Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Acceleration profiles and processing methods for parabolic flight

Acceleration profiles and processing methods for parabolic flight www.nature.com/npjmgrav BRIEF COMMUNICATION OPEN Acceleration profiles and processing methods for parabolic flight 1,2 1 1 3 2 1 Christopher E. Carr , Noelle C. Bryan , Kendall N. Saboda , Srinivasa A. Bhattaru , Gary Ruvkun and Maria T. Zuber Parabolic flights provide cost-effective, time-limited access to “weightless” or reduced gravity conditions, facilitating research and validation activities that complement infrequent and costly access to space. Although parabolic flights have been conducted for decades, reference acceleration profiles and processing methods are not widely available. Here we present a solution for collecting, analyzing, and classifying the altered gravity environments experienced during parabolic flights, which we validated during a Boeing 727-200F flight with 20 parabolas. All data and analysis code are freely available. Our solution can be integrated with diverse experimental designs, does not depend upon accelerometer orientation, and allows unsupervised classification of all phases of flight, providing a consistent and open-source approach to quantifying gravito-inertial accelerations (GIA), or g levels. As academic, governmental, and commercial use of space advances, data availability and validated processing methods will enable better planning, execution, and analysis of parabolic flight experiments, and thus facilitate future space activities. npj Microgravity (2018) 4:14 ; doi:10.1038/s41526-018-0050-3 INTRODUCTION research section (Fig. 1b-c). Direct Current (DC) acceleration was recorded at 411 Hz. Additional data, calibration, and calibration Parabolic flights are cost-effective, ground-based analogs that error ( < 2%) assessment are described in Methods. achieve variable g (Earth-relative GIA) level environments that 1,2 In any given experiment, one accelerometer orientation may be recreate conditions experienced during space flight. Specialized more appropriate than another. Thus, our phase of flight aircraft can maintain approximately 20–30 s of a 0 g, freefall identification is based on a metric that is independent of environment before an increased GIA recovery phase (Fig. 1a). accelerometer orientation: the Euclidean norm of the acceler- Modified trajectories can achieve reduced g levels experienced on ometer (x,y,z) axes, which we refer to as g level or g (Methods). the lunar surface or on Mars (0.17 and 0.38 g, respectively). To facilitate our analysis (Fig. 1d, Methods, Supplementary Figs. Parabolic flights serve as valuable proving grounds for experi- 1-5), acceleration data were filtered (Fig. 1e) using a zero-delay, mental efforts to maximize the research potential of the 3,4 low-pass filter prior to parabola identification using change point International Space Station and to accommodate increasing 7,8 5,6 detection. Conceptually, this process finds the point for which a interest in commercial space flight. statistical property (e.g., mean), has minimum total residual error Here we address the limited availability of open access summed across two groups, e.g., before and after the change acceleration datasets containing parabolic flight profiles and point. Here residual error is the difference between an observed enable unsupervised and precise characterization of timing and value and the statistical property for the group. g levels for all flight phases. We demonstrate this approach using a Change point detection was first applied to the filtered g level, small (65 g) battery powered commercially available acceler- g , to identify differences in mean g levels in an unsupervised filt ometer and vibration measurement system. Together, these tools manner (Fig. 1f). To break down the flight into regions of stable g and products reflect a comprehensive solution for experiment levels, data within 10 s of each change point was subjected to planning, execution, and analysis of g level and vibrations during secondary change point detection using a linear slope metric, parabolic flight. which segmented the flight into regions of rapid “transition” (indicated by dotted lines, Fig. 1g-i) and more stable regimes. Non- transition periods were subsequently classified into “parabola,” RESULTS “hypergravity,” and “other” based on their duration and g level Flight operations were conducted on November 17, 2017 onboard (Fig. 2a-b; Methods). These “hypergravity” periods result from a Boeing 727-200F aircraft (G-Force One®, Zero Gravity Corpora- entry into and exit from the “parabola” periods. tion). Four sets of parabolas were performed with 5, 6, 4, and 5 parabolas, respectively. The first set targeted, in order, Mars g, DISCUSSION Mars g, Lunar g,0 g, and 0 g. All other parabolas targeted 0 g. Data were collected for 1.77 h during all phases of flight from a Slam Parabolic flights provide the opportunity to perform simulated Stick X™ (Mide Technology Corp.) mounted in the rear of the space research in a cost-effective manner. Recently, Lambot and 1 2 Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA, USA; Department of Molecular Biology, Massachusetts General Hospital, Boston, MA, USA and Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA, USA Correspondence: Christopher E. Carr (chrisc@mit.edu) Received: 4 January 2018 Revised: 25 June 2018 Accepted: 3 July 2018 Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA Acceleration profiles and processing methods for parabolic flight CE Carr et al. to aircraft nose a b to aircraft tail c 3 m 45° 20-30 s 1.5-3 km 50.8 cm "Zero G" Baseplate AN6 bolt points 18 m d Raw Data Calibrate Low Pass Filter Segment flight into period s Characterize Classify Other Compute g level (norm) g level, Optimize Compute filtered Change point detection Hypergravity Filter g level (norm) of g level transitions duration Parabola Power Spectral Density 1.5 0.5 0 1000 2000 3000 4000 5000 6000 Time (s) 1.5 gh i 0.5 1000 1500 2000 2500 3000 3500 Time (s) g