Analysis of acoustic emission in cohesionless soil

Analysis of acoustic emission in cohesionless soil

Material Information

Analysis of acoustic emission in cohesionless soil
Mathiyaparanam, Jeyisanker
Place of Publication:
[Tampa, Fla]
University of South Florida
Publication Date:


Subjects / Keywords:
Fast Fourier Transform
De-noised Signals
Dissertations, Academic -- Civil Engineering -- Masters -- USF
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )


ABSTRACT: Acoustic emission is a widely used nondestructive technique for identification of structural damage. The AE technique relies on transient energy waves generated by the materials during their failure. As for soils, the basic causes of acoustic emission are the mechanisms which are responsible for shearing of soils. Mobilization of shear strength within a soil itself and the interaction of the soil with the adjacent natural or construction materials are directly related to the level of acoustic emission in soils. It is envisioned that acoustic emission signals in deforming soils can be used as an early warning sign in real time landslide-monitoring systems.This thesis study uses a laboratory experimental setup to record the acoustic emission signals emitted during the shearing of cohesionless soils. Several tests were performed with different rates of shearing with parallel (horizontal) and perpendicular (vertical) placement of the AE mote- sensor with respect to the shear plane. Since the original raw signals recorded contain large amounts of noise, it is necessary to de-noise them. The current study uses wavelet and FFT to de-noise the original signals. The filtered signals obtained using wavelet analysis and FFT are compared to determine the suitability of the two techniques. The peak AE values and the time taken to observe an initial visible peak under different conditions are reported in this study. It is observed that relatively faster rates of shearing generate more AE signals compared to slower rates of shearing. In addition, the rapid shearing produces initial visible peak AE activities within a short period of time than in slow rate of shearing.
Thesis (M.A.)--University of South Florida, 2006.
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 80 pages.
Statement of Responsibility:
by Jeyisanker Mathiyaparanam.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
001910832 ( ALEPH )
173609137 ( OCLC )
E14-SFE0001715 ( USFLDC DOI )
e14.1715 ( USFLDC Handle )

Postcard Information



This item has the following downloads:

Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001910832
003 fts
005 20071002083446.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 071002s2006 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001715
1 100
Mathiyaparanam, Jeyisanker.
0 245
Analysis of acoustic emission in cohesionless soil
h [electronic resource] /
by Jeyisanker Mathiyaparanam.
[Tampa, Fla] :
b University of South Florida,
3 520
ABSTRACT: Acoustic emission is a widely used nondestructive technique for identification of structural damage. The AE technique relies on transient energy waves generated by the materials during their failure. As for soils, the basic causes of acoustic emission are the mechanisms which are responsible for shearing of soils. Mobilization of shear strength within a soil itself and the interaction of the soil with the adjacent natural or construction materials are directly related to the level of acoustic emission in soils. It is envisioned that acoustic emission signals in deforming soils can be used as an early warning sign in real time landslide-monitoring systems.This thesis study uses a laboratory experimental setup to record the acoustic emission signals emitted during the shearing of cohesionless soils. Several tests were performed with different rates of shearing with parallel (horizontal) and perpendicular (vertical) placement of the AE mote- sensor with respect to the shear plane. Since the original raw signals recorded contain large amounts of noise, it is necessary to de-noise them. The current study uses wavelet and FFT to de-noise the original signals. The filtered signals obtained using wavelet analysis and FFT are compared to determine the suitability of the two techniques. The peak AE values and the time taken to observe an initial visible peak under different conditions are reported in this study. It is observed that relatively faster rates of shearing generate more AE signals compared to slower rates of shearing. In addition, the rapid shearing produces initial visible peak AE activities within a short period of time than in slow rate of shearing.
Thesis (M.A.)--University of South Florida, 2006.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 80 pages.
Adviser: Manjriker Gunaratne, Ph.D.
Fast Fourier Transform.
De-noised Signals.
Dissertations, Academic
x Civil Engineering
t USF Electronic Theses and Dissertations.
4 856


AnalysisofAcoustic EmissioninCohesionless Soil by J eYisanker athi aparanam A thesis submittedinpartial fulfillmentofthe requirements for the degreeofasterofScienceinCi il Engineering DepartmentofCivil and En ironmental Engineering CollegeofEngineering Uni ersityofSouth Florida ajor Professor: anjriker Gunaratne Ph.D. Gra ullins Ph.D. Ashok Kumar Ph.D. DateofApproal:July142006 Ke ords: Fast Fourier Transform, Wavelet De-noised Signals Filtering Threshold CoPYright 2006 JeYisanker athi aparanam


AC OWLEDGEsFirstofall I ould like to express m most sincere gratitude to my ad isor Prof. anjriker Gunaratne for guiding and helping me to produce thisI ould also like to extend mythanksto the thesis committee members Dr. Ashok Kumar and Dr. Gray ullins for their in aluable suggestions time and also letting me use their equipment. Special thanks go to Ke in Johnson Raghu Mudhiarthiand Ragha a Kakiredd for their support during this experimental thesis study. Finall I ould like to thank m wifeJ.Kal ani m parents and brother for their encouragement, support and help in countless as for me to achie e this milestone in m life.


TABLEOFCO TE SLIST OF TABLESiiiLIST OF FIGURES ABSTRACT CHAPTER 1 TRODcnoIX11.1Acoustic Emission 1 1.2 Acoustic Emission in Soils 3 1.2.1 Useofa e-Guides 41.3Research Objectives 5 CHAPTER 2 WAVELET TRANSFORMS APPLIED TO IDE TIFY THE ACOUSTIC EMISSIO SIG ALS SOILS 62.1Wavelet Transforms 6 2.2 ComparisonofWaelet Analysis with Fourier Anal sis 8 2.3 The Continuous a elet Transform 9 2.4 Scaling112.5 SequenceofSteps to Obtain a Continuous a elet Transform (CWT)122.6 Relationship Between Scale and Frequenc142.7 The Discrete a elet Transform142.8 One-Stage Filtering: Approximations and Details182.9 Multiple-Le el Decomposition212.10 umerical Example232.10.1 The Forward Transform232.10.2 The Inverse Transform 25 2.10.3 Thresholding Decomposed Signal 26 CHAPTER 3 EXPERIEMTAL SETUP AND DATA COLLECTIO 303.1Introduction 30 3.2 Shearing Procedure 30 3.3 Data Acquisition 32 3.3.1 PCS64 i Oscilloscope33


3.3.2 Acoustic Emission Piezoelectric Sensors 3.3.3 Universal Micro-Tribometer (UMT) CHAPTER 4 AL YSISOFDATA 3435374.1Introduction 37 4.2 Useofthe MATLAB a elet Toolbox for AE Signal Processing394.3Fast Fourier Transformation Using MATLAB 46 4.4 ComparisonofFiltered Signals Obtained Using Wavelet with thatofFFT494.5Comparison Between Data Obtained from Chemical echanical Planarization (CMP) and Delamination Processes604.6 Comparison Between the Experimental PlotofShear Force Versus Time and Plot Obtained Using AE Signals614.7 Suggestion for Future Work62CHAPTER 5COCLUSIO S635.1Conclusions635.2 Limitationsofthe Test655.3Potential Practical Applications65REFERECES67APPENDICES69AppendixA:PlotsofData with Filtered Signals7011


LIST OF TABLES Table2.1A Sample Acoustic Emission Data Table 2.2 Complete Transformationofthe Signal Table 2.3 ReconstructionofData Using a Threshold Value 2 Table 2.4 ReconstructionofData Using a Threshold Value 3 Table 2.5 ReconstructionofData sing a Threshold Value 4 Table 2.6 ReconstructionofData Using a Threshold Value10Table3.1Test Designations11124 25 27 27 27 28 36


LIST OF FIGURES Figure2.1Typical Wavelet Transforms Figure 2.2 Common Wavelet Functions (Bruce 1996) Figure 2.3 IllustrationofFourier Transforms Figure 2.4 EffectofShifting Figure 2.5 IllustrationofWaelet Transforms Figure 2.6 EffectofScale FactorsonSinusoids Figure 2.7 EffectofScale Factor on Wavelets Figure 2.8 Comparisonofa Wavelet with a Segmentofa Signal Figure 2.9 ComparisonofShifted Wa elet with Signal Figure 2.10 ComparisonofScaledWaelet with Signal Figure 2.11 Different Scale Wa elets Figure 2.12 Smooth ApproximationofaSineWae Using a Block Pulse Scaling Function Figure 2.13 How an Infinite SetofWaelets is ReplacedbyOne Scaling Function Figure 2.14 One Stage Filter Figure 2.15 One Stage Filter with Thousand Samples Figure 2.16 Down Sampling Figure 2.17 Schematic Diagram with Real SignalsIV6 891011 111213 131314 1718191920 20


Figure 2.18 ultiple Le el Decomposition Figure 2.19 Multile el Filtering Figure 2.20 ReconstructionofSignal Figure2.21PlotsofOriginal and Filtered Signal Figure3.1Shear Box with Electric Screw Jack Figure 3.2 Shear Box with Manual Screw Jack Figure3.3Uni ersal MicroTribometer Figure 3.4 ote Acoustic Emission Sensors with Circuit Board Figure3.5PCS64i Digital Storage Oscilloscope Figure 3.6 ide Band Acoustic Emission Sensors Figure4.1Data Obtained with Horizontal PlacementofMote-Sensor (Shearing Rate: Fast) -HF1Figure 4.2 Data Obtained with Vertical Placementofote-Sensor (Shearing Rate: ormal) Figure4.3Data Obtained nder0Shear Condition Figure 4.4 tartup indoofthe MATLAB Programtoelect a elet Tool Box2122 222830313131333538 38 38 39 Figure 4.5 Selection indo for 1-D and 2-D Continuous and Discrete Signals 40 Figure 4.6 PlotofDaubechies Mother Wa elet Function (db5)41Figure 4.7 PlotofDaubechies (5) Scale Function 42 Figure 4.8 Loaded Original Acoustic Emission Signal Obtained in HS2 with the Horizontal PlacementofMote-Sensor 42 Figure 4.9 a elet Tree and Approximation atLeel Fi e Obtained for HS2 with the Horizontal Placementofote-Sensor43Figure 4.10 Approximate and Detail Coefficient Plot for Test 5 with the Horizontal Placementofote-Sensor 44


