<%BANNER%>

Characterization and Optimization of Radiography by Selective Detection Backscatter X-ray Imaging Modality


PAGE 1

CHARACTERIZATION AND OPTI MIZATION OF RADIOGRAPHY BY SELECTIVE DETECTION BACKSCATTE R X-RAY IMAGING MODALITY By BENJAMIN TEICHMAN ADDICOTT A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ENGINEERING UNIVERSITY OF FLORIDA 2006

PAGE 2

Copyright 2006 by Benjamin Teichman Addicott

PAGE 3

The Dude

PAGE 4

ACKNOWLEDGMENTS I would like to thank my friends and fam ily, especially my parents and brothers, my Aunt Deborah, Dr. Alan Jacobs, Dr. Sa mim Anghaie, Dr. Alireza Haghighat, Dr. Edward Dugan, The Balta-Cooks, The Dude (and family), and my research group. Iwould also like to thank NANT, Lockheed-Martin Space Systems Co. and NASA and the University of Florida Department of Nucl ear and Radiological Engineering for support and funding of this project.

PAGE 5

v TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iv LIST OF TABLES...........................................................................................................viii LIST OF FIGURES...........................................................................................................ix ABSTRACT.....................................................................................................................xi x CHAPTER 1 INTRODUCTION........................................................................................................1 New Foam Imaging Backscatter Modality...................................................................1 Motivation.............................................................................................................1 Spray On Foam Insulation.....................................................................................2 Purpose of Investigation...............................................................................................3 Radiography by Selective Detection............................................................................4 2 BACKGROUND..........................................................................................................5 Premise........................................................................................................................ .5 New Backscatter Radiography System.........................................................................7 System Description and Configuration.........................................................................7 3 THEORY....................................................................................................................10 Photons.......................................................................................................................1 0 General Physics of Photons.................................................................................10 Photon Interactions..............................................................................................11 Photoelectric effect.......................................................................................13 Compton scattering......................................................................................14 Rayleigh scattering.......................................................................................16 Photon Production and Energy Distribution Spectrum.......................................17 Attenuation..........................................................................................................19 Detection Modalities...................................................................................................20 Detector Components..........................................................................................20 Scintillator....................................................................................................20 Photo multiplier............................................................................................21

PAGE 6

vi Preamplifier..................................................................................................21 Modes of Operation.............................................................................................21 Pulse mode...................................................................................................21 Integrating mode..........................................................................................22 NaI.......................................................................................................................22 Plastic Scintillator................................................................................................23 Detector Comparison...........................................................................................23 Compton Backscatter Radiography............................................................................25 Radiography by Selective Detection..........................................................................26 Monte Carlo Methods (MCNP5)................................................................................27 SABRINA Supplemental Track Plotter......................................................................28 4 RADIOGRAPHY BY SELECTIVE DETECTION...................................................29 RSD............................................................................................................................ .29 Photon Transport Model.............................................................................................48 Introduction and Concept....................................................................................48 Support /Evidence and Characteristics................................................................51 Image contrast (bright vs. dark images).......................................................53 Collimation trends (optimization)................................................................64 Image pixel shifts and shadows....................................................................73 Signal intensity.............................................................................................75 Applications and Limitations......................................................................................76 5 BACKSCATTER FIELD DISTRI BUTION AND DETECTOR PLACEMENT .. ..77 Backscattered X-ray Signal Profile............................................................................77 Detector Placement Considerations............................................................................80 6 RSD OPTIMIZATION AND IMAGING CHARACTERISTICS.............................84 Optimization Principles..............................................................................................84 SOFI Foam..........................................................................................................85 Voids in foam...............................................................................................89 High density absorber and scattering type flaws..........................................94 Shadows.......................................................................................................97 Alumin um..........................................................................................................101 Plastic................................................................................................................112 Concrete and Gypsum.......................................................................................114 Reactor Insulation..............................................................................................118 Space Shuttle Thermal Protection Sh ield (TPS) Insulation Tiles.....................121 7 SPECTROSCOPY....................................................................................................123 Flaw Type and Orientation.......................................................................................124 Flaw Material.....................................................................................................124 Flaw Depth........................................................................................................130 Target Material.........................................................................................................134

PAGE 7

vii Detector Material......................................................................................................140 Detector and Collimation Configuration..................................................................142 Detector to sample spacing................................................................................142 Collimation extension........................................................................................144 Monte Carlo Verification..........................................................................................149 Applications..............................................................................................................149 8 CONCLUSIONS......................................................................................................151 Applications..............................................................................................................151 Recommendations for Further Developments..........................................................152 Detector Configuration......................................................................................152 Detector Materials a nd Operation Modes..........................................................153 MCNP5 Simulations to be Considered..............................................................154 Optimizing Geometrical Variables....................................................................155 APPENDIX A SELECTED SABRINA GENERATE D PHOTON TRACK PLOTS......................156 B ONE SCATTER SIMPLIFIED PHOTON TRANSPORT MODEL........................182 C COLLIMATION DEPENDENT C ONTRAST TREND ANALYSIS.....................187 D IMAGED SAMPLE DESCRIPTIONS....................................................................193 E SPECTROSCOPIC TRENDS..................................................................................195 F PHOTON CROSS-SECTION DATA......................................................................209 LIST OF REFERENCES.................................................................................................219 BIOGRAPHICAL SKETCH...........................................................................................222

PAGE 8

viii LIST OF TABLES Table page 4.1 Relative contrasts as calcula ted and observed experimentally...................................64 7.1 Percent contrast for various flaw types in aluminum................................................129

PAGE 9

ix LIST OF FIGURES Figure page 1.1 Mean free path of various peak ener gy photon beams in SOFI and in aluminum.........3 2.1 Simple schematic of landmine detection system...........................................................6 2.2 Tiled sample landmine scans. VS-1.6 antipersonnel mine, 2.5 cm depth-of-burial, 15 mm resolution........................................................................................................6 2.3 Schematic of detector components. Gray arrow represents phot on beam direction. Center of detector is 9 cm from photon beam center.................................................8 2.4 System signal flow chart for image acquisition.............................................................9 2.5 Yxlon x-ray tube head and four NaI collimated detectors .............................9 3.1 Oscillating electric (E) and magnetic (M) fields about a propagating photon............10 3.2 Dominant interaction type as a func tion of energy and material Z number................13 3.3 Photoelectric cross-section for aluminum. Log / Log scale.......................................14 3.4 Compton interaction between photon and stationary electron....................................15 3.5 Relative Klein-Nieshena Compton collisi on cross-section as a function of energy and scattering angle..................................................................................................16 3.6 Typical bremsstrahlung x -ray energy spectrum ..........................................................18 3.7 Bremsstrahlung spectrum with characteristic x-rays..................................................18 3.8 Scatter components of Compton backscatter radiography signal................................26 3.9 Scatter component contributions to co llimated RSD backscatter imaging signal......27 4.1 First collision components in SOFI reaching the detector with collimator extension...................................................................................................................32 4.2 Multiple collision components in SOFI reaching the detector with collimator extension...................................................................................................................32

PAGE 10

x 4.3 First collision components in aluminum......................................................................33 4.4 Multiple collision components in aluminum...............................................................34 4.5 First collision components...........................................................................................35 4.6 Second collisions components.....................................................................................35 4.7 Third collisions components.......................................................................................36 4.8 Fourth collisions components......................................................................................36 4.9 Higher order (fifth and gr eater) collisions components..............................................37 4.10 Scatter component contribution to detector current tally for geometry of Figures 4.5 to 4.9 .....................................................................................................38 4.11 Relative contrast by scatter order...............................................................................39 4.12 Directional distribution of ta lly components by scatter order..................................40 4.13 Percent signal, relative contrast and total contra st contribution by scatter components for 1 cm collimator extension..............................................................41 4.14 Percent signal, relative contrast and total contra st contribution by scatter components for 1.32 cm collimator extension.........................................................42 4.15 Percent signal, relative contrast and total contra st contribution by scatter components for 1.5 cm collimator extension...........................................................42 4.16 Percent signal, relative contrast and total contra st contribution by scatter components for 2.32 cm collimator extension.........................................................43 4.17 Schematic of CRP. Photons at and below CRP can pass under collimator and enter detector............................................................................................................45 4.16 Severely under-collimated. first scatters 5.08 cm radius NaI 1.14 cm from SOFI. Dark line indicates CRP above which no fi rst scatters can en ter the detector.........46 4.19 First Scatters. NaI 6.14 cm from SOFI 5 cm collimator extension...........................47 4.20 First Scatters. NaI 11.14 cm from SOFI 10 cm collimator extension. .....................47 4.21 First Scatters. NaI 14.15cm from SOFI 14cm collimator extension. .....................48 4.22 One scatter photon transport model. Double lines indicate upper and lower bounds from important scatters................................................................................49

PAGE 11

xi 4.23 SABRINA generated photon track plot. Aluminum plate with shallow scattering type flaw...................................................................................................................52 4.24 SABRINA generated photon track plot. Aluminum plate with shallow scattering type flaw...................................................................................................................53 4.25 Difference is important photon exiti ng paths caused by a focusing collimator extension...................................................................................................................54 4.26 VShL: Void shallow long........................................................................................56 4.27 VShI: Void shallow incident.....................................................................................57 4.28 VShE: Void shallow exit..........................................................................................57 4.29 VDL: Void deep long................................................................................................58 4.30 VDI:Void deep incident............................................................................................58 4.31 VDE: Void deep exit................................................................................................59 4.32 ScShL: Scatterer shallow long..................................................................................59 4.33 ScShI: Scatter shallow incident...............................................................................60 4.34 ScShE: Scatter shallow exit......................................................................................60 4.35 ScDL: Scatter deep long. .......................................................................................61 4.36 ScDI: Scatter deep incident. ...................................................................................61 4.37 ScDE: Scatterer deep exit........................................................................................62 4.38 MCNP data plots of cont rast vs. collimation length.................................................65 4.39 Experimental data points of contrast vs. collimation length....................................65 4.40 Analytically derived contrast trend as CRP is moved from sample surface to flaw top.....................................................................................................................67 4.41 Analytically derived contrast tr end as CRP moves below flaw bottom...................68 4.42 Contrast vs collimator extension trend. MCNP5 simulations and 1st scatter model........................................................................................................................69 4.43 Contrast vs collimation extension. MCNP5 simulation da ta and 1st scatter model approximation................................................................................................70

PAGE 12

xii 4.44 Contrast vs collimation extension. MCNP5 simulation da ta and 1st scatter model approximation................................................................................................70 4.45 Geometrical considerations for shadow pixel shifting relative to detector position.....................................................................................................................74 4.46 Bright shadow cast by void in Al seen by detector in lower right hand corner........75 4.47 Bright shadow cast by void in Al seen by detector in lower left hand corner..........75 5.1 Klien-Nishiena differential sc attering cross-section for 55 keV photon..................78 5.2 Backscattered photon flux across a plan e parallel to SOFI sample surface................79 5.3 Percent difference in signal due to void fl aw in SOFI as a function of scatter field component. ..............................................................................................................80 5.4 RSD focusing. Each collimation confi guration A, B, C selects for photons originating at and below each specif ic depth A, B, C, respectively.........................82 6.1 Two important paths of a backscattered photon..........................................................86 6.2 MCNP5 simulated geometry. Four, two inch thick layers of SOFI on aluminum substrate....................................................................................................................88 6.3 Detector contribution by cell as a function of co llimation. 60 keV incident spectrum...................................................................................................................88 6.4 Detector contribution by cell as a function of co llimation. 75 keV incident spectrum...................................................................................................................89 6.5 Void-type flaw in SOFI. CRP is optim ally set to be just above flaw..........................91 6.6 Images of foam calibration panel with varying degrees of collimation. Collimation increases clockwise from lower left.........................................................................93 6.7 Void-type flaw in SOFI. Thin arrow de monstrates the path difference induced by the lack of scatter at the void site.............................................................................94 6.8 Two mechanisms (1-lack of scatter, 2increased attenuation) for generating low intensity signal from an ab sorber-type flaw in SOFI...............................................95 6.9 Scan of depth staggered aluminum insert s in SOFI. Bright inserts are above CRP and dark inserts are below CRP...............................................................................97 6.10 Mechanism for shadow image generati on. Dashed line represents true flaw position, solid arrow indicated shad ow image detection position............................98 6.11 SOFI panel with contoured aluminum substrate. B&W...........................................99

PAGE 13

xiii 6.12 SOFI ramp panel with contoured aluminum substrate. .......... ...............................100 6.13 Bolt of flange panel under SOFI..............................................................................101 6.14 Clockwise from upper left, detectors 1,2,3,4...........................................................103 6.15 Photon exit paths across a void channel.................................................................104 6.16 Shadow shifting with detector position....................................................................105 6.17 Bright shadow images of aluminum fl aw plate. 5 cylindrical void flaws at various depths are imaged as brig ht with severe over-collimation........................106 6.18 Schematic of flaw shadow and de tector orientation relationship............................107 6.19 Correlated (processed) image of sample aluminum plate........................................108 6.20 Uncollimated image of aluminum flaw plate...........................................................109 6.21 Collimation set to discriminate just abov e shallowest flaw. Flaw depth increases from lower right, counterclockwise to center.........................................................110 6.22 Over-collimated image of aluminum samp le plate..................................................111 6.23 Over-collimated image of small channel aluminum plate.......................................112 6.24 Under-collimated image of small channel aluminum plate.....................................112 6.25 Plastic flawed plate #1, uncollimat ed on left, collimated on right...........................113 6.26 Correlated image of LANL block...........................................................................115 6.27 LANL block color image........................................................................................115 6.28 Clock radio, glass tube, wire, and acrylic rod inside cinder block..........................116 6.29 Various objects behi nd 1 inch of gypsum................................................................117 6.30 Miniature stereo, glass, fiber optic ca ble, copper wire, behind 1 inch of gypsum (drywall).................................................................................................................117 6.31 Reactor insulation panel image showi ng steel nameplate and shadow also corrugated interior foil structure evident................................................................118 6.32 Steel reactor insulation pane l with boric acid residue.............................................119 6.33 Color image of insulation panel, bor ic acid on far side clearly evident.................119

PAGE 14

xiv 6.34 Reactor insulation panel correlation imag e. Plastic bag with boric acid on far side of panel. Interior stru cture of foil also apparent............................................120 6.35 Side view of reactor insulation panel, showing several layers of corrugated foil..120 6.36 Internal structure of corrugated foil inside reactor insulation panel.......................120 6.37 Space shuttle insulation tile. Density vari ation in surface adhesive evident. Dark spots are glue conglomerations between ceramic tile and laminate covering........121 6.38 Space shuttle insulation tile, under-col limated. Bright smears are glue below CRP, dark circles on left are drilled holes..............................................................122 7.1 MCNP5 generated backsca tter spectra of various flaw materials in Al, high collimation..............................................................................................................126 7.2 MCNP5 generated backsca tter spectra of various flaw materials in Al, low collimation..............................................................................................................127 7.3 Unresolved experimental spectral trends Experimentally acquired data for the spectral shift observed for void ty pe flaw at various depths..................................132 7.5 Unresolved simulated spectral trends. MCNP5 simulations for spectral shifts observed from void flaws at various indicated depths...........................................133 7.6 Resolved simulated spectral trends. MCNP5 simulations for spectral shifts observed from void flaws at various indicated depths...........................................134 7.7 Normalized (first moment) experimental data 75 keV backscatter spectra from a plastic target...........................................................................................................136 7.8 Normalized (first moment) experimental data 75 keV backscatter spectra from an aluminum target......................................................................................................136 7.9 Normalized (first moment) experimental data 75 keV backscatter spectra from a steel target..............................................................................................................137 7.10 Normalized (first moment) experimental data 75 keV backscatter spectra. No collimator extension, 6 cm from NaI to Sample surfaces......................................137 7.11 Normalized (first moment) experimental data 75 keV backscatter spectra. ...........138 7.12 Photoelectric cross-sections for lead........................................................................139 7.13 Photoelectric cross-sections for aluminum..............................................................140 7.15 Normalized experimental spectral shifts for various detector to sample distances over alum inum........................................................................................................143

PAGE 15

xv 7.16 Collimator extension and CRP.................................................................................146 7.17 Normalized experimental spectral shifts as a function of collimation extension....148 A1 Photons having four or le ss collisions (void flaw).....................................................156 A2 Photons having one collision (void flaw)..................................................................157 A3 All photons entering detector (histo ry filtered for clarity-void flaw)........................157 A4 All photons entering detector. Scatter si tes marked with black X (plastic flaw).......158 A5 All collision components, history filtered (plastic flaw)............................................158 A6 No flaw first through f ourth collision components....................................................159 A7 Plastic flaw first through fourth collision components.............................................159 A8 No flaw all collision components, zoomed in view..................................................160 A9 Plastic flaw all collision components zoomed in view..............................................160 A10 Plastic flaw first collision components....................................................................161 A11 Plastic flaw first collision component......................................................................162 A12 Plastic flaw first co llision scatter points..................................................................162 A13 Plastic flaw, all scatter components ( note many scatters off the NaI surface and down back into Al).................................................................................................162 A14 First and second scatter points.................................................................................163 A15 Multiple scatter sites................................................................................................163 A16 First, second, third scatter points.............................................................................164 A17 First, second, third, fourth scatter points..................................................................164 A18 All scatter points......................................................................................................165 A19 Void flaw first collision components.......................................................................165 A20 Void flaw all collision components.........................................................................166 A21 First collision components.......................................................................................166 A22 Second collision components...................................................................................167 A23 Third collision components......................................................................................167

PAGE 16

xvi A24 Fourth collision components....................................................................................168 A25 All collisions............................................................................................................1 68 A26 First collision............................................................................................................ 169 A27 All collisions............................................................................................................1 69 A28 No flaw................................................................................................................... .170 A29 SSL....................................................................................................................... ...170 A30 SSL zoomed in view of tracks across flaw and aluminum....................................171 A31 SSL. Flaw is above CRP........................................................................................171 A32 SSL....................................................................................................................... ...172 A33 ADL flaw is below CRP........................................................................................172 A24 ADL zoomed in view..............................................................................................173 A35 VDL. Flaw is below CRP......................................................................................173 A36 VDL zoomed in view..............................................................................................174 A37 VSL. Flaw is above CRP.......................................................................................174 A38 VDL, flaw shown opaque.......................................................................................175 A39 VDL....................................................................................................................... .175 A40 VDL, flaw is transparent.........................................................................................176 A41 VSL zoomed in view...............................................................................................176 A42 VDE. Aluminum and flaw are both set to invisible to emphasize simple photon tracks......................................................................................................................177 A43 ASL. Photon on far ri ght are reflecting off NaI, going up into it...........................177 A44 ASL zoomed in view.............................................................................................178 A45 ADL...................................................................................................................... .178 A46 VDL....................................................................................................................... .179 A47 VDL....................................................................................................................... .179 A48 VSL....................................................................................................................... ..180

PAGE 17

xvii A49 SSL....................................................................................................................... ...180 A50 SDI zoomed in view of flaw and photon tracks. Aluminum is set to transparent.181 B.1 Typical photon path across a flaw.............................................................................182 B.2 Parameters used to determine optimal collimation length for focusing to a specified depth........................................................................................................186 C.1 Signal intensity as a function of depth and across a void type flaw..........................187 C.2 Contrast versus collimator extension for collimator from zero to critical (optimal) extension. This optimal corresponds to collimating to top of flaw......................189 C.3 Contrast versus collimator extension for collimator from critical (optimal) extension to flaw bottom........................................................................................190 C.4 Contrast versus collimator extension for collimator extension corresponding to flaw bottom to maximum extension.......................................................................191 C. 5 Contrast versus collimator extension .....................................................................192 E1 Spectroscopic trend for flaw depth by incident energy spectra.................................195 E2 Trends for flaw depth. Flaw s depth increases from A to E......................................196 E3 Scatter component break down for spectra..............................................................197 E4 Spectral break down by scatter component...............................................................197 E5 Spectral break down by scatter component. MCNP5 current tally...........................198 E6 Current tally spectra for various flaw depths and collimation configurations..........198 E7 Collision component signal contribution for various flaw depths in aluminum.......199 E8 Collision component contributions for various flaw depths with uncollimated detector in aluminum..............................................................................................199 E9 Collision component contributions to 0.6cm void flaw in aluminum with various collimations configurations....................................................................................200 E10 Spectroscopic trend with flaw depth. Flaw depth increases from A to E................200 E11 Experimental spectroscopic trend with detector height...........................................201 E12 MCNP5 current tally spectroscopic trend with detector height...............................201

PAGE 18

xviii E13 Scatter component contribution fo r various detector and collimator configurations.........................................................................................................202 E14 Spectral breakdown by collision component............................................................202 E15 Spectral breakdown by scatter component...............................................................203 E16 Spectral breakdown by scatter component...............................................................203 E17 Spectral breakdown by scatter component...............................................................204 E18 Experimental trend with flaw depth.........................................................................204 E19 Experimental trend with flaw depth.........................................................................205 E20 MCNP5 trend with collimation extension...............................................................205 E21 Experimental trend with collimation extension........................................................206 E22 Scatter component contribution tr ends with varying collimation............................206 E23 Scatter component contribution by collimation.......................................................207 E24 Experimental spectral trend with flaw depth............................................................207

PAGE 19

xix Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Engineering CHARACTERIZATION AND OPTIMIZATION OF RADIOGRAPHY BY SELECTIVE DETECTION BACKSCATTER X-RAY IMAGING MODALITY By Benjamin Addicott May 2006 Chair: Edward Dugan Cochair: Alan Jacobs Major Department: Nuclear and Radiological Engineering Backscatter x-ray imaging techniques have been developed at the University of Florida for applications ranging from the detection of buried landmines to the non-destructive-examination (NDE) of various industrial materi als such as aluminum and carbon-carbon composites. Recently, a new ba ckscatter x-ray imaging system with geometries and components different than th at employed in previous systems has been developed under contract w ith Lockheed Martin. The primary purpose of this new system is the NDE of the foam thermal insulation material used by NASA on the space shuttle external fuel tank. The imaging modality has also been applied to evaluate other obj ects such as samples made from aluminum, plastics, steel, concrete, gypsum and titaniu m as well as to reactor vessel head steel insulation panels. Detailed subsurface imag es were acquired on each of these various materials indicating the wide range of a pplicability of this imaging modality.