h i 1.5 1.5 1.5 z z z 1 1 1 0.5 0.5 0.5 y y y 0 0 0 x x x 1250 1260 1270 1280 1290 1330 1340 1350 1360 1370 1410 1420 1430 1440 1450 Time (s) Time (s) Time (s) Fig. 1 Parabolic flight acceleration data and analysis methodology. a Typical flight path during a single parabola. b Research section of aircraft with baseplate location during flight. c Accelerometer orientation on baseplate. d Overview of analysis method. e Measured accelerations after low-pass filtering (g ). f Change points (vertical lines) for mean g levels as measured by g . g–i Second-level linear change points filt filt (vertical dotted lines) define transition regions for a Mars, a lunar, and a 0 g parabola, respectively. Individual parabolas corresponding to g–i are labeled in f. For accelerations during each of the 20 parabolas, see Supplementary Figs. 6–7 Ord (2016) evaluated data from over 400 parabolic flights and addition, the code implementing our methodology to categorize assessed the quality of reduced g datasets. Although consider- all phases of flight and characterize g levels and durations of able effort was dedicated to identifying the highest quality, low g parabolas is publicly available in order to facilitate future parabolic time periods (with variations less than ± 0.01 g) from these flights, flight research. In addition, our methods could be adapted to neither the acquisition hardware, the raw data, nor the code analysis of data from suborbital flights, drop towers, or studies implemented for analysis, are currently available to the public. involving launch and landing accelerations. Indeed, when reviewing the literature, we found no such easily Our hardware solution, the Slam Stick X™, offers a compact, accessible parabolic flight acceleration data nor published analysis flexible, low-power, high resolution solution for acceleration and methodology. vibration monitoring with favorable comparison to alternatives Here, we provide the following: (1) a commercially available (Methods). The small size facilitates integration into experiments hardware solution for data acquisition; (2) raw and calibrated data and measurement of the local GIA environment, which is not for all phases of flight; (3) data analysis methodology that is constant across the aircraft. In addition, mounting with double- independent of accelerometer orientation, and (4) characteriza- sided tape is simple, robust, and does not impact the frequency tion of g levels and durations achieved for 20 parabolas. In response (Methods). npj Microgravity (2018) 14 Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA 1234567890():,; g g filt filt filt filt filt Acceleration profiles and processing methods for parabolic flight CE Carr et al. Lambot and Ord identified a “sweet zone” in the middle of the sides, centered on the baseplate. Washers (McMaster Carr parabola with low acceleration deviation. Our data are consistent 92503A230) were used for mounting in combination with AN-6 with this conclusion (Supplementary Fig. 8), although Lambot and steel bolts (3/8 inch) provided by ZGC. The Slam Stick X™ was Ord used, after unspecified low-pass filtering, a much more mounted to a standard baseplate with double-sided sticky tape stringent tolerance (± 0.01 g) that was not met by our filtered data. (3 M 950), which is the preferred mounting method due to its We encourage future parabolic flight experimenters to release raw vibration frequency response (near unity) and robustness: this data as well as to provide data processing details (ideally including method has previously been validated during vibration testing at in executable form) to facilitate validation and improvement of over 75 g at 1 kHz . The Slam Stick X™ was configured using Slam processing algorithms, as well as refine expected g levels and Stick Lab 1.8. Acquisition was initiated and terminated manually support planning and analysis that can be tailored to specific using the control pad on the device. Sampling rates were 5 kHz experimenter needs. (piezoelectric vibration sensor), 411 Hz (DC acceleration), and 1 Hz Due to the limited availability and high cost of actual space (pressure and temperature). environments, it is imperative that we continue to utilize parabolic flights as a means to simulate space – and to understand the Data calibration accuracy and limitations of this modality. By making our data and The raw IDE file generated by the Slam Stick X™ was converted to methods available we hope to enable others to better plan, calibrated MAT (MATLAB, The Mathworks, Natick, MA) files using execute, and analyze parabolic flight experiments, and thus to the ide2csv.exe command line utility (Mide Technology Corp.) help facilitate future space activities. using the factory calibration. Note that data calibration and export functions can also be performed directly using Slam Stick Lab. Here we focus on data from the DC accelerometer. We note that METHODS temperature varied less than 1 °C and pressure showed a typical Device Selection regulated profile and was highly stable during parabolas The Slam Stick X™ (Mide Technology Corp., www.mide.com) was (Supplementary Fig. 5). selected based on its size (76 mm × 30 mm × 15 mm), low mass (65 g), integrated battery, manual and USB interfaces, and Orientation-independent approach combination DC (Analog Devices ADXL345) and piezoelectric (TE 832M1) accelerometers to enable accuracy at both low (e.g., down In any given experiment, one accelerometer orientation may be to 0 Hz) and high frequencies (up to 20 kHz sample frequency). more appropriate than another. Thus, we based our phase of flight The aluminum body was selected to provide improved high identification method (below) on a measure that is independent frequency response. Additional integrated sensors included of the accelerometer orientation: the Euclidean norm of the x, y, temperature and pressure (NXP MPL3115) and control pad and z axes, which we hereafter refer to as the g level or g. As this temperature and pressure (TE MS8607). variable is a positive scalar, it does not capture directional