Assignment Writing Servicehttp://StudyHub.vip/A-Residual-Acceleration-Data-Analysis-A

1. Adv. Space Res. Vol. 13, No. 7, pp. (7)261-.(7)265, 1993 0273—1177/93$24.00
Printedin Great Britain. All rights reserved. Copyright © 1993 COSPAR
A RESIDUAL ACCELERATION DATA
ANALYSIS AND MANAGEMENT SYSTEM
R. Wolf, M. J. B. Rogers and J. I. D. Alexander
Centerfor Microgravity and Materials Research, University ofAlabama in
Huntsville, Huntsville, AL 35899, U.S.A.
ABSTRACF
Efficient analysis,management, and disseminationof large, complex, and sometimes poorly understood
residual acceleration datasets obtained from low earth orbit space laboratories are necessary in order to
maximize the scientificreturnfrom microgravityexperiments. In view of the largeamounts of datathat will
be collected in future missions, the need for an organized approach to the reduction, analysis, and
dissemination is anticipated. Thus, an acceleration data processing plan was developed at the Centerfor
Microgravityand MaterialsResearch. Toward thatgoal weare developing a data analysis and management
system that employs a variety of pattern recognition, data visualization, database management, Fourier
analysis, expert systems, and vibration and ergodic windowing techniques.
INTRODUCFION
It has beenrecognized forsome time that the low-gravityenvironmentof a low earth orbitspacecraft can
be used as a laboratory forthe studyof a varietyofphysical phenomena underreduced(equivalent) gravity
conditions. It has also been recognizedthat the residual accelerations arising from gravity gradient tides,
atmospheric drag, thruster firings, crew motions, etc. are sufficient to cause measurable deviations from
“zero-gravity” conditions. The needto measure and record the ever-changinglow-gravity environment in
order to provide important“environmental” data forthe post-flightassessment of low-gravityexperiments
and the physical condition of astronauts has led to the developmentof several acceleration measurement
systems since the early 1970s. Lately, particularly since dedicated spacelab missions began, the interest
inresidualacceleration datahas increased. Indeed withthe now regular use ofNASA’s Space Acceleration
MeasurementSystem (SAMS), the amount of acceleration databeing collected is increasing rapidly. The
first threeflights ofSAMS (SLS-l, STS-43, and IML-l) yielded several gigabytes of data. In responseto
the need foran organized approachto the reduction, analysis, and dissemination of data, an acceleration
dataprocessing plan was developed at the Center forMicrogravity and Materials Research /1,2/.
We havedeveloped a database management system to handle the large quantityof residual acceleration
data that results from a typical low-gravityOrbitermission. The systemmanages a large, graphic database
insupport of supervised and unsupervisedpatternrecognition /3,4/. Useof patternrecognition techniques
allows identification of specific classesof accelerations so that theseclasses can be easily recognized in
any set of acceleration dataretrieved from spacecraft accelerometer systems. The data willbe partitioned
following the ANSI/SPARC model /5/. The entire mission time history forms the internal layer of the
model. Data reductiontechniquesidentify limited timewindows of interest. Time and frequency domain
representations of these windows compose the conceptual level of the model. The graphics aspect of the
management system includes several data visualization techniques to help the user betterunderstand the
nature of the acceleration signal being studied. The PRIDE (Eattern Recognition Vjsualization
~atabas~ system was developed on a UNIX-based Stardent Titan computer, and is being tested
on Spacelab 3 (SL-3) residual acceleration data. When fully developed, itwill be suitable foruse with
other residual acceleration data bases and will be portable to otherUNIX-based workstations.
PATFERN RECOGNITION AND DATA VISUALIZATION SYSTEM
Both supervised and unsupervised patternrecognition techniques are incorporatedintothe PRIDEsystem.
Supervised pattern recognition uses human knowledge of datato aid in classificationwhile unsupervised
(7)261