PAGE 20

xx This investigation aims at describing the physics and phenomena associated with this new imaging modality. The photon transpor t and interaction processes leading to an acquired image are explored. Both Monte Carlo simulations and analytical calculations are used, in conjunction with experimental results, to develop an analytical model detailing the mechanics of the photon tran sport process governing the imaging modality. Trends in detector response and spectr oscopic profiles due to variations in parameters such as detector collimation le ngth, detector to target-object spacing, flaw type (i.e., void, scattering or absorbing type flaws) and orientation and media type is cataloged and explained in terms of this model. System variations su ch as detector type and detector mode of operation are also investigated. Consideration of these trends as well as the simple analytical model developed is then applied, in conjunction with experiments and previously collected data, to identify key parameters in detection efficiency and apply them towards system optimization and an overall assessment of the applicability and limitation of the imaging modality.

PAGE 21

1 CHAPTER 1 INTRODUCTION This research endeavor details the de sign and optimization of a unique Compton backscatter type non-destructiv e x-ray imaging modality deve loped over the past several years at the University of Florida .1-4 The origin of the system dates back to systems initially designed for the dete ction of buried landmines. 58 Success of this imaging modality at acquiring definitive images of subs urface landmines had led to its application in the NDE (non-destructive examination) of various industrial materials. Currently, a new state-of-the-art system has been devel oped at the University of Florida designed specifically for the NDE of the foam therma l insulation on the space shuttle external tank 9, but applicable, with modification, to a wi de range of industrial type materials. New Foam Imaging Backscatter Modality Motivation The new system is essentially a derivative of the original landmine oriented system. The present system, funded largely by a contract from Lockheed Martin, 10 is designed for and applied primarily to the imaging of the special foam like material used for thermal insulation of the space shuttle fuel ta nks. This foam insulation, or spray-on-foam insulation (SOFI),11, 12 because of its extremely low de nsity, represents a unique imaging challenge. The integrity of the insulating SOFI barrier must be ensured before shuttle launch is permitted. Imaging this material and optimizing the system for detecting certain classes

PAGE 22

2 of flaws in this material has consequently be come a priority and mu ch of the optimization methods and investigations have been conducted towards this end. Spray On Foam Insulation The SOFI material used by Lockheed Martin and NASA roughly comprises a mixture of carbon, nitrogen, chlorine, fl uorine, oxygen, and hydrogen. The exact chemical composition was deemed proprieta ry and therefore had to be estimated 13 from the composition of similar industrial foam in sulation materials. The density of the material is given as 0.03g/ cm3, about two orders of magn itude lower than that of previously imaged materials such as aluminum (2.7 g/ cm3), plastics and composites (~1 g/ cm3). Due to the extremely low density of the foam, a photon, on average, will have significantly fewer interactions, or travel a greater distance in the medium, before it scatters and returns to the de tector. The mean free path in this material (SOFI), roughly inversely proportional to the density, is orders of magnitude larger than that of any previously investigated materi als. It is therefore reas onable to expect the physical processes resulting in the imaging of this materi al to be somewhat different than those for other higher density materials, such as alum inum or landmines. Figure 1.1 displays the average mean free paths obtained from MCNP5 14 Monte Carlo calculations of several incident photon spectra in both aluminum and SOFI.

PAGE 23

3 45 keV 60 keV 75 keV foam Al 66.5 91.5 112.8 0.3 0.6 0.8 0 20 40 60 80 100 120 mfp (cm) EmaxAvg MFP (cm) vs Peak Energy foam Al Figure 1.1 Mean free path of various peak energy photon beams in SOFI and in aluminum Purpose of Investigation This discussion aims at describing the physics and phenomena which govern the new imaging modality. Detector response and spectral features of this imaging modality and the parameters which affect them will be investigated and their mechanisms detailed. Also covered will be the methods employe d to investigate these phenomena and how they are applied to the optimization of the present system as well as suggested modifications for further development of fu ture generation backscatter type imaging modalities. Careful observations and experimentati on, complemented by MCNP5 Monte Carlo simulations have afforded a comprehensiv e understanding of the phot on interactions and detector responses indicated by particular images. This knowledge has allowed simple

PAGE 24

algorithms to be established to optimize imag e quality and increase system efficiency. Methods have also been developed to system atically and accurately focus the detector collimators to target a specific depth in materi al as well as to disti nguish between types of flaws in various materials and, in some situ ations, to describe flaw depth and threedimensional orientations. Radiography by Selective Detection Detector response and image acquisition obs erved throughout this investigation are described by a concept referred to as Radi ography by Selective Detection (RSD). The theory of RSD is that by preferentially selecting specific components of a scattered photon field, information relating to specific locations and properties of an imaged sample can be extracted. That is, sens itizing the detector towards selected photons within a specific volume will result in larger relative signal fluctuations caused by flaws and features within that volume. In this imaging modality, collimators are us ed in conjunction with the detectors to discriminate against certain components of th e reflected scatter field. This effectively enhances the relative contribution and, thus, the contrast resulting from more important and often deeper regions of a sample where a fl aw or region of interest lies. In this study, a simplified analytical model of the RSD im aging modality is developed and supported. Detector responses and imaging characteristics of the system are then described in terms of this proposed model.

PAGE 25

5 CHAPTER 2 BACKGROUND Premise Backscatter radiography methods were de veloped at the University of Florida 15 as a means to effectively detect and positively id entify buried landmines. The method was developed in response to problems arising from other land mine detection systems available at the time. 16-18 Most significant among thes e problems was that of false positive identification of buried landmines. Previous methods identified landmines only as subsurface irregularitie s making it difficult and often impossible to distinguish between true landmines and landmine si zed objects (e.g. rocks, debris, etc.).19 In the system designed at the Univ ersity of Florida, positive landmine identification via realistic image acquisiti on was accomplished by placing some of the detectors behind a series of collimators. 20 The collimators were positioned so that the majority of the shallow firs t scattered photons would be pr evented, geometrically, from reaching the detectors. This essentially provi ded the detectors with a less cluttered view of the subsurface features of the landmine s and other buried objects. The landmine detection system configurat ion and a demonstration of its imaging capabilities is presented below in Figures 2.1 21 and 2.2. 22

PAGE 26

Figure 2.1 Simple schematic of landmine detection system Figure 2.2 Tiled sample landmine scans. VS -1.6 antipersonnel mine, 2.5 cm depth-ofburial, 15 mm resolution

PAGE 27

Previous research endeavors centere d upon understanding this process and optimizing this system initially for landmine detection and, later, for other industrial nondestructive examination applications have b een carried out over the past several year. 2128 Analysis and evaluation of these along with the inevitab le progress of technology have led to a gradual and steady advancement of this imaging method and the understanding of the mechanisms which govern it. New Backscatter Radiography System Recently, the backscatter x-ra y system has been acutely modified and redesigned to meet the specific needs of imaging SOFI a nd other drastically different materials than the original system was designed for. The new system features di fferent detector and collimator configurations, different photon beam energies as we ll as new electronic components and detector geometries. Thes e differences, combined with the fresh perspective afforded by optimizing and testi ng a new system configuration on a new and unique material (i.e., SOFI), have led to new imaging approaches and concepts as well as a comprehensive analytical model which accurately describes the photon transport process and resultant detector responses. System Description and Configuration The system used in this series of inves tigations consists of four sodium iodide [NaI (Tl)] scintillation detectors a nd an Yxlon MCG41 x-ray generator 29 mounted onto a scanning table with X Y scan motion capabi lities. The detectors are positioned at the corners of an eighteen by eighteen centimeter square, centered on the x-ray beam. Each detector comprises a two inch diameter by two inch thick NaI scintillation crystal mounted onto a photomultipler tube (PMT) and a fast preamplifier sp ecifically designed to handle high count rates. The customize d, ultra-low-noise, high count rate preamps

PAGE 28

have a maximum noise level of about 5 mV for a 1 volt pulse output, while operating in close proximity to a strong electro-magnetic fi eld (x-ray generator tube). The preamp pulses have a typical rise time of 100 na no-seconds and fall time of about 1000 nanoseconds, yielding a total pulse width of about 1.1 micro-seco nd. This specific pulse width (1.1 micro-seconds) allows sufficient lig ht and charge collection time from the NaI and PMT (about five time constants), while al lowing the detectors to measure backscatter fields up to 700,000 counts per second, without experiencing statistically significant pulse pile-up. A schematic of the RSD det ector components and their configurations is presented below in Figure 2.3, followed by a fl ow chart of the entire image acquisition process from detection to display in Figur e 2.4. Figure 2.5 is a picture of the x-ray generator tube head and co llimated detectors. Figure 2.3 Schematic of detector components. Gray arrow represents photon beam direction. Center of detector is 9 cm from photon beam center.

PAGE 29

Figure 2.4 System signal flow chart for image acquisition Figure 2.5 Yxlon x-ray tube head and four NaI collimated detectors Scanning System Flow Diagram NI-Motion Breakout Box LabView/Computer X-ray PMT PreAmp Current Current FastAmp Analog pulse SCA Is the pulse in the Voltage Window Analog pulse Na I Visible Light Pulse Train BNC 2121 yesDigitalPulse NI-Daq PCI 6602 OR NI-Motion PCI 7344 Y-Motor Amps Y-Motor Limit/Home Switches Active Active Step Step Complete Pulse Train Complete Pulse Train Image Complete Pulse Train Ocilloscope Counter/Timer MCA X-Motor X-Motor Amps Active Step Dir Dir Y-Axis Y-Axis X-Axis X-Axis Y-Axis X-Axis Analog pulseAnalog pulse DigitalPulse yesAnalog pulse Photo-Cathode X-ray scattered toward detector

PAGE 30

10 CHAPTER 3 THEORY Photons General Physics of Photons Photons are a form of electromagnetic radi ation. Like all forms of electromagnetic radiation, photons travel in waves and are accompanied by oscillating electric and magnetic fields. These fields are perpendicular to each other as well as to the direction of photon propagation and rotate about the axis of travel. A qualita tive representation of propagation of electromagnetic waves and their a ssociated electric and magnetic fields is presented in Figure 3.1. Figure 3.1: Oscillating elect ric (E) and magnetic (M) fiel ds about a propagating photon.

PAGE 31

Although photons are necessarily accompanied by an electric field, they have no net electric charge and, to a first approximati on, are not influenced by the electric charges and fields of neighborin g particles or waves. 30 Photon Interactions Although photons, like all electromagnetic radi ation, are most completely described by their quantum mechanical wave functi ons and probabilities, for our purposes, considering the energies (45-100 keV) and th e processes that we are interested in (Compton scattering and photoele ctric absorption), they can be well represented by discrete particles with ener gies dictated by a classical and definite wavelength. Photon interactions, then, ma y be accurately and relative ly completely described by five major interaction types. These are, photoelectric effect, Co mpton Scattering, pair production, Rayleigh (coherent) sc attering, and photonuclear intera ctions. Each of these interaction types is described by a quantum mechanical or empirical microscopic crosssection detailing its probability of occurrence for a photon in a particular phase space. These cross-sections, modified by certain atom and energy dependent incoherent scattering functions and form factors, give the probability of each particular type of interaction per atom in a sample. The sum total of these, or the total cross-section, represents the total in teraction probability per atom fo r a photon in a particular phase space. Multiplying these microscopic crosssections by atom density produces the macroscopic cross-sections which represen t the probability per unit path length of interaction of a particular type. Typically, microscopic cross-secti ons are given in units of atom cm /2 or in barns (atom cm / 102 24 ) per atom, depending upon whether they

PAGE 32

have been appropriately modi fied. Macroscopic cross-sections are typically given in units of M / or1 3 2) / ( ) / )( / ( cm g atoms cm g g cm. Once these cross-sections have been define d, another parameter, the mean free path (mpf) can then be introduced. This is a t ypical attenuation length or free flight travel length given as 1/ = cm. This is the length that the uncollided portion of a beam of photons will be attenuated by a factor of1e(2.718). The MFP can also be interpreted as the distance in a material that a photon of certain energy would have a probability 1eof traveling through wit hout interaction. 31 Each of these photon cross-se ctions is a function of at least the energy of the photon and Z of the material. For the typica l materials (low Z < 15) and photon energies, <~ 100 keV, the dominant interaction type s are Compton scatte ring and photoelectric effect. For lower energies, < 30 keV cohere nt scattering of photons can also become important. Figure 3.2, 31 below, plots the three important cross-section, i.e. photoelectric, Compton, and pair production, as a functi on of both energy and material Z number. Clearly, the first two interaction mechanisms photoelectric and Co mpton scattering, are the dominate interaction mechanisms fo r the typical Z range and energy range [highlighted in the (red) circle] that we are concerned with. The rest of our treatment of photons in this discussion will be for finite wave/particles with de finite energies and knowable pre and post collision velocities.

PAGE 33

Figure 3.2 Dominant interacti on type as a function of en ergy and material Z number Photoelectric effect Photoelectric interaction is an intera ction between a photon and a bound electron. In these interactions, a photon is completely absorbed by a bound atomic electron. This electron is then ejected from the atom. It is therefore necessary for the incident photon to carry at least the energy required to free the el ectron. This is eff ectively the electrons binding energy. If the photon imparts less th an the binding energy to the electron, the electron will merely be excited and cons equently release a photon upon de-excitation rather than being ejected form the atom. A typical photoelectric cros s-section, shown in Figure 3.332 for aluminum, is a strong function of energy. In fact it varies roughly as (1/E^3) .32 This results in a very large absorp tion cross-section at low energies.

PAGE 34

Figure 3.3: Photoelectric cross-section for aluminum.32 Log / Log scale. Compton scattering Compton scattering is best conceptualized as an elastic billiard ball type collision between a photon particle a nd a free electron. In this process, a photon with known direction and energy collides with an assumed stationary (unbound) atomic electron. After the collision, both the photon and elec tron velocity vectors (energy and position vectors) can be accurately described via cl assically considered conservation laws of energy and momentum.

PAGE 35

Figure 3.4: Compton intera ction between photon (thick, re d) and stationary electron (thin, blue) It is, for the most part, these types of collision which concern us, as it is these collisions which predominately result in phot ons being backscattered from the target to the detector. The Compton Scattering probability is descri bed in terms of a differential crosssection which relates the probability in cm2 / steradian per electron for a photon of a specific energy to scatter into a specific solid angle about a specified angle. This differential probability distribution is the Klei n-Nieshena differential cross-section. The Klein-Nieshena cross-section uniquely relates initial ener gy, scattering angle and final energy of a photon. This relationship is given as: ) sin ' ( '2 v v v d d where, ) cos 1 ( 1 1 v is the ratio of scattered to initially photon energy and 2c m ho. Is the reduced initial photon energy. As th is relationship dict ates, the scattering probabilities will be peaked (have local ma ximums) in the directly forward (0) and directly backward (180) directions. The scattering will also follow a forward peaking ehv hv

PAGE 36

trend as the energy is increased The Klein-Nieshena relative scattering probabilities for several incident photon energi es are plotted below in Figure 3.5 for 0 to 180 degrees scatters. Figure 3.5: Relative Klein-Nieshena Compt on collision cross-section as a function of energy and scattering angl e, eta (in radians). Rayleigh scattering Rayleigh scattering, pronounced mainly at low energies (h < 30 keV) and high Z materials, is the highly forward elastic s cattering which occurs between an incident photon and a bound electron. The important diff erence between coherent (Rayleigh) and incoherent (Compton) scattering is that in Co mpton scattering, the collision is between a photon and an essentially unbound electron, whereas in coherent scatte ring the collision is between a photon and the target atom as a whole. This difference allows for the vast majority of the momentum and energy to be retained by the incident photon while the Incident photon energy

PAGE 37

target atom receives only the minimum amount of momentum to ensure conservation. Due to the relatively huge mass of a target atom relative to a single electron (as in Compton scattering), the photon doe s not transfer much energy to the atom. This is very similar to bouncing a ball off of a wall. In this case, they wall, being fixed, does not absorb a significant amount of energy or momentum and the ball rebounds with essentially the same amount of energy as it ha d before the collision. Coherent scattering is highly forward peaked. Similar to incohe rent (Compton) scattering this peaking is accentuated at higher energies. Photon Production and Energy Distribution Spectrum Photons utilized in research are produced vi a an x-ray generator. X-ray generators operate on the principle that high energy el ectrons incident upon a tungsten target produce a spectrum of bremsstralung 33 photons. The initial pr oduction of these photons is more or less isotropic. The energy is di stributed, initially, according to a typical bremstraulung spectrum as shown in Figure 3. 6. The addition of aluminum of copper filters hardens the primary bremstraulung spectrum. Figure 3.7 shows target characteristic x-rays superimposed on th e continuous bremstrahlung spectrum. For a tungsten target, the generator voltage has to ex ceed at least 69.5 keV, the K-shell binding energy, for the characteristic x-rays to be gene rated. These are the type of spectra used in the investigations pres ented in Chapters Five through Seven.