Alternative data acquisition systems include many commercial fluctuations in the gravity vector. Thus, for characterization of off the shelf (COTS) accelerometers, as well as the NASA Suborbital phases of flight, vector-based statistics should be used. For Flight Environment Monitor (SFEM) . There may be potential example, we estimated the mean g level during a 0 g parabola by benefits of using the SFEM, although the Slam Stick X™ offers averaging the x, y, and z components, and then computing the comparable or longer recording time, DC and piezoelectric norm. accelerometers (enabling both g level and high frequency vibration measurements), higher sampling frequencies, a wider Calibration verification operating temperature range (− 40 °C to + 80˚C) and much lower The expected value of the g level is unity on Earth under non- ( > 10 × ) mass and volume. Another COTS option is the Lansmont accelerated conditions; as a verification of our accelerometer 3X90 , although the Slam Stick X™ specifications provide benefits calibration, we found the norm under lab bench conditions (14.2 s in several areas (size, mass, sampling rates, and temperature recording) to be 0.9840 (rms) and 0.9840 ± 0.0055 (mean ± SD), range). consistent with < 2% error. This is a lower bound when vibration One consideration for parabolic flight experiments is that the or specific force other than that caused by gravity is present, GIA environment is not constant across the aircraft. In some cases, consistent with the rms value (1.07) observed during flight. it may be adequate to have a single reference flight profile to be Specific force was concentrated in the z axis as measured by root used by multiple experiment teams. However, some applications mean square (rms) values (0.0466, 0.0775, 1.0662 for x, y, and z may be better served through measurement of the local GIA axes, respectively), consistent with the accelerometer orientation environment of a given experimental apparatus. Here, the small (Fig. 1c). Calibration accuracy was also assessed after filtering (see size of our solution facilitates direct incorporation into a payload, below). as well as placement in the desired location or orientation. When selecting a data acquisition solution, it is also important to consider how the mounting of the accelerometer itself may Phase of flight characterization impact the frequency response; in our case, use of double-sided For parabola identification, we first filtered the raw data using a sticky tape represents both an extremely practical and low bias zero-phase 12th order Butterworth filter using the designfilt() option, enabled by the low device mass. Because no additional function using a half power frequency (HPF) as described below. materials separate the accelerometer from the aircraft, there is no 7,8 Next, we utilized change point detection as implemented by the need to correct for the frequency response of the mounting MATLAB FindChangePts() function. interface. Change point detection was first applied to the filtered g level g to identify differences in mean g levels. A known number of filt Device mounting and data acquisition change points was specified based on the parabola number within Zero Gravity Corporation (ZGC) utilizes a standard system of each set, e.g., two times the number of parabolas, plus two mounting hardware to the aircraft structure consisting of a additional transitions (first pre-parabola pull up; last post-parabola baseplate (61 cm × 61 cm × 1.27 cm aluminum plate, e.g., McMas- pull up) for each set of parabolas. In our case sets of 5, 6, 4, and 5 ter Carr 86825K25) bolted to the aircraft structure using four (20 total) parabolas become 12, 14, 10, and 12 change points. This clearance holes at the corners of a square with 50.8 cm (20 in) total number of change points (48) was specified and Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA npj Microgravity (2018) 14 Acceleration profiles and processing methods for parabolic flight CE Carr et al. however, based on the wide separation between “parabola,” a 4 10 Straight & “hypergravity,” and “other” classes, this is not expected to present level flight Transition a significant challenge to standard unsupervised classification Parabola Turns & approaches (e.g., k-means). Hypergravity setup for 10 Other last set of parabolas Filter optimization To optimize the filter, we selected a HPF based on the g level power spectral density (PSD, Supplementary Fig. 2a). The PSD was computed via the MATLAB pwelch() function with default parameters. To select the HPF, we examined the cumulative sum 1 of the PSD (Supplementary Fig. 2b), which revealed a sharp increase in power above 0.01 Hz. We chose this value (HPF = 0.01 Hz) to maximize the low frequency content of the filtered data while rejecting as much spectral power from higher frequencies as possible. As an example, filtering at HPF = 0.01 Hz 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 preserves parabola dynamics, whereas filtering at HPF = 0.001 Hz Mean g level does not (Supplementary Fig. 2c). Our selected value provides the smoothest data for identifying parabolas, while still accurately representing g level transitions. A manual procedure identified 0 g Lunar Mars similar values, e.g., adjusting the HPF toward DC until the rapid (N=17) (N=1) (N=2) transitions between g levels showed systematic bias, then setting the HPF to 10 × this value, also gave HPF = 0.01 Hz. Filtering reduced the root mean square specific force in the 15 lateral (x) direction but little in other directions (0.0161, 0.0681, 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 1.0623 for x, y, z, respectively), consistent with low frequency Mean g level aircraft accelerations mainly due to pitch maneuvers. The g level was near unity during periods of relative calm (Supplemental Fig. Fig. 2 Phases of flight classification and characterization. a After identifying transitions, non-transition events were classified as 3a-b), including the first 1000 s of data collected during largely duration < 100 s (horizontal line) and