Figure4.11PlotsofDetail Coefficients with Deoised Signal for HS245Figure 4.12 PlotofDe-noised Signal for HS2 46 Figure 4.13 Sample PlotofPower Spectral Density (PSD) Versus Frequenc 47 Figure 4.14 agnitudeofTransformed Complex Vector Versus Frequenc (Identity Test HS2) 48 Figure 4.15 Sample Filtered Signal sing FFT (Identity Test HS2)48Figure 4.16 De-noised Signal Using a elet and FFT (TestHFl)Figure 4.17 De-noised Signal Using Wavelet and FFT (Test HF2) Figure 4.18 De-noised Signal Using Wa elet and FFT (Test HF3) Figure 4.19 De-noised Signal sing a elet and FFT (TestHS1)Figure 4.20 De-noised Signal sing Wa elet and FFT (Test HS2) Figure4.21De-noised Signal Using Wa elet and FFT (TestS)Figure 4.22 AE Signals Obtained with Horizontal Placementofthe Sensor Filtered sing a eletAnal sis Figure 4.23 AE Signals Obtained with Horizontal Placementofthe Sensor Filtered Using FFT Figure 4.24 De-noised Signal sing Wa elet and FFT (Test Figure 4.25 De-noised Signal smg a elet andFIT(TestSI)Figure 4.26 De-noised ignal sing Wa elet and FFT (Test V 2) Figure 4.27 De-noised Signal sing Wa elet and FFT (Test VF1) Figure 4.28 De-noised Signal sing a elet and FFT (Test VF2) Figure 4.29 De-noised Signal sing a elet and FFT (TestS)Figure 4.30 AE Signals Obtained with Vertical Placementofthe Sensor Filtered Using Wa eletVI49 49 505051 5152 5253 53 5354 5455 55


Figure4.31AE Signals Obtained withVertical Placementofthe Sensor Filtered Using FFT Figure 4.32 PlotofaximumAmplitudeofAE Signals with the Horizontal Placementofote-Sensor for Different RatesofShearing Figure 4.33 PlotofMaximum AmplitudeofAE Signals with the Vertical PlacementofMote-Sensor for Different RatesofShearing Figure 4.34 PlotofA erage Maximum AmplitudeofAE Signal with Slow and Fast RateofShearing Figure 4.35 PlotofaximumPositi e and Absolute Maximumofegati e Peak Values Obtained From De-noised Signals U sing a elet and FFT Figure 4.36 PlotofTime at Which the Positi e aximum Amplitude Recorded for Different Tests 56 57 5758 5859 Figure 4.37 PlotofTime Taken For OccurrenceofInitial VisiblePeak:Amplitude in Different Tests 60 Figure 4.38 Sample PlotofAE Signal from Polishing and Delamination Process 60 Figure 4.39 Sample PlotsofShear Force Versus Time and AE Versus Time61FigureA.lPlotofHFlwith Filtered Signal sing a elet Figure A.2 PlotofHF2with Filtered Signal Using a elet Figure A.3 Plot ofHF3 with Filtered Signal Using Wa elet Figure A.4 PlotofHSlwith Filtered Signal Using Wa elet Figure A.5 PlotofHS2with Filtered Signal Using Wa elet Figure A.6 PlotofHNSwith Filtered Signal Using Wa elet 707172737475Figure A.7 Plotofwith Filtered Signal sing a elet 76 Figure A.8 PlotofVSlwith Filtered Signal Using a elet Figure A.9 PlotofVS2 with Filtered Signal sing a elet117778


Figure A.I 0 PlotofVFIwith Filtered Signal Using Wa elet FigureA.IIPlotofVF2with Filtered Signal Using WaveletVlll7980


AL YSISOFACO STIC EMISSIO COHESIO ESS SOILJeyisankerathiaparanamABSTRACTAcoustic emission is a widely used nondestructi e technique for identificationofstructural damage. The AE technique relies on transient energywaes generated by the materials during their failure.Asfor soils the basic causesofacoustic emission are the mechanisms which are responsible for shearingofsoils. Mobilizationofshear strength within a soil itself and the interactionofthe soil with the adjacent natural or construction materials are directly related to theIeelofacoustic emission in soils. It is en isioned that acoustic emission signals in deforming soils can be used as an earl warning sign in real time landslide-monitoring systems. This thesis study uses a laboratory experimental setup to record the acoustic emission signals emitted during the shearingofcohesionless soils. Se eral tests were performed with different ratesofshearing with parallel (horizontal) and perpendicular ( ertical) placementoftheAEmotesensor with respect to the shear plane. Since the original raw signals recorded contain large amountsofnoise, it is necessary to de-noise them. The current study uses wa elet and FFT to de-noise the original signals. The filtered signals obtained using a elet analysis and FFT are compared to determine the suitabilityofthe two techniques. The peak AE alues and the time taken to observe an initial visible peak under different conditions are reported in this study.Itis observed that relatively faster ratesofshearing generate more AE signals compared to slower ratesofshearing.Inaddition the rapid shearing produces initial visible peak AE acti ities within a short periodoftime than in slow rateofshearing.lX


CHAPTER1INTRODUCTIO 1.1 Acoustic EmissionAcoustic Emission (AE) is oneofthe widel used nondestructi e testing DT) techniques among the man emerging NDT techniques. The acoustic emission technique pro ides earl indicationsofan small deformation that takes place during progressi e failure. Oneofthe ad antagesofAE compared to other NDT techniques is the possibility to observe damage processes during the entire loading history without an disturbance to the materials. AE related monitoring techniques reI on the detectionoftransient elastic a es emanating from stressed materials during their acti eflo. The induced stress can be tensile compressi e or shear. A certain strain is associated ithany stress and hence under the stress the material can expand contract and/or shear. Depending on the stressIeel the strain can be elastic strains or permanent (plastic) strains. A material under stress accumulates energy and a sudden deformation con erts this energy to elastic a es hich generate acoustic emission signals. Acoustic emission is almost al as associated with permanent strains. Sourcesofacoustic emission can be classified as folIos:1.Dislocation mo ements2.Phase transformation3.Friction mechanisms4.Crack formation and extension Elastic a es are sound a es generated within a material. These elastic a es are called stress ae emissions microseisms micro seismic-acti ity and acoustic emissions. B detecting and quantifying the AE from an acti e stressed material hich1


subsequentl deforms one can assess the stabilityofthe material. ostoftheAE a es are short-time transient e ents or burst signalsofsignificant energ and propagate long distance in circles in all possible directions. Becauseofthe abo e reason AE testing must co er large often inaccessible monitored areas. Often AE signals are not audible becauseoftheir10amplitude or high frequency or both. Ho e er b using an appropriate transducer hich con erts mechanical energy into electrical en erg AE a es can be detected. There are arious typesofAE transducers. Among them piezoelectric transducers are ell-pro en and b far the most del used for AE testing. Piezoelectric transducers produce electrical signals proportional to the amplitudeofthe AE signals or ibration being detected. The sensor may be located some distance from the source and made to detect the signal subject to attenuation. Typical 1 AE s stems operateinthe frequenc rangeof1kHzto 2 MHz or greater. The10er frequenc limit is imposed b background noises such as friction outside impacts or process generated signals that tend to mask acoustic emission. The upper frequenc limit is imposed b attenuation hich tends to limit the rangeofdetectionofacoustic emission signals. A critical partofthe AE application process is the selectionofa suitable frequenc range for AE detection and signal processing. It must be abo e the non-AE related background noises hile pro iding the necessary detection range (distance/frequenc ) and sensiti ity to AE related signals. This is accomplished b the selectionofAE sensors that operate in arious narro -band or ide-band frequenc ranges and adoptionofelectronic signal filtering to remo e un anted noise. When the acoustic emission source generates pulses the detectedsignal-is dependent on the characteristicofae propagation between the sensor and the source and thatofthe sensor. Wa e modes elocitreflection multiple path and re erberation are important factors in the propagationofAE a es. The ae elocity depends on material ae modes (compression shear surface etc.) and thicknessofthe material in hich the ae tra els. When the ae tra els the peak amplitude drops due to attenuation. The most important causesofattenuation are listed belo .2


1.Geometric spreadingofthe a eform. The amplitude decreases in ersel ith distance in three-dimensional media or with the square rootofdistance in two-dimensional media Dominant close to the source2.Absorption or damping in the propagation medium. The amplitude usuall decreases exponentiall with distance Dominant far from the source aterial properties are dependent on the frequenc (higher frequenc AE signals undergo higher attenuation than10er frequenc signals.)3.Leakingofthe ae energ into adjacent media such as contained fluids.1.2 Acoustic EmissioninSoilsThe basic generatorsofacoustic emissions in soils are the mechanisms hich are responsible for the shear strengthofsoils. obilizationofshear strength components within the soil itself and due to the interactionofsoil with other adjacent natural or construction materials is direct! related to theIeelofacoustic emission. When soils are stressed the respond b reorganizing constituent particles and changing their relati e positions th the consequent generationofstress a es. This is also the main objecti eofthe AE monitoring techniques hich enables the detectionofthe occurrenceofdistress in soil before the de elopmentofsignificant mo ements. For granular soils the shearing resistance rolling friction and dilatation are the fundamental attributesofAE.Incaseofcohesi e soils AE acti ities depend on both friction and cohesion.Insoils frictional componentsofshear strength are more emitti e than the cohesion components. Particle size shape gradation and mineral types also ha e major roles in the frictional componentsofthe acoustic emission. On the other hand ater content plasticity and stress history ha e major roles in cohesi e componentofthe AE.3