PAGE 38

Figure 3.6 Typical bremsstra hlung x-ray energy spectrum. 33 Figure 3.7. Bremsstrahlung spectru m with characteristic x-rays.33

PAGE 39

Important parameters describing this sp ectrum are the most probable energy, the average energy, and the peak energy. The spectral analysis completed as part of this investigati on makes use of the scattering-to-absorption ratio. It is this ratio, whic h includes both coherent and incoherent scatters, that governs the spect ral shifts and trends realized in our measurements and simulations. This can be visualized by co nsidering a uniform spectrum of photons traveling th rough and being attenuated by a particular material. Since for lower energies the attenuation coeffi cients are higher, due to the photoelectric effect, it is logical to predict that the em ergent beam spectra will be shifted towards higher energies. This is a result of the lower energy photons being preferentially absorbed. Other shifts in ener gy can likewise be explained. For our particular situation where we are interested mainly in backscattered photons, an appropriate down shift in energy, due to energy lost in scattering can be observed. This downshift is theoretically dependent upon the init ial photon energy, the scattering angle, and the number of scatters encountered before reaching the detector. Attenuation Linear attenuation coefficients are desc ribed for monoenergetic uncollided photons beams as the sum of all interaction cross-sections multiplied by the appropriate atom density. This coefficient, given in terms of 1/cm is then multiplied by the photon path in cm and the exponential function of this product, given as e-N x then represent the fractional uncollided photon intensity as a f unction of distance tr aveled in a medium. This simple equation can easily be expa nded into a series with appropriate coefficients and weighting parameters to provide a description of the uncollided photon intensity of a particular photon energy distribution.

PAGE 40

Detection Modalities Detector Components Scintillator Scintillation detectors function by turning radiation energy into visible light, which is subsequently collected and converted into an electrical signal. The process by which visible light is produced from incide nt radiation takes place in the scintillation material, the portion of the detector which inte racts with the radiati on, and the process is referred to as fluorescence. This process i nvolves the absorption of some portion of the incident radiations energy by an electron. Th e electron is then elevated from its normal energy state into an excited state. The exci ted state is necessarily less stable that the original ground state of the el ectron and thus de-excites back to this more stable ground state. With this de-excitation comes a phot on of light of wavelength determined by the energy gap that the initial elec tron traversed in its excitation.34 In inorganic scintillation crys tals, such as the sodium iodide (NaI) used in most of these investigations, electrons are excited from the valence band to a conducting band across an energy gap called th e forbidden band. In this energy gap, no sublevels are found so that there is no real probability of finding an electron be tween the valence and conduction bands. In pure inorganic crystals the de-excita tion of an electron to the valance band with the proper photon emission is not realistically e fficient for practical detection requirements. To compensate for this, and to increase the probability of the resultant photon being in the us eful visible range of the el ectromagnetic spectrum, small amounts of impurities, referred to as activators, are added to the crystal. These impurities

PAGE 41

have an energy band within the forbidden band gap of the pure crystal. With the appropriate activator added, an electron-hol e pair will migrate though the crystal until they reach an activator impurity site where th ey will quickly de-excite with the release of a useful photon of visible light. This light is then funneled into a photomultiplier tube where it is turned into a measurable electronic signal.34 Photo multiplier Photomultiplier tubes collect the visibl e scintillation photons and convert them into a measurable electric signal. The total process progresses in three distinct stages. Photons from the scintillation crystal im pinge upon the photocathode region of the tube where they are converted to electrons. These electrons are then channeled down the electron multiplier where they are proportionately multiplied by several (typically 5-7) orders of magnitude. After this multiplication process, the electrons are then collected at the anode end of the tube where they have e ffectively become a now measurable electric signal proportionate to the incident scintillation photons and thus, to the original radiant energy deposition.34 Preamplifier The preamplifier serves as an intermedia te signal amplification step between the detector and the analytical circuit used to process the detected signal. The circuit components and time constant of the preamplifier have important implication on the detector behavior as a whole. Modes of Operation Pulse mode The detection of signal energy distribu tion (spectroscopy) or true count rates requires the detector to be used in pulse m ode. In this mode, each incident quantum of

PAGE 42

radiation produces a pulse. Each pulse in turn is individua lly collected by the detector and processed as a count. The voltage height of this pulse is pr oportional to the energy deposited. This is the most common mode of operation and for most of the experiments conducted in this investigations pulse mode operation is used.34 Integrating mode In situations where the detected count rate is high so that pulse pile up occurs, current (or integral) mode of operation can be used to mitigate such detector saturation problems. In integral mode, the total charge generated over a set time is collected. This set time, known as the detector response tim e, is large compared to the time between individual events, and thus the association between ch arge created and individual interactions is lost. The benefit, however, is that the detector doe s not need a recovery time between individual pulses and thus is capable of hand ling much higher count rates. One important property of inte gral mode of operation is that the current measured or detected is not exactly equi valent or necessarily strictly proportional to the true count rate. The reason for this is that it is the to tal charge deposited in the detector over a set time which is measured. This charge is dependent upon the number of particles interacting, the type of interaction, as well as the energy of the interacting particles. In other words one particle which deposits most of its energy will be detected the same as two particles each of which deposit half as much energy.34 NaI The NaI is the standard by which the othe r materials are measured. The system was originally designed with NaI detector s and most of the experiments and images acquired on the system have been with this dete ctor type. The advantage of NaI is its fast response time and large photoelectri c cross-section. This property allows NaI to be used

PAGE 43

as thin crystals and still collect full energy deposition. Additionally, because most of the photons which impinge upon NaI deposit their full energy, it has good energy resolution and is appropriate for spectroscopic investigatio ns. Disadvantages of NaI include that it is brittle and hygroscopic. Plastic Scintillator The advantage of the plastic scintillation detectors is that they are more resilient, cheaper, not hygroscopic and can be produced in almost any desired shape. The plastic detectors are also capable of handling higher co unt rates. The mean free path (mfp) of a photon in plastic is on the order or 2 cm, much larger than th at of NaI crystals(~mm). However, the relative photoelectric cross-section for plastic type scintillators is smaller than that of NaI (Appendix F). Cons equently, many photons do not deposit their full energy within the plastic scintillation material. For this reason, energy resolution is poor in plastic type scintillators and they are of almost no use in spectroscopic analysis. Plastics are more conducive to integrat ion mode operation at high count rates. Detector Comparison Both the NaI and the plastic detectors may be operated in both pulse and integrating mode. The concept is that inte gral mode is not sus ceptible to pulse pileup problems encountered at high count rates. Since pulse mode count s each individual photon by collecting the appropria te charge deposited in the detector, pulse mode detectors are limited, in count rate, by an a ssociated dead time whic h is the time required to collect and dissipate the ch arge of an individual photon. This dead time is ultimately limited by the charge collection time of the detect or material itself which in NaI is about 0.23 s, indicating a maximum ideal (assuming cps could be limited only by this time constant ) count rate in pulse mode of about 800,000 cps (5 time constants 35 ) before

PAGE 44

significant pulse pile up and dead time effect s result. In reality, however, the maximum count rate is governed by the time constant of the entire detector circuit, including the pre-amplifier. Current or integrating mode, however, co llects the charge produced by multiple photons incident over a predetermined integration period. In integral mode, it is the total energy deposited over a period of time and not the energy deposited per interaction that is important. Because cu rrent mode operates on a voltage produced by the incident photons and the number of electro n-hole pairs they gene rate, greater weight is given to a higher energy photon since they generally de posit more energy and thus create more electron-hole pairs in the detect or material. This results in images which show biasing towards high ener gy photon detection. This ef fect is more pronounced in NaI than it is in the plastic scintillators. This is due to the fact that more energy is deposited in NaI than in the plastic material (per unit distance tr aveled by a photon). A photon impingent on a NaI crystal is likely to deposit all or most of its energy within that crystal resulting in a proportional amount of electron-hole pairs being produced. That same photon incident upon a plas tic scintillation material (o f comparable thickness) is less likely to deposit the majo rity of its energy in the de tector. A lower energy photon, however, will deposit a larger portion of its en ergy since its MFP is smaller. The result of this phenomenon is that there is esse ntially a maximum energy deposition limit above which no single photon will usually deposit the re st of its energy. Consequently, unlike NaI, most photons, regardless of their en ergy deposit about the same amount of energy and therefore produce about the same number of electron-hole pairs in a relatively thin plastic scintillation material.

PAGE 45

A problem with this feature of the integrating mode (primarily in the NaI) is that the change in signal strength resulting from photoelectric attenuation, which is dominant at low energies, is not observed as well. Often this differential photoelectric signal is a significant portion of the overall contrast and discounting it re sults in image degradation. Another problem with neglecting the low energy portion of the signal is that often the high energy portion of the incident spectrum doe s not interact as intimately with the medium and consequently contributes to noise rather than signal. Compton Backscatter Radiography Compton Backscatter Imaging (CBI) Radi ography is a non destructive imaging modality used to image objects when transmi ssion type methods are not feasible. The important feature of CBI is that access to onl y one side of an object is required. For transmission type radiography a detector is placed on the opposite side of an object relative to the source and is sensitive to photons which tr aveled through the object; CBI techniques place the detector and source on the sa me side of the object and the detector is then sensitive to photons which interacted in the object and scattered backwards into the detector. Many traditional Compton methods employ a large unobstructed detector which is sensitive to the entire distri bution of backscattere d photons (at least those scattering into the relevant solid angle subtended by the det ector). Since the majo rity of backscattered photons will suffer a scatter near the surface (w ithin half of a mean free path) of the substrate and then return to the detector, the signal and the image generated directly reflects electron density varia tion within this surface and shallow subsurface region of the sample. An example of this configurati on is illustrated in Figure 3.5 below.

PAGE 46

Figure 3.8 Scatter components of Comp ton backscatter radiography signal. Here, the shallow first scatte red component of the backsc attered field, represented by the thick (red) arrow is the dominant signa l contribution and clearly overwhelms the contribution of the deeper penetrating and/or multiple-collided photons which are usually considered as noise. This is intuitive if the exponential atte nuation of photons is considered, necessitating that each additiona l distance traveled into the sample by the photon beam results in exponent ially less photons available to scatter back into the detector. Additionally, the contributions of multiple-scattered photons are further reduced since these photons must both have a nd survive multiple collisions without being absorbed. Radiography by Selective Detection Radiography by Selective Detection (RSD) techniques are similar to CBI methods in that they rely on Compton (mostly) backsc attered photons to generate an image of the investigated object. The difference is th at RSD techniques employ collimators and detector target

PAGE 47

calculated detector positioning to select for specific regions of a backscattered photon field. The effect of this is to enhance the sensitivity to the detector response function to variations of specific components of the backscatter field. A simple illustration of this principle is shown in Figure 3.9 below. Figure 3.9. Scatter component contributi ons to collimated RSD backscatter imaging signal. Here, the shallow first scattered field, represented by the thick (red) arrow is effectively discriminated against by the collimat or (lower (pink) cyli nder). The resultant image is thus generated largely form deep er penetrating photons. In RSD methods detector and collimator geometries and orie ntation are governing para meters in selecting the portion of the detected backscattered photon field, and hence the region of the sample, viewed by the detector. Monte Carlo Methods (MCNP5) Throughout this investigation Monte Carl o methods, implemented via MCNP5, are used to simulate experimental setups and detector responses Monte Carlo methods are a detector collimator

PAGE 48

means of solving a problem through statistica l sampling of probabilities and are used when deterministic methods are not desirable. Effectively, Monte Carlo methods arrive at a particular solution by tracking partic les and tallying individual events until enough information has been obtained to infer a reasona ble answer. Each event is determined by sampling from a pool of random numbers di stributed according to the appropriate interaction probabilities. SABRINA Supplemental Track Plotter SABRINA36 is an application code which, in conjunction with MCNP, graphically displays the simulated geometry and/or the photon tr acks and interactions mechanics. It utilizes the MCNP geometry input deck and a special PTRAC card which causes MCNP to generate a file in which selected history data (location, interactions, and velocity components) of photons run in the MCNP simulation are recorded.

PAGE 49

29 CHAPTER 4 RADIOGRAPHY BY SELECTIVE DETECTION RSD Radiography by Selective Detection (RSD) produces images via a signal differential due to a linear attenuation di fference experienced by single and multiple scattered photons as they traverse various regions of a sample. The photons essentially travel in a simple reflection path between the source and the detect or (much like optical photons), with an appropriate backwards scatte ring occurring in the target object. This approximation is roughened by photons which in teract more than once in the sample before scattering into the detector. Photons having more than one scatter in the target material deviate, to varying degrees, from those having only one collision. For the examined configurations, analysis and experiments indicate that these multiple collided photons behave essentially as single scat tered photons (for the purpose of providing contrast in imaging modalities) in that they transverse flaws directly, as the once scattered photons do, on the way from their last scatter to the detector. That is, the mechanism for generating flaw contrast is essentially the sa me for single and multiple scattered photons. The contrast, regardless of the number of scatters, is a function of the attenuation difference afforded by the flaw as the photons impinge upon it and exit, after scattering, towards the detector. The effect of these multiple scattered photons, to a first approximation, is tantamount to a broadening of the initial impingent photon beam. Photons having more than one scatter lose some degree of their original incident directionality. Thus,

PAGE 50

depending upon how many scatters the photons ha ve suffered, their backscattered field distribution is skewed from the primary once scattered backscatter photon field. Additionally their final scattering points ar e necessarily displaced from the incident photon beam axis. The contrast observed by these photons, however, is generated by the same mechanism responsible for single scatte r photon contrast. That is, the attenuation differential afforded by a flaw as a phot on traverses it. Fo r photons having many collisions (usually 4 or more) this attenua tion differential can become negligible in comparison to the total photon path length in th e target material and thus the contrast for these very high order photons is often much lower than for the primary and secondary scattered photons. This concep t of effective beam spreading can be observed in Figures 4.1 and 4.2, below. Each of these figur es is a SABRINA generated photon track plot from an MCNP5 simulation. The simulati on models a one inch aluminum substrate below eight inches of SOFI foam. The detect or is 2 inch diameter NaI and is located 9 cm, centerline-to-centerline, from the impi ngent beam (2mm in diameter) and 5.14 cm above the foam surface. The collimator is ex tended 4 cm past the NaI surface, or 1.14 cm from the SOFI surface. Figure 4.1 shows first collision components of the backscattered radiation field that re ach the detector. Intuitively, all thes e collisions occur along the axis of the im pingent beam, shown by the black arrow. Figure 4.2 is the SABRINA generated plot of the same MCNP5 simulation displaying multiple (second order and higher) scattered photons. As demonstrated in the figure, except for a few outliers, the effect of multiple scattered photons can be approximated by an effective broadening of the impingent photon beam. That is, the result of total scatters from a narrow beam can be approximated well by

PAGE 51

considering only single scatters from a broader, diverging beam. This concept is meant to augment the understanding of the contra st generating mechanism involved in the imaging modality rather than to be used as a quantitative model for system optimizations. The multiple scattered photons make up an effective source distribution within the target material which has a wider distribution than the effective source distribution of the once scattered photons. Thus, intuitiv ely, the last scatter in the target material, before photon detection, of the multiple scattered photons does not occur along the impingent beam axis as the first scatter does. If the initial beam were diverging, however, then the first scatter site distribution for the diverging beam would be similar to the multiple scatter sites of the line source which is actually impingent. The dark arrows in Figure 4.2 demonstrate the concept of effective impi ngent beam widening which woul d account for the effect of multiple scatters. Many of the scatters occu r within this area and upon their final scatter are directed towards the detector. These ar rows are meant to indicate approximately the effective multiple scattered photon source distribution and are not a quantitative representation of an actua l beam divergence. (Note: the right half of the area within the arrows does not display scatters because the SABRINA plot was filtered to only include photons scattered into the detector shown.)

PAGE 52

Figure 4.1 First collision components in SO FI reaching the detector with collimator extension. Figure 4.2 Multiple collision components in SOFI reaching the detector with collimator extension. Diverging arrows demonstrate approximate area of multiple scattered source distribution. The concept of effective beam broadeni ng is further demonstrated, in aluminum targets, by the following figures, 4.3 and 4.4. Figures 4.3 and 4.4 are SABRINA plots of an aluminum target with a 75 keV impingent photon beam. The detectors are 5.08 cm

PAGE 53

radius NaI positioned 9 cm from the photon beam and are 5 cm from the aluminum surface with a 4.4 cm collimator extension. Figures 4.3 and 4.4 show the first collision and multiple collision backscattered com ponents which reach the NaI detectors, respectively. As again illustrated by the bl ack arrows, the first collision components all originate along the impingent beam axis while the higher order components can be modeled as originating from a radial axis of a broader, diverging beam. The validity of this approximation is based upon the fact th at, upon suffering a scattering collision, a photon is necessarily deflected at some angl e away from the initial photon beam. In order for the photon to be detected, it must eith er scatter directly in to the NaI, or suffer another event that scatters it into the detector. The photons which do not scatter directly into the NaI, as the figures illustrate, eff ectively make up a distributed source within the target itself. This effective distributed source of multiple scatters is very similar in effect to the primary scatters of an initially broader, diverging, beam having a spread approximated by the black arrows in Figure 4.4. Figure 4.3 First collision components in aluminum

PAGE 54

Figure 4.4 Multiple collision components in aluminum. Arrows indicate approximate multiple scattered photon source distribution. The following set of figures, 4.5 4.9, show the effective beam spreading as a function of scatter components. These plot s demonstrate that the dominant mechanism for image contrast generation regardless of th e number of collisions is the attenuation difference provided by the flaw as the photons directly traverse it. The figures are again generated by the SABRINA application us ing MCNP5 simulation data. Each figure models a 10 x 10 x 1 inch aluminum plate wi th a void type flaw running along the axis from beam to detector. The flaw is 0.4 cm in height and 1 cm wide and 1 cm below the aluminum surface. The 2.54 cm radius detect or is positioned at 9 cm from the beam center and 2.9 cm above the aluminum surface w ith a 1.5 cm lead collimator extension. This collimation is configured so that the CRP (critical reference plane, see Figure 4.17) is located just below the flaw channel botto m. In each figure the black arrows again roughly indicate the effective si ngle scatter beam di vergence that would approximate the multiple scattered photons. As each of thes e demonstrate the vast majority of photons directly and linearly traverse the flaws and consequently the resulting acquired image

PAGE 55

contrast can be directly rela ted to the difference in attenu ation properties induced by the flaw. This is the same effect we would observe if the signal comprised all first scattered photons originating from a wider range than th e initially impingent beam. That is the dominant mode for image generation is the sa me regardless of the scatter order of the photon. Figure 4.5. First collision components. Bl ack arrow indicates impingent beam axis and line of scatter origination. Note that photons dire ctly traverse the flaw. Figure 4.6 Second collisions components. Bl ack arrows indicate approximate effective beam divergence. Note that photons directly trav erse the flaw

PAGE 56

Figure 4.7 Third collisions components. Black arrows indicate a pproximate effective beam divergence. Note that photons directly trav erse the flaw Figure 4.8 Fourth collisions components. Bl ack arrows indicate approximate effective beam divergence Note that photons directly traverse the flaw

PAGE 57

Figure 4.9 Higher order (fifth and greater) collisions compon ents. Black arrows indicate approximate effective beam divergence. Note that photons directly traverse the flaw The degree to which the effective beam di vergence is observed is a function of the relative scatter component contribution to the detector, as dictated geometrically by the collimation configuration, as well as the mean free path of the target material. For relatively high density materials such as alum inum, shown in the figures above, most of the second and third order sc atters occur close enough to the initial beam so that neglecting them in an approximation is valid. The relative importance of the first seven scatter components for this simulation, give n as percent contribution to the detector current tally, for the simulations depicted above in Figures 4.5 4.9, are plotted below in Figure 4.10. Even for this highly collimated situation, contributions of the first three scattering components make up over 70% of the total signal. Furthermore, comparison of this data against a similar MCNP5 run without the flaw channel reveal s that the majority of the contrast contribution a nd thus the important part of the signal comprised mostly first and second scatters as shown below in fi gure 4.11. This is indicative of the proposed

PAGE 58

contrast providing mechanism. As scatte r order increases so does the total path length of a photon in the target material. As pa th length increases (w ith scatter order) the relative attenuation differential afforded by the flaw (since it remains the same size) decreases. Thus higher order scatter com ponents have lower relative contrasts even though they may represent larger por tions of the total signal. 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 60.00% 70.00% 80.00% 90.00% 100.00%relative contribution 1st2nd3rd4th5th6th7thtotal scatter componentRelative Scatter Component Contribution to detector Tally Figure 4.10. Scatter component contribution to detector cu rrent tally for geometry of Figures 4.5-4.9.

