) = (| G_{m m^{'}, t} | - | G_{m m^{'}, e} |) e^{- i {f Φ}_{m m^{'}}}

(5)

The resulting $G_{m m^{'}, s b} (f)$ is the cross-spectrum from the combination of the desired sound sources and the background noise.

To further reduce the influence of the background noise for the Functional beamforming scheme, the eigenvalues of the $G_{m m^{'}, s b}$ matrix was inspected via an eigenvalue decomposition (dropping the mm′ part of the subscript):

G_{s b} = U Σ U^{*} = \sum_{m = 1}^{M} σ_{m} u_{m} u_{m}^{*},

(6)

where, U is a unitary matrix whose columns are the eigenvectors (u₁, u₂,…. u_M) of G_sb and Σ is a diagonal matrix whose diagonal elements are real valued eigenvalues (σ₁, σ₂,… σ_M) of G_sb. Figure 7 shows the distribution of the eigenvalues obtained from the present test at frequencies of 510 hz and 1025 Hz. Note that most eigenvalues are very small even for the distributed nature of the noise sources expected in the engine test. Because the intention was to investigate the peak noise sources within a given dB range, a threshold level can be used based on the highest eigenvalue σ_max. For the present application the intention was to find the sources responsible for creating at most the top 10 dB range of pressure fluctuations. Therefore, all eigenvalues below the following criteria were considered due to the background noise sources and were equated to zero.

σ_{m} < \frac{σ_{\max}}{40} \to 0 .

(7)

For the two frequencies 510 Hz and 1025 Hz shown in Figure 7, respectively, 6 and 12 eigenvalues were retained – rest were equated to zero. These eigenvalues, σ_s, s = 1,2…S are due to the desired sound sources and were used to reconstruct the cross-spectral matrix G_s free of the background noise

G_{s} = U Σ_{S} U^{*},

(8)

where, Σ_s is a diagonal matrix composed of the non-zero eigenvalues, rest equated to zero. Note that this procedure of zeroing out the weaker eigenvalues was used only with the Functional beamforming. For conventional and Spectral Element Method (SEM) schemes the noise subtracted cross spectra of equation (5) was used.

Conventional beamforming (CB)

To create maps of the noise sources, first, the region where such sources are expected to be present is divided into a set of grid points. Typically, a planar phased array is insensitive in the depth direction (i.e., along the normal to the array plane). The current hemispherical shape is expected to provide improvements, yet the large separation distance between the pad and the array location is expected to make such improvements insignificant. Therefore, a two-dimensional set of grid points (also called interrogation points) is set up. Let j = 1, 2, 3…N be the interrogation points. The radial distance r_jm, from an interrogation point to an individual microphone, determines the phase shift and the relative amplitude measured by the microphone. The steering vector is a column matrix defined to incorporate these properties

w_{j, m} = \frac{r_{j m}}{r_{j c}} e^{- i k r_{j m}}

(9)

Here r_jc is the distance of each interrogation point to the center of the array. Note that:

w_{j}^{†} w_{j} = 1

The elements of the cross-spectral matrix of equation (5) above are summed up, preceded by a phase adjustment by the steering vectors, to interrogate the individual grid points. These two steps are combined in a matrix manipulation which leads to the conventional beamform map. In the following $G_{m, m^{'}}$ is a shorter representation of the matrix $G_{m m^{'}, s b}$ in equation (5).

b_{j, c b} (f) = {w_{j, m}}^{†} G_{m, m^{'}} w_{j, m^{'}}

(10)

Zeroing out the diagonal elements of $G_{m, m^{'}}$ improves the beamformed map.

Modified functional beamforming (FB)

This method proposed by Dougherty^11,12 uses the eigenvalue and eigenvector decomposition of the cross-spectral matrix described earlier. For the present case, the electronic and background noise is removed from the cross-spectrum. Then the eigenvalue decomposition of this cross-spectrum (Equation (8)) is used. Note that the functional beamforming proposed by Dougherty did not account for noise reduction via zeroing out of the smaller eigenvalues. That idea was a part of the Orthogonal beamforming scheme.¹³ Since the current method is a mix of the two, it is referred to as either modified Functional or Functional-Orthogonal scheme. The central idea of functional beamforming is to reduce the spot size (point-spread function) and the alias points (sidelobes – fake noise sources) by reducing the magnitude of the cross-spectral matrix via an exponent 1/ν, where, ν > 1, before adjusting the phase and summing up the levels. This is followed by raising the beamformed levels back to their original levels. The reductions in the magnitude of the cross-spectral matrix can be accomplished by operating on the non-zero eigenvalues: ${σ_{s}}^{\frac{1}{ν}}$ . The expression for the functional beamforming is the following.