by mean g level (vertical straight and level flight (rms 0.9919 and 0.9856, raw and filtered, dotted lines) into “parabola,”“hypergravity,” and “other.” The subset respectively). This unfiltered estimate is 0.8% higher than under of “hypergravity” periods with lower duration were identified as lab bench conditions, and both are consistent with accurate those at the start or end of a set of parabolas. All phases of flight sensor calibration at DC to lower than 2% error, based on the were unambiguously classified. b Parabola characteristics (see also: factory calibration. Supplementary Information). Error bars: mean ± SD Regression analysis FindChangePts() identified all g level change points in an Regression of parabola g level on duration was performed using unsupervised manner (Fig. 1f). the MATLAB fitlm() function. Confidence intervals were deter- mined using the MATLAB coefCI() function. We desire to break down the flight into regions of stable g levels. Thus, for each change point, we used a secondary change point detection to identify differences in the slope of the g level Code availability vs. time curves. Data within 10 s of each change point was The MATLAB scripts implementing our analysis are available at: subjected to this secondary change point detection using a linear https://github.com/CarrCE/zerog. slope metric. This step successfully segmented the flight into regions of rapid “transition” (indicated by dotted lines, Fig. 1g-i) Data availability and more stable regimes. This resulted in 97 flight periods (2 × the Raw and calibrated data are available via the Open Science number of change points + 1). Framework at: https://osf.io/nk2w4/. Classification of non-“transition” flight periods into “parabola,” “hypergravity,” and “other” (which includes straight and level flight as well as standard rate turns) was then performed, first by ACKNOWLEDGEMENTS categorizing any periods with duration > 100 s as “other”, then by We thank the MIT Media Lab Space Exploration Initiative for providing the parabolic segmenting data according to g level (“parabola” ≤ 0.9 g, flight. This work was supported by NASA award NNX15AF85G. N.C.B. was supported 0.9 < “other” ≤ 1.1 g, “hypergravity > 1.1 g). Despite its simplicity, by NASA Postdoctoral Fellowship award 80NSSC17K0688. this classifier achieved good separation between classes (Fig. 2a). Parabola durations (mean ± s.d.) were 19.5 ± 1.4 s (0 g, N = 17, AUTHOR CONTRIBUTIONS range 17 to 24 s), 23.7 s (Lunar g; N = 1), and 28.9 ± 0.7 s (Mars g; N C.E.C. designed the experiment. C.E.C., K.S., S.A.B., and N.C.B. built and tested the = 2). The g levels achieved were 0.041 ± 0.005 g (0 g) and 0.159 g hardware. N.C.B. and M.T.Z. collected the data. C.E.C. and N.C.B. processed the data. G. (lunar g). Both Mars parabolas achieved 0.356 g, indicating high R. advised on the experiment design. C.E.C. and N.C.B. wrote, and all authors edited consistency between parabolas targeting similar g levels. Higher g and approved, the paper. C.E.C. is the guarantor. levels were significantly associated with longer-duration parabolas (Supplementary Fig. 4a), although not when lunar and Mars data were excluded (Supplemental Fig. 4b). ADDITIONAL INFORMATION Some limitations are inherent in our study, which focused solely Supplementary information accompanies the paper on the npj Microgravity website on one flight and 20 parabolas. If analyzing multiple flights, with (https://doi.org/10.1038/s41526-018-0050-3). parabolas performed under more varied conditions, it is possible a Competing interests: The authors declare no competing interest. slightly more complex classification strategy might be required; npj Microgravity (2018) 14 Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA Duration (s) Duration (s) Acceleration profiles and processing methods for parabolic flight CE Carr et al. REFERENCES Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, 1. Karmali, F. & Shelhamer, M. The dynamics of parabolic flight: flight characteristics adaptation, distribution and reproduction in any medium or format, as long as you give and passenger percepts. Acta Astronaut. 63, 594–602 (2008). appropriate credit to the original author(s) and the source, provide a link to the Creative 2. Shelhamer, M. Parabolic flight as a spaceflight analog. J. Appl. Physiol. 120, Commons license, and indicate if changes were made. The images or other third party 1442–1448 (2016). material in this article are included in the article’s Creative Commons license, unless 3. Rai, A. et al. Expanded benefits for humanity from the International Space Station. indicated otherwise in a credit line to the material. If material is not included in the Acta Astronaut. 126, 463–474 (2016). article’s Creative Commons license and your intended use is not permitted by statutory 4. Shelhamer, M. Why send humans into space? Science and non-science motivations regulation or exceeds the permitted use, you will need to obtain permission directly for human space flight. Space Policy, https://doi.org/10.1016/j.spacepol.2017.10.001 from the copyright holder. To view a copy of this license, visit http://creativecommons. (2017). org/licenses/by/4.0/. 5. Smith, L. Commercial space flight. Congr. Dig. 96,30–31 (2017). 6. Musk, E. Making humans a multi-planetary species. New Space 5,46–61 (2017). 7. Lavielle, M. Using penalized contrasts for the change-point problem. Signal Pro- © The Author(s) 2018 cess. 85, 1501–1510 (2005). 8. Killick, R., Fearnhead, P. & Eckley, I. A. Optimal detection of change points with a linear computational cost. J. Am. Stat. Assoc. 107, 1590–1598 (2012). 9. Lambot, T. & Ord, S. F. Analysis of the quality of parabolic flight. Next-Generation Suborbital Researchers Conference, https://ntrs.nasa.gov/search.jsp?R=20160007932 (2016). Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA npj Microgravity (2018) 14 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png npj Microgravity Springer Journals

Acceleration profiles and processing methods for parabolic flight

Download PDF
5 pages

Loading next page...