PAGE 59

0% 5% 10% 15% 20% 25% 30% 35% 40%contrast 1st2nd3rd4th5th6th7thtotal scatter orderPercent Contrast by Scatter Components Figure 4.11. Relative contrast by scatter order. This is the contrast that would be observed if each scatter component could be isolated and individually considered. Additionally, a directional di stribution of the tally co llision component breakdown (for the simulation discussed above) reveals that for each scatter component considered, up to seventh, the vast majority of the dete cted photons impinge upon the detector at an angle of thirty degrees or le ss to the horizontal, just as the once scattere d photons do. This indicates that even high order scatter events do not bring the photon significantly far away (geometrically: i.e. the scattering angles do not vary by more than a few degrees) from the impingent beam axis. If photons in ge neral, were to scatter farther before being deflected into the detector, we would observe a more sign ificant deviation in angular direction components as scatter order increased. Ther e is, as expected, a noticeable increase in detected photons entering furthe r away from the horizontal with increasing

PAGE 60

scatter order. However, the majority of photons enter at relative ly the same range of angles as the once scattered photons, implying th at they traverse the flaw at a similar angle and are thus similarly attenuated by th e flaw. The important photons, as defined above and in Figure 4.10 and Figure 4.11, are shown in Figure 4.12 to be composed of more that 90% photons entering within th irty degrees of the horizontal. 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0.870.940.9841 cosine angle from horizontal (mu)Scatter Component Detector Entry Angle 1st 2nd 3rd 4th 5th 6th 7th Figure 4.12. Directional distribution of tally com ponents by scatter order. The following four figures, 4.13 4.16, are plots of data taken from MCNP5 simulations. Each shows the pe rcent signal, relative contrast and contrast contribution of each scatter component (up to seventh) of the signal. The four plot s are taken from four separate simulations each of a 40 x 40 x 5.08 cm aluminum target with a 0.08 cm high and 1 cm wide flaw channel 0.1 cm below the aluminum surface. Each simulation was modeled with a 5.08 cm diameter NaI detector 2.9 cm from the target aluminum surface

PAGE 61

and offset from the impingent beam (cente r-to-center) by 9 cm. The collimation in each of these runs was varied as 1 cm, 1.32 cm 1.5 cm and 2.32 cm extension past the NaI surface. In these plots, the percent signal of each component is calculated by dividing the signal strength of that compone nt by the total signal strength. The relative contrast is calculated by dividing the diffe rence between the nth componen ts of flawed versus nonflawed scenarios by the appropriate component of the non-flawed scenario. This relative contrast represents the contra st that would be observed if the detector was only sensitive to the nth scatter component of the backscattere d field. The contrast contribution in these plots is calculated by dividing the signal di fference of the nth s catter component by the total signal difference. This represents the contribution to the cont rast generated by each scatter component. 40X40X5.08 AL PLATE 9 CM TO DET CENTER 1 CM COL EXT 2.9 CM TO DET SURFACE 0.08cm high flaw 1 cm wide 0.1 cm below surface 0.00% 20.00% 40.00% 60.00% 80.00% 100.00% 120.00% 1ST2ND3RD4TH5TH6TH7THTOTAL Scatter Component % SIGNAL RELATIVE CONTRAST CONTRAST CONTRIBUTION Figure 4.13. Percent signal, re lative contrast and total co ntrast contribution by scatter components for 1 cm collimator extension.

PAGE 62

40X40X5.08 AL PLATE 9 CM TO DET CENTER 1.32 CM COL EXT 2.9 CM TO DET SURFACE 0.08 cm high flaw 1cm wide 0.1 cm below surface 0.00% 20.00% 40.00% 60.00% 80.00% 100.00% 120.00% 1ST2ND3RD4TH5TH6TH7THTOTAL Scatter Component % SIGNAL RELATIVE CONTRAST CONTRAST CONTRIBUTION Figure 4.14 Percent signal, rela tive contrast and total contrast contribution by scatter components for 1.32 cm collimator extension 40X40X5.08 AL PLATE 9 CM TO DET CENTER 1.5 CM COL EXT 2.9 CM TO DET SURFACE 0.08 cm high flaw 1cm wide 0.1 cm below surface -20.00% 0.00% 20.00% 40.00% 60.00% 80.00% 100.00% 120.00% 1ST2ND3RD4TH5TH6TH7THTOTAL scatter component % SIGNAL RELATIVE CONTRAST CONTRAST CONTRIBUTION Figure 4.15. Percent signal, re lative contrast and total co ntrast contribution by scatter components for 1.5 cm collimator extension

PAGE 63

40X40X5.08 AL PLATE 9 CM TO DET CENTER 2.32 CM COL EXT 2.9 CM TO DET SURFACE 0.08cm high flaw 1 cm wide 0.1 cm below surface -40.00% -20.00% 0.00% 20.00% 40.00% 60.00% 80.00% 100.00% 120.00% 1ST2ND3RD4TH5TH6TH7THTOTAL Scatter Component % SIGNAL RELATIVE CONTRAST CONTRAST CONTRIBUTION Figure 4.16. Percent signal, re lative contrast and total co ntrast contribution by scatter components for 2.32 cm collimator extension As these plots demonstrate, the relative contrast of the first collision component is usually (with exceptions for extremely overcollimated and under-collimated cases) the largest. This is because for these cases, the first collision path length is the shortest and thus the flaw represents the largest relative attenuation difference. As scatter order increases, relative contrast pe rcentages generally decrease. This is due to the relatively longer path length of multiple scattered photons in the target materi al and consequently lessened effect of the attenuation difference cau sed by the flaw. In fact, as the figures also indicate, for scatter components on the or der of 5 or more, th e resultant signal is considered noise and can detract from the desired contrast. The above plots also indicate, as expected, that increased coll imation increases the signal contribution of higher order scatters. This is accomplished mostly by eliminating shallow low order

PAGE 64

scatter components geometrically from entering the detector. The contrast contribution is a function of both the rela tive contrast and the signal c ontribution of each scatter component of the signal. Consequently, ev en though a particular scatter component may have the highest relative contra st, it may not represent the do minant contrast contribution if it does not compose a significant percentage of the total signal. Similarly, the fact that a particular scatter component dominates the si gnal or even the cont rast does not imply that it necessarily produces the largest relative contrast. The justification for including higher order (second, third, and fourth) s catter components even though they may have lower relative contrasts than th e first scatter component is that the advantage of decreased scanning time provided by the stronger si gnal outweighs the disadvantage of lower contrast. The scattered photons viewed by an RSD configuration are those specific photons which interacted in or traveled through a specific region of intere st. In radiography by selective detection, collimat ors and detector placement ar e coordinated so that only certain components of a backscattered signal ar e detected. In many cases, this amounts to using the collimators to discriminate agains t all interactions occurring above a specific region of interest. This allows a signal originating from deeper within a sample to be collected and relative differences caused by small or deep flaws to become observable. In the RSD imaging modality, this collimati on-induced specificity for signal components is referred to as focusing. By focusing to a specific depth, the modality effectively discriminates against all photons having scatters above this de pth. The depth to which a RSD configuration is focused is described by a critical reference scattering plane (CRP) which is an imaginary plane located at the depth at which the first significant primary

PAGE 65

scatter contribution to the detector occurs. The concept of a CRP is demonstrated by the schematic presented in Figure 4.17. In this figure, below, the CRP is shown as a dark horizontal line. The effect of the collimator on once scatte red photons originating from above this plane and below this plane is shown. Figure 4.17 Schematic of CRP. Photons at and below CRP can pass under collimator and enter detector. Photons scattering from above CRP are blocked by the collimator. Figures 4.18 4.21 demonstrate how a part icular collimation configuration focuses to a specific depth by discriminating agains t shallower components of the returning scatter field. These figures are ag ain SABRINA photon track plots of MCNP5 simulations. The target is eight inches (20.32 cm) of SOFI foam on an aluminum substrate. Each figure shows the CRP location by a dark horizontal line at the site of the first important scatter event. Each scenario simulated has a collimator sleeve to sample separation of 1.14 cm. Thus a 5 cm collimator extension implies a distance of 6.14 cm from detector (NaI) surface to sample surface. In Figure 4. 18, the collimator is fully withdrawn so that photons scat tering from all depths in the target may impinge upon the

PAGE 66

detector. In Figure 4.19, the collimator is extended 5 cm past the NaI. This forces the CRP to a depth of 5.2 cm below the foam surface. In this figure it is evident that no scatters occurring above the CRP are reaching th e detector. Figure 4.20, shows the effect of further increasing the collimator extension to 10cm past the detector surface. Simple geometrical calculations reveal that the CRP is now 11.6 cm below the SOFI surface and, as indicated in the figure, this is the mini mum depth that photons must penetrate before being able to directly scatter into the dete ctor. Figure 4.21, features still further collimation as the collimator is extended 14 cm past the NaI surface. Here the CRP is moved to a depth of 16.7 cm and, as the SABRINA track plot demonstrates, no primary scatter events occur above this plane and enter the detector. Figure 4.18. Severely under-col limated. first scatters. 5.08 cm radius NaI 1.14 cm from SOFI. Dark line indicates CRP above which no first scatters can enter the detector.

PAGE 67

Figure 4.19. First Scatters. Na I 6.14 cm from SOFI 5 cm collimator extension. Dark line indicates CRP above which no fi rst scatters are tallied. Figure 4.20. First Scatters. NaI 11.14 cm from SOFI 10 cm collimator extension. Dark line indicates CRP above which no first scatters are tallied.

PAGE 68

Figure 4.21. First Scatters. NaI 14.15cm from SOFI 14cm collimator extension. Dark line indicates CRP above which no first scatters are tallied. RSD modalities can often be modeled as one scatter phenomena in that the photons behave much like the once scattered photons in a traditional Compton imaging system. Essentially the physics of RSD can be consider ed similar to tradit ional CBI, except that the photons having scatters above a selected depth are discrimi nated against. In this idealization, RSD methods eff ectively remove a specified amount of material from the surface of a sample and thereby view the lower layers, below the CRP. Photon Transport Model Introduction and Concept A rough analytical model has been develope d under consideration of experimental observations, MCNP simulation, and photon tr ansport physics. The purpose of this model is to facilitate visu alization and understanding of the phenomena leading to an image as well as to approximate detector re sponses and system optimization parameters

PAGE 69

for various types of materials and flaws. Th e model used to describe the current system is a simplified one scatter model, shown below in Figure 4.22. In this model, each poten tial first scatter si te along the length of the impingent beam axis is considered and the integrated relevant path length leading from each element to the detector is then calculated. The difference between the attenuation of this path for flaw versus no flaw conditions is then taken to approximate a contrast ratio. This model facilitates quick calculations and se rves as a good analytical model upon which to base optimization parameters. Figure 4.22 One scatter photon transport model. Double lines indicate upper and lower bounds from important scatters. Here the signal difference between flawed and non flawed regions of a sample can be visualized as a difference in attenuati on. The photon beam poten tially crosses a flaw twice, as shown above, once impingent and once after scattering. The difference in attenuation of the photon beam over its entire path (incid ent and scattered) between flawed and non-flawed regions is the prim ary contributor to the signal difference observed in the detector. In this model the lack of scattering within a volume due to a Detector Collimator Flaw 1st significant scatter that passes under collimator as determined by CRP Deepest scatter that reaches detector, determined by photon penetration.

PAGE 70

low density flaw can be shown to be equiva lent to an increase in attenuation of the photon beam that otherwise woul d have scattered within that volume and returned to the detector. That is, the photons which do not in teract in the low density flaw continue to penetrate through the sample where eventually th ey will interact. When they do interact, it will necessarily be deeper within the samp le and they will necessarily have a longer distance to travel though the material to the detector. Consequently these photons will have a greater probability of being attenua ted on the way to the detector than their counterparts which interacted in the shallo wer volume of a non-flawed sample. [This is, however, contingent upon flaw orientation and detector and collimation configurations. If the system is configured such that th e selected portion of the back scattered field traverses the flaw again on the way out, the overall effect on a ttenuation may be an increase or decrease, depending upon flaw height and material interaction characteristics, as discusse d later in this chapter]. Analytical calculations can easily be performed based upon the simplified one scatter model. In this one significant scat ter model, the attenuated photon intensity at each point along the incident beam path (shown as the thick, red arrow) is scattered and further attenuated towards the detector. The in tegral of this path over all scattering points then results in a final once scattered intensity at the detector. The detected signal intensity then has the following form:) (y B e Aeb a. Here A is the origin al photon intensity and the exponential terms, a and b, account for the a ttenuation between the source and the first scatter and between the first scatter and the de tector, respectively and B(y) is a function which accounts for the angular cross-section of the scatter and the solid angle subtended by the detector. The percent difference between this integrated intensity for a flawed

PAGE 71

versus a non-flawed region in then taken to be the percent contrast. More mathematical detail of this treatment is in Appendix B. While in reality a detected signal is com posed largely of multiple scattered ( 2nd order and higher) photons, the usefulness of th e simplified once scatte red transport model is that it provides a means of understa nding and visualizing contrast generating mechanisms and thus augments our ability to predict and understand trends, optimization and relevant image features such as shifts and shadows. Support /Evidence and Characteristics This photon transport model of the RSD modality is based largely upon experimental observations and supporte d with MCNP5 simulations, SABRINA track plots, and analytical calcula tions. Trends observed in de tector response arising from parameter variations such as flaw type and orientation and collimation configuration strongly suggest a scatter model of this type. Specifically, the image contrast positive and negative (depending upon flaw type and or ientation) as well as collimation induced contrast trends and pixel shifts observed in acquired images all lead to and are well described by the proposed transport model. Figure 4.23 and Figure 4.24 are SABRINA generated track plots of MCNP5 simulations The simple linear tracks shown here support the premise of the once sc attered transport model, i.e. that contrast is primarily generated by a differential in attenuation expe rienced by photons as th ey directly traverse a given flaw. Figure 4.23 is a plot of a highl y collimated detector over an aluminum plate with a scattering-type fl aw channel (modeled as C2H802 plastic). As the plot shows, many of the detected photons suffer multiple scatters. However, for the purpose of generating image contrast, even these high or der scatter photons be have essentially as primary or first scatter photons. That is, the majority of the photons regardless of scatter

PAGE 72

order, traverses the flaw in approximately the same manner with approximately the same angle and consequently results in similar contrast. Figure 4.24 is a similarly produced SABRINA track plot except th at the collimation is much less pronounced. In this scenario the photons similarly traverse the fl aw directly and thus generate an image contrast based upon the differential in scatte ring and attenuation char acteristics provided by the flaw. As the Figures 4.23 and 4.24 demonstrate, the attenuation process and contrast mechanics are essentially the same for all important photons, regardless of their scatter angle and order. Figure 4.23. SABRINA genera ted photon track plot. Alum inum plate with shallow scattering type flaw. Severe collimation.

PAGE 73

Figure 4.24. SABRINA genera ted photon track plot. Alum inum plate with shallow scattering type flaw. Less collimated. Image contrast (bright vs. dark images) Images generated with the RSD imaging modality display flaws as either bright (high intensity) or dark (low intensity) region s of a sample. Bright and dark regions are a result of relatively more or relatively less photons, respectively, reaching the detector due to the presence of a flaw. This, in turn, is usually caused by a lack of or increase in photon beam attenuation resulting from the pres ence of a flaw. In the proposed transport model, this differential attenuation is a func tion of both flaw type and orientation as well as detector and collimator conf igurations. That is, the intens ity of a detected signal is not uniquely a function of electron density of th e region of the sample where the incident photon beam impinges. The explanation for th is is that, RSD techniques are sensitive to the photons exit paths though the media, wh ich depending upon system geometry and flaw depth, can be the source of significant attenuation. Th e contrast between exit paths

PAGE 74

of important contributing photons due to co llimation in RSD imaging modalities is illustrated in Figure 4.25 Figure 4.25 Difference is impor tant photon exiting paths caused by a focusing collimator extension. The additional distance traveled in th e material by collimated RSD contributing photons, highlighted by the (red) circle in the figure above, provides additional influences on the contrast presented by collimated RSD images. Considering relevant parameters in fl aw type and orientation and collimation configurations, most relevant imaging scen arios can be idealized by eighteen simple models. The parameters are critical reference scattering plane (a function of collimation configurations), flaw type, and flaw orientat ion (length). The collimator can be set so that the flaw lies eith er above or below the critical sc attering reference plane. (This imaginary plane, again, represents the depth at which the first important scatter occurs that can pass under the collimator and enter the de tector.) The flaw, relative to the object Uncollimated Collimated

PAGE 75

media, can be low density (void), high density scatterer or absorber. The orientation of the flaw can be such that both the incident and exiting photon fiel d pass through it, only the incident field passes though it, or only the exiting field passes though it. These eight characterizing parameters henceforth refe rred to by the following letter designations: D deep, the flaw is below the critical reference plane Sh-shallow, the flaw is above the critical reference plane Vvoid, the flaw is a void Scscatter, the flaw is a scattered (higher density) *low density scattere rs are treated as voids A-absorber, the flaw is an absorber (higher density) *qualitatively, low density absorbers ar e either treated as weak absorbers or as voids and treatment is conti ngent upon flaw dimensions and relative attenuation characteristics between flaw and target medium. This scenario was not considered in detail. L-long, the flaw is oriented so that bo th incident and reflected photon beams pass though it I-incident, the flaw is orie nted so that only the incide nt photon beam passes through it E-exit, the flaw is oriented so that only the reflected beam passes through it on its way to the detector. These parameters can be organized into ei ghteen permutations re lative to a control, non flawed sample. They are specified by th ree letter designations such that the first letter indicates flaw type (V/A/Sc), the second letter indicates the collimation

PAGE 76

configurations (D/Sh) and the third letter i ndicates the flaw orie ntation (L/I/E). The permutations are thus: VShL, VShI, VShE, VDL, VDI, VDE, ScShL, ScShI, ScShE, ScDL, ScDI, ScDE, AShL, AShI, AShE, ADL, ADI, ADE They are depicted in the following Figures 4.26 4.37: Figure 4.26. VShL: Void Shallow Long. More photons reach a depth at which they can scatter to the detector (below refere nce plane) due redu ced attenuation on the way down. More photons reach the detect or on the way out due to lack of attenuation on the way out. BRIGHT

PAGE 77

Figure 4.27. VShI: Void Shallow Incident. More photons reach a depth at which they can scatter to the detector (below refe rence plane) due to reduced attenuation on the way down. BRIGHT Figure 4.28. VShE: Void Shallow Exit. More photons reach the detector on the way out due to reduced attenuation on the way out. This is also the cause of shifts and shadows: BRIGHT

PAGE 78

Figure 4.29. VDL: Void Deep Long. This scen ario can either be vi ewed as bright or dark and is contingent upon the combin ation of two main mechanisms. The first mechanism is the lack of scattering at the flaw site due to the lack of material. This mechanism is similar to traditional Compton backscatter mechanism with a specifie d amount of material re moved form the top by preferential discrimination of the collimat or. This mechanism tends to lead to a dark image as expected by a void type flaw in traditional Compton backscatter radiography. The other majo r mechanism is the increase signal intensity due to the decr eased attenuation of the exiting photons due to the flaw. In this mechanism, the incident photons that traverse the flaw scatter below it and experience a lessened atte nuation on the way out due to the presence of the flaw. This mechanism tends to produce a bright image. The interplay between these tw o mechanisms and thus the overall contrast (bright or dark) of the image is very sensitive to parameters such as flaw height and depth, detector and collimator confi gurations and the attenuation properties (scattering to absorption ratio and over mfp) of the target material Figure 4.30. VDI: Void Deep Incident Almo st same effect as VDL but dark because photons are effectively transported deeper by the flaw and now travel further out through the material. Again similar to traditional CBI: DARK

PAGE 79

Figure 4.31 VDE: Void Deep Exit. More photons reach the detect or on the way out due to reduced attenuation on the way out: BRIGHT Figure 4.32. ScShL: Scatterer Shallow Long. Less photons reach a depth at which they can scatter to the detector (below reference plane) due to additional attenuation on the way down. Less photons reach the detector on the way out due to additional attenuati on on the way out. DARK (not e: if scatterer is pure scatterer or very low de nsity, the opposite effect may be observed and these flaws are treated as voids.)