b_{j, f b} (f) = {[{w_{j, m}}^{†} {\sum_{m = 1}^{M} {σ_{s}}^{\frac{1}{ν}} u_{m} u_{m}^{*}} w_{j, m^{'}}]}^{ν}

(11)

The exponent ν is to be selected by the user. For a completely noise free data the dynamic range (reduction of the side lobes) and the reduction in the spot size (i.e., improvement in the resolution of the beamformed map) is directly related to ν: the higher the ν the smaller is the spot size, and wider is the dynamic range. In practice, the residual noise level and the lack of convergence due to the limited time extent of the microphone data plateau the spot size to a fixed level. Typically, ν in the range of 10–100 is used in the literature. Marino-Martinez et al.¹⁴ found an improvement of the array resolution by a factor of 6 compared to the conventional beamformed value by applying ν = 100 to clean data from an airframe noise test. For the present work ν = 20 is used. One observation was that the absolute beamformed levels from FB were much lower than that found from the conventional beamforming. As ν was increased and more of the lower eigenvalues were neglected, the discrepancies grew larger. Therefore, the peak beamformed levels from the FB scheme were adjusted to those calculated from the CB scheme. In other words, calculations for functional beamforming needed to be preceded by the conventional one.

Direct spectral estimate – spectral element method (SEM)

The resolving capability and the dynamic range available by a phased array of a given size and given number of microphones are limited by the width of the point spread function (psf). As discussed earlier the psf and the sidelobe make a point source of sound to appear over a wide spatial extent and at pseudo–locations, respectively. Functional beamforming does make significant improvements of the resolvability, but the best results can be obtained if the psf can be deconvolved out of the beamformed maps. There are a host of different schemes available¹⁴ towards that end. An important limitation of many such schemes is that they break down a distributed source into small points which makes the results appear unphysical. The SEM also suffers from this limitation but to a lesser extent.¹⁴ Like other techniques, SEM also assumes that the sound sources are made up of uncorrelated monopoles. However, instead of making phase adjustment via the steering vectors, the method attempts to find the strength of the monopoles directly from the measured cross-spectral matrix, G_sb. Like before, the source region is divided into N_s interrogation points (also called grid points) and the steering vectors are used to calculate a model of cross-spectrum between two microphones m and m^′. The unknown in the model is the strength of the monopole sources α_j² at each of the interrogation point j = 1, 2 … N_s. The model cross-spectrum between microphones m and m^′ is the following

G_{m m^{'}}^{\mod} = \sum_{j = 1}^{N} w_{j, m} α_{j}^{2} w_{j, m^{'}}^{*} = \sum_{j = 1}^{N} (\frac{r_{j m}}{r_{j c}} e^{- i k r_{j m}}) α_{j}^{2} {(\frac{r_{j m^{'}}}{r_{j c}} e^{- i k r_{j m^{'}}})}^{*}

(12)

The positive values of α_j² assure elimination of unphysical negative source strengths. In the method proposed by Blacondon & Élias¹⁵ the source distribution is determined via minimization of the error between the measured and the modeled cross-spectra using an iterative least-square minimization scheme.

E (α_{j}, f) = \sum_{m, m^{'} = 1}^{J} {| G_{m m^{'}} - \sum_{j = 1}^{N} w_{j, m} α_{j}^{2} w_{j, m^{'}}^{*} |}^{2}

(13)

An advantage of the SEM is that the diagonal terms of the measured cross-spectrum can be avoided in the minimization scheme. Without the diagonal terms, the measured cross-spectral matrix G_sb contains (M²-M)/2 number of independent, complex values. Counting the imaginary and the real numbers separately, the total number of independent variables to be minimized becomes J.

J = M^{2} - M

(14)

Similar to a procedure followed by Casalino et al.,¹⁶ the calculations were carried out in Matlab® using routine lsqnonneg that uses a non-negative least square process.