 
/lp/springer-journals/acceleration-profiles-and-processing-methods-for-parabolic-flight-xx4I0lJJmo

References (16)

Publisher
Springer Journals
Copyright
Copyright © 2018 by The Author(s)
Subject
Life Sciences; Life Sciences, general; Classical and Continuum Physics; Biotechnology; Immunology; Space Sciences (including Extraterrestrial Physics, Space Exploration and Astronautics) ; Applied Microbiology
eISSN
2373-8065
DOI
10.1038/s41526-018-0050-3
Publisher site
See Article on Publisher Site

Abstract

www.nature.com/npjmgrav BRIEF COMMUNICATION OPEN Acceleration profiles and processing methods for parabolic flight 1,2 1 1 3 2 1 Christopher E. Carr , Noelle C. Bryan , Kendall N. Saboda , Srinivasa A. Bhattaru , Gary Ruvkun and Maria T. Zuber Parabolic flights provide cost-effective, time-limited access to “weightless” or reduced gravity conditions, facilitating research and validation activities that complement infrequent and costly access to space. Although parabolic flights have been conducted for decades, reference acceleration profiles and processing methods are not widely available. Here we present a solution for collecting, analyzing, and classifying the altered gravity environments experienced during parabolic flights, which we validated during a Boeing 727-200F flight with 20 parabolas. All data and analysis code are freely available. Our solution can be integrated with diverse experimental designs, does not depend upon accelerometer orientation, and allows unsupervised classification of all phases of flight, providing a consistent and open-source approach to quantifying gravito-inertial accelerations (GIA), or g levels. As academic, governmental, and commercial use of space advances, data availability and validated processing methods will enable better planning, execution, and analysis of parabolic flight experiments, and thus facilitate future space activities. npj Microgravity (2018) 4:14 ; doi:10.1038/s41526-018-0050-3 INTRODUCTION research section (Fig. 1b-c). Direct Current (DC) acceleration was recorded at 411 Hz. Additional data, calibration, and calibration Parabolic flights are cost-effective, ground-based analogs that error ( < 2%) assessment are described in Methods. achieve variable g (Earth-relative GIA) level environments that 1,2 In any given experiment, one accelerometer orientation may be recreate conditions experienced during space flight. Specialized more appropriate than another. Thus, our phase of flight aircraft can maintain approximately 20–30 s of a 0 g, freefall identification is based on a metric that is independent of environment before an increased GIA recovery phase (Fig. 1a). accelerometer orientation: the Euclidean norm of the acceler- Modified trajectories can achieve reduced g levels experienced on ometer (x,y,z) axes, which we refer to as g level or g (Methods). the lunar surface or on Mars (0.17 and 0.38 g, respectively). To facilitate our analysis (Fig. 1d, Methods, Supplementary Figs. Parabolic flights serve as valuable proving grounds for experi- 1-5), acceleration data were filtered (Fig. 1e) using a zero-delay, mental efforts to maximize the research potential of the 3,4 low-pass filter prior to parabola identification using change point International Space Station and to accommodate increasing 7,8 5,6 detection. Conceptually, this process finds the point for which a interest in commercial space flight. statistical property (e.g., mean), has minimum total residual error Here we address the limited availability of open access summed across two groups, e.g., before and after the change acceleration datasets containing parabolic flight profiles and point. Here residual error is the difference between an observed enable unsupervised and precise characterization of timing and value and the statistical property for the group. g levels for all flight phases. We demonstrate this approach using a Change point detection was first applied to the filtered g level, small (65 g) battery powered commercially available acceler- g , to identify differences in mean g levels in an unsupervised filt ometer and vibration measurement system. Together, these tools manner (Fig. 1f). To break down the flight into regions of stable g and products reflect a comprehensive solution for experiment levels, data within 10 s of each change point was subjected to planning, execution, and analysis of g level and vibrations during secondary change point detection using a linear slope metric, parabolic flight. which segmented the flight into regions of rapid “transition” (indicated by dotted lines, Fig. 1g-i) and more stable regimes. Non- transition periods were subsequently classified into “parabola,” RESULTS “hypergravity,” and “other” based on their duration and g level Flight operations were conducted on November 17, 2017 onboard (Fig. 2a-b; Methods). These “hypergravity” periods result from a Boeing 727-200F aircraft (G-Force One®, Zero Gravity Corpora- entry into and exit from the “parabola” periods. tion). Four sets of parabolas were performed with 5, 6, 4, and 5 parabolas, respectively. The first set targeted, in order, Mars g, DISCUSSION Mars g, Lunar g,0 g, and 0 g. All other parabolas targeted 0 g. Data were collected for 1.77 h during all phases of flight from a Slam Parabolic flights provide the opportunity to perform simulated Stick X™ (Mide Technology Corp.) mounted in the rear of the space research in a cost-effective manner. Recently, Lambot and 1 2 Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA, USA; Department of Molecular Biology, Massachusetts General Hospital, Boston, MA, USA and Department of Aeronautics and Astronautics, Massachusetts Institute of Technology, Cambridge, MA, USA Correspondence: Christopher E. Carr (chrisc@mit.edu) Received: 4 January 2018 Revised: 25 June 2018 Accepted: 3 July 2018 Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA Acceleration profiles and processing methods for parabolic flight CE Carr et al. to aircraft nose a b to aircraft tail c 3 m 45° 20-30 s 1.5-3 km 50.8 cm "Zero G" Baseplate AN6 bolt points 18 m d Raw Data Calibrate Low Pass Filter Segment flight into period s Characterize Classify