PAGE 80

Figure 4.33. ScShI: Scatter Shallow Incide nt. Less photons reach a depth at which they can scatter to the detector (below reference plane) due to additional attenuation on the way down. DARK Figure 4.34. ScShE: Scatter Shallow Exit. Less photons reach th e detector on the way out due to additional attenuation on the way out: DARK

PAGE 81

Figure 4.35. ScDL: Scatter Deep Long. More photons are scattered at a shallower depth: BRIGHT Figure 4.36. ScDI: Scatte r Deep Incident. More photons are scattered at a shallower depth: BRIGHT

PAGE 82

Figure 4.37. ScDE: Scatterer Deep Exit. A dditional attenuation from denser material on the way out: DARK The six scenarios considered with an ab sorber type flaw re sult in photon paths identical to those for the six scatter type flaws de picted above and are, for the sake of reducing unnecessary redundancy, not shown. Flaws of dense absorber material will always produce a dark image because whethe r the dominant process is transmission or reflection dense absorber materials always decrease intensity relative to no flaw. As the figures above indicate, the relative intensities, high or low, are a primary result of the differential in attenuation pr ovided by a flaw. The mechanisms by which a flaw can perturb the degree of attenuation, as illustrated above, include both primary attenuation differences resulting from the portion of the flaw traversed by the photon field (incident and/or exit) as well as sec ondary differences resulting from the effective translation in scattering depth, either deeper (void) or shallower (scatterer), resulting from the presence of a flaw. Besides changing the path-lengt h (and consequently the degree of attenuation) of an exiting photon beam, tr anslating the scattering position (in depth) changes the solid angle subte nded by the detector and thus produces second order effects on the photon intensity reach ing the detector.

PAGE 83

The scenarios presented above are first scat ter simplifications meant to aid in the understanding of important contrast gene rating mechanisms and physical processes occurring within the imaging modality. In real world applications, results deviate to varying degrees from these models due to a number or second order effects. Namely, geometries which select for higher order co llision components and geometries which are more accurately described by a combination of two or more of the scenarios described above rather than just one. In these situations, bright and da rk images as well as expected absolute contrast become very sensitive to flaw, target, photon beam, and detector characteristics and dimensions. Thus, while these scenarios accurate ly describe a wide range of true applications, th ere are many which are not perf ectly described by this very simplified one scatter linear model. Further support of this model is provide d by the agreement, between predictions based upon rough analytical calculations, expe riments, and MCNP5 simulations, of the percent contrast for the simplified scenarios described above. The percent contrast (relative to a control no flaw scenario) fo r each experimental, analytical and MCNP5 simulations for a few of the scenarios is list ed in Table 4.1 below. Several important differences between the calculations, MCNP5 mo dels and experimental results should be noted however. The transport model, upon which the calculations are based are 2-D geometry and only roughly approximate the re al world 3-D scenarios. Additionally, for ease of calculations, the impingent photon b eam is treated as monoenergetic. The experimental results also incl ude real world uncertainties a nd efficiencies that neither MCNP5 the calculated results consider.

PAGE 84

Table 4.1. Relative contrasts as calculated and observed experimentally Configuration 1 scatter MCNP 5 Experiment Bright/Dark Agreement VDL* 33.8% 22.3% Bright Yes VShL 54.2% 29.7% Bright Yes VDE 15.9% 17.1% Bright Yes VDI -22.9% -21.1% dark Yes 2.54 cm radius NaI detector 2.9 cm fr om sample surface, 9 cm centerline-tocenterline form impingent photon beam. 0.15 cm high void channel flaw 0.65 cm deep, 1.5 cm collimator extension. ** MCNP5 data has error of less than 1%. Collimation trends (optimization) Both experiments and Monte Carlo simulati ons reveal inverse parabolic plots of contrast versus collimator extension for a wide range of flaw types and system configurations. The trend shows contrast to increase exponentia lly as collimation extension is increased from no extension up to a critical optimal extension length. The contrast then remains roughly constant as th e collimation is increased slightly and then begins to decrease exponentially as the collimator extension is further increased. Figures 4.3826 and 4.3928 are plots from previous publications of MCNP and experimental data, respectively, and both rev eal this trend.

PAGE 85

Figure 4.38 MCNP data plots of contrast vs. collimation length. This data is taken from an MCNP run with a 2 inch square de tector located 5.96 cm center-to-center from the impingent beam (radius 0.5 cm). The detector height is 6 cm and the void flaw channel is 0.3 cm below the aluminum surface. Analytical calculations based upon first scatter geom etry predict an optimum collimation length of 1.98 cm which matches well with the MCNP derived current tally optimum. Figure 4.39 Experimental data points of cont rast vs. collimation length. This data was taken experimentally and the results have not been veri fied in this study. The trend, however, assumes the predicted shape for contrast vs collimation curves.

PAGE 86

This trend also supports and is well descri bed by the simple direct scatter transport model. To illustrate, consider the simple, ideal case of a regular void type flaw of arbitrary dimensions positioned anywhere within a sample. ( Note that as the collimator is extended, for any particular geometry, th e critical reference scattering plane is moved accordingly deeper and the combination will hen ceforth be referred to as a movement of the reference plane.) As the reference plane is lowered from th e surface of the object to the top of the flaw, image contrast should increase since th e scatters in the mate rial above the flaw contribute only to noise and thus reduce th e relative contributi on of the important scatters. This increase in attenuation should be roughl y logarithmic (inversely exponential) since the photon is attenuated exponentially and thus each additional unit of depth into the sample contri butes exponentially less photons to the signal. The trend then, as the reference plane moves from the sample surface to the top of the flaw should look approximately like Figure 4. 40. This trend is derive d analytically, based upon the first scatter model, and is fu rther detailed in Appendix C.

PAGE 87

Figure 4.40 Analytically derived Contrast tr end as CRP is moved from sample surface to flaw top. As the reference plane moves from the t op of the flaw towards the bottom of the flaw, the contrast should remain about consta nt since no scatters occur within the void region of the flaw and consequently no phot ons are eliminated by varying the reference plane within this region. The trend will not be exactly flat in this region however, because increasing the collimator extension with in this region does affect the contribution of important multiple scattered photons and thus we expect, and observe, a slight downward slope within this region. Also, of note is that if the flaw is anything other than a void, i.e. a scatterer or an absorber, this downward slope will be more exaggerated since important scatters will occur within the flaw height region. Once the reference plane reaches the flaw (void) bottom and is moved towards the bottom of the sample or to the maximum pe netration depth of the incident photon beam, Flaw top CRP % Contrast

PAGE 88

we expect the contrast to d ecrease. This decrease is due to the fact that the photons below the sample, which are now being disc riminated against by the collimator, are important photons in that they tr averse the flaw and contribute to the signal differential. Since a relatively constant (assuming little variance in the K-N cross-section and solid angle) portion of these photons which interact below the flaw contribute to the signal, we expect and observe the contrast to taper off approximately as the signal does, i.e. exponentially. This trend is shown below in Figure 4.41 as the reference plane moves from the flaw bottom towards the maximum photon penetration depth. Figure 4.41 Analytically deri ved contrast trend as CRP moves below flaw bottom Combining the three segments described a bove results in a similarly shaped curve with similar inflection points observed in both experiments and MCNP5 simulations. Furthermore, it can be demonstrated that th e peak in the curve, the optimum collimation

PAGE 89

length, corresponds to the depth of the top on the flaw, and th at the width of the plateau in the trend corresponds approximate ly to the height of the flaw. Contrast vs Collimator Extension 40X40 cm Al Plate 9 cm to Det Radially 2.9 cm to Det Surface 0.08cm high flaw void channel 1cm wide .1cm below surface 0.00% 5.00% 10.00% 15.00% 20.00% 25.00% 30.00% 35.00% 40.00% 45.00% 00.511.522.5 col ext (cm)contrast MCNP5 cont 1scat cont 1st MCNP5 Figure 4.42 Contrast vs Collimator extension trend. MCNP5 simulations and 1st scatter model.

PAGE 90

Contrast vs Collimator Extension 40X40 cm Al Plate 9 cm to Det Radially 2.9 cm to Det Surface 0.18 cm high flaw void channel 1 cm wide .3 cm below surface -20.00% 0.00% 20.00% 40.00% 60.00% 80.00% 100.00% 120.00% 00.511.522.5 collimator ext (cm)percent contrast MCNP5 cont 1scat cont 1st MCNP5 Figure 4.43 Contrast vs collimation extensi on. MCNP5 simulation data and 1st scatter model approximation. 40X40X5.08 AL PLATE 4 CM TO DET CENTER 2.9CM TO DET SURF 0.08cm high void flaw 1cm wide .1cm below surface -10.00% 0.00% 10.00% 20.00% 30.00% 40.00% 50.00% 00.511.522.533.5 collimator ext (cm)contrast MCNP5 cont 1scat cont 1st MCNP5 Figure 4.44 Contrast vs collimation extensi on. MCNP5 simulation data and 1st scatter model approximation.

PAGE 91

Figures 4.42 4.44 reveal this trend for two flaw geometries for both MCNP5 data and for 1st scatter analytical calculations as desc ribed in Appendix B. Data for Figure 4.42 was acquired for a 40 x 40 cm aluminum plat e, 5.08 cm thick. The detector is a NaI crystal located 9 am centerline-to-center line from the beam and 2.9 cm above the aluminum surface. The flaw was a .08 cm ta ll void channel 1 cm wide and located at 0.1 cm below the surface. The collimator extens ion for the trend was varied from 1 cm to 2.32 cm. The MCNP5 data plotte d is from a current tally cro ssing the detector surface. The 1st scatter data is calculated analytically as described in appendix B. The mfp used was taken from the average mfp as calculate d by MCNP5. Figure 4.43 was similarly derived except that the flaw channel was change d to a height of 0.18 cm at a depth of 0.3 cm. Geometry for data shown in Figure 4.44 is identical to that of Figure 4.42 except that detector to photon beam cen terline-to-centerline spacing was changed from 9 cm in Figure 4.42 to 4 cm in Figure 4.44. As the plots demonstrate, the trend, in cluding inflection point s (as described in appendix B) and optimal collimation length assumes the predicted shape for both the analytical calculations and theMCNP5 si mulations. Further the trends, MCNP5 and calculated display identical optimal collima tion length, indicating that the contrast generating mechanism is indeed a first scatter type phenomena. That is, the contrast is generated by the attenuation differe nces as the flaw is traverse d directly by the photons. The calculated data and the MCNP5 acquired data, while displaying the same trend, do not match up identically. This is expected. The discrepancy between the first scatter analytical model and the MCNP5 data is attributed to two main factors. The first being that while MCNP5 uses an impingent x-ray beam with the true energy distribution the

PAGE 92

analytical calculations assu me an appropriate mono-energetic photon beam. Thus all properties and attenuation charac teristics, which are treated as continuous in MCNP5, are treated as a one group approximation in the ca lculations. The other reason is that MCNP5 data is for total detected signal while the analytical model considers only 1st scattered photons. Thus, as previously dem onstrated and thus expected, the analytical calculations display higher contrasts that the MCNP5 simulation because the analytical model only considers 1st scattered photons which typica lly have the highest contrast. Also of note is that for Figure 4.38 and Figures 4.42, 4.43 and 4.44 (Figure 4.39 is excluded because it is used only to demonstrat e the trend and the exact experimental data points have not been reproduced or confirmed in this study) the optimal overall contrast is achieved at the same collimation configur ation dictated by the optimal first scatter configuration. This is true even for situa tions when the true first scatter contribution is below 10% of the signal (e.g. for data in Figure 4.44 1st scatter component at the optimum collimation length represents only 1. 2% of the total signal) This supports the two main premises of the proposed photon transport mechanism. The first scatter component of a signal has the highest relative contrast and thus cont ributes a larger than expected (from overall signal frac tion) contrast contribution. The significant higher order components of the signal behave essentially th e same as first scat ter components for the purposes of providing det ectable image contrast. A more comprehensive analytic explanati on of this trend in contrast versus collimation, as handled by the transp ort model is included as Appendix C.

PAGE 93

Image pixel shifts and shadows Pixel shifts observed in the images between various detectors serve as an important proof of principle tool for va lidating the transport model. A clear angle and thus a projection is realized between the detector and the inve stigated object. The model predicts that, much like optic s, a relative shift will be observed between images taken from different view points, i.e. different de tectors. The phenomena responsible for both pixel shifts and shadows, or more appr opriately, pixel shifts of shadows, are demonstrated in the following schematic labele d Figure 4.45. As the illustration indicates there are four important variab les which determine the pixel shift observed by a particular detector. These include the thickness of th e sample (or the pene tration depth of the effective impingent photon beam), the depth of the flaw within the sample, the position of the detector (both x and y components) and the collimation extension past the detector surface. In Figure 4.45 these vari ables are labeled as t, d, R, and c, respectively, and the shift in shadow relative to the true flaw location is given as s.

PAGE 94

Figure 4.45 Geometrical considerations for sh adow pixel shifting relative to detector position. The following set of images, Figures 4.46 and 4.47, demonstrate, according to the above schematic, the shadow pixel shifts as observed by the various detectors. In each image, the arrow point in the direction that th e effective shadow is cast. This is away from the position of the detector. That is, th e shadow image is formed when the flaw is between the incident beam and the detector thus the image appears on the opposite side of the flaw as the detector. Thus the arrow points from th e detector position towards the beam.

PAGE 95

Figure 4.46 Bright shadow cast by void in Al seen by detector in lower right hand corner. Figure 4.47 Bright shadow cast by void in Al seen by detector in lower left hand corner. Signal intensity Total signal intensity in the RSD imag ing modality is observed to decrease markedly with increased collimation. C onsidering the exponentia l attenuation of the incident signal this should be expected based upon the same arguments presented above

PAGE 96

in the collimation extension section. As collimation is increased and the critical scattering reference plane is consequently m oved deeper into the sample, more of the scattered signal is being di scriminated against. As the scattered signal varies as interaction rates which vary with incide nt photon flux, the characteristic exponential attenuation is also observed in the signal intensity decrease with collimation extension. Applications and Limitations To date, RSD imaging modalities have been successfully applied to various materials including, SOFI foam, aluminum, plastics, landmines buried in soil, concrete, drywall, and reactor insulation. Each of these materials possesses unique interaction and attenuation properties and presents specific important flaw types. Therefore, each specific imaging task represents its own unique problems and inherent limitations.

PAGE 97

77 CHAPTER 5 BACKSCATTER FI ELD DISTRIBUTION AND DETECTOR PLACEMENT Backscattered X-r ay Signal Profile The backscattered photon field is dist ributed according to the appropriate differential scattering cross-se ction. Thus, the backscatte red photon distribution should be symmetric about the axis of the impingent beam. The signal profile should also be peaked at 180 degrees (direct backscatter) and sinusoidally taper off towards a minimum at 90 degrees. The sharpness of the peak a nd speed of the tapering are functions of the incident photon energy, with the scatter profile becoming more isotropic as energy is decreased. The unmodified (i.e. without c onsideration of atomic form factor) KlienNishiena cross-section for a 55 keV photon beam is presented below (Figure 5.1) as a function of scattering angle. The backscattered photon flux through a plane parallel with the surface of foam target (w/ Al substrate) is presented below as Figure 5.2. This plot is taken from data acquired via MCNP5 simulations The foam target in Figure 5.2 is eight inches thick with a one inch aluminum substr ate. The tally plane is 1 cm above the surface of the foam. The impingent photon beam is 75 keV peak spectra with a 0.5 cm diameter. In Figure 5.2, as expected from the shape of the differential scattering crosssection, the highest backscat tered photon flux is directly above the source (180 degrees scattering angle). This plot, however, repr esents the flux across a surface and is thus more exaggerated than the Klien-Nishiena cr oss-section because the flux is dependent upon both the relative fraction of phot ons scattered into a particul ar solid angle as well as

PAGE 98

the relative orientation of the reference pl ane to the incident photons. Since photons scattered at smaller angles (t o the horizon), i.e. closer to 90 degrees than to 180 degrees across the reference plane at more severe angl es (further away from perpendicular), these surfaces have less effective area and thus w ill display lower fluxes than the scattering cross-sections would otherwise dictate. Additionally, sin ce the tally is taken across a plane rather than a sphere, the photons reach ing the outer mesh voxels are geometrically attenuated by the 1/r^2 law and t hus the fluxes are further reduced. Figure 5.1 Klien-Nishiena differential scattering cross-section for 55 keV photon

PAGE 99

Figure 5.2 Backscattered photon flux across a plane parallel to SOFI sample surface As the transport model indicates (chapter 4) the primary cause of a detected signal differential (i.e. image contrast) is a change in the scattering/att enuation characteristics of the photon field as a result of a flaw in the imaged target material. Since to a first approximation, the presence or absence of a flaw changes only rela tive interaction rates within a specific volume (about the flaw) and has no effect on the directional distribution of the scattered photon field, we would expect that the same signal (i.e. same relative contrast) could be detected from any pa rticular solid angl e component of the backscattered photon field. In fact, for a perfectly symmetrical flaw in a uniform medium, this is indeed the case as illustrated in Figure 3. Figure 3 is a plot of relative differences (% contrast) in photon fluxes across a meshed tally plane oriented parallel to a target sample surface. The target, as in Figure 5.1, is a SOFI foam material on an aluminum substrate. The data for the plot is taken from 2 MCNP5 simulations. The first simulation utilizes a 15 x 15 cm mesh tally take n over an eight inch th ick sample of SOFI on a one inch aluminum substrate. The sec ond simulation is identical except that a one

PAGE 100

cm diameter spherical void flaw is placed tw o and a half cm below the sample surface. In both cases the incident photon beam is 55 ke V peak and the mesh tally plane is located 1 cm above the surface of the foam. The b ackscattered photon flux though the mesh tally for simulation 1 is shown above as Figure 5. 2 and the backscattered flux for simulation 2 (with the flaw) would have an identical rela tive distribution. The percent difference between these two simulations is presented be low as Figure 5.3. As the plot shows, relative differences in the signals are essentia lly the same across all voxel elements of the mesh plane. The high and low peaks observe d towards the outer edges of the plot are statistically insignificant and ar e a result of the relatively fe w number of photons crossing the outer mesh elements. Figure 5.3 Percent difference in signal due to void flaw in SOFI as a function of scatter field component. Notice that the cont rast is evenly di stributed across the entire scatter field. Detector Placement Considerations The ideal scenario described above in whic h the same relative contrast is obtained regardless of scatter component selected doe s not indicate that detector placement and orientation are irrelevant for image acquisition. On the contrary, in real situations, detector orientation and a ppropriate selection of scat ter-field components are of

PAGE 101

paramount importance in generating meaning im ages and accurate detection of flaws. The reasons for this are manifold. In many real imaging scenarios, the signal perturbation caused by a flaw, particularly a deep flaw, is masked by the relative contribution of the material above the flaw. Thus the signal originating from this region must be effectively discriminating against. In the RSD imaging modality, this discrimination is accomplished by selecting the appropriate scatter component based upon geometrical collimation as shown in Figure 5.4. Simple trigonometry indicates that the further away (laterally) from the incide nt photon beam, and the smaller the diameter of the detector, the easier it is to effec tively collimate to a specific depth below the surface of a sample. This concept, howeve r, is checked by the fact that the signal intensity, as shown in Figures 5.1 and 5.2, is strongest closer to the photon beam. Additionally certain scanning applications necessitate tightly packed detector configurations in order to fit into specific geometries. Furthermore, most realistic flaws are not uniform and symmetrical and thus w ill exhibit varying degrees of differential attenuation (and thus various percent contrasts) depending upo n the specific scatter field components selected for and its specific (and uni que) path through the flaw and the target material. Detector configuration is also cr ucial because in a many situations, an imaged sample will have natural artifact and contour s in it. These artifacts can, depending upon detector positioning and configuration, obscure important flaws either geometrically, or by saturating the local contrast or by casting a shadow (also geometric in nature) over an important region.