2. (7)262 IL Wolf etal.
pattern recognition does not. In the PRIDE system, supervised pattern recognition is primarily achieved
through the application ofvarious data visualization techniquesto raw accelerationdata. In addition, these
techniques are applied to various transformations ofthe raw data, such as vectormagnitudes and Fourier
transforms. Additional supervision will be possible through use of the CLIPS expert system /6/. This
allows the increasing expertise of the system users to be stored in the data base through CLIPS.
Unsupervised pattern recognition inPRIDE isachieved through the use of the ISODATA algorithm /4/.
Thisalgorithmis typically used toclassify dataforwhich thereisno known classification. Thisalgorithm
is applicable tothe acceleration databasedirectly by an operator, and is also available for controlleduse
by a specially written CLIPS shell. This will make it possible to run ISODATA without human
intervention. In addition, an ergodicsearchtechniquehas beendeveloped which identifies thelocation and
duration of high energy acceleration events. Note that this technique offers refined control of the
application of Fourier techniques. For example, information such as the start of a potentially interesting
acceleration event, or the optimum length ofthe datawindow forspectralanalysis, canbe readilyprovided.
Datavisualization isused notonly insupervisedpatternrecognition as discussed above, butit isalso used
forgeneral evaluation ofthe datacharacter. Analysis ofthe data characterleadstoa clearerunderstanding
ofthe natureofthe problem and shouldlead tothe developmentofnew approachesto accelerationanalysis.
Note that this use of data visualization is independent of the supervised pattern recognition support
function.
A relational data base has been installed on the CMMR Stardent computer system and has been fully
integratedwithall the functions ofPRIDE. Useof this common databaseform will enhancethe transport
ofPRIDEto otherplatforms. Additionally, the use ofa database providesseveral advantages over flat file
formats, forexample data integrity, maintainability, and large data basemanipulation.
A ..
‘ ~
Fig. 1. 3D-FFT representation showing a projection from 4-tuples (A~(f),A~(t),
A~(f),
t) to 3-tuples
(A~(f),
A~(t),
A~(t))
plottedin3-D spaceandlinked bylines. Highenergyeventsfrom 30minutesofdata
during SL-3.

3. Data Analysis and Management System (7)263
1712 I I ‘ I I
1539 - -
1368-
I— z
0
1026.
U)
855- -
~ 684 -
513- -
342- -
171 ~
0 30 60 90 120 150
FREQUENCY [Hz]
Fig. 2. Amplitudespectrum foreachofthe threeaccelerationcomponents. Data
are taken from one of the eventwindows shownin Figure 1.
10 i i i
~
o 3 6 9 12 15 18 21 24 27 30
TIME [mm]
Fig. 3. Spikes inthis ergodic plot indicatethe temporal location ofhighenergy events.