1.2.1 UseofWave-GuidesInmostofthe non-destructi e tests the transducers are attached direct! on the structure being monitored.Insoils on the other hand attenuationofthe elastic a es as the tra el through the soil can be relati el high depending on the frequencofthe AE a es. Therefore in soils installationofae guides is important to a oid the signal strength loses in detection. onitoring soils using high frequencyAEtechniques is largel affected b the highIeelofattenuation in soils. Since metalshaea three to four timesofmagnitude10er attenuation than soils metal a e-guides ha e been found ery useful in conducting the signals from within the soil mass to be recei ed b AE sensors. Therefore theusesofa e-guides to monitorAEin soils ha e become commonplace (Dixon 1999). Aa e-guide is a de ice that couples elastic energy from a structureorother test objects to a remotel mounted sensor during AE monitoring. It is possible to dri e the a e-guide into the host soil for short distances. For deeper slopes it is necessary to install a e-guides in pre-drilled boreholes. The latter method requires a backfill material to impro e the contact between the host soil and a e-guide. Depending on the backfill material two possible s stems are a ailable passi e and acti ea e-guide s stems. For passi es stems the annulus around the a e-guide has to be backfilled with10AE acti ity material such as cIa s so the installation does not introduce additional sourcesofAE into the a e-guide.Anrecorded AE signal is assumed to emanate from the deforming host soil itself. Dri en a e-guide s stems can also be defined as passi e as a resultofthe a e-guide being in direct contact ith the in-situ material. Acti ea e-guide s stems are installed hen the monitoring site consistsofcohesi e material. ince the emissionIeels generated in such soils are10it is difficult to obtain strong AE records. Therefore the annulus can be backfilled ith granular soils such as sandorgra el hich produce strongAEsignals and enhance the original signal. Although the recordedAEdata will not relate direct! to the stress stateofthe host soil it rna be possible to calibrate the s stem such that the recorded enhanced AE signal can be related to the magnitudeofthe general ground deformationofthe host soil.4


teel tubes are also used for a e-guides and steel threaded rings are used for connecting sectionsofa e-guides. The choiceofthe steel tube AE a e-guide has the ad ant ages that it can be easil fabricated to the required cross section and length and also possesses a10ultrasonic attenuation coefficient. The steel a e-guide s stem transmits the emission generated b the frictional motionofthe soil particles which are in contact with or close to the a e-guide. Experiments conducted b Dixon (1999) with two backfill types gra el and sand sho ed that the former backfill as noisier under small displacements than the latter. The sand backfill appears to generate10magnitude emission0er a long time scale hereas graelbackfill generates high magnitude emission0er a short time scale. In general the acoustic emission signals obtained using transducers are wide band and consistofbroad rangeofhigh frequenc components and significant time-based characteristics and noise. This condition creates difficultiesinisuall characterizing the AE signals in a meaningful manner. Therefore one has to use a sophisticated signal processing method to analyze the non-stationary AE signals.Inthis regard the a elet based multi-resolution approach pro ides a con enient approach for anal zing transient or time arying signals.1.3 Research ObjectivesThe objecti esofthis research are to1.detect and record the acoustic emission (AE) signals emitted during the shearingofa cohesionless soils (sand)2.anal ze the recorded AE signals using the a elet theory the propertiesofde-noised signals obtained from a elet anal sis with thoseofFast Fourier Transformation (FFT).5


WAVELETTCHAPTER 2 SFORMS APPLIEDTOIDEIFYTHEACO STIC EMISSIO SIG ALS SOILS 2.1Waelet TransformsWhile the Fourier transform deals with transforming the time domain componentstothe frequenc domain and frequenc anal sis the a elet transform deals th scale anal sis. The later approachisknown as multi-resolution anal sis. To approximate chopp signals for man decades scientists ha e anted more appropriate functions than the sine and cosin(base functions) hich comprise Fourier anal sis. cale analSISdeals with creating mathematical structures that pro ide arying time/frequenc / amplitude slices for anal sis. This transform is a portion or one or afecdesofa complete a eform hence the term a elet.Lo\I\rfrequencieA _arebetter ''-"re01edinfrequenTime Highfrequenciesarebetterre01edintimFigure 2.1 Typical Wavelet Transforms6


The a elet transfonn has the ability to identify frequenc (or scale) components and ith their location(s) in a time scale. Additionall computations are direct! proportional to the lengthofthe input signal. In a elet anal sis the scales pIa a major role. a elet algorithms process data at different scales or resolutions.Ifone looks at a signal with a large "windo"oneould notice gross features. imilarlifone looks at a signal with a small "windo"oneould notice relatielsmall discontinuities as shown in Figure 2.1. One a to achie e thisisto ha e short highfrequenc fme scale functions and long10frequenc ones. B defmition sine and cosine functions are non-local stretching out to infmity and therefore do a poorjobin approximating sharp spikes. On the other hand with a elet anal sis one can use approximate functions that are contained neat! in fmite (time/frequenc ) domains. a elets are ell-suited for appro imating data with sharp discontinuities. The a elet anal sis procedure usuall adopts aa elet prototype function called an "anal zing aelet" or "mother a elet." Temporal anal sis is perfonned with a contracted highfrequenc ersionofthe prototype a elet hile frequenc anal sis is perfonned ith a dilated10-frequenc ersionofthe prototype ae1el.Because the original signal or function can be represented in tennsofaa elet expansion (using coefficients in a linear combinationofthe a elet functions) data operations can be perfonned using onI the corresponding a elet coefficients. The flexibilityofthe anal zing a elet is a major difference between the0typesofanal ses and is important in detennining the resultsofthe analsis. The wrong a elet rna be no better (or e en far orse than) than the Fourier anal sis. Hence a successful application presupposes some e pertise on the partofthe user. orne prior kno ledge about the signal must generallbe a ailable in order to select the most suitable distribution and adapt the parameters to the signal. Someofthe most common a elet types are showninFigure 2.2. 7


----'111Figure 2.2 Common Wavelet Functions (Bruce, 1996) 2.2 Comparisonofa elet Anal sis ithFourierAnal sis While a typical Fourier transform pro ides frequenc content information for samples within agien time interval a perfect a elet transform records the startofone frequenc (or e ent)and then the startofa subsequent e ent th amplitude added to or subtracted from the base e ent. The main disad antageofa Fourier transform is that it has on! frequenc resolution and no time resolution hich means that although one might be able to determine allofthe frequencies present in a signal' one ould not kno at hat time the appear.Ina elet anal sis on the other hand the useofa full scalablemodulated windo sol es the signal-cutting problem. The indo is shifted along the signal and for e ery position the spectrum is calculatd.Then this process is repeated man times th a slight! shorter (or longer) windo for e ery ne ccleoInthe end the result'11be a collection oftime-frequenc representationsofthe signal all with different resolutions. Becauseofthis collectionofrepresentations one can speakofa multi-resolution anal sis.Inthe caseofa elets one normall does not speak about timefrequenc representations but abouttime-scale representations.8


2.3TheContinuous WaveletTransformathematicall the processofFourier anal sis is represented b the Fourier transform:cof (w)=Jf(t)e -jwtdt-co(2.1) Which is the sum0er the entire time domainofthe signal f(t) multiplied ba comple e ponential hich contains sine and cosine functions. The resultsofthe transform are the Fourier coefficientsf()hich hen multiplied b the corresponding sinusoidoffrequenc ield the constituent sinusoidal componentsofthe original signal. Graphical I the transformation process can be shown as in Figure 2.3.-'-"...,'"',I.'"...''.:...:',:..,',I.I"III.:.........:',.,."Fourier...,.I"ITransfonn",:':/'0..,...,"I II ---,-'-,, ,"',.'to.....:-:'"I..I',:'.'I,I,It,tI'"II'."'",,.,'"-It""I,I,I,.:-:':-:...'"SignalConstituent sinusoidsofdifferent frequenciesFigure 2.3 illustrationofFourierTransformsimilarl the continuous a elet transform (CWT is defined as the sum0er the entire timeofthe signal multiplied b scaled shifted ersionsofthe a elet functionwet)(Equation (2.3) The resultsofthe CT are man a elet coefficients Ct(Equation (2.2)) hich are functionsofscale and position. When using a real functionC.,=If(t)'!',(t)dt92.2)


(2.3)Wt.sis the mother a elet used for scaling and translation. The ariables s and-rare scale factor and translation factor respecti el The primary requirementsofthewet)are that the integralofthe a elet0er the entire time domain must be zero (Equation (2.4)) and square integralofwet)must satisfy the admissibility condition (Equation (2.5)).cof\}J(t)dt=0 (2.4)'(1)1dt (00(2.5) Shifting aa elet simpl means dela ing or hastening its onset. athematicall dela ing a function f(t) b-ris represented b f(t--r). Figure 2.4 illustrates the a eletwet)and shifted a eletW(t--r).'(t-r+tWayatfunctionShiftedwaveffunctionFigure2.4EffectofShifting ultipl ing each coefficient b the appropriatel scaled and shifted a elet ields the constituent a eletsofthe original signal f(t) (Equation (2.6)). f(t)=ffCs .t\}JS,t(t)dsd'[10 (2.6)


2.4 Scaling Scaling allo s one to either narro down the frequenc bandofinterest or determine the frequenc contentina narro er time interval. s illustrated in Figure 2.5 scaling aa elet simpl means stretching or compressing it. athematicall this can be achie ed b introducing a scale factor often denoted ba.For example the effectofthe scale factor on sinusoids is seen in Figure 2.6.,.,...\.oIII'.'..........I..III.,"Wavelet+.':.:Transform...+Signal Constituent wall8/etsofdifferent scalesandpositionsFigure 2.5 lliustrationofWaveletTransformsI --.........c/"{(O=sin(O:a=1'.......-I-..............-0s1-/'-'"'jct'/(t)=sin2t);a=....."--'",.;--1 -..:!1-,/"'\: J,I\,./(t)=sin(4t);a=,-'1-0Z6CFigure 2.6 EffectofScale Factors on Sinusoids The scale factor a is applicable to a eletsina similara.The smaller the scale factor the more "compressed" the a eletis.11


1a=.4ft)=",(4t);ft=",(fa=1It:-o9t)4:-00ft",(2t1=a=2.11()..:.-Figure 2.7 EffectofScale Factor on a eletsItis clear from Figures 2.6 and 2.7 that, for a sinusoid sin(cotthe scale factor is in ersel related to the radian frequencco.irnilarl related to the frequencofthe signal. th a elet anal sis the scale is2.5 SequenceofSteps to Obtain a Continuous Wa elet Transform (CWT)As shown in Equation (2.2) the continuous a elet transfonn is the sumofthe segmentsofthe original signal multiplied b scaled shifted ersionsofthe a elets0er the entire time domainofthe signal. This process produces a elet coefficients that are functionsofscale and position. The following are the steps to be folIo edincreating a CWT:1.Consider agien a elet and compare it to a section at the startofthe original signal.2.Calculate a number C that represents ho closel correlated the a elet is with this sectionofthe signal. The higher C is the higher the correlation is. More preciselifthe signal energy and the a elet energ are equal to one C rna be interpretedasa correlation coefficient.Itmust be noted that the results will depend on the shapeofthe a elet one chooses (Figure 2.2).12


Sgne.1VI"cVVWV\r-2Figure 2.8 Comparisonofa Wavelet with a Segmentofa Signal3.hiftthe a elet to the right and repeat steps 1 and 2 until one has co ered the entire signal (Figure 2.9).SignaJWavelet..Figure 2.9 ComparisonofShiftedWaelet with Signal4.Change the scale (stretch)ofthe a elet (Figure 2.10) and repeat steps 1 through4.5.Repeat steps 1 through 4 for all selected scales.--..C=0.2247Figure 2.10 ComparisonofScaled Wavelet with Signal13