Matlab^® implementation

The above beamforming schemes were implemented in the commercially available Matlab® platform. Once the cross-spectral matrix (Equation (5)) was calculated, the rest of the beamforming operation for conventional, functional and SEM methods completed quickly. To facilitate direct identification of the noise sources, the interrogation grid was created over a photograph of the region of interest captured via either the visible band or the infra-red band video camera. The photographed region was divided into a uniformly spaced grid. To select the number of grid points, Rayleigh criterion was applied to the highest frequency of interest. Typically, the grid spacing obtained by Rayleigh criterion was refined by a factor of 10. The beamformed colored maps were superimposed on a video frame from either the visible band or the IR-band camera. Implementation of SEM typically required fewer grid points. To limit the computing time to a reasonable duration, the maximum number of iterations in lsqnonneg routine was limited to 500.

Results

Validation of the phased array operation via identification of a single speaker source

Correct registration of the beamformed map and the video frames required a calibration process. The look-angle of the video camera changed slightly every time the array was rebuilt. To correct for such a small but important change, a speaker was placed at various locations in the camera field of view, and the video image was rotated in pitch, yaw and/or roll directions for correct superposition of the beamformed maps. These are the first series of test conducted days before the engine burn – just after lifting and securing the array on the scaffold structure. A high intensity speaker (Long Range Acoustic Device, LRAD) was mounted on the bed of a truck and was placed at multiple locations around the test site (Figure 8(a)), and on the test stand. The cameras inside the array canister were used to photograph the speaker location (Figure 8(b)) and the acoustic data was recorded. The beamformed map superimposed on the camera images were used to determine the slight tilt angles of the camera with respect to the array axis and to verify the camera calibration. The speaker produced sharp tones mostly >1 kHz frequencies. Figure 9 shows sample beamformed maps at 3 kHz superimposed on a frame of the IR camera for two different speaker positions. Similar maps were created for all other speaker locations. All maps were created using the Functional-Orthogonal scheme. The look angles were calculated from the center of the image and adjusted for the camera tilt angles. The color scale on the right shows the top 10 dB range of the beamformed map used for plotting. Correct juxtaposition of the beamformed map on the speaker location verified the calibration and brought confidence on the accuracy of noise maps created by the array.

RS-25 static burn

The phased array was used during the hot-fire test of a redesigned RS-25 engine on Dec 14, 2022. The engine would be used in the future Artemis missions. Microphone data and camera images were collected for the entire test duration of 209 s. The array hardware was setup 5 days prior to the actual burn and was exposed to high wind, multiple thunderstorms, and wind driven rain for four consecutive days. The exposure to the weather element was also part of the validation test. Two of the 70 microphones showed unstable signal due to exposure to moisture and were unusable. The rest of the instruments worked without any issue.

Figure 10 shows a sample time series from microphone number 1 along with the engine start and stop signal. An examination of Figure 10 shows a signal level higher than the background noise floor but significantly weaker than the expected level from an Artemis launch. The lower level is expected since only one engine - out of 4 liquid engines and two solid boosters – was used in this test. It also provides assurance that the sensors pressure range would be adequate for the full launch vehicle event. Nonetheless, the longer duration signal provided good data for beamforming. The time series was truncated and data from the duration of the test was used to calculate auto- and cross-spectra.

Figure 11 shows an auto and a cross-spectra and the impact of background subtraction. The blue lines of Figure 11 are spectra from data collected before the engine burn and represent the background noise. The green lines are the spectra from data collected during the burn and represent as measured sound spectra. The black lines are as-measured sound spectra minus the spectra of background noise. The black lines mostly hug the green lines except at the high frequency end. This shows that the background subtraction mostly influenced the high frequency (>2.5 kHz) part of auto and cross-spectrum. An examination of the spectral shapes shows that the very low frequency part: 10 Hz ≤ f ≤ 120 Hz has significantly higher level of energy compared to all other frequencies. The test stand generated high level of low frequency noise despite the water deluge. Note that this part of the spectra might also be affected by the wind noise over the Kulite sensors. Another interesting observation is that the spectrum levels fall very quickly beyond ∼3 kHz frequency. It is suspected that the high amount of water injection attenuated the high frequency part of the acoustic spectrum but was less efficient in attenuating the very low frequency noise.