PAGE 102

Figure 5.4 RSD focusing. Each collimation configuration A, B, C selects for photons originating at and below each specif ic depth A, B, C, respectively. Thus each specific imaging task requires a unique balance to be reached between maximizing signal intensity (thus reducing s can time) and optimum detector placement for focusing. Realistic spatial constraints, which are also task-specific, must also be considered. Consequently each new imaging ta sk presents a unique problem and requires a unique solution with tailored detector configurations for optimum image generation and flaw detection. The more freedom the dete ctor assembly provides for varying detector configuration the easier and more effective imaging optimization becomes. A new detector assembly should allow for detector s to be able to move both laterally and vertically with respect to the photon beam and should also provide a means for the detector surface to be pivoted about an axis perpendicular to its longitudinal axis. This pivoting will allow for the detector face to always be oriented so that the important scatter field components impinge upon it normall y. Consequently, the detected photon

PAGE 103

flux will be increased, ultimately leading to a decrease in scanning time. Additionally, if the photons strike the detector normally, more of the photons which s catter at the surface of the detector will scatter into rather than out of the detector and thus more efficient energy deposition will result.

PAGE 104

84 CHAPTER 6 RSD OPTIMIZATION AND IMAGING CHARACTERISTICS Optimization Principles RSD modality optimization is achieved w ith consideration given to both sample material and flaw type and configuration. The degree to which each of these commands specificity varies based upon the physical pa rameters such as MFP and scattering to absorption ratio of the photons interacting in the sample material as well as the relative interaction properties and orientation of the fl aw. Since a detected signal contrast is due primarily to a differential in attenuation caused by a flaw, certain flaw types, such as high density absorbers and scatterers in a low density media, are detectable in a variety of configurations and thus need not be perfectly optimized for. Other flaw types such as small cracks and delaminations or slight density fluctuations are more difficult to detect and can only be done so under optimized conditions. Variable parameters of the system include detector spacing, collimation extension and rotation, incident photon energy, and phot on beam spot size. Geometric variables such as collimation and detector configura tions, as previously described, function to select specific components of the backscat tered photon field originating from scatters occurring in specific regions of a sample. These parameters are usually adjusted to focus a detector on an important region where a flaw is suspected or crucial to sample integrity. Photon beam spot size is directly related to the resolution required to detect a flaw. The photon beam spot size must be approximately equal to the pixel size to achieve meaningful results. Situations wher e the spot size is much larger than the

PAGE 105

pixel size result in blurred images due to an effective averaging out of the region impinged upon by the larger beam. Such c onfigurations lead to an inability to effectively detect small flaws. Since decrea sing the spot size and w ith it the pixel size of the acquired image, increases total scan time, a balance must be reached between resolution and realistic scan time. Photon beam energy is optimized both for flaw type and depth as well as for substrate material. The correct degree of penetration and scatting is important to image a flaw. The incident spectrum must pe netrate to and scatter back from at least the depth of the flaw and al so interact enough to provide a meaningful backscatter signal. SOFI Foam The spray on foam insulation (SOFI) ma terial provided by Lockheed Martin presents a unique imaging challenge. The SO FI is applied over a contoured aluminum substrate. The aluminum s ubstrate contains bolts, nuts, flanges and other structural components which effectively absorb contrast away from the actual foam and thus make flaws more difficult to detect. The lo w density of the foam, roughly 0.03 g/ cm3, results in a large fraction of the inci dent photon beam penetrating all the way to the aluminum substrate. The imaging process thus become s complicated by a strong transmission-type phenomena with an effective source distribut ion located at the aluminum substrate surface. Figure 6.1 depicts the two important pathways for a photon to reach the detector. Photons can either interact in the foam and s catter directly into the detector (B) or they can penetrate the foam and reflect off the al uminum substrate towards the detector (A). The relative probability of each of these pathways is a function of flaw parameters, foam thickness, incident energy spectra, and detect or and collimator configurations. For high

PAGE 106

density flaws, photons may also interact with in the flaw itself and thus be scattered towards the detector. This enhances flaw contrast and greatly simplifies the imaging process, however in practical ap plications there is rarely a high density flaw in the SOFI material. Figure 6.1 Two important path s of a backscattered photon Figures 6.2 -6.4, below, catalog the eff ect of both collimation and energy on the relative detector contributions from each of th ese pathways for a few scenarios. The data for this plot is taken from several MCNP5 simulations. Each simulation models an eight inch think sample of SOFI divi ded into four, two inch layers. The SOFI is mounted to a one inch thick aluminum plate. The paramete rs varied were incident beam spectra peak energy, detector to sample distance (height ) and collimation extension. The data presented below comprises 6 MCNP5 simulations. Three different detector and collimator configurations were each run at 60 keV and 75 keV peak spectra. The detector and collimation configurations ar e described in the plots below as X-Y. In this format, X A B

PAGE 107

describes the distance from the NaI surface to the SOFI surface (in cm) and Y describes the collimation extension (in cm) past the NaI surface. The absence of a Y indicates that the collimator is flush to the NaI surface. Thus, the three geometries simulated: 5.14, 5.14-4 and 1.14, describe 5.14cm from NaI to SOFI with flush collimator, 5.14 cm from NaI to foam with 4 cm collimator extension and 1.14 cm from NaI to SOFI with flush collimator, respectively. Figure 6.2 is a SABRINA generated plot of the geometry and shows the numbering scheme for foam layer reference. Figures 6.3 and 6.4 plot the relative detector (energy de position) contribution from photons having their deepest collision (thus max penetration depth) in each layer of foam as a function of collimation configuration, for 60 keV and 75 keV incident photon spectra, respec tively. These plots reveal, as expected, that increasing collim ation causes the contribution from photons scattered in the deeper layer of the SOFI to become more important. It is important to realize that the net backscattered photon field (for each incident beam spectra) is identical for each geometrical variation. What change s is the portion of that field which reaches the detector and thus becomes the detected si gnal. That is, the geometrical configuration does not change the actual backscattered fi eld, merely which part of it we detect.

PAGE 108

Figure 6.2 MCNP5 simulated geometry. F our, two inch thick layers of SOFI on aluminum substrate. 0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% AlFoam 1Foam 2Foam 3Foam 4Detector Energy Deposition by cell (60 keV) 8" Foam 1.14. 5.14-4 5.14. Figure 6.3. Detector contribution by cell as a function of collimation. 60 keV incident spectrum.

PAGE 109

0% 10% 20% 30% 40% 50% 60% AlFoam 1Foam 2Foam 3Foam 4Detector Energy Deposition by cell (75 keV) 8" Foam 1.14. 5.14-4 5.14. Figure 6.4. Detector contribution by cell as a function of collimation. 75 keV incident spectrum Voids in foam Void-type flaws are the most difficult to image in the SOFI material because the relative density between void (air) and foam is very slight, 0.03 g/ cm3 versus 0.001g/ cm3. For these flaw types, optimization of th e system is crucial in order to acquire meaningful images. The source of the signal difference between void and non-void regions of a foam sample is th e lack of interaction within the flaw volume relative to the non-flaw region. This lack of interaction due to the void flaw has two effects: it causes less scatters at the flaw loca tion to be directed towards th e detector and it causes more scatters from lower regions, in cluding reflection off the aluminum substrate, to occur and thus be directed towards the detector. Consequently, depend ing upon the interplay between these two mechanisms, dictated by co llimation configuration, a void-type flaw can be imaged as either a bright or a dark re gion. That is, the system can focus on either

PAGE 110

the weaker scatter field retu rning from the flaw region or the stronger scatter field returning from below the flaw region. This is completely determined by collimation configuration. If the flaw is above the CPR, then the scatters occurr ing at the flaw site and above are discriminated against and thus the dominant mechanism is the increased signal originating from below the flaw. Again, the reason for this increased signal is the lack of attenuation caused by the flaw, effec tively transporting more photons deeper into the sample. If the flaw is below the CRP then (because of exponential attenuation of the signal) the dominant contrast generating mechanism is the lack of scatters occurring at the flaw site. In this situation the flaw is imaged as dark. In an uncollimated configuration, the flaw (assuming it is not too deep or too small, in which case it would not show up on the uncollimated image) will be imaged as a dark region. As the collimator is increased so that the CRP appr oaches the top of the flaw, the flaw will be imaged as an increasingly dark image. Once the collimator is increased so that the CRP is below the flaw, the flaw will be imaged as a bright region as the secondary (increased penetration) effect begins to be dominant.

PAGE 111

Figure 6.5 Void-type flaw in SOFI. CRP is optimally set to be just above flaw. With collimation set as in figure 6.5, the scatters originating from above the flaw, which are identical in both flawed and non-flawed situa tion, are eliminated from the signal. The detected signal is thus composed of photons inte racting at the flaw depth and below. In a void flaw situa tion virtually no photons interact at the flaw depth and thus the photon beam is transported, unattenuated, d eeper into the sample by the height of the flaw. This, as mentioned above, results in a dark image of the void-type flaw. The mechanism can be thought of as either of two ways. The firs t idea is that, relative to the non-flawed sample, there are less interactions and thus less s catters occurring at the flaw region and thus there is a weak er signal coming from this region. The second concept is that the photons that do not intera ct at the flaw depth interact instead at a distance deeper into the sample, and this, re lative to where they otherw ise (in a non-flawed sample) would have interacted, the have a longer pa th and thus experience greater attenuation CRP

PAGE 112

upon exiting the sample towards the detector. This represents the optimum configuration for imaging void-type flaws in SOFI. Ov er-collimation and relying upon the secondary increased signal (bright image) is not as e ffective since it represents a significantly weaker signal. The following Figure 6.7 illustrates several void type flaws imaged in foam at the indicated depth and collimation settings. Sin ce a void is usually imaged as a dark region and is effectively filtered out once the critical scattering reference plane is below it, it is believed that the dominant effect which leads to detection of flaws is their lack of scattering rather than their increased transmission. If, however, a void in SOFI is sufficiently large and collimation is prope rly configured, the increased transmission effect can be observed and a void flaw will be imaged as a subtle bright area. This can be seen below in Figure 6.6. This figure repr esents a foam calibration panel (Appendix D). In this panel the upper two holes are at subs trate level (eight inches) and the lower two holes are at mid-plane level (four inches) be low the foam surface. In each of these images, dark regions represent flaws below the CRP and bright regions represent flaws above the CRP (over-collimated) as discussed above.

PAGE 113

Figure 6.6 Images of foam calibration pa nel with varying degrees of collimation. Collimation increases clockwise from lower left. In Figure 6.7, below, the scattering phenom ena caused by a void-type flaw in SOFI foam are shown. Scatters which occur above the flawed area are identical in either flawed or non-flawed cases and thus contribute to noise if th ey are not filtered out. The area between the two thick (dar k) converging arrows, originating at the void, represents the region from which no scatters occur. The lack of scatter from this region due to a void-type flaw represents lessened detected si gnal intensity and thus a dark image of the flaw. These un-scattered photons will, howev er, suffer a collision somewhere deeper within the foam. From these deeper coll isions, the photons not only see the detector through a smaller solid angle (to a first approxi mation due to increased distance), but also must travel through more material on the wa y towards the detector. The path difference caused by a void on a photon which consequently interacts deeper within the sample is shown by the two (color) thin arrows.

PAGE 114

Figure 6.7 Void-type flaw in SOFI. Thin a rrow demonstrates the path difference induced by the lack of scatter at the void site. High density absorber and scattering type flaws Detection of high density, especially absorb ing, materials in a SOFI-like substrate is a trivial task. The density of such material s, lead or aluminum for example, is several orders of magnitude larger th an the substrate and consequen tly stands out as a contrast regardless of the configuration setting. The detector respons es to optimization routines and various geometrical confi gurations used for high density flaw detection is, however; of significant academic interest as it elucid ates the photon transport processes occurring in this RSD imagine modality. Unlike void type flaws in SOFI, whose presence can often be effectively filtered out by lowering the critical reference scatte ring plane sothat they lie above it, high density type flaws, in SOFI, because they have strong affects on both transmission and reflection, can not be similarly filtered out. This is shown below in figure 6.8. If the collimator is set to discriminate above the blue line (labeled 1), the photons shown in blue

PAGE 115

(top arrow) will still experien ce significant relative attenuation as they pass through the absorber and upon subsequent reflection (off th e AL) they will be detected as a decrease in signal intensity. If the collimator is se t so that it discriminates above the red line (labeled 2), the initial image will be generate d by the lack of (absorber) or increase in (scatterer) scattering at the fl aw location, shown in red (low er arrow), relative to the non flawed section. Figure 6.8 Two mechanisms (1-lack of scatte r, 2increased attenuation) for generating low intensity signal from an absorber-type flaw in SOFI. The dual nature of image generation and the relative importance of each photon pathway are demonstrated in Figure 6.6. Th is is an image of several thin aluminum inserts in a foam media. Each insert is set at a different depth. Notice that some of the inserts appear as dark and some as bright images. This is due to the phenomena illustrated above in figure 6.8. Each insert necessarily produces both images generated by the photon path represented by a solid li ne and the photon path represented by the 2 1 1 2

PAGE 116

dashed, respectively. For the case of a scatte ring media, such as aluminum, the solid path (1) will result in an intensity increase (highe r scattering) and the red path will generate an intensity decrease (higher atte nuation). When under collimated (shown in by the solid scattering reference line) the soli d (high intensity scattering) effect is dominant. This is due mainly to the fact that it represents a shorter total path through the material and thus is attenuated less. Other factors increase this effects dominancy including the reduced solid angle subtended by the detector, the additional scattering that must occur for the dashed effect, and the high scattering to absorption ratio of aluminum. When the collimator is set such that the red line now becomes the critical scattering reference plane, the bright increased scattering signal no l onger reaches the det ector. Under these settings, the dark, low intensity, signal due to the increased attenuation as the beam passes though the flaw is now observed. In th e figure below, Figure 6.9, the collimator is positioned such that the critical refere nce scattering plane is located between the aluminum inserts. The bright flaws lie be low the plane and thus increase scattering towards the detector and the dark flaws are shallower and lie above the plane and thus represent only an increase in attenuation.

PAGE 117

Figure 6.9 Scan of depth staggered aluminum inserts in SOFI. Bright inserts are above CRP and dark inserts are below CRP. Shadows Shadows observed in images acquired of th e foam material are a result of high density flaws in the material through which an exiting photon traverse s on its way back to the detector. These are different than th e dark images observed from over-collimated images of high density flaws. The over-col limation produced dark images, as described above, are a result of attenua tion of the primary photon beam before it scatters back towards the detector. These images occur when the incident beam impinges directly

PAGE 118

above the actual flaw. Dark shadow imag es, on the other hand, are produced by the scattered photon beam on its way toward the detector. These images are displaced by a distance dependent upon the detector geometry Since these shadow images result from photon exit path attenuation, they can not be filtered out by collimation. The mechanism describing both the shadow phenomena and its relative displacement is illustrated below in figure 6.10. In this schematic, the double arrow indicates the shift between the position of the actual flaw (dashed line) and the position of the shadow image (impingent beam). As shown, shadows are not created wh en the beam is directly over the flaw, but rather when the flaw is between beam and detector. In this geometry, the photon beam traverses the flaw, after scattering, as it travels towards the detector. Figure 6.10. Mechanism for shadow image gene ration. Dashed line represents true flaw position, solid arrow indicated sh adow image detection position.

PAGE 119

These shadows can be readily observed in figure 6.9 above as well as in many of the following foam images. In the image of the aluminum inserts in SOFI, Figure 6.9, a dark shadow can be seen associated with each insert. In these images it is evident that regardless of flaw position relative to the CPR, the shadow is always a dark image. Theoretically, based upon the proposed mechanisms, voids in foam should produce bright shadow type affects. However, experiment s done with the current system have not yet been able demonstrate this. Most likely the s light change in attenuation due to the flaw on the exiting photon beam is on the order of magni tude of the inherent noise in the system and thus can not be resolved. The following figures represent typical images acquired of the SOFI foam material. Descriptions are included to point out certain features a nd characteristics. Figure 6.11 SOFI panel with contou red aluminum substrate. B&W

PAGE 120

Figure 6.12 SOFI ramp panel with cont oured aluminum substrate. Color. Figures 6.11 and 6.12 demonstrate the di fference in contrast between gray scale and color scale images of the same scan. Each is taken at 60 keV on ramped flange panel 2 (Appendix D) with artificially inserted flaws. Notice, however, that some flaws shown in the gray scale image are not in the color image. This is due to the different projection of the detectors. The flaw in the middle stringer, for example, is not seen in the color image because from that detectors viewpoint the aluminum stringer hides the flaw. Figure 6.13 is an image of a bolt and stringer panel under SOFI.