When the entire process is complete one has the coefficients produced at different scales b different sectionsofthe signal. The coefficients represent the resultsofregressionofthe original signal performed on the a elets. 2.6 Relationship Between ScaleandFrequencThe higher scales correspond to the most "stretched" a elets. The more stretched the a elet the longer the portionofthe signal with hich it is being compared and thus the coarser the signal features being measured b the a elet coefficients.+SignaJWavelet(a)Lowscale(b)HighscaleFigure 2.11 Different ScaleWaelets Thus the following correspondence e ists between a elet scales and frequenc asreealed by a elet anal sis: Loscale corresponds to a compressed a elet.Itis useful for rapidl changing details and high frequenc(co). High scale corresponds to a stretched a elet.Itis useful for slo I changing coarse features and10frequenc(co).2.7TheDiscrete WaveletTransformCalculating the a elet coefficients at e ery possible scale using continuous a elet transform (CWT) is a tedious exercise that generates excessi e data. To o ercome this problem and get a manageable countofdata discrete a elets ha e been introduced. Discrete a elets canonIbe scaled and translated in discrete steps. This can be achie ed b modifying the mother a elet representationinEquation (2.3) to Equation (2.7).14


\}I(t)=_1_\}I(t-kTotJ1.kG1VSo0(2.7) Wherejandkare integers andSo>1)is a fixed dilation step.Ingeneral the translation factor('to)depends on the dilation step. common choice forSoand'toare 2 and 1 respectiel(Equation (2.8 hich lead toadadic sampling grid. (2.8) Then the discretized a elet transform (D T) pair can be gi en in Equation (2.9).d 1.k=fJ(t).\}I1ok(t)dt(2.9 Where d j k are the discrete a elet transform alues gi en on a scale-location gridofindicesjk.For the discrete a elet transform the aluesdj,kare known as a elet coefficientsordetail coefficients.Anarbitrary signal f(t) can be reconstructed b summing the orthogonal a elet basis functions eighted b the a elet transform coefficients djok.CQJ(t)=LLdjk\}I1k(t)1=--Jk=--J(2.10 Orthonormal d adic discrete a elets are associated th scaling functions and their dilation equations. The scaling functionis associated th smoothingofthe signal and has the same form as the a elet gi en b (2.11)15


The ha e the propertycofo,o(t)dt=1-co(2.12) Whereo,o(t)=(t) is sometimes referred to as the father scaling function or father a elet. The scaling function is orthogonal to translationsofitself but not to dilationsofitself. The scaling function can be con01ed th the signal to produce appro imation coefficients as folIos:(2.13) A continuous approximationofthe signal at scalejfj(t)) can be generated b summing a sequenceofscaling functions at this scale factored b the appro imation coefficients as folIos:coft)=Lajokjok (t (2.14)k=-


scaling function spectrum(

1GJ}FiltersIIhlgh-passFigure2.14OneStageFilterInFigure 2.14 the original signal S passes through0complementary filters and emerges as two signals. nfortunatelifone actuall performs this operation on a real digital signal th 1000 samples then the resulting signals will each ha e 1000 amples for a totalof2000(Figure 2.15). B observing the computation carefull one rna keepon!one data point outof0in eachofthe01000-length samples to obtain the complete information. This is the notionofdown sampling in hich one produces o sequences called cA and cD (Figure 2.16). These are also called discrete a elet transform coefficients.-1000samQIJ::i1000samples-1000samplesFigure2.15OneStageFilterwithThousandSamples19


CD-troosamples-500wets-500wetsFigure 2.16 Down Sampling As an example a one-stage discrete a elet transformofa signal will be a pure sinusoid ith high-frequenc noise added toit.A schematic diagram th real signals inserted into it is showninFigure 2.17.cD High FrequencyS-mowooZI\IV\'emdatapodscA Low FrequencyLM-CD-f\M-500DWTcoefficientsFigure 2.17 Schematic Diagram with Real SignalsIti noticed that the detail coefficients cD are small and consist main!ofahighfrequenc noise hile the approximation coefficients cA contain much less noise than does the original signal. One rna observe that the actual lengthsofthe detail and approximation coefficient ectors are slightl more than half the lengthofthe original signal. This has todowith the filtering process hich is implemented b con01ing the signaltha filter. Thus the con olution "smears" the signal introducing se eral extra samples into the result.20


2.9 ultiple-Le el Decompositiona elet transform is constituted b differentIeels. The major factor that affects the numberofIeels one can reach to acme e the satisfactory noise remo al results is the signal-to-noise ()inthe original signal. Generall the signals from the piezoelectric Therefore to process the acoustic emission data, one needs moreIeelsofa elet transform to remo e mostofits noise. The decomposition process can be carried out so that one signal is broken down into man10er resolution components. This process is called the multi-Ieeldecomposition and the product is termed the a elet decomposition tree.Figure 2.18 ultiple Level DecompositionThe a elet decomposition treeofa signal can ield aluable information regarding the signal. oise in the signal is captured in the detail coefficients particularl in the small coefficients at higherIeels in the decomposition. B zeroing or shrinking these coefficients one can get smoother reconstructionsofthe input signal. This is done b specifying a threshold alue for eachIeelofdetail coefficients and then zeroing or shrinking all the detail coefficients belo this threshold alue.21


Figure 2.19 ultile el FilteringThe coefficients cAlcA2andcA3shown in Figures 2.18 and 2.19 are the approximate coefficients computed using the a elet anal sis atIeel1Ieel2andIeel 3 respecti el imilarl the coefficients cDIcD2and cD 3 are the detail coefficients computed using the a elet anal sis atIeel 1Ieel2andIeeI3respectieI.Ithas been e plained abo e ho the discrete a eIet transform can be used to anal ze or decompose signals. This process is called the decomposition or anal sis. The otherhalfofthe story is ho those components can be assembled back into the original signal without lossofinformation (Figure 2.20). This process is called the reconstruction or s nthesis. The mathematical manipulation that effects synthesis is called the in erse discrete a elet transforms (Equation 2.15).Figure 2.20 ReconstructionofSignal22


This chapter e plains both continuous a elet transform and discrete a elet transform. Ho e er due to large amountofdiscrete and non-stationary AE data obtained in this stud a fast algorithm is needed to anal ze the data. Therefore the particular techniqueofdiscrete a elet transform (D T is selected for acoustic emission data anal sis in Chapter4.2.10 umerical Example 2.10.1 The Forward TransformThe forward Haar a elet transform can be described in termsofa eraging and differencing adjacent pairsofdata alues at a sequenceofgeometricall increasingIeels. The computationofa erages (appro imate coefficients) and differences (detail coefficients) can be obtained using Equations 2.9 2.14 2.16 and 2.17. Haar mother a eletlj/{t):111lj/{t)=2-1Haar scaling function(t):ot(1otherwise2.16) (2.17) Practicallifa data set{soSI...S-I}contains elements there ill be /2 approximate coefficientsaj(a erages) and /2 detail coefficients dj(differences) Addition 2002). The a erages become the input for the nextIeel in the decomposition calculation. The recursi e iterations continue until a single a erage and a single coefficient arecalculated. One can deri e the Equations (2.18) and (2.19) using the Equations (2.9) (2.14) (2.16) and (2.17) to compute the coefficients from an odd and e en element in the data set.23


5,+1+1a---i-2d=5,-11I2WhereSjandSj+1are the ithand (i+ 1thelementsinthe data set respecti elTable 2.1 A Sample Acoustic Emission Data(2.18) (2.19) Sample o. 1234567 8 Signal (mV) 37 35 28 28 58182115Table2.1sho sa sample data set consistingofacoustic emission signals in mY. Considering the sample data from Table2.1and using the Equations (2.18) and (2.19) the appro imate coefficients(ajand detail coefficients (dj) are computed as folIos:First select0adjacent odd and e en elements. Let sj=37 andSj1=35sing Equation (2.18) and the selected dataofSjandSj1 the approximate coefficient (aj) is computed as folIos:= 37 + 35 = 36ai2Similarl using the Equation (2.19) and the selected dataofSjandSj1 the detail coefficient (d j ) is computed as folIos:d= 37 -35=1i224


Table 2.2 represents the completed transformationofthe data set in Table 2.1. The first ro in Table 2.2 is the original signal. The second ro in Table 2.2 is generated b using Equations (2.18) and (2.19). The coefficients computed using Equation (2.18) are written in the first four places and the coefficients computed using Equation (2.19) are written in the remaining four places consecuti el Computations are repeated on the appro imate coefficients bile the detail coefficients are retained in each step.Table 2.2 Complete Transformationofthe SignalSignal (mV) 373528 2858182115Le el(1)36 28 38181 0 20 3 Leel(2) 32 284101 0 20 3 Le el (3) 30 24101 0 20 3Appro imate Coefficients (aj) Detail Coefficients (dj )2.10.2 The Inverse TransformOne can deri e the Equations (2.20) and (2.21) using the Equations (2.15) (2.16) and (2.17) to reconstruct the original and truncated signals. s shown in Equation (2.20) the ithapproximate coefficient(aj)and the ithdetail coefficient (d j ) are added to reconstruct the ithdata(Sj)in a particularIeel. Theseajand d j are computed using the forward transform described in ection 2.7. (2.20) (2.21) Considering the transformed data inIeel(1)from Table 2.2 and using the Equations (2.20) and (2.21) the original dataSjandSjIare computed respecti el For e ample select an approximate coefficient (aj) and the corresponding detail coefficient (di )as folIos:25


sing the Equation (2.20) and the selected coefficientsofaiand djthe original data(Sj)is computed as follos:= 36+1=37Similarl using the Equation (2.21) and the selected coefficientsofajand d j the original data(Si+l)is computed as follos:,I=36-1=35 From these sample calculations one can understand ho the Haar a elet transforms ork in casesofdecomposition and reconstruction.Bcontinuing this decomposition until a single approximate coefficient and a single detail coefficient are calculated one can obtain the complete transformationofthe sampledata as shown in Table Thresholding Decomposed SignalThe processofthresholdingiscarried out on detail coefficients. There are man methods for setting the threshold. The most time-consuming a is to set the threshold limit on a case-b -case basis. The limit is selected such that satisfactory noise remoalis acme ed. For a particular threshold alue the detail coefficientsinTable 2.2 hich are less than or equal to that particular threshold alue are replaced b zeros. Consider a thresholdof2 for this transform. Thus detail coefficients hich are less than 2 or equal to 2 become 0 as shown in theIeel3ofTable 2.3. These updated alues are shown in blue-shaded cells. Once the thresholding is completed one can reconstruct the signal using Equations (2.20) and (2.21) from the coefficients in theIeel(3T)ofTable 2.3. 26