Noise sources identified through third octave beamform maps

Figure 12 shows the beamformed source maps at six different frequencies obtained using Functional-Orthogonal method. The frequencies are the third-octave center frequencies. The frequency range covers a wide band from 32 Hz to 2048 Hz demonstrating the impressive ability of the relatively small sized array to provide information over all frequencies of interest. Similar maps at two other frequencies are shown below in Figure 13. Once again only the top 10 dB range of the beamformed levels are plotted on one of the IR camera images collected during the RS-25 burn. Note that a higher value of ν = 100 was used to reduce the spot size in the lower two frequencies of 32 Hz and 64 Hz. An examination of Figure 12 shows that fundamentally there are two noise sources: the flame deflector and the gap in decks 4 and 5 of the test stand. A compact region on the ground in front of the test stand also shows up as noise source in some of the noise maps, but this is interpreted as a reflection of the actual source on the test stand. The sound waves generated from the test stand have two different paths to reach the array: via direct propagation and via reflection on the ground. The solid ground acts as an acoustic mirror with varying reflectivity. The location of the sound reflection – marked as the ground reflection - rightfully shows up in the noise maps and ought to be ignored for source interpretation.

The plume out of the aspirator deck plunges on the deflector making the deflector the primary source at all frequencies. At the lower frequency where the acoustic sources are expected to be wider the entire deflector brightens up as the source. As noted earlier the large amount of water injection significantly attenuates the high frequency sound. Sound maps at 1 kHz and above shows the top part of the deflector to be primary source. Noise form the plume shear layer, yet to be quenched by the water flow, is believed to be the source that radiates out from the top of the deflector. The same is believed to be the cause for openings of the 4th and the 5th deck to appear as the high frequency noise sources. The shear layer from the nozzle exit and various vibration induced noises from the engine emanate out of these openings making them the noise sources.

Comparison between beamforming schemes

Figure 13 shows a comparative study of the spot sizes for two different third-octave center frequencies using the three different beamforming schemes. The large spot size of the CB map of Figures 13(a) and (d) followed the expected Rayleigh criteria. The FB map of Figures 13(b) and (e) used ν = 20 and is found to reduce the spot size. However, the reduction was found to be frequency dependent. For the lower frequency (203 Hz) the reduction was by a factor of 3, while the reduction for the higher frequency (1625 Hz) is perhaps by a factor of 1.5. The CB has lower resolution (larger spot size) for the lower frequencies and the improvement of resolution via the FB scheme is particularly encouraging. The SEM maps of Figures 13(c) and (f) farther reduced the beamformed maps to a distribution of very small spots. This behavior was also observed in the past for other deconvolution schemes.

Summary and conclusion

A new, readily deployable, phased array of microphones built at NASA Ames Research Center specifically for the harsh environment encountered in launch pads of rocket vehicles shows progress in the state of the art for beamforming mapping. The 70-microphone array is built on a 10.5 ft diameter open frame dome structure, which was light yet robust enough to sustain the high wind load of typical seaside launch pads and the blast and acoustic loads from the launch. A 200-ft long cable bundle carries the microphone signals to a weather-protected cabinet containing the data systems and allows for the placement of the array on tall structures. The array was tested during an RS-25 engine test at the Fred Haise A-1 test stand of NASA Stennis Space Center. The engine was held vertically, and the hot plume came down on a deflector which was supplied with heavy water flow for thermal protection. Subsequently the plume and steam flowed horizontally on an open trench. The array was elevated 50 ft above ground on scaffold structure, which was erected 460 ft away from the test stand.

At first a set of validation tests was conducted using a high intensity single speaker positioned at many locations on and around the test stand. Correct superimposition of the beamformed maps on the speaker location brought confidence on the accuracy of the noise maps. Before the actual engine test, all the array hardware except for two of the microphones survived through 4 days of rain, thunderstorms, and high wind. This proved the feasibility of using the array at rocket launch sites where prolonged exposure to the outdoor elements are expected. Good quality microphone data was collected from the test.

The data processing was conducted using three different beamforming schemes: conventional, Functional-Orthogonal and SEM with a goal of improving the resolution of the beamformed map. To reduce the impact of the electronic noise, first a background subtraction process was employed on the calculated cross-spectrum. Additionally, the weaker eigenvalues of the cross-spectrum were zeroed out to further reduce the impact of noise. Functional-Orthogonal beamforming was found to reduce the spot size by a factor of three, at the desirable low frequency end of spectrum.

The relatively small sized array was found to provide interesting insights into the sound sources on the test stand over a wide range of frequencies 34 Hz ≤ f ≤ 2 kHz. It was found that the deflector at the bottom of the test stand was the primary noise source. For the low frequencies the entire deflector is the noise source. Impingement of the engine plume on the deflector is the cause of noise generation. A huge amount of water injection particularly attenuated the high frequency part of the spectra. The openings in the nozzle and aspirator decks were found to be the other noise sources. The noise generated from the plume shear layer and engine vibrations were radiated out of these openings. Interestingly, unlike the expectations from standard models,¹ the flume through which the plume and the steam mixture flowed downstream was not found to contribute towards the top 10 dB of noise sources. We believe that the noise maps presented in this report are first such measurements from a full-scale rocket engine test stand. Finally, the present work validate the array system design for deployment in a forthcoming Artemis launch.