PAGE 121

Figure 6.13 Bolt of flange panel under SOFI. Aluminum System optimization for imaging aluminum objects precedes in much the same manner as for foam objects. Geometrically the mechanisms are the same and thus similar collimation length and detector positi oning must be considered. The density of aluminum, however, is orders of magnitude larg er than that of the SOFI material thus the incident photon energy must be accordingly adjusted and the beam is not expected to penetrate as far. Additiona lly, in aluminum the important flaw types are almost always void type flaws, manifested as cracks, holes, or fractures. The photon spectrum used to investigate aluminum samples was typically 75keV peak energy. The effective meanfree-path of this spectrum in aluminum calculated by MCNP5, is about 1 cm. Since the density differential between aluminum (2.7g/ cm3) and void (air .001 g/ cm3) is so large, imaging large (~1cm photon beam chord) voids is easily accomplished with this modality and collimator focusing is often not necessary unless the flaw is deeper

PAGE 122

than about half of a centimeter. These large void type flaws while often trivial to image, are of fundamental import because of the in sight they lend to the imaging process and photon transport mechanics involved. In aluminum objects, it is easier to see how important flaw alignment and detector geometry is and what effects they have on the resultant images. The same flaw depending upon configuration can be imaged as e ither a dark or bright region. This is related to whether the flaw provides an incident photon beam with relatively more or relatively less aluminum to travel through before reaching the detector. This phenomenon is illustrated in the following sequence of figures. The following images were taken of a 1 inch thick aluminum plate with three channels, each 1 cm wide by .1 cm high and located at0.3 cm, 0.5 cm and 0. 7 cm below the surface respectively. Each image in Figure 6.14, below, was acquired from a different detector during the same scan. The incident energy was 75 keV peak. The flaw s are oriented at 45 de grees so that they either subtend a direct line between the sour ce and the detectors (detectors 2 and 4) or they lie perpendicular to such a line (detectors 1 and 3). Detector s 2 and 4 view the flaw much as the VDL scenario described in chapte r 4. In this confi guration, the flaw is imaged as a high intensity, bright region because, the shadow effectively overlaps directly the flaw so that the combined en try and exit path of the photon, due to the presence of the void channel, represents a le sser attenuated path. The image of detector 4 is a highly over-collimated case and image degr adation is obvious. In this view (detector 4) only a portion of the flaw ch annel is evident. This is due to the collimation induced geometry. The flaw will only be imaged as br ight as long as the reflected beam passes through only the flawed channel on its way to wards the detector. As the collimation

PAGE 123

increases, the angle becomes more severe a nd this distance, across the increases. At a point the detector will be positioned so th at the exiting photon beam goes through not only a portion of the channel but also a portion of the unflawe d aluminum plate. At this point the contrast will be either lost or the flaw will be imaged as dark This is illustrated below in figure 6.15. Essentially, when scanning a channel flaw aligned with the detector, at some point the mechanism changes from VDL to VDSh (as described in chapter 4). At this point the high intensity signal will be lost. As collimation increases and the angle becomes more severe, this changing point becomes a larger part of the flaw, much as optical shadow length increa ses with angle of light projection. Figure 6.14 Clockwise from upper left, detectors 1,2,3,4.

PAGE 124

Figure 6.15. Photon exit paths across a void channel. Detectors 1 and 3 are oriented perpendicularly to detectors 2 and 4. In Figure 6.14 the channels are imaged as dark low intensity regions in these detectors. In this situation the channels, relative to the detector and source orientati on are no longer long channels since they are not aligned along the detector to source axis. This configuration forces the flaws to be images by the VDSh mechanism. Also of interest is the bright shadow flanking these images from detectors 1 and 3 in Figure 6.14. These bright shadows are produced by the mechanisms previously di scussed (also see VDE, chapter 4) and are located on the opposite side of th e flaw as the detector is. That is, they are produced when the flaw is between the detector and the source and thus appear to be directed away from the detector. Similar to absorbers in foam, voids in aluminum cast shadow type images. The images are similarly produced and similarly displaced form the actual flaw position. The

PAGE 125

shadow images of a void, however, are a bri ght region. This br ight region of high relative photon intensity is caused by the lack of attenuation afford ed by the void when it is between a photon scatter site and the detect or. This is the same phenomenon illustrated in Figure 6.7 above for foam. Since the sha dow is caused by the void being between the detector and the incident beam, it is necessary and expected that each detector will see the shadow at a different location. The shadow should extend directly from the flaw towards the detector. This is observed and shown be low in Figures 6.16. This again is an aluminum sample plate. This plate had holes drilled into it. The four images correspond to the four different detectors and the bright shadow is accordingly displaced in each one. Figure 6.16 Shadow shifting with detector position

PAGE 126

As noted in the foam example, since th e shadow effect is related to the exiting photons path, it can not be collimated out. Th e actual image of the flaw however, since it is the incident path that affects this, can be collimated out. This is seen below in Figure 6.17. Here the collimator is set so that the critical scattering reference plane is below the deepest flaw. Thus the dark image of th e flaw is not observed and only it shadow projection is seen. Figure 6.17 Bright shadow images of aluminum flaw plate. 5 cylindrical void flaws at various depths are imaged as brig ht with severe over-collimation. The shadow effect is the mechanism re sponsible for producing bright images of the channel-like flaw observed above in Figur e 6.14. When the detector and flaw are thus aligned, the shadow of the flaw identically overlaps the actual flaw and thus a bright image (if the attenuation deficit is significan t) can be obtained. Notice that in Figure 6.11, there is a bright area at one end of the channel and a dark image at the other end. The shape and intensity profiles of the fl aw shadows are direc tly related to the

PAGE 127

dimensions of the flaw and the offset is dir ectly related to the depth of the flaw. These relations, as detailed in the schematic gi ven as Figure 6.18, are contingent upon the detectors viewpoint relative the flaw as well as to the effective pene tration depth of the impingent photon beam. Figure 6.18 Schematic of flaw shadow and detector orientation relationship

PAGE 128

Figure 6.19 Correlated (processed) im age of sample aluminum plate Figure 6.19 demonstrates the correla tion routine of the LABVIEW image processing program used to interpret the images 33. This represents a simple correlation where each detector is weighted equally. If each detector is similarly collimated so that they each have a symmetric view of the im aged object (as in Figure 6.19, above) the effect of the correlation is to effectively remove shadowing effects. The long, streaking shadows observer about each flaw in the sing le detector views becomes a small halo around the flaws in the correlated view. This is essentially an effect of adding the contrast of each separate detector togeth er to generate a final correlated image. The following set of Figures, Figures 6. 206.22, demonstrates the effect of increasing collimation on the acquired image in aluminum samples. The target, again, is

PAGE 129

sample plate #2 (Appendix D). Each Figure, 6.20-6.22, is acquired with from the same detector (detector 4) with the same incident energy spectrum (75 keV peak). Figure 6.20 is taken with the collimators withdrawn so that the CRP lies above the aluminum surface. This is essentially an uncollimated image. The figure shows that the shallower two flaws are imaged as dark regions with high intens ity shadow regions. The next two deepest flaws are again imaged as dark regions, but th is time (although barely visible in flaw C) there is no associated high inte nsity shadow. This is becau se the flaws are deep enough that the signal from the aluminum above th em, since it is not collimated out, overwhelms the slight high intensity signal that originates from below the flaw. Flaw E, the deepest one, is not even observed, for the same reason. Figure 6.20 Uncollimated image of aluminum flaw plate.

PAGE 130

Figure 6.21 is acquired with th e collimator set so that the CRP is just above the bottom of flaw A. Here each flaw, as optim ally expected is presented as a dark, low intensity region, with an associated high inte nsity, bright shadow. Notice that flaw A, which should have the most inte nse bright and the most intense dark image, has a barely observable dark image. This is due to the fact that the collimator is set of just barely above the bottom of this flaw. Such collim ation, by discriminating against all scatters above the CRP, effectively images the flaw as if it were only as high as the distance between the CRP and the flaw bottom. The bright shadow, however is not degraded with over attenuation since, as mentioned before it is an effect of the emergent scattered beam, not the initial impingent beam. Figure 6.21 Collimation set to discriminate just above shallowest flaw. Flaw depth increases from lower right, counterclockwise to center.

PAGE 131

Figure 6.22 shows an image acquired with the geometry severely over-collimated. In this image, the CRP is set below the botto m of the deepest flaw, Flaw E. As shown, each flaw is observed only as a bright region. This is because the true image of the void, the dark region is completely collimated out, thus only the bright shadow is detected. The bright, high intensity regions are slightly offset, since they are shadows, from the actual void flaw location. Figure 6.22 Over-collimated image of aluminum sample plate.

PAGE 132

Figure 6.23 Over-collimated image of small channel aluminum plate. Figure 6.24 Under-collimated image of small channel aluminum plate. Figures 6.23 and Figure 6.24 are taken with 75 keV peak incident spectra. They are of a small channel aluminum plate (A ppendix D) with voidtype flaw channel machined at various depths. Figure 6. 23 and 6.24 are overand undercollimated, respectively. Due to previously discussed m echanism, the channels are thus imaged as bright and dark regions for the over-collimated and under-collimated cases, respectively. Plastic The extremely high scattering to absorption ratio of most plastic materials makes imaging subsurface flaws without collimation fo cusing nearly impossible. Even without surface features, the initial scattering from a plastic surface often saturates the signal, masking the effects of any subsurface feat ures. The addition of a collimator, however,

PAGE 133

allows for this surface and shallow scattering to be removed from the detected signal increasing the relative signal contribution of deeper penetrating photons which interact with the void region. The drastic effect of proper collimation focusing is shown below in Figure 6.25. Both of these images are of the same plastic sample plate with a 1 cm wide by .15 cm high flaw running down its center. For the first image on the left in Figure 6.25, no collimator is used and only the surface is imaged. In Figure 6.24. a collimator is set to focus on the flaw, by moving the critical references scattering plane to just above the flaw surface. In this conf iguration, the subsurface channel now becomes apparent. Figure 6.25 Plastic flawed plate #1, unc ollimated on left, collimated on right. The high scattering and lack of absorp tion in plastics also requires low photon energy be used. Typically 55 to 60 keV peak in cident spectra provide the best results for imaging plastic type materials. Despite the di fferences in interacti ons properties between plastics and aluminum, the governing mechan ism and hence the optimization processes

PAGE 134

are similar to those of aluminum. Dark sha dow type regions on the close end of flaws are not as dark in plastic since the absorption is markedly less pronounced in this material. Concrete and Gypsum Imaging of concrete and gypsum sample s usually indicated scanning for objects concealed behind walls. For this type of scenario, higher photon energies (up to 100 keV peak) are typically used and collimation setti ng is configured so as to focus past the inch to 1 inch of wall material which is us ually represents impedance to imaging. Additionally, for instance, when collimation ex tension alone is not sufficient to focus to the desired depth or when higher count rates are desired, above that which the properly collimated detector receives, closing the det ector to sample gap has a similar effect, but functions by a different mechanism, as in creasing the collimation length. The KleinNishina cross-section dictates scattering angle distribution peak towards zero (forward) and 180 (backward) degrees. Thus as the sample to detector distance is decreased, deeper penetrating photons, having scatters closer to 180 degrees are preferentially detected. As energy is increased near the 100 keV peak wh ich is often used form imaging concrete (represents the limit of the current Nylon ge nerator capabilities) the backscatter angular distribution peaking towards 180 degrees be comes more pronounced and this method of focusing becomes more effective. The followi ng figures are indicativ e of typical results achieved with the current RSD imaging modality on various concrete and gypsum targets.

PAGE 135

Figure 6.26 Correlated image of LANL block Figure 6.27 LANL block color image.

PAGE 136

The two images above (Figures 6.26 and 6.27) are taken of a concrete sample block from Los Alamos National Laboratory. The sample is re fereed to as LANL block and is described in Appendix D. The various flaws present and dete ctable include nylon wire, lead arrows, plastic and steel at depths from almost flush to up to 2 inches below the surface. Figure 6.28 is an image of various objects (radio, glass graduated cylinder, acrylic rod) placed within a concrete cinder block. Figures 6.29 a nd 6.30 are images of similar objects behind 1 inch of drywall. Figure 6.28 Clock radio, gla ss tube, wire, and acrylic rod inside cinder block.

PAGE 137

Figure 6.29 Various objects behind 1 inch of gypsum. Figure 6.30 Miniature stereo, gl ass, fiber optic cable, coppe r wire, behind 1 inch of gypsum (drywall).

PAGE 138

Reactor Insulation The problem of imaging the reactor insu lation panels, provided by Westinghouse, is similar to that of typica l concrete and gypsum problems in that there is a layer behind which the infrastructure is to be investigated. For the reactor insulation, the layer is a thin stainless steel sheet. The structure of the pa nel is such that behind this sheet is several layers of corrugated steel foils. Important fl aws in this material were described to be boric acid residue on the opposite side and crushed or disengaged foil components on the interior. The following figures, Figures 6.31 thr ough 6.36, demonstrate various views and characteristics of the reactor insulation pa nel and the systems imaging capabilities. Figure 6.31 Reactor insulation panel image s howing steel nameplate and shadow also corrugated interior foil structure evident.

PAGE 139

Figure 6.32 Steel reactor insulation panel with boric acid residue. Figure 6.33 Color image of insulation pane l, boric acid on far side clearly evident.

PAGE 140

Reactor Insulation The problem of imaging the reactor insu lation panels, provided by Westinghouse, is similar to that of typica l concrete and gypsum problems in that there is a layer behind which the infrastructure is to be investigated. For the reactor insulation, the layer is a thin stainless steel sheet. The structure of the pa nel is such that behind this sheet is several layers of corrugated steel foils. Important fl aws in this material were described to be boric acid residue on the opposite side and crushed or disengaged foil components on the interior. The following figures, Figures 6.31 thr ough 6.36, demonstrate various views and characteristics of the reactor insulation pa nel and the systems imaging capabilities. Figure 6.31 Reactor insulation panel image s howing steel nameplate and shadow also corrugated interior foil structure evident.

PAGE 141

Figure 6.29 Various objects behind 1 inch of gypsum. Figure 6.30 Miniature stereo, gl ass, fiber optic cable, coppe r wire, behind 1 inch of gypsum (drywall).

PAGE 142

The two images above (Figures 6.26 and 6.27) are taken of a concrete sample block from Los Alamos National Laboratory. The sample is re fereed to as LANL block and is described in Appendix D. The various flaws present and dete ctable include nylon wire, lead arrows, plastic and steel at depths from almost flush to up to 2 inches below the surface. Figure 6.28 is an image of various objects (radio, glass graduated cylinder, acrylic rod) placed within a concrete cinder block. Figures 6.29 a nd 6.30 are images of similar objects behind 1 inch of drywall. Figure 6.28 Clock radio, gla ss tube, wire, and acrylic rod inside cinder block.

PAGE 143

Figure 6.26 Correlated image of LANL block Figure 6.27 LANL block color image.

PAGE 144

are similar to those of aluminum. Dark sha dow type regions on the close end of flaws are not as dark in plastic since the absorption is markedly less pronounced in this material. Concrete and Gypsum Imaging of concrete and gypsum sample s usually indicated scanning for objects concealed behind walls. For this type of scenario, higher photon energies (up to 100 keV peak) are typically used and collimation setti ng is configured so as to focus past the inch to 1 inch of wall material which is us ually represents impedance to imaging. Additionally, for instance, when collimation ex tension alone is not sufficient to focus to the desired depth or when higher count rates are desired, above that which the properly collimated detector receives, closing the det ector to sample gap has a similar effect, but functions by a different mechanism, as in creasing the collimation length. The KleinNishina cross-section dictates scattering angle distribution peak towards zero (forward) and 180 (backward) degrees. Thus as the sample to detector distance is decreased, deeper penetrating photons, having scatters closer to 180 degrees are preferentially detected. As energy is increased near the 100 keV peak wh ich is often used form imaging concrete (represents the limit of the current Nylon ge nerator capabilities) the backscatter angular distribution peaking towards 180 degrees be comes more pronounced and this method of focusing becomes more effective. The followi ng figures are indicativ e of typical results achieved with the current RSD imaging modality on various concrete and gypsum targets.

PAGE 145

allows for this surface and shallow scattering to be removed from the detected signal increasing the relative signal contribution of deeper penetrating photons which interact with the void region. The drastic effect of proper collimation focusing is shown below in Figure 6.25. Both of these images are of the same plastic sample plate with a 1 cm wide by .15 cm high flaw running down its center. For the first image on the left in Figure 6.25, no collimator is used and only the surface is imaged. In Figure 6.24. a collimator is set to focus on the flaw, by moving the critical references scattering plane to just above the flaw surface. In this conf iguration, the subsurface channel now becomes apparent. Figure 6.25 Plastic flawed plate #1, unc ollimated on left, collimated on right. The high scattering and lack of absorp tion in plastics also requires low photon energy be used. Typically 55 to 60 keV peak in cident spectra provide the best results for imaging plastic type materials. Despite the di fferences in interacti ons properties between plastics and aluminum, the governing mechan ism and hence the optimization processes

PAGE 146

Figure 6.23 Over-collimated image of small channel aluminum plate. Figure 6.24 Under-collimated image of small channel aluminum plate. Figures 6.23 and Figure 6.24 are taken with 75 keV peak incident spectra. They are of a small channel aluminum plate (A ppendix D) with voidtype flaw channel machined at various depths. Figure 6. 23 and 6.24 are overand undercollimated, respectively. Due to previously discussed m echanism, the channels are thus imaged as bright and dark regions for the over-collimated and under-collimated cases, respectively. Plastic The extremely high scattering to absorption ratio of most plastic materials makes imaging subsurface flaws without collimation fo cusing nearly impossible. Even without surface features, the initial scattering from a plastic surface often saturates the signal, masking the effects of any subsurface feat ures. The addition of a collimator, however,

PAGE 147

Figure 6.22 shows an image acquired with the geometry severely over-collimated. In this image, the CRP is set below the botto m of the deepest flaw, Flaw E. As shown, each flaw is observed only as a bright region. This is because the true image of the void, the dark region is completely collimated out, thus only the bright shadow is detected. The bright, high intensity regions are slightly offset, since they are shadows, from the actual void flaw location. Figure 6.22 Over-collimated image of aluminum sample plate.

PAGE 148

Figure 6.34 Reactor insulation panel correlation image. Plastic bag with boric acid on far side of panel. Interior structure of foil also apparent. Figure 6.35. Side view of r eactor insulation panel, showing several layers of corrugated foil. Figure 6.36 Internal struct ure of corrugated foil inside reactor insulation panel.

PAGE 149

Figure 6.32 Steel reactor insulation panel with boric acid residue. Figure 6.33 Color image of insulation pane l, boric acid on far side clearly evident.

PAGE 150

Figure 6.38 Space shuttle insulation tile, under-collimated. Bright smears are glue below CRP, dark circles on left are drilled holes.

PAGE 151

Table 7.1 Percent contrast for various flaw types in aluminum Flaws are .25 inch high by .5 inch diameter cylinder 0.236 inches below aluminum surface. Collimation is configured as in Figure 7.1A. Void Plastic Lead 5 cm to NaI 4.4 cm ext (Fig 7.1A) 6.06% 13.11% -77.02% 5cm to NaI 2.4 cm ext (Fig 7.1B) -8.11% 4.03% -21.5% Figure 7.2 shows data from the same th ree flaws taken at di fferent collimation geometry. In this set of simulations, all para meters are identical except the collimator is extended only 2.4 cm past the NaI surface. As shown in the figure, this affects the shapes and relative shifts of the vari ous spectra. In this set of simulations a small but obvious shift is observed between each spectrum. This is because, for the collimation configuration modeled, the CRP is above the to p of the flaw. This allows for differences in both scattering and absorption properties to affect the spectra. The shift is expected based upon interaction characteristic s and is similar to that obse rved in the previous set of simulations as far as the order of spectra. The plastic and void flaws, however, in this configuration, are shifted relative to one another. This is mo stly due to the fact that in the plastic flaw, more photons will interact and s catter towards the detector form within the flaw. In the void flaw, the photons which trav erse the flaw do not scatter towards the detector until they have reached the alumin um of the flaw botto m. This mechanism affects the spectral shape in two ways. It m oves some of the scatters shallower in the material (within the flaw rather than below it) which results in a shorter exiting

PAGE 152

In this geometry (Figure 7.1) ther e is no significant difference between the backscattered spectra taken over the void flaw and that taken over the plastic flaw. This is because, due to the collimation conf iguration, most photons impinging upon the detector have their first scatter at the flaw depth and below, effectively eliminating the increased signal due to primary scatters in th e flaw. Additionally, the geometry is such that the significant attenuation (filtering) is caused by the aluminum traversed by the photon as it exits the target. That is, in this situation th e spectral shift caused by the attenuation difference between the plastic-type flaw (modeled as C5O2H8) and the void type flaw is not significant when compared to the spectral filtering due to the aluminum target material traversed. The differential signal intensities, given as percent contrast versus non-flawed simulation, are given for each flaw type as modeled (in Figure 7.1) as Table 7.1. As the data indicates, the relativ e contrast is much higher and thus a more intense shift is expected for the lead flaw relative to both the plastic and the void flaw. The percent contrast ( differential signal intens ity) is directly and intuitively correlated to the number of interactions and thus the degree of filtering occurring within a sample. Thus, a geometry which provides higher relati ve contrasts will necessarily enhance the relative spectral shifts. In the case of the plastic and void flaws in the scenario modeled in Figure 7.1, the contrast is on the order of a few percent. This contrast, when distributed over the entire range of the spect rum does not present itself as an observable shift. The spectral shape then, filtered mostly by the aluminum target medium is thus relatively the same for the three cases (no flaw, plastic flaw a nd void flaw) and it is merely the overall intensity that results in the detected signal.