Table2.3ReconstructionofDataUsing aThresholdValue2320 3 Le el (3T) 30 410Leel2) 30 30 410Le el (1) 34 26 40 20 Filtered 34 3 26 26 signal (mV)o o60oo20 20 20233 317Repetitionsofthe abo e reconstruction procedure 'th threshold alueof34 and10ield results showninTables 2.4 2.6.Table2.4ReconstructionofDatasing aThresholdValue 3 3 Set Zeros with a Threshold Value 3 Leel(3T) 30 4 10 Le el (2) 3030 41000 20 0 Le el (1) 34 26 40 20 0 0 20 0 Filtered 34 34 26 26 60 20 2020 signal (mV)Table2.5ReconstructionofDataUsing aThresholdValue4 3 Le el (3T) 30 Le el (2) 30 30 0 10 00 20 0 Leel(1) 30 30 40 20 0 0 20 0 Filtered 30 303030 60 202020 signal (mV) 27


Table 2.6 ReconstructionofDatasing a Threshold Value 103(JI ,--,: -()'(,:I! :j -, -Leel(3T) 30 Le el (2) 30Leel(1)30 Filtered 30 signal (m ) 30 30 30o30 30o30 30oo50oo1020 20 20 30oo30 Finall Figure 2.28 graphicall represents the original signal (Table 2.1) and the filtered signals obtained from Tables 2.3 2.6 with different threshold alues.80105o60-f----+-----!---...;------+---i"f---+-----O----+---+-------;ample umberFigure 2.21 PlotsofOriginal and Filtered SignalAs shown in Figure 2.21 the smaller threshold alues 23 and 4 smoothen the small peak alues with no significant change in the ae pattern. Also no significant change occurs on the maximum peak alueoforiginal signal. On the other hand the threshold alue10not onI changes the peak alue significantl but also changes the 28


ae patternofthe original signal. From these results one can conclude that the selectionofan appropriate threshold alueisa ital stepinthe filtering process.29


CHAPTER3EXPERIMEAL SETUP AND DATACOLLECTIO3.1 IntroductionThis section describes the detailsofthe different experimental setups and shearing conditions under hich acoustic emission data as collected.3. 2 Shearing ProcedureA12x12x16ooden shear box (Figures3.1and 3.2) as constructed for inducing direct shear on a sand sample. To maintain a constant rateofshear an electric screjackas first attached to the upperhalfofthe shear box (Figure 3.1). The operationofelectricjackcreated noise hich made it difficult to differentiate the actual AE signals due to the shearing from the total AE signals that contained the surrounding noise.Inorder to a oid this problem the electric screjackas replaced ba manuall operated screjack(Figure 3.2). Approximate constant shearing ratesofone cdeofre olutionofthe screjackper enty seconds and one cdeofreolutionofthe screjackper ten seconds ere respecti el used in slo and fast shearing modes.Figure 3.1ShearBox with Electric ScrewJack30


Figure 3.2ShearBox withanualScrewJackFigure 3.3 Universal MicroTribometer Figure 3.4 Mote Acoustic Emission Sensors with Circuit Board31


The following steps ereused for the data collection process: I. The bottom partofshear boasfilled ith sand placed in successi e la ers. Each la er as compacted using surface ibration.2.The sensoraslaid just belo the shear plane. Then sandasfilled up to the topofthe shear box th frequent surface ibration. 3. Approximatel 80 lbsofsurchargeasapplied on the sand.4.The built-in sensor circuits ere remo ed from theuniersal icro Tribometer (Fig. 3.3) and the externalAEsensor circuit board(Fig.3.)asconnected to the abo e machine.5.To collect continuousAEsignals the T-2 s stem softwareasset up.6.The ertical pins from the shear bo ere remo ed and then the manual scre jackasoperated ith an approximate constant rate.7.TheAEdata sa edinthe computerasdownloadedaste t files. 3.3 Data Acquisition T0different data acquisition s stems ere emplo ed to acquire the acoustic emission signals.Inthe first data acquisition s stem the following instruments ere used:I.PCS64i Oscilloscope (Figure 3.5)2.idebandAESensors (Figure 3.6) 3. ooden Shear Box (12 x12'xl6) (Figure 3.2)4.Computer Becauseofthe following difficulties encounteredinthe abo e experimental setup an alternati e (second) option as selected to complete the task.I.oise created b instrumentation itself2.oise created b connectors 3. Unsupporti e software platform (DOS) to record the transient signals4.eed for a higher triggering alue 32


Inthe second data acquisition s stem the following instruments ere used:1.Uni ersal cro-Tribometer (Figure 3.3)2.oteensor th circuit board (Figure 3.4) 3. ooden ShearBox(12' x12x16)(Figure 3.2) 4. Computer3.3.1 PCS64i OscilloscopeInthe first experimental setup A PCS64i digital storage oscilloscope (Figure 3.5) as used to observe the signals hich ere obtained from two different typesofwide band sensors designed for acoustic emission monitoring.Theare D9204 and ISR3 bought fromPhsical Acoustics Corporation. During the first stageofthe e perimental process along with the PCS64i digital storage oscilloscope and the abo e two typesofacoustic emission sensors a band pass filter the ooden shear box (Figure 3.2) and a computer th icrosoft indo s 98 platform ere used. ThePC64i is a digital storage oscilloscope hich uses a compatible computer and its monitor to displa acquired a eforms. All standard oscilloscope functions are a ailable in the supplied DOSorindo s programs. Its operation resembles thatofa normal oscilloscope with the difference that most operations in the former can be performed using the mouse. Apart from its use as an oscilloscope the unit can also be used as a spectrum analzerup to 16MHzand also as a transient signal recorder for recording oltage ariations for the comparisonoftwo oltages0era longer period.Theoscilloscope and transient recorder operations can be performed on two completel separated channels with a sampling frequenc up to 32MHzin real time.(a) Front (b) Rear Figure 3.5 PCS64i, Digital Storage Oscilloscope33


3.3.2 Acoustic Emission Piezoelectric SensorsTheAEsensor is a detection de ice such as a piezoelectric transducer that transforms the particle motion produced b an elastic ae into an electrical signal. The piezoelectric sensor consistsofa piezoelectric element encapsulated in e truded aluminum housing. ince piezoelectric sensors are the most appropriate forAEtesting irtuall all acoustic ae de ices and sensors use piezoelectric material. The are also robust and more sensiti e thanother sensing techniques. Since the AE signals are relati eleaka preamplifier is generall connected to the sensor to minimize the noise interference and pre ent the signal loss. Sometimes the sensor and the preamplifier are built as one unit. The acoustic propertiesofthe medium here the measurement is made are ery important in the design and selectionofappropriate transducers. Transducers must also thstand the se ere effectsofeather changes biological acti ity h drostatic pressure and extreme temperature conditions. AE sensors are either single-ended (S)ordifferential(D)in construction. The single-ended design emplo sa single crystal to pro ide high sensiti ity and omni directional response to the AE excitation regardlessoforientation. The differential designs pro ide common mode rejectionofun anted signals in en ironmentsofhigh electromagnetic interference. Common-mode rejection is the abilityofa balanced (or differential) input to reject the partofthe incoming signal hich has the same amplitude and phaseonboth input terminals referenced to ground. Hence this is the ability to respond to onI differences at the input terminals. The sensors used in the first experimental setup are deband differential and piezoelectric with a integral cable (Figure 3.6). ideband sensors are typicall used in research applications or other applications here a high fidelity AE response is required.Inresearch applications here frequenc anal sisoftheAEsignal is required debandAEsensors are useful in helping determine the predominant frequenc bandofAEsources for noise discrimination and selectionofa suitable10ercost general purposeAEsensor.Inhigh fidelity applications ariousAEa e-modes can be detected using 34


ide band sensors thus pro iding more information about theAE source and distanceofthe AE e ent. In the second experimental setup an AE piezoelectric mote-sensor as used in conjunction ith the ni ersal icro-Tribometer (Figure 3.3). The later has been manufactured to monitor the nanomachining process including the delamination defects.Figure 3.6 WideBandAcoustic Emission Sensors 3.3.3 niversal Micro-Tribometer(UMT)This multi-specimen test s stem (Figure 3.3) has been de eloped for stud ing the chemical mechanical planarization (CMP). It is equipped with acoustic emission (AE) and coefficientoffriction (CoF) sensors and the required data acquisition s stems. During the current experimental process these built-in AE and friction sensors ere disconnected from the data acquisition s stems and an external AE sensor ith the circuit board Figure 3.4 as connectedtothe dataacquisition s stems.Ineach trial AE data ere collected and downloaded in the text fonnat. Se eral tests ere completed using the abo e test setup ith horizontal and ertical placementofthe sensor as ell as different ratesofshearing. Table3.1pro ides the designations used to identify different testing conditions.35


Table 3.1 Test Designations PlacementofSensor RateofShearing Test Designation FastHFlHF2 HF3 Horizontal Placementof10H1 HS2 ensor0hear S Fast VFI VF2 Vertical Placementof10VSI V2 Sensor ormalVNo Shear S 36


CHAPTER 4ALYSISOFDATA4.1 IntroductionThis chapter describes the methods used to anal ze the acoustic emission (AE) data collected from ni ersal icro Tribometer T) (Chapter 3). The conceptsofa elets and Fast Fourier Transform are the two primary methods used for data processing. First the a elet theory as used for de-noising the acoustic emission signals and then thede-noised signals ere compared th the similarl de-noised signals obtained using Fourier anal sis. Data obtained from the AE mote-sensor as sa ed in the computer and then downloaded into text files. These text files aregien in the AppendixA.The downloaded text format data are transformed into Excel spreadsheets and used for the anal sis. For signal processing TLAB built-in functions with codes de eloped b the author ha e been used. ostofthe common mother a elet functions (Section 2.1) used for data anal sis with differentIeels are built-in ith the TLAB a elet toolbox. Data ere anal zed using discrete a elet transform ith Daubechies mother a elet and scale functions (Addison 2002). Figures4.1to 4.3 are the sample plotsofdownloaded data obtained under different conditions. Figures4.1and 4.2 illustrate the plotofacoustic emission with noise ersus time obtained with the horizontal and ertical placementofthe mote-sensor during the processofshearing respecti el Figure 4.3 sho s the data obtained under thesame experimental setup with no shear condition. These plots are compared in detail in ection 4.4. 37