Other Compute g level (norm) g level, Optimize Compute filtered Change point detection Hypergravity Filter g level (norm) of g level transitions duration Parabola Power Spectral Density 1.5 0.5 0 1000 2000 3000 4000 5000 6000 Time (s) 1.5 gh i 0.5 1000 1500 2000 2500 3000 3500 Time (s) g h i 1.5 1.5 1.5 z z z 1 1 1 0.5 0.5 0.5 y y y 0 0 0 x x x 1250 1260 1270 1280 1290 1330 1340 1350 1360 1370 1410 1420 1430 1440 1450 Time (s) Time (s) Time (s) Fig. 1 Parabolic flight acceleration data and analysis methodology. a Typical flight path during a single parabola. b Research section of aircraft with baseplate location during flight. c Accelerometer orientation on baseplate. d Overview of analysis method. e Measured accelerations after low-pass filtering (g ). f Change points (vertical lines) for mean g levels as measured by g . g–i Second-level linear change points filt filt (vertical dotted lines) define transition regions for a Mars, a lunar, and a 0 g parabola, respectively. Individual parabolas corresponding to g–i are labeled in f. For accelerations during each of the 20 parabolas, see Supplementary Figs. 6–7 Ord (2016) evaluated data from over 400 parabolic flights and addition, the code implementing our methodology to categorize assessed the quality of reduced g datasets. Although consider- all phases of flight and characterize g levels and durations of able effort was dedicated to identifying the highest quality, low g parabolas is publicly available in order to facilitate future parabolic time periods (with variations less than ± 0.01 g) from these flights, flight research. In addition, our methods could be adapted to neither the acquisition hardware, the raw data, nor the code analysis of data from suborbital flights, drop towers, or studies implemented for analysis, are currently available to the public. involving launch and landing accelerations. Indeed, when reviewing the literature, we found no such easily Our hardware solution, the Slam Stick X™, offers a compact, accessible parabolic flight acceleration data nor published analysis flexible, low-power, high resolution solution for acceleration and methodology. vibration monitoring with favorable comparison to alternatives Here, we provide the following: (1) a commercially available (Methods). The small size facilitates integration into experiments hardware solution for data acquisition; (2) raw and calibrated data and measurement of the local GIA environment, which is not for all phases of flight; (3) data analysis methodology that is constant across the aircraft. In addition, mounting with double- independent of accelerometer orientation, and (4) characteriza- sided tape is simple, robust, and does not impact the frequency tion of g levels and durations achieved for 20 parabolas. In response (Methods). npj Microgravity (2018) 14 Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA 1234567890():,; g g filt filt filt filt filt Acceleration profiles and processing methods for parabolic flight CE Carr et al. Lambot and Ord identified a “sweet zone” in the middle of the sides, centered on the baseplate. Washers (McMaster Carr parabola with low acceleration deviation. Our data are consistent 92503A230) were used for mounting in combination with AN-6 with this conclusion (Supplementary Fig. 8), although Lambot and steel bolts (3/8 inch) provided by ZGC. The Slam Stick X™ was Ord used, after unspecified low-pass filtering, a much more mounted to a standard baseplate with double-sided sticky tape stringent tolerance (± 0.01 g) that was not met by our filtered data. (3 M 950), which is the preferred mounting method due to its We encourage future parabolic flight experimenters to release raw vibration frequency response (near unity) and robustness: this data as well as to provide data processing details (ideally including method has previously been validated during vibration testing at in executable form) to facilitate validation and improvement of over 75 g at 1 kHz . The Slam Stick X™ was configured using Slam processing algorithms, as well as refine expected g levels and Stick Lab 1.8. Acquisition was initiated and terminated manually support planning and analysis that can be tailored to specific using the control pad on the device. Sampling rates were 5 kHz experimenter needs. (piezoelectric vibration sensor), 411 Hz (DC acceleration), and 1 Hz Due to the limited availability and high cost of actual space (pressure and temperature). environments, it is imperative that we continue to utilize parabolic flights as a means to simulate space – and to understand the Data calibration accuracy and limitations of this modality. By making our data and The raw IDE file generated by the Slam Stick X™ was converted to methods available we hope to enable others to better plan, calibrated MAT (MATLAB, The Mathworks, Natick, MA) files using execute, and analyze parabolic flight experiments, and thus to the ide2csv.exe command line utility (Mide Technology Corp.) help facilitate future space activities. using the factory calibration. Note that data calibration and export functions can also be performed directly using Slam Stick Lab. Here we focus on data from the DC accelerometer. We note that METHODS temperature varied less than 1 °C and pressure showed a typical Device Selection regulated profile and was highly stable during parabolas The Slam Stick X™ (Mide Technology Corp., www.mide.com) was (Supplementary Fig. 5). selected based on its size (76 mm × 30 mm × 15 mm), low mass (65 g), integrated battery, manual and USB interfaces, and Orientation-independent approach combination DC (Analog Devices ADXL345) and piezoelectric (TE 832M1) accelerometers to enable accuracy at both low (e.g., down In any given experiment, one accelerometer orientation may be to 0 Hz) and high frequencies (up to 20 kHz sample frequency). more appropriate than another. Thus, we based our phase of flight The aluminum body was selected to provide improved high identification method (below) on a measure that is independent frequency response. Additional integrated sensors included of the accelerometer orientation: the Euclidean norm of the x, y, temperature and pressure (NXP MPL3115) and control pad and z axes, which we hereafter refer to as the g level or g. As this temperature and pressure (TE MS8607). variable is a positive scalar, it does not capture directional Alternative data acquisition systems include many commercial fluctuations in the gravity vector. Thus, for characterization of off the shelf (COTS) accelerometers, as well as the NASA Suborbital phases of flight, vector-based statistics should be used. For Flight Environment Monitor (SFEM) . There may be potential example, we estimated the mean g level during a 0 g parabola by benefits of using the SFEM, although the Slam Stick X™ offers averaging the x, y, and z components, and then computing the comparable or longer recording time, DC and piezoelectric norm. accelerometers (enabling both g level and high frequency vibration measurements), higher sampling frequencies, a wider Calibration verification operating temperature range (− 40 °C to + 80˚C) and much lower The expected value of the g level is unity on Earth under non- ( > 10 × ) mass and volume. Another COTS option is the Lansmont accelerated conditions; as a verification of our accelerometer 3X90 , although the Slam Stick X™ specifications provide benefits calibration, we found the norm under lab bench conditions (14.2 s in several areas (size, mass, sampling rates, and temperature recording) to be 0.9840 (rms) and 0.9840 ± 0.0055 (mean ± SD), range). consistent with < 2% error. This is a lower bound when vibration One consideration for parabolic flight experiments is that the or specific force other than that caused by gravity is present, GIA environment is not constant across the aircraft. In some cases, consistent with the rms value (1.07) observed during flight. it may be adequate to have a single reference flight profile to be Specific force was concentrated in the z axis as measured by root used by multiple experiment teams. However, some applications mean square (rms) values (0.0466, 0.0775, 1.0662 for x, y, and z may be better served through measurement of the local GIA axes, respectively), consistent with the accelerometer orientation environment of a given experimental apparatus. Here, the small (Fig. 1c). Calibration accuracy was also assessed after filtering (see size of our solution facilitates direct incorporation into a payload, below). as well as placement in the desired location or orientation. When selecting a data acquisition solution, it is also important to consider how the mounting of the accelerometer itself may Phase of flight characterization impact the frequency response; in our case, use of double-sided For parabola identification, we first filtered the raw data using a sticky tape represents both an extremely practical and low bias zero-phase 12th order Butterworth filter using the designfilt() option, enabled by the low device mass. Because no additional function using a half power frequency (HPF) as described below. materials separate the accelerometer from the aircraft, there is no 7,8 Next, we utilized change point detection as implemented by the need to correct for the frequency response of the mounting MATLAB FindChangePts() function. interface. Change point detection was first applied to the filtered g level g to identify differences in mean g levels. A known number of filt Device mounting and data acquisition change points was specified based on the parabola number within Zero Gravity Corporation (ZGC) utilizes a standard system of each set, e.g., two times the number of parabolas, plus two mounting hardware to the aircraft structure consisting of a additional transitions (first pre-parabola pull up; last post-parabola baseplate (61 cm × 61 cm × 1.27 cm aluminum plate, e.g., McMas- pull up) for each set of parabolas. In our case sets of 5, 6, 4, and 5 ter Carr 86825K25) bolted to the aircraft structure using four (20 total) parabolas become 12, 14, 10, and 12 change points. This clearance holes at the corners of a square with 50.8 cm (20 in) total number of change points (48) was specified and Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA npj Microgravity (2018) 14 Acceleration profiles and processing methods for parabolic flight CE Carr et al. however, based on the wide separation between “parabola,” a 4 10 Straight & “hypergravity,” and “other” classes, this is not expected to present level flight Transition a significant challenge to standard unsupervised classification Parabola Turns & approaches (e.g., k-means). Hypergravity setup for 10 Other last set of parabolas Filter optimization To optimize the filter, we selected a HPF based on the g level power spectral density (PSD, Supplementary Fig. 2a). The PSD was computed via the MATLAB pwelch() function with default parameters. To select the HPF, we examined the cumulative sum 1 of the PSD (Supplementary Fig. 2b), which revealed a sharp increase in power above 0.01 Hz. We chose this value (HPF = 0.01 Hz) to maximize the low frequency content of the filtered data while rejecting as much spectral power from higher frequencies as possible. As an example, filtering at HPF = 0.01 Hz 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 preserves parabola dynamics, whereas filtering at HPF = 0.001 Hz Mean g level does not (Supplementary Fig. 2c). Our selected value provides the smoothest data for identifying parabolas, while still accurately representing g level transitions. A manual procedure identified 0 g Lunar Mars similar values, e.g., adjusting the HPF toward DC until the rapid (N=17) (N=1) (N=2) transitions between g levels showed systematic bias, then setting the HPF to 10 × this value, also gave HPF = 0.01 Hz. Filtering reduced the root mean square specific force in the 15 lateral (x) direction but little in other directions (0.0161, 0.0681, 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 1.0623 for x, y, z, respectively), consistent with low frequency Mean g level aircraft accelerations mainly due to pitch maneuvers. The g level was near unity during periods of relative calm (Supplemental Fig. Fig. 2 Phases of flight classification and characterization. a After identifying transitions, non-transition events were