4. (7)264 R. Wolf eta!.
DATA MANAGEMENT AND EXPERT SYSTEMS
The manipulation of large data bases is achieved through the application of various data reduction
techniques. Two main datareduction techniquesused are vibration windowing and ergodicwindowing.
Vibration windowing is used to identify and select windows that fit the definition of a damped high
magnitude oscillatory vibration. Ergodic windowing identifies and selects windows basedon a measure
of the energy attributes of a portionof the accelerometer time series. These provide a way to make the
processing of such huge databases more practical. Additionally, the use oftwo independent techniques
makes the resulting classification more dependable.
Expertsystemstechnology providestwo basiccapabilities. First ofall,itmakes the use ofISODATAmore
practical. Normaluse ofISODATA requires ahuman operatorand,evenwithdatareduction, the data sizes
thatmust be handled ina finite time are enormous. A CLIPS shell to control the ISODATA algorithm is
useful. Second, and more importantly, the availability of an expert system built into the system
with a data base form of access to the data will provide in the future (when a complete
understanding of the processes involved has been achieved) a way to store and use this knowledge to
drive the characterization process.
Figures 1-3 give examples of typical screen images that can be obtained using PRIDE. Figure 1 is an
exampleof the 3D-FFF techniqueapplied to a 30 minute acceleration timeseries from SL-3. Thesample
timeseries contains 420,000 data points peraxis. Visual display ofso many datapoints tends tomask the
characterof the acceleration and isoftennot veryuseful. The 3D-FFF,combinedwithvibration orergodic
windowing, enablesidentification and displayof acceleration events that fit predeterminedspecifications
(either thresholds or particular patterns). This reducesthe amount ofdata that isactually displayed,while
supplying more information than visualization ofthe timeseries itselfprovides. Figure 1 displays3D-FFF
information for windows which exceeded specified acceleration thresholds. Each axis represents the
acceleration component foreach of the x-, y-, and z-directions. For each window, the discrete Fourier
transformof the acceleration dataforeach axis produces a discrete set offrequencies separatedby a fixed
interval, Af. Eachfrequency f
1 isconnected by a straight line segment tothe f~..1= f1 - Afand
1i+1 = f~
+ Al
frequencies. Note that the events in this figure are dominated by accelerations associated with the z-
direction. Figure 2 is the standard FFT for one of the events (windows) used to produce Figure 1.
Figure 3 isa graph drawnfrom an ergodic analysis. Thisgraph isproduced by usinga variable sizesliding
window to createa mapping from an acceleration vectorto its associated energy. Subsequently, an energy
threshold is selected. The temporallocations of all windows with associated energy greaterthan or equal
to the characteristic energy ofthe threshold are recorded. The time segments associated withthese high
energy events within the overall time interval are used to determine the number of high energy events
occurring at any given time.
SUMMARY
PRIDE is inits initial stages of development. Preliminary results suggest that it will be a useful basis for
futureacceleration processing strategies. It is currently being used to examine recentlyobtained SAMS 17/
dataas well as Spacelab-3 accelerometer data. Correlation of experimentevents withacceleration events
as well as general characterizationof space laboratory environments can readily be accomplished using
organized approaches such as PRIDE. The development of such data processing and dissemination
packages combined withNASA’s program of acceleration measurements planned underthe ACAP/8/
program will enable future low-gravity experiment investigators to obtain a clear description of the
residual acceleration conditions prevailing during their experiment.
ACKNOWLEDGEMENTS
Thisprojectis supported by theNational Aeronautics and SpaceAdministrationthrough grantNAG8-759.
The authors would like to thank Dr. Robert Snyder and Charles Baugher for their encouragement and
continuing interest inthis work.

5. Data Analysis and Management System (7)265
REFERENCES
1. M.J.B. Rogers and J.I.D. Alexander, Residual acceleration data analysis for Spacelab-3 missions,
Microgravizy Sci. Technol. V,43 (1992).
2. M.J.B. Rogers and J.I.D. Alexander, A strategy for residual acceleration data reduction and dissemi-
nation, Adv. Space Res. 11, #7, (7)5 (1991).
3. A. Stevens, C Database Development, MIS Press, Portland, Oregon, 1991.
4. J.T. Tou and R.C. Gonzales, Pattern Recognition Principles, Addison-Wesley, Reading, Massachu-
setts, 1974.
5. C.J. Date, An Introduction to Database Systems, Addison-Wesley, Reading, Massachusetts, 1990.
6. J.C. Giarrantano,CUPS User’s Guide,Version 5.1, Artificial IntelligenceSection,Lyndon B. Johnson
Space Flight Center, JSC-25013, 1991.
7. R. DeLombani, B.D. Finley and C.R. Baugher, Development of and flight results from the Space
Acceleration Measurement System (SAMS), AIAA paper 92-0354, 30th Aerospace SciencesMeeting,
Reno, January 1992.
8. Acceleration Characterization and Analysis Project, C.R. Baugher, Marshall Space Flight Center,
personal communication, 1992.