PlotofDataobtainedinHFI with Horizontal Placementofote-nsoro102030Time(s)4060Figure4.1Data Obtained with Horizontal Placementofote-Sensor (Shearing Rate: Fast) HFIPlotofData Obtainedin\\;th ertical PIa ernentofote-n or 120 100 8060Time(s)4020;;0.0oZ0-0.0<-0.1+----+--iHillt-t-ttt--ttt-tt--t+-iHIIt-++---tt-tt--+--i-----t-t--t-t---+------ft++f-if-0.15+------.---------r-------.-------r---------r--------ioFigure 4.2 Data Obtained with Vertical Placementofote-Sensor (Shearing Rate: ormal) -PlotofDataObtained tmder0Shear Condition604030Time(s)2010o-0.1o0z-0.05<:_ 0.05GFigure 4.3 Data Obtained o Shear Condition38


4.2 seofthe MATLAB Wavelet ToolboxforAcoustic Emission Signal Processing The a elet Toolbo extends the TLAB technical computing en ironment"thgraphical tools and command-line functions for de eloping a elet-based algorithms for the anal sis de-noising and compressionofsignals. a elet anal sis pro ides more precise information about signal data than other signal anal sis techniquessuch as the Fourier transforms. The a elet Toolbox supports the interacti ee plorationofa elet properties and I-D and 2-D applications in communications and geoph sics. The a elet Toolbox supports a full suiteofa elet anal sis and synthesis operations listed belo : Achie e high ratesofsignal or image compression with irtuall no lossofsignificant data Restore nois signals Disco er trendsinnois or faulty data Stud the fractal propertiesofsignals Extract information-rich features for useinclassification and pattern recognition applications The following figures illustrate the major steps folIo ed during the data analSISusing MA TLAB. The startup windo is showninFigure 4.4.LaunchPad0@?CurrertDirectory:IC:-..cATLA86p5\wortcCommandl4ndow0@U:5inqToolboxPathtorITo"KATLABHelp"fro.theHelp.enu..Current DirectoryCommandHistOl'"YFigure 4.4StartupWindowofthe TLABProgramto Select Wavelet Tool Box 39


S-rdT110peclaIze00S-SWTOe-noi-g1-0DensityE.-.-1-0Regression E-lion1-0WaveletCoefficiesSelection1-0 uous Wavelet 1-0____IIWavelet Packet2-D,SpecializedTools2-DSWTDe-no"-g2-DWaveletCoefficiesSelection2-D__II1----Ex-te......Figure 4.5 Selection Window for I-D and 2-D Continuous and Discrete SignalsAfter selecting the a elet tool box the windo in Figure 4.5appear on the screen. From this windo one can ha e two selections for 1D with different graphical options such as a elet I-D and a elet packet I-D. The data must be in the mat format in ordertoimport them into the MATLAB tool box. Using the TLAB codesdeeloped b the author the data transferred into the e cel sheet is first changed into the mat format.Inthe literature se eral a elet basis functions such as Haar s Daubechies Coiflets and Symmlets etc. are a ailable. One should select the mother a elet carefull to approximate and capture the transient spikesofthe original signal. Mother a eletnoton!determine ho ell one estimates the original signal in termsofthe shapeofthe spikes but also it will affect the frequenc spectrumofthe de-noised signal. Someofthe desirable propertiesofthe basis functions aregood timefrequenc40


localizations arious degreesofsmoothness and large numberofanishing moments. Although the Haar a elet algorithm has the ad antageofbeing simple to compute with compact support and easier to understand it does not ha e better time frequenc localization. Further it is unsuitable for representing classesofsmoother functions due to its discontinuities. The Daubechies algorithm has a slight! higher computational o erhead and is conceptuall more complex. Since the transformation using the Daubechies mother a elet0erlaps be een the iterations that0erlap allo s the Daubechies algorithm to pickupdetail that is missed b the Haar a elet algorithm. The most widel used a elet is the Daubechies basis function. The Haar s filterisbest suited to represent step signals or piecewise constant signals hereas the Daubechies filter is better for smoother signals.-F-Way."tfunctionpsi-10.5-4.5Figure 4.6 PlotofDaubechies other Wavelet Function (db5)Figures 4.6 to4.11illustrate a sample test data processing procedure using the abo ementioned graphical options.Inthis signal processing the Daubechies mother a elet (db5) and scale functions withIeelfie are used.41


ScloIIngfuncdonpHE--I-.....,b...--=\U,-....."'"e.zFigure 4.7 PlotofDaubechies(5)Scale Function Figure 4.8 Loaded Original Acoustic Emission Signal ObtainedinHS2 with the Horizontal Placementofote-Sensor42


sing the dialog bo shown in the Figure 4.8 one can select the a elet mother function and the numberofIeels required to anal ze the signal. Once the anal sis is completed using the selected mother a elet the same dialog box can be used for the de noising process. sing different displa modes one can ie the different modesofsynthesized signal to define the threshold alues. Defining an appropriate threshold alue is oneofthe crucial stepsin01ed in the de-noising process. Therefore the user has to carefull reiethe synthesized signal and adjust the threshold alue in order to omit the lossofan useful information.DNT:WaveletTree0.10.05o-0.1Signal2000400060008000100001200014000Approximationatlevel5(reconstructed).2000400060008000100001200014000.,.....,...:[1-..a.-Figure 4.9 Wavelet Tree and Approximation at Level Five Obtained for HS2 with the Horizontal PlacementofMote-SensorFigure 4.9 is oneofthe built-in displa windo s used to erify the approximate plotofthe signal atIeelfie.B changing the numberofIeels and the mother a elet one can refme the approximate s nthesized signal based on judgment. De-synthesisofthe signal (s) intoajand d j coefficients (Figure 4.9) is described in detail in Chapter 2 (Section 2.7).43


Signal andApproxlmatlon(s)....,....cfsLMeICoats,Signal and DetaiJ(s)54 3210.1o-0.1L..-_-----""'-----'-..:...:...:..:----'-'-=_-.....J10000 15000s0-0.10.10G4135-0.02-O.G4____3 4:1-0.050. 0513 20-0.05____0. 051330-0.055000Figure 4.10 Approximate and Detail Coefficient Plot for Test 5 with the Horizontal PlacementofMote-Sensor The options shown in Figure 4.10 can be used to obtain the compressed and denoised signals. 44


..(15524)r-..-::---,55s ........'Ft.........r...('.....S........1buduIeUnsc*e4.....Ln1111S'"51...-!..I4f13.!.f3f13jJ1JLd2J.........................1R.....IFigure 4.11 PlotsofDetailCoefficients with Deoised Signal for HS2Finall Figure4.11is used to adjust the threshold alues during the de-noising process. One can adjust the threshold alue b observing the plots for detail coefficientsinorder to anticipate with the rapid ariations. Figure 4.12 illustrates the de-noised signal obtained using TLAB tool box with ae1etanal sis.45


0.10.05-.'0.05.1Figure 4.12 PlotofDe-noised Signal for HS2 4.3 Fast Fourier Transformation UsingMATLAB As partofthe objecti esofthis research the AE signals ere anal zed using Fast Fourier Transforms and compared'ththe corresponding anal sis obtained from a elet transfonnation. TLAB built-in codes and codes de eloped b the author ha e been used to complete task. The following steps ere folIo ed in the Fast Fourier Transformation (FFT):1.Export the signals into Excel spreadsheet and rearrange them into column ector2.Load the ector using the TLAB programming language, 3, Obtain the FFTofthe loaded data(X)and obtain the resultsofthe transformed complex ector4.Calculate the magnitudeofthe complex ector5.Plot the estimated magnitudesofthe complex ector ersus frequenc (Figure 4.13). 6. From the plot in step 5 identify the range hich should be remo ed from the transfonnation.46


7.Replace the selected rangeofthe comple ectors ith zeros.8.Obtain in erse Fourier transform (Z).9.B obtaining the real partofector (Z) reconstruct the de-noised signal (Figure 4.14). Among the steps described abo e the step 7 is the most important one. Using FFT a signal is transformed from the time domain to the frequenc domain.Inthe frequenc domain the10frequenc components contain mostofthe information (approximation). This is shown b the large peak in the amplitude en elope (Figure 4.14) calculated from the transformed signal. On the other hand the background noise from surrounding is ide band dominated b high frequenc components. This is shown b small peak in the amplitude en elope(Figure 4.14) calculated from the transformed signal.Ifthe signal does not contain an background noise the amplitude en elopeofhigh frequenc componentofthe signal is ery close to zero. Therefore the selectionofan appropriate rangeofthe transformed ector using Figure 4.14 to introduce zerosina certain frequenc range is a crucial step in order to omit the lossofuseful information. Po er spectral density (Figure 4.13) is oneofthe useful methods to select the dominant frequenciesofa signal...SpectrumViewerGJ(Q)rg)II I IIrII I15.51'6172q,;.3!584651e-006PSD21.50.5P.amebn-:;::::::;::====;11MelhodJmiJ14096Ilnheri&om8Ea,tl.XI.H4S9'lat HFl----[467J.b-1realFigure 4.13 Sample PlotofPowerSpectral Density Versus Frequency47


lagnitudeof20,...,.------,------,----r------,-----.......-------,-----r-------,16....;141210Small PeakFrequenc.(HzFigure 4.14 agnitudeofTransfonnedComplex Vector Versus Freqoenc (Identity Test HS2)0.05 120Figure 4.15 Sample Filtered Signal singFFT(Identity Test HS2)48


4.4 ComparisonofFilteredSignals Obtained sing Wa elet with thatofFFTAll the test data obtained from shearing the soil ith the horizontal and ertical placementofAE mote-sensor are de-noised using both the a elet transform and FFT. Figures 4.16 to4.21illustrate the comparisonofde-noised signals obtained with horizontal placementofsensor and Figures 4.24 to 4.29 illustrate the comparisonofde noised signals obtained th ertical placementofthe sensor.Denoised ignal iog FFT0.05 0.030.01-0.0110202530Tim()34045055Figure 4.16 De-noised Signal sing Wa elet and FFT (TestHFl)Denoisedignaliog FFT4030Time()20100.08-r--------------------------------,0.06+--------------_____1"----------------10.04-f--------------.,---___+--------------__10.02,.-0.02-0.04#-----v----9-""'-----+t'+l--------------------------i-0.06oFigure 4.17 De-noised Signal Using Wavelet and FFT (Test HF2)49


avelet and FFT(festHF3)DenoisedSignalingFFT0.08 0.06 0.040.02t:..l0<-0.02 -0.04 -0.06 01020Figure 4.18 De-noised Signal4050Figures 4.16 to 4.18 illustrate the de-noised signals obtained with a relati el fast rateofshearing th horizontal placementofsensor. ignals de-noised using FFT and a elets folIo the same pattern.Inall three tests the maximum positi e peak AE alue occurs within the fIrst35secondsofthe testing period. Also the maximum peak alues are higher than 0.5V.100 avelet80 60 40 200.00.03+-----------------/1--------------------10.02+---J1r---:-------------dtT------------------I,....0.0 12:-<-0.0 1-0.02+---------t-------------t---+-----'-------I-0.03+---------t-----------------------I-0.04-t---------r------r--------...-------.-----------.---JoTime ()Figure 4.19 De-noised Signal Using Wavelet and FFT(festHS1)50