classified as 3a-b), including the first 1000 s of data collected during largely duration < 100 s (horizontal line) and by mean g level (vertical straight and level flight (rms 0.9919 and 0.9856, raw and filtered, dotted lines) into “parabola,”“hypergravity,” and “other.” The subset respectively). This unfiltered estimate is 0.8% higher than under of “hypergravity” periods with lower duration were identified as lab bench conditions, and both are consistent with accurate those at the start or end of a set of parabolas. All phases of flight sensor calibration at DC to lower than 2% error, based on the were unambiguously classified. b Parabola characteristics (see also: factory calibration. Supplementary Information). Error bars: mean ± SD Regression analysis FindChangePts() identified all g level change points in an Regression of parabola g level on duration was performed using unsupervised manner (Fig. 1f). the MATLAB fitlm() function. Confidence intervals were deter- mined using the MATLAB coefCI() function. We desire to break down the flight into regions of stable g levels. Thus, for each change point, we used a secondary change point detection to identify differences in the slope of the g level Code availability vs. time curves. Data within 10 s of each change point was The MATLAB scripts implementing our analysis are available at: subjected to this secondary change point detection using a linear https://github.com/CarrCE/zerog. slope metric. This step successfully segmented the flight into regions of rapid “transition” (indicated by dotted lines, Fig. 1g-i) Data availability and more stable regimes. This resulted in 97 flight periods (2 × the Raw and calibrated data are available via the Open Science number of change points + 1). Framework at: https://osf.io/nk2w4/. Classification of non-“transition” flight periods into “parabola,” “hypergravity,” and “other” (which includes straight and level flight as well as standard rate turns) was then performed, first by ACKNOWLEDGEMENTS categorizing any periods with duration > 100 s as “other”, then by We thank the MIT Media Lab Space Exploration Initiative for providing the parabolic segmenting data according to g level (“parabola” ≤ 0.9 g, flight. This work was supported by NASA award NNX15AF85G. N.C.B. was supported 0.9 < “other” ≤ 1.1 g, “hypergravity > 1.1 g). Despite its simplicity, by NASA Postdoctoral Fellowship award 80NSSC17K0688. this classifier achieved good separation between classes (Fig. 2a). Parabola durations (mean ± s.d.) were 19.5 ± 1.4 s (0 g, N = 17, AUTHOR CONTRIBUTIONS range 17 to 24 s), 23.7 s (Lunar g; N = 1), and 28.9 ± 0.7 s (Mars g; N C.E.C. designed the experiment. C.E.C., K.S., S.A.B., and N.C.B. built and tested the = 2). The g levels achieved were 0.041 ± 0.005 g (0 g) and 0.159 g hardware. N.C.B. and M.T.Z. collected the data. C.E.C. and N.C.B. processed the data. G. (lunar g). Both Mars parabolas achieved 0.356 g, indicating high R. advised on the experiment design. C.E.C. and N.C.B. wrote, and all authors edited consistency between parabolas targeting similar g levels. Higher g and approved, the paper. C.E.C. is the guarantor. levels were significantly associated with longer-duration parabolas (Supplementary Fig. 4a), although not when lunar and Mars data were excluded (Supplemental Fig. 4b). ADDITIONAL INFORMATION Some limitations are inherent in our study, which focused solely Supplementary information accompanies the paper on the npj Microgravity website on one flight and 20 parabolas. If analyzing multiple flights, with (https://doi.org/10.1038/s41526-018-0050-3). parabolas performed under more varied conditions, it is possible a Competing interests: The authors declare no competing interest. slightly more complex classification strategy might be required; npj Microgravity (2018) 14 Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA Duration (s) Duration (s) Acceleration profiles and processing methods for parabolic flight CE Carr et al. REFERENCES Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, 1. Karmali, F. & Shelhamer, M. The dynamics of parabolic flight: flight characteristics adaptation, distribution and reproduction in any medium or format, as long as you give and passenger percepts. Acta Astronaut. 63, 594–602 (2008). appropriate credit to the original author(s) and the source, provide a link to the Creative 2. Shelhamer, M. Parabolic flight as a spaceflight analog. J. Appl. Physiol. 120, Commons license, and indicate if changes were made. The images or other third party 1442–1448 (2016). material in this article are included in the article’s Creative Commons license, unless 3. Rai, A. et al. Expanded benefits for humanity from the International Space Station. indicated otherwise in a credit line to the material. If material is not included in the Acta Astronaut. 126, 463–474 (2016). article’s Creative Commons license and your intended use is not permitted by statutory 4. Shelhamer, M. Why send humans into space? Science and non-science motivations regulation or exceeds the permitted use, you will need to obtain permission directly for human space flight. Space Policy, https://doi.org/10.1016/j.spacepol.2017.10.001 from the copyright holder. To view a copy of this license, visit http://creativecommons. (2017). org/licenses/by/4.0/. 5. Smith, L. Commercial space flight. Congr. Dig. 96,30–31 (2017). 6. Musk, E. Making humans a multi-planetary species. New Space 5,46–61 (2017). 7. Lavielle, M. Using penalized contrasts for the change-point problem. Signal Pro- © The Author(s) 2018 cess. 85, 1501–1510 (2005). 8. Killick, R., Fearnhead, P. & Eckley, I. A. Optimal detection of change points with a linear computational cost. J. Am. Stat. Assoc. 107, 1590–1598 (2012). 9. Lambot, T. & Ord, S. F. Analysis of the quality of parabolic flight. Next-Generation Suborbital Researchers Conference, https://ntrs.nasa.gov/search.jsp?R=20160007932 (2016). Published in cooperation with the Biodesign Institute at Arizona State University, with the support of NASA npj Microgravity (2018) 14

Journal

npj Microgravity – Springer Journals

Published: Aug 7, 2018

There are no references for this article.