Abstract An automated system wasdevelopedtomonitorthehealthgangsofcompositesItusesthevibrationcharacteristicsofcompositestoidentifyacomponcnt'sdamage conditionThe vibration r_sponscs are characmrized by a set of signalfcaan-es _ in thetime,frequencyandspatialdomains.The identificationof these changes m the vibrationcharacteristicscorrespondingtodifferenthealthconditionswas performedusingpatternrecognitionprinciples.Thisallowsefficientdatareductionandinterpretationofvastamounts of information Test components weremanufactured flom isogrid panels to evaluam performanceofthemonitoringsystem.The componentsweredamagedby impacttosimulatedifferenthealthconditions.Freevibrationresponsewas inducedbyatapt_tonthetestcomponents. Themonitoringsystemwas U'ained usingthesericevibrationresponsestoidentify three differenthealthconditions. They areundamaged vs.damaged,damagelocation,anddamage zonesize.Highreliabilityinidentifyingthecorrectcomponent healthconditionwasachievedbythemonitoringsystem._fonltorin_ PrinciplesThe damage monitoring of composite using patternrecognition principleshasbeenshown to be feasibleIwitha limit_t amountofdatafroma compositecantileverbeamThe changes in structural vibration can bc associated withthedamagein amonitored su'ucture.2"6Thesechangescanbe efficientlyinterpretedthroughtheuse of patternrecognition method. The application of pattern recognitionmethod, I, 7, S requires prior knowledge in the correctclassification of an output class using available inputinformation of a monitored structure. The knowledge canbe acquired through a training process. This process usesa database of relevant input information that corresponds toa defined monitored health condition of the structure. Toobtain the necessary information, the input data can beacquired from a network of suitable sensors. This inputinformation can be described as a feature vector. Thefcana'es are defined according to a specific application. Thefeature information is used in the training of a monitoringsystem to obtain an optimum feature set for a specificclassification of output. This optimum feature set is used bythe classifiers to perform the output classification. Thecommonly used classifiers in pat_cm recognition arcNearest Neighbor Criteria (NNC), Gaussian and Fisher. gComposite _'ucture Health Monitoring SystemA health monitoring system for composite structures,Figure 1, was developed on a microprocessor computer toimplement the above principles in the classification ofstructural component's health conditions. A schematic ofth# monitoring sysama is presented in Figure 2. The systemconsists afa 16 c.hanne.l signal conditioner, a post-amplifierwithnoisefill=,and an analog-to-digital(A/D) cardplugged'into a rack mounted 486/33MHz personalcomputer.TheA/D cardiscapableofdigitizingdataupto150KHz forone channel.An integratedsoftwarewasdevelopedforthesystem,Figure3.ThissoRwareismenudriven It'scapabilitiesincludedataacquisition,signalprocessing, feature extraction, classification, and filemanagement On screencalibrationprocedures are alsoprovided.Classificationresultsonthecomponent healthconditionare provided at the end ofdataacquisition. Datacan be saved in files for further training, evaluation orarchive.
[1]
R. L. Thomas,et al.
Nondestructive determination of fatigue crack damage in composites using vibration tests.
,
1972
.
[2]
Keinosuke Fukunaga,et al.
Introduction to Statistical Pattern Recognition
,
1972
.
[3]
Mark Richardson,et al.
FAULT DETECTION IN STRUCTURES FROM CHANGES IN THEIR MODAL PARAMETERS
,
1989
.
[4]
Robert D. Adams,et al.
Vibration Testing as a Nondestructive Test Tool for Composite Materials
,
1975
.
[5]
Keinosuke Fukunaga,et al.
Introduction to statistical pattern recognition (2nd ed.)
,
1990
.
[6]
Julius T. Tou,et al.
Pattern Recognition Principles
,
1974
.