100806040200.0-0.04+------------------if------------------t0.03t---------------ttfttt-----------------f0.020.01<:-0.01-0.02-0.03+-------------------------t--------1-0.04+--------r-------,r-------r--------r--'------r--......oFigure 4.20 De-noised SignalFigures 4.19 to 4.20 illustrate the de-noised signals obtained th a relati el slo rateofshearing th horizontal placementofsensor. ignals de-noised using FFT and a elet folIo the same pattern.Inboth tests the maximum positi e peak AE alue occurs after 40 secondsofthe testing period. Also the maximum peak alues are higher than 0.03V.,

HFI-HF2-HF3-HII0.08 0.06 0.04 0.02<0 -0.02-0.04-0.06 01020304060 Time(5)Figure 4.22 AE Signals Obtained with Horizontal Placementofthe Sensor Filtered sing a elet Analysis-HFI-HF2-HF3-HII0.08 0.06 0.04E0.02"<0 -0.02 -0.04 -0.06 01020 30 40Time(5)Figure 4.23 AE Signals Obtained with Horizontal Placementofthe Sensor FilteredsingFFT52


0.06.,----------------------------,120 1008060o20-0.04oFigure 4.24 De-noised SignalIIUaI.1IJf.di1.,INlJJIjrII.i..",nr.,'Ilr"0.05 0.04 0.03 0.02>':"' 0.01.......0.01

Figures 4.25 to 4.26 illustrate the de-noised signals obtained ith a relati el slo rateofshearing with ertical placementofsensor. Signals de-noised using FFT and a elet folIo the same pattern.Inboth tests the maximum positi e peak AE alue occurs after 60 secondsofthe testing period.Also the maximum peak alues are higher than 0.04V.0.04 0.02

0.05 0.04 0.03 0.02s:-0.01"0

-I-2-VFI-VF2-Q.0310090807060Time ()40302010-Q.04oFigure 4.31 AE Signals Obtained with Vertical Placementofthe Sensor FilteredsingFFTAs seen in Figures .22 and 4.23 as ell as Figures 4.30 and4.31the signals filtered usingFITand a elet folIo the same pattern and more or less coincide with each other.Inboth casesofhorizontal and ertical placementofthe sensorshoe er one can distinguish the difference between the abo e0anal sis hen the signal contains rapid changes. One can observe ho the FFT smoothens the signals here er the data contains rapid changes hereas the a elet anal sis pro ides a clearer ieofthe spikes or rapid changesofAE. oreo er it is found that a elet anal sisisa less time consuming approach than FFT.56


O.l4,..-------------------------------L....----,0.120.1>:g0.08.=0.060.04 0.02oHFI HF2 HF3 HITDesignationH2Figure 4.32 PlotofaximumAmplitudeofAE Signals with the Horizontal Placementofote-Sensor for Different RateofShearingaximumAmplitud0.140.12-t-----0.1>.g0.080.06<:0.04 0.02oV2TestsVFIVF2Figure 4.33 PlotofaximumAmplitudeofAE Signals with the Vertical Placementofote-Sensor for Different RateofShearingFigures 4.32 and 4.33 represent the maximum positi e amplitudes and the absolute aluesofthe maximum negati e amplitudes obtained during testing with horizontal and ertical placementsofthe mote-sensor respectielfor different ratesofshearing. Furthermore Figures 4.32 and 4.33 clearl sho that the maximum amplitudesofAE signals recei ed are greater than the maximum amplitude that is obtained during the no shear condition. These maximum amplitudes are also higher than0.1V under shearinginboth horizontal and ertical placementofthe sensor.Inaddition for both the 57


horizontal and ertical placementofthe sensor the a erage maximum positi e amplitudes obtained during the faster rateofshearing seem to be slight! higher than thatofslo er rateofshearing (Figure 4.34).0.13;;;-':""'0.11

Figure .35 sho s the peak alues obtained from the de-noised signals using a elet and FFT. E cept in HFI the peak alues obtained from de-noised signals using FFT are smaller than the peak alues obtained from de-noised signals using a elet. ince these peak alues depend on the threshold alues used in a elet anal sis and the cut-off range (numberofzeros introduced) in the FFT anal sis it is difficult to obtain a definiti e relation be een the peak alues. As shown in Figures 4.32 4.33 and 4.35 the positi e peak alues obtained from the original signals and the filtered signals using FFT and a elet are larger for the faster rateofshearing than for the slo er rateofshearing.Tim(s) From DenoisedignaJingFIT8070600.......t)40EE=3020100t:..N==-Test DesignationFigure 4.36 PlotofTime at Which the Positive aximum Amplitude Recorded for Different TestsThe time taken for the occurrenceofmaximum amplitude is obtained from the de noised signals using FFT and a elet. From Figure 4.36 the occurrenceofthe maximum positi e spikes (peak amplitude) in the faster rateofshearing is quicker than in the slo er rateofshearing. This trend is observed for both horizontal and ertical placementofthe sensors. 59


35HFIHF2HF3HIH2 2VFI VF2Tt DesignationFigure 4.37 PlotofTime Taken For OccurrenceofInitial Visible Peak Amplitude in Different TestsFrom Figure 4.37 it can be concluded that there is no correlation between the initiationofpositi e and negati e initial isible peak AE acti ities in each test with different ratesofshearing. Ho eerthe initial AE acti ities begin sooner in the faster rateofshearing than in the slo er rateofshearing. This is observed in both horizontal and ertical placementofthe mote-sensor. The AE data obtained from this stud sho s that there are se eral isible AE peak alues recorded before the reach their maximum AE alues. This beha ior ould lead one to obtain an earl arningofthe soil rno ernent ell in ad anceofthe actual rno ement.4.5 Comparison between Data Obtained from Chemical echanical Planarization (CMP) and Delamination ProcessesFiltered\iitha eletI0.50.4+-----+----------+------------,-------------1200Time (s) I100500.10.3t.l.l-<0.2-I-+l....---:-+-+-+-+--+-r-+--If---------+------,-------r--+-----r--+--+--+--+--+--+--,--+--+---+--iFigure 4.38 Sample PlotofAE Signal from Polishing and Delamination Process60


B comparing Figures4.1and 4.38 one can recognize the AE signal obtained from the shear test to be basicall similartothe AE signal obtained during the processofpolishingofa computer chip.Inboth cases friction is the major causeofacoustic emission. ince the amplitudeoftheAE signal obtained during the delaminationofcomputer chips is higher thaninthe polishing process the signal obtained during the delamination process has sudden peak alues. In soils delamination is analogous to shear failureofsoil after the peak shear alue is reached. This particular situation can be clearl observed in Figure Comparison Between the Experimental PlotofShear ForceVersus Time and Plot Obtained Using AE Signals--HF2Filtered--hearForce--HI Filtered--hearForce20Time ( )40 0.125-p---------,--.,..--r1000.07580-t--I--t-I-I"MM-it-o-:--+--t---+60oI.;..40caIl,)..c-o.OT++-----'-------'---'----+--f----+200600.0880W0.04 60;e........>0.0250';;........0400<-0.0230;Il,)-0.0420..c-0.0610-0.08 0o20Time( )40 60(a) Fast RateofShearing (b) Slow RateofShearing Figure 4.39 Sample PlotsofShear Force Versus Time and AE Versus TimeFrom Figure 4.39 one can conclude that the maximum AE acti ity seems to be happening around the time at hich the soil reaches its peak stress. Since shear force ersus time and AE ersus time data ere obtained from completel different tests it can not be direct! correlated. Ho e er in both cases one can observe the time-lag between the peak alues.61


4.7 Suggestion for Future WorkFollowing suggestions can be offered for the enhancementofthe AE signal obtained from shear tests:1.Introduce a pre-amplifier: The output signal from the acoustic emission sensor is picallofery10magnitude and hence the signal-to-noise ratio (8 ) is relatielsmall. A pre amplifier boosts this signaltoa higher oltage to enable it to be sent down through the cables without suffering from further attenuation or being significant! affected b the electronic background noise.2.Introduce a differential sensor with digital oscilloscope hich pro ides0input channels: The differential sensor can be used to capture the e temal noise and eliminate that noise from the signal obtained from AE sources in a real time basis. Ho e er to introduce this sensor one has to ha ea data acquisition s stem with a minimumoftwo channels.3.Introduce a band pass ftIter: The band pass filter remo es un anted frequencies and known sourcesofbackground noise from known sources. A band pass filter can be selected to complement the sensor in producing a refined signal.4.Increase the sampling rate: Increasing sampling rate pro ides more detailsofthe captured signal. It helps one to identify thenatural frequencofthe AE signal. B knowing the frequencofthe acoustic emission one can select suitable sensors and a band pass filter for accurate data acquisition s stem.5.Introduce a data acquisition s stem hich can control the triggering alue oltage: B introducing an appropriate triggering alue one can reduce the numberofun anted data captured b the AE sensors.62


CHAPTER5COCLSIO S5.1 ConclusionsAcoustic emission acti ities depend on the releaseofenerg as transient elastic a es emanating from the stressed materials.Anexperimental stud as performed in this research on cohesionIess soil to sho that the acoustic emission (AE) acti ities result upon direct shearingofsoil. The testing as conducted at two different ratesofshearing and with horizontal and ertical orientationofa mote-sensor especiall designed to sense AE signals. The original signals ere filtered using both Fast Fourier Transform and a elet anal sis. According to the data obtained from the AE mote-sensor under direct shear the numberofAE e ents and their amplitudes increase ith time and reach peak alues before graduall diminishing to ards the endofshearing. The folIo ng additional conclusions can be drawn from the test results:1.ComparisonofAE signals obtained from shearing and no shear conditions re ealed that a significant amountofAE acti ity is created for fast shearing rates bile the sand as being sheared.2.ComparisonofAE signals obtained from fast ratesofshearing and slo ratesofshearing conditions re ealed that the10frequency acoustic emission acti ity isonIsignificant under fast ratesofshearing and it seems to occur around the time at hich the soil mobilizes its peak shear stress.3.Visible AE acti ity hich is due to particle relocations can be observed ell before the AE acti ity produces rapid changes and the maximum peak AE alues.63