Footnotes

Acknowledgements

The authors are thankful to NASA Engineering and Safety Center (NESC) for supporting the building of the hardware and performing the verification and validation tests. Help from many technicians and support staff of NASA SSC for lifting and mounting of the array played a vital role. We also acknowledge help from Mr. Mike Smiles, NESC chief engineer at SSC in coordinating the Stennis activities. Mr. David Keil of Jacobs, Mr. Matt McKay of NASA Ames, and Mr. Rene Formoso and Mr. Malay Shah of NASA Kennedy Space Center participated in various preparatory and data collection stages of the operation. The first author had the privilege of knowing prof. Ahuja for many years. It's a joy to be able to contribute to this edition of IJA honoring prof. Ahuja.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Jayanta Panda

Appendix

References

Eldred

Jones

Jr . Acoustic load generated by the propulsion system. NASA SP-8072, 1971.

Panda

Mosher

. Microphone phased array to identify liftoff noise sources in model-scale tests. J Spacecr Rockets 2013; 50(5): 1002–1012. DOI: 10.2514/1.A32433

Panda

Mosher

Porter

. Noise source identification during rocket engine test firings and a rocket launch. J Spacecr Rockets 2014; 51(4): 1761–1772. DOI: 10.2514/1.A32863

Gély

Elias

Bresson

, et al. Reduction of supersonic jet noise. Application to the Ariane 5 launch vehicle. In: 6th AIAA/CEAS Aeracoustics Conference, AIAA paper 2000-2026, 2000.

Mortain

Clero

. Full-scale acoustic source identification on Vega Launch pad at lift-off. In: 26th International Congress on Sound & Viberation, July 2019.

Lubert

Gee

Tsutsumi

. Supersonic jet noise from launch vehicles: 50 years since NASA SP-8072. J Acoust Soc Am 2022; 151(752): 752. DOI: 10.1121/10.0009160

Panda

Nguyen

Keil

, et al A microphone phased array for launch acoustics application. In: Presented at AIAA SciTech Foru, 23–27 Jan 2023, National Harbor, MD: AIAA paper 2023-0790.

Burnside

Horne

. Development of instrumentation and methods for time-domain measurement of rotor-type noise sources in a hard wall wind tunnel. In: 2020 presented at the VFS aeromechanics for advanced vertical flight testing meeting, San Jose, CA, Jan 21–23.

https://www.nasa.gov/sites/default/files/atoms/files/fred_haise_test_stand_v1.pdf

10.

Bahr

Horne

. Subspace-based background subtraction applied to aeroacoustic wind tunnel testing. Int J Aeroacoust 2017; 16(4–5): 299–325.

11.

Dougherty

. Functional Beamforming. In: Berlin Beamforming Conference, BeBeC 2014-1, 2014.

12.

Dougherty

. Functional Beamforming for aeroacoustic source distributions. AIAA Paper 2014-3066, 2014.

13.

Sarradj

. A fast signal subspace approach for the determination of absolute levels from phased microphone array measurements. J Sound Vib 2010; 329: 1553–1569.

14.

Merino-Martinez

Sijtsma

Snellen

, et al. A review of acoustic imaging methods using phased microphone arrays. CEAS Aeronaut J 2019; 10: 197–230.

15.

Blacondon

Élias

. Level estimation of extended acoustic sources using a parametric method. J Aircraft 2004; 41(6): 1360–2136.

16.

Casalino

Santini

Genito

, et al. Rocket noise sources localization through a tailored beam-forming technique. AIAA J 2012; 50(10): 2146–2158. DOI: 10.2514/1.J051479

Acoustic sources identified using a microphone phased array during a rocket engine test

Abstract

Keywords

Introduction

Array hardware

Test setup

Phased array data processing

Conventional beamforming (CB)

Modified functional beamforming (FB)

Direct spectral estimate – spectral element method (SEM)

Matlab® implementation

Results

Validation of the phased array operation via identification of a single speaker source

RS-25 static burn

Noise sources identified through third octave beamform maps

Comparison between beamforming schemes

Summary and conclusion

Footnotes

Acknowledgements

Declaration of conflicting interests

Funding

ORCID iD

Appendix

References

Matlab^® implementation