4.The occurrenceofthe initial isible AE peak alues as ell as the maximum positi e peak amplitude is quickerinfast ratesofshearing hen compared to that in slo ratesofshearing.5.ComparisonofAE signals obtained from the tests performed under different ratesofshearing sho ed that the positi e peak alues obtained th relati el faster ratesofshearing are significant! greater than that obtained with slo er ratesofshearing perhaps duetothe releaseofmore elastic energy in rapid shearing.6.The filtered signals ere obtained using the FFT and a elet anal sis. The energ lost in both cases is more or less the same. Typicall the results obtained from these0anal ses folIo ed the same patternhoe er FFT smoothens the curve here er rapid changesoccurhereas the a elet approach retains the observed spikes in the original signal. Especiall for large amountsofra data, a elet anal sis approach is more efficient because it ields more accurate filtered signals and also consumes less time for the filtering process.7.When AE signals obtained after de-noising and the original signals are considered it is observed that the peak AEalues obtained using FFT is smaller than that obtained from a elet anal sis. Since the peak alues depend on the appropriate threshold alue selected in a elet anal sis and the cut-off introduced in the FFT one needs to be a areofthe sensiti ityofthe threshold alue usedina elet anal sis and the cut-off in FFT.8.The discrete a elet transformation(DT) algorithm pro ides a fast and efficient meansofanal sis. The a elet decompositionofa time series offers a arietyofapplications on statistical signal anal sis not on! on periodical signals here Fourier transformation is typicall used but also in the anal sisoftransient64


pe signals.Itcan be concluded that the a elet transform is a relati el fast eas and effecti e tool for filtering and de-noising acoustic emission signals. 5.2 Limitationsofthe Test1.s seen in the po er spectral density (PSD) obtained ith a relatielsmall sampling rate the dominant frequenc rangeofthe signal is ery10(less than 1Hz). Ho e er since the sampling rate usedinthe e periments as small it cannot be concluded that the dominant frequencofacoustic emission is less than 1Hz.Ifone uses a much higher sampling rate one rna perhaps obtain a PSD with another range dominated b higher frequenc AE signals.2.The shear force ersus time plots obtained from fast and slo ratesofshearing sho that there is a significant time-lag be een the occurrencesofshear failure. ince the shear force ersus time and the acoustic emission ersus time plots ere obtained using complete! different tests the peak AE and peak shear force obtained cannot be correlated. B introducing an e perimental setup hich includes both a load cell and an AE sensor one could come up ith a correlation be een the occurrencesofpeak AE and shear failureofthe soil.3.From the filtered signals one can clearl observe the occurrenceofpeak AE acti ity. Though the original signals sho some initial isible peaks one cannot clearl identify them from the filtered signals probabl due to their high frequenc Thus the implementationofthis AE technique in a real time soil mo ement monitoring s stem rna be difficult ithout more extensi e laboratory experimentation and data anal sis. 5.3 PotentialPracticalApplications DetectionofAcoustic Emission (AE) signals is potentiall useful for real time monitoringofimpending landslides here the failure planes are predetermined. ince the65


field data is generall contaminated ba large amountofsurrounding noise differential sensors must be introducedinorder to eliminate the noise. On the other hand hen an AE monitoring s stem is implementedinthe field' one could e pect a large numberofAE signals. Therefore a prior kno ledgeoftheAE frequenc at earl shearing stages can be used th sophisticated filtering techniques like a elet anal sis in real time de noisingofthe recorded signals. To enhance the effecti enessofthe monitoring s stem an AE monitoring s stem can be combined ith other traditional monitoring s stems hich use altemati e triggering factors such as moisture content pore ater pressure suction and soil displacement. Thus an appropriate monitoring s stem can be designed depending on the geologic conditionsofthe site.66


REFERE CESDixon. Ka anagh.J and Hil1.R onitoring landslide acti ity and hazard b acoustic emissionJournalofthe geological ocietyofChinaOct. 1996 ppDi on. Ka anagh.JHil1.Rand Kousteni.A Acoustic emission technique for monitoring soil and rock slope instabiliSlope tability Engineering1999 ppFuji ara.T and Ishibashi.A Applicationofacoustic emission method to shirasu slope monitoringlope tability Engineering1999 pp149. allatSHangL Singularity Detection and Processing with a eletIEEE Trans on Information Theory1992 38(2) ppDi on. Ka anagh.J and Hil1.R Acoustic emission monitoringofslope instabili and de elopmentofan acti ea eguide s stemProceedingofthe In titutionofCi il Engineering2002 ppHaar ."ZurTheorie der orthogonaJen Funktionens sterne"ath. Ann1910 331-371. all at (1989) "A theory for multiresolution signal decomposition: the a elet representation"IEEE Pattern AnalAndMachine Intel01.11no. 7 pp.674-693. AntoniadisA.(1994) "Smoothing nois data with coiflets "Statistica Sinica4 (2) pp. 651-678. Basil auriceEansAEapplicationtociil engineering orkAcoustic Emissio icro ei mic Acti ity in Geologic Structure and Materials1981pp319. 67


Rajesh.G. Tapas.K.D. and iekanand..a elet-based multi scale statistical process monitoring: literaturereielIE Transactions (2004) 36 787-806. Rajesh.G.Process onitoring and Feedback Control sing ultiresolution AnalSISand achine Learning Ph.D.Dissertation 2005. Bruce Donoho and Gao a elet AnalSISIEEE Spectrum article October 1996. Paul. S.A. The Illustrated a elet Transform Handbook: Introductory Theory and Applications in Science Engineering medicine and Finance InstituteofPh sics Publishing 2002. Lord. A.E. and Koerner.R.. Acoustic Emissions in Soils and Their se in Assessing Earth Dam Stabili JournalofAcoustic ocietyofAmerica,01.570.2February 1975. KoernerR..Lord A.E. and cCabe .. Acoustic Emission onitoringofSoil Stability Journalofthe Geotechnical EngineeringDiision a 1978. Graps.A.An Introduction to a elets IEEE Computational cience and Engineering Vol. 2 number 2 summer 1995. Basic Kno ledge about Acoustic Emission ethodhttp://cmsnt.frne.tbr.czJuk/odborv/a/E2001/contrib/pazdra.htm.Acoustic Emission easurement Technologies ?go=company. sing non-traditional tool discrete a elet transformation to anal sisofacoustic emission signal http://cmsnt.frne.vutbr.czJuk/odbory/a/AE2001/contrib/pazdera2.htm.68




AppendixA:PlotsofData with Filtered SignalsI-HFI-HFIFiltrdI0.11 -0.03 -0.13o102030 Time ()405060Figure A.I PlotofHFI with Filtered Signal sing Wavelet70


AppendixA:(Continued)I-HF2-HFFeredI6050o30Tm(s)o10-0.15+-------r-------r--------,.-----r--------r-------!oFigure A.2 PlotofHF2 with Filtered Signal Using Wavelet71


AppendixA:(Continued)1-HF3 HF3 FilteredI60504030Time (s)20100.040.094-----------0.01-0.06-0.11+--------r---'----:.--r--....:....---........,.---.........oFigure A.3 PlotofHF3 with Filtered Signal sing Wavelet72


AppendixA:(Continued)604030 Time () 20100.07-i----------tt-+----t-+----H___f-+----t--+--r-10.12-r--------------------r----------,-0.08-0.13+-------,.----"""'T"""----r-------,------r--------.,r-'o0.02-<-0.03FigureA.4PlotofHSlwith Filtered Signal sing Wa elet73


AppendixA:(Continued)o60403020100.12+-------------------t--.------1H----------it-+--------i0.07+---------------II---+HI+-I--+---+-4...........H-----4I-Hf-lH+lHHl-f--+--+-I0.02 -0.03-t+--t+ii..--0.08+-------------'-------'--f---f-+---------------''---+--+-------HI------i-0.13o

AppendixA:(Continued)0.10.08 0.06 0.04 0.020UJ<:-0.02 -0.04 -0.06 -0.08-0.10101-20 30Time () FilteredI405060Figure A.6 PlotofS with Filtered Signal Using Wavelet75


AppendixA:(Continued)0.11-FilteredI0.17060o40Time ()302010-0.15oFigure A.7 Plotofwith Filtered Signal sing Wa elet76


AppendixA:(Continued)1-I FilteredI0.1--r--------------------------------,1201008060Time(5)4020-0.15oFigure A.8 PlotofVSlwith Filtered Signal Using Wavelet77


Appendix A: (Continued)1009080o40 060Time ()302010-0.1+------+-+-+++'--+--'-f-t-t+---Lf-!f-H+--t--H---fi-....L..:ff-L--''F-----'----+--'----I-t+-+-t---I-H-----I-1I---'-t-+-+----t0.1+-----,I-II--I------II+-----I-------II++---------4'----+---+----+-----Ho0.1-..,...-----------------------------,0.05 -0.05-0.15oFigure A.9 PlotofVS2 with Filtered Signal78


AppendixA:(Continued)I-VFI-VFIFilteredI0.15-r--------------------------------,0.1+---+-I---140120 10080604020-0.15oTime ()Figure A.lO PlotofVFlwith Filtered Signal sing Wavelet79


AppendixA:(Continued)1-F2-VF2Filteredl0.1....-------------================-------------.0.1<0.05o+-----++--J--HlI-+--+-IIt+-t---+---tlft--t--f-tt-+--.+iI-+-.---H-f-lHt-f--+-.......-....-.........f-f-tlt--------1o204060Time(5)80100120140FigureA.IIPlotofVF2with Filtered Signal sing Wa elet80


Download Options

Choose Size
Choose file type
Cite this item close


Cras ut cursus ante, a fringilla nunc. Mauris lorem nunc, cursus sit amet enim ac, vehicula vestibulum mi. Mauris viverra nisl vel enim faucibus porta. Praesent sit amet ornare diam, non finibus nulla.


Cras efficitur magna et sapien varius, luctus ullamcorper dolor convallis. Orci varius natoque penatibus et magnis dis parturient montes, nascetur ridiculus mus. Fusce sit amet justo ut erat laoreet congue sed a ante.


Phasellus ornare in augue eu imperdiet. Donec malesuada sapien ante, at vehicula orci tempor molestie. Proin vitae urna elit. Pellentesque vitae nisi et diam euismod malesuada aliquet non erat.


Nunc fringilla dolor ut dictum placerat. Proin ac neque rutrum, consectetur ligula id, laoreet ligula. Nulla lorem massa, consectetur vitae consequat in, lobortis at dolor. Nunc sed leo odio.