PAGE 153

Spectral Shift due to Flaw Material 5 cm to NaI 2.4 cm Ext0% 2% 4% 6% 8% 10% 12% 14% 16% 18% 20% 0.000.010.020.030.040.050.060.070.080.09 Energy (MeV) void platicv lead Figure 7.2 MCNP5 generated backscatter spec tra of various flaw materials in Al, low collimation The figures above demonstr ate, not only how flaw mate rial can affect spectral shapes, but also the sensitivity to relative pho ton track lengths through the flaw and target medium. In Figure 7.1, the system is configur ed so that the CRP is below the flaw. In this situation, most phot ons must completely traverse the height of the flaw in order to scatter into the detector. Thus the phot on beam must cross the flaw twice, once impingent and once upon exiting, in order to re ach the detector Additionally, the severe angle subtended by the detector increases (for the flaw geometry modeled) the exiting photon track length through the fl aw as well as through the alum inum target. As shown, the lead flaw has a marked spectral shift relati ve to the plastic and void type flaws. This is due to the large attenuation, specifically absorption, differences between lead and the other flaw materials as well as to the relativ e attenuation of the alum inum target medium.

PAGE 154

within each energy bin, howev er, range from about 2-3% for the important relevant energy bins (25-60 keV) to around 10% for the lower and upper limit energy bins. Spectral Shift due to Flaw Material 5 cm to NaI 4.4 cm Ext0% 5% 10% 15% 20% 25% 30% 35% 0.000.010.020.030.040.050.060.070.080.09 Energy (MeV) void pb plastic Figure 7.1 MCNP5 generated backscatter spectr a of various flaw ma terials in Al, high collimation

PAGE 155

marked than that between alum inum and plastic. This cause s the spectral shift of the flawed region (the lead in the foam and the pl astic in the aluminum) to be more distinct in the lead/foam case. Additionally, the relative density differential between lead/foam is much more severe than between aluminum/pla stic. Thus the lead flaw in the foam represents a more significant por tion of the backscattered signal that does the plastic flaw in the aluminum and thus the lead/foam scenar io will have a much more exaggerated shift than then aluminum/plastic. This shift, fl aw orientation remaini ng the same, is thus a function of the total relative difference in attenuating propert ies of flaw versus target material. That is, it is dependent upon the t ype of filtering caused by the flaw (related to scattering and absorption prope rties within the re levant energy continuum) and to the amount of filtering (related also to th e size and density of the flaw). The following figures, 7.1 and 7.2 demonstrate the ideal shifts (MCNP5 simulations) observable for various flaw materi al in an aluminum target medium. The simulation models a 10 inch by 10 inch, 2 inch thick aluminum plate with a inch high by inch diameter cylindrical flaw at a dept h of 0.236 inches (0.6 cm) below the surface. The various flaw materials considered ar e air (void), lead (Pb) and plastic (C5O2H8). The simulation in Figure 7.1 models a NaI to sample spacing of 5 cm and a collimator extension of 4.4 cm. Figure 7.2 models a 5 cm NaI-to-sample spacing with a 2.4 cm collimator extension. In both configurations the detector is a 2 inch diameter NaI crystal offset, centerline-tocenterline, from the photon beam by 9 cm. The incident photon beam is 75 keV peak (75kVp) and 5 mm in diameter. MCNP5 calculations for these data sets were done with 2E8 particles yielding total uncertainties within 0.5%. Uncertainties

PAGE 156

Plastic Backscatter Spectra0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0102030405060708090 energy (keV)Relative Intensity plastic 0, 6cm pllastic 2, 2.5cm Figure 7.7 Normalized (first moment) experi mental data 75 keV back scatter spectra from a plastic target. Aluminum Backscatter Spectra0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0102030405060708090 energy (keV)Relative Intensity aluminum 0, 6cm aluminum 2, 2.5cm Figure 7.8 Normalized (first moment) experi mental data 75 keV back scatter spectra from an aluminum target.

PAGE 157

Steel Backscatter Spectra0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0102030405060708090 energy (keV)Relative Intensity steel 0, 6cm steel 2, 2.5cm Figure 7.9 Normalized (first moment) experi mental data 75 keV backscatter spectra from a steel target. Normalized Backscatter Spectra Low Collimation0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0102030405060708090 energy (keV)Relative Intensity plastic aluminum steel lead plastic aluminum steel Pb Figure 7.10 Normalized (first moment) experime ntal data 75 keV backsc atter spectra. No collimator extension, 6 cm from NaI to Sample surfaces.

PAGE 158

Normalized Backscatter High Collimation0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0102030405060708090 energy (keV)Relative Intensity plastic aluminum steel lead Al steel Pb Plastic NaI K-edge Figure 7.11 Normalized (first moment) experime ntal data 75 keV back scatter spectra. 2 cm collimator extension, 4.5 cm from Na I to Sample surfaces Arrow indicated k-edge in photoelectric cross-section of iodine. The figures above, especially the sideby-side comparisons pr esented in Figures 7.10 and 7.11, show a marked spectral shift as a result of target material. As previously noted, the higher the relative absorption to scatte ring characteristics of the target material, the more pronounced the effective spectral ha rdening will be. The aluminum and steel spectra above clearly show significant hardeni ng compared to the spectrum taken off of the plastic sample. The slightly more compli cated spectrum backscattered from the lead target material is a result of the absorption edges in the photoelectric cross-section of lead. These edges, shown below as Figure 7. 12, result in the lower energy peaking (at ~ 10 keV) shown in Figures 7.10 and 7.11 for lead. Similar to other spectroscopic characteristics, this low energy peaking is accentuated as collimation is increased.

PAGE 159

Additionally, the relative flatne ss of the lead spectrum, comp ared to those of aluminum and steel, is a result of the relative slopes of each of the absorption cross-sections (versus energy) these plots, for aluminum and lead are shown below as figures 7.12 and 7.13. The photoelectric cross-section for aluminum is almost three orders of magnitude higher at the low energy end (~10keV) of the plot th an it is at the high energy end (~100 keV). In contrast, the cross-section for lead varies by only about two orders of magnitude across the same relevant energy range. This indicates that in aluminum the relative absorption preference for low energy photons is stronger than it is in lead. For this reason, relatively more low energy photons are absorbed in al uminum per high energy photon absorbed and consequently we have a much sharper hi gh energy photon peak in the backscattered spectrum of aluminum relative to lead. Figure 7.12 Photoelectric cr oss-sections for lead

PAGE 160

Figure 7.13 Photoelectric cros s-sections for aluminum. Detector Material A detected photon must have at least one interaction in the detector material. Usually, if spectroscopic information is to be retrieved from that photon, several interactions or a photoelectri c absorption must occur within the detector to ensure significant energy deposition. Of the detector materials and modes considered in this investigation, only the NaI crystal in pulse mode is capable of collecting useful spectroscopic data. This, as described earlier is because current m ode detects a voltage caused by total energy deposition and does not assign any portion of that deposited energy to a particular photon. Additionally, the plastic scintillation material, due to its intrinsic physical prope rties (also mentioned previous ly) does not induce many photons to deposit their full energy. The consequen ce of this is most photons deposit about the same energy in the detector regardless of their actual wavelength.

PAGE 161

Energy distributions collect ed by the NaI crystals are subject to the inherent resolution limit of the crysta l, generally taken to be about 5-10 keV. This energy resolution limit is a function of the crystal itself, related to the excitation energy state gaps available within the doped crystal and to the inherent de-e xcitation or charge collection time of the crystal. Conseque ntly, regardless of front-end or back-end electronics or MCA (multi-cha nnel analyzer) components, en ergy differences below this limit can not be accurately measured with NaI. An important feature of the NaI crysta l is the sharp negative peak (valley) observed in all experimental and MCNP5 sp ectroscopic data collected. The valley is observed at about 34 kV in all spectra collected with the NaI and is shown in each of the above figures, but marked with an arrow in Fi gure 7.11 for clarification. The invariance of this valley to other parameter perturbati ons, led to the conclusion that it must be a property of the detector mate rial and how it interacts with the incident photons. Consideration of the various cr oss-sections for both sodium and iodine revealed that the valley corresponded identically and thus must be a result of the K-edge in the photoelectric cross-section of iodine. Figure 7.14 shows th e photoelectric cross-section for iodine32. The sharp k-edge coincides identically with the valley observed in the NaI acquired spectra. Further investigation of documented NaI sp ectroscopic properties confirmed this conclusion 34.

PAGE 162

Figure 7.14 Photoelectric cross-section for Iodine K-edge at ~30keV. Detector and Collimation Configuration Detector to Sample Spacing Detected signal spectra were taken fo r several different sets of system configurations. Each set had a different but constant (within the set) incident energy distribution and collimator extension and a va riable detector (NaI surface) to sample spacing. Similar trends for each set were observed. The dominant general trend is a sharpening and hardening (incr ease in most probable energy) of the detected spectra as detector to sample distance is decreased. This is mostly due to the fact that as distance from sample to detector is decreases the effective collimation is increased and photons from deeper within the sample are preferentially selected fo r. These photons, since they have traveled further thought the target material, are filt ered to a higher degree by the

PAGE 163

target material and thus demonstrate stronger filtering effects in the spectral shapes. For the case in Figure 7.10, the dominant filtering effect is absorption of the low energy end of the spectrum which produces the observed spec tral hardening. To illustrate this trend one section of a set in the se ries is shown below in Figur e 7.15 which depicts trends observed experimentally for target to detector surface distance variation. In this set the collimator is fixed at 1.5 cm past the NaI surface. The collimator fins are fixed at zero degrees. The set is take n with a 75 keV incident spectrum on an aluminum sample. Spectra were taken as th e NaI to sample surface distance was varied from 8.5 cm to 2.5 cm. In this particular set the dominant trend is the shift towards higher energy. Other sets show similar tr ends and are included in Appendix E for completeness along with complimentary MCNP5 simulations. NaI to Sample Distance0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0102030405060708090 energy (keV)Relative Intensity 25 45 55 75 85 increasing distance Figure 7.15 Normalized experime ntal spectral shifts for va rious detector to sample distances over aluminum.

PAGE 164

Collimation Extension Parallel experiments were done to inves tigate the effect of varying collimation extension. In these sets, incident energy di stribution and detector to source spacing were held constant while collimation extension was varied. These experiments revealed similar, if more exaggerated, trends as the experiments done by varying detector to sample spacing. In these sets the same sort of sharpening of the detected spectra was observed as the collimator extension was increa sed. This sharpening was similar to that observed previously in that it is apparent ly caused by both a hardening of the low energy end and a softening of the high energy end of the spectrum. The mechanisms responsible for spectrosco pic shifts and trends in both detector spacing and collimation length extension expe riments are identical. The same physical process leads to the energy shifts observed in both experiments and the degree to which the process alters the spectrosc opic properties of the detected signal is determined almost purely by geometrical configurations. The hardening in energy observed on th e low energy end of the detected spectrum is due to filtration by photoe lectric attenuation. As show n for aluminum in Figure 7.13 above, the photoelectric effect is the dominate mode of inte raction by several orders of magnitude at the low energies considered in th ese investigations (up to about 30 keV). When a poly-energetic photon beam passes through a media, the lower energy components are preferentially removed since their removal (i.e., phot oelectric absorption) cross-section is much larger than that of the high energy phot ons. Consequently, it is the higher energy photons which have a greater prob ability of interacting in a media and then returning to the detector. The further a photon beam travels through a material or the more interactions a photon has in a material, ultimately the same thing, the greater the

PAGE 165

probability of a photoelectric absorption event o ccurring. Thus, we would expect deeper penetrating and higher order scat tering detected photon fields to reveal larger up-shifts in energy for the low energy components of their distributions. Th e up-shift referred to here is actually a preferential elimination of the low energy portion of the spectrum which causes the renormalized distribution to appear shifter upwards in energy. Decreasing detector-to-sample spacing as well as increasing the collimator extension both function to select high order scattering and deeper penetrating backscattered photons. Increasing collimator length, based upon previously verified geometrical arguments, results in the detected signal comprising increasing contributions of deeper penetrating photons. This is shown below in Figure 7.16. As the collimator extension is increased, the critical scatte ring reference plane moves deeper below the object surface. All scatters occurring above this plane are discriminated against allowing the deeper penetrating photons to be a more prominent part of the detected signal. Similarly, as the detector to sample spaci ng is decreased, as shown in Figure 7.12, the angle and solid angle substension to the detector change. The change in the angle is such that photons, scattering in the sample and exiting towards the detector must, geometrically, travel further and thus be at tenuated more in the sample before impinging upon the detector. The change in solid angl e also functions, though less dramatically, to select for deeper penetrating photon s as illustrated in Figure 7.11.

PAGE 166

Figure 7.16 Collimator extension and CRP. As collimator is extended from position A to B to C, the critical scattering reference plane moves from A to B to C, respectively. All primary scatters occurring above the CSP do not reach the detector. The slight softening of the high energy side of the spectra is a result of the energy lost in scattering. Compton scattering, which is the more dominant and relevant phenomena for backscatter scenarios require s that a recoil electron, for momentum conservation, inherit a portion of the incident photons ener gy. Thus, the backscattered distribution is expected to be downshifted in energy by at le ast the amount of energy lost during scattering. Furthermor e, this energy loss should increase with increased number of scatterings, increased b ackscattering angle, and increa sed initial photon energy. This shift and the trends describing it ar e much less obvious than the up-shift in energy at the opposite end of the spectra. At the energies c onsidered in these investigations, up to 75 keV peak, energy lost in a Compton collision is slight. Even in the limiting case of the largest energy photon (75keV) backscattered at 180 degrees, the

PAGE 167

energy lost is only 12 keV, and thus the scattered photon returns with an energy of 58keV. Since this shift is barely within the energy resolu tion of the NaI detectors and because the detected scatters, being mostly first and second order collisions suffer collisions of less than 18 0 degrees, the shift due to scattering angle is not well distinguished. The shift due to number of scatterings, however, does appear to be detected by the NaI resolution limit and is observed in the expected direction. As collimator extension is increased or as sample-to-detector spacing is decreased, as described above, higher order scatter component s are preferentially se lected. Thus the softening in energy, due to multiple scatters is observed to increase with both of these parameters. The following Figures 7.17 and 7.18 show the spectral shifts caused by increasing collimation as NaI-to-sample distance is he ld constant. Figures 7.17 and 7.18 show the effect as measured and as si mulated via MCNP5, respectively. In this set, the NaI to sample distance was held fixed at 8.5cm. The collimator fins were fixed at zero degrees. The spectra were again taken off an alumin um sample with a 75 keV incident photon beam spectrum. The collimator sleeve, for this set, was extended from 1.5 cm to 5.5 cm past the NaI surface.

PAGE 168

Collimator Extention0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0102030405060708090 energy (keV)Relative Intensity 1.5 2.5 3.5 4.5 decreasing collimation extension 7.17 Normalized experimental spectral shifts as a function of collimation extension. Collimation Trend on Al 0% 5% 10% 15% 20% 25% 30% 35% 0.010.020.030.040.050.060.070.08 energy (MeV) 2cm 3cm 4cm 5cm 6cm 7cm decreased collimation extensionFigure 7.18. Normalized MCNP5 spectral shifts as a function of collimation extension

PAGE 169

Monte Carlo Verification Monte Carlo simulations were conducted to analyze the trends caused by detector and collimator configurations initial photon energy spectra, fl aw type and configuration and target material. The trends observed fr om Monte Carlo collected data closely agree with those experimentally collected. Additionally, tall y component break down by both penetration depth and scattering order suppor t the mechanism proposed above for causing the spectral shifts. Tally break down by collision component show s that spectra composed of larger fractions of higher order scat tering components demonstrate la rger shifts, on both ends of the spectra. For completeness, MCNP5 generated spectra along with tally scatter breakdown are compared with similar experime ntally recorded spect ra in Appendix E. Applications Spectroscopic information can be used a guide to visualize the degree of collimation and thus approximate the fo cusing depth set by the collimation and geometrical configurations. Since the mechan isms resulting in trends due to collimation and detector geometrical configurations are well understood and since the degree of collimation has already been correlated with depth focusing through the previously developed RSD transport model, it is intuitive that the spectral trends observed will serve well as an approximate depth focusing gauge. This gauge of course is relative and must be appropriately calibrated. That is, it is the shift in spectral shape rather than merely the shape itself which is roughl y related to focus depth. Additionally, spectral information can be used to assess the relative absorption and scattering properties of a material. Material compositions can theoretically be estimated based upon the relative shift observed in a backscatte red spectrum. This can be

PAGE 170

accomplished by establishing a relative scal e based upon backscattered spectra form materials of known properties taken at a spec ific and constant energy and collimation configuration. The work done in evaluating the spect roscopic trends and the mechanisms responsible for them in this study represent only an initial qualita tive investigation. There is much more analysis to be carried out to further detail these mechanism and possible quantitatively relate spectral tre nds to the variables described above. Additionally, the use of spectro scopy to ascertain relative impo rtance (to image contrast) of various energy regions within the detected backscatter signal is an application that should be investigated in further detail. The concept is that certain regions of a spectrum will exhibit greater shifts and this higher rela tive contrast between flawed and non-flawed regions of a sample, thus using energy windows to preferentially select these regions may increase overall contrast and decrease scan time. One of the important obstacle s hindering a complete analysis of the backscattered spectra and its trends is that there are always several mechanisms which are filtering the detected spectra simultaneously. It is expe rimentally impossible to truly isolate one mechanism at a time as each is a strong func tion of system geometry and thus when one parameter is varied other parameters are of ten varied as well. If one particular mechanism is not completely dominant, d econvolving the trends w ith respect to the appropriate mechanism can become very complicated.

PAGE 171

151 CHAPTER 8 CONCLUSIONS Development of an accurate qualitative photon transport model and the indicative comprehensive understanding of the phenomen a involved in this new imaging modality have led to more effective use of the system and increased awareness of possible future modifications to further increase the perf ormance of the system and its range of applicability. The system has been proved e ffective on a variety of materials for a wide range of flaws and defects. Additionally future modifications should increase the systems applicability to other ma terials and imaging tasks. Applications Currently the backscatter X-ray imaging sy stem has proven to be effective in the NDE of materials ranging from ultra-low de nsity materials such as SOFI to higher density materials such as aluminum, steel, and concrete. Key limiting factors in the performance and applicability of the system include scanning speed, detector and x-ray generator head size and impinge nt x-ray beam energy. The scanning speed is limited by the availability and delectability of an ade quate backscattered photon signal. Ideally each pixel should comprise at least 30,000 counts for good im age quality and statistical validity. This signal in turn is limited by the current available from the x-ray generator. Image resolution which is a function of pixel si ze is also a factor in determining scanning speed. Increasing number of pixels a part icular image is divided into increases resolution, obviously an arti fact smaller or about the si ze of a pixel will not be well detected, and increases scanning time. Thus a balance must be reached for each

PAGE 172

particular scanning application between speed and resolution. The size of the detector and x-ray generator assembly determine the t ype of areas which can be scanned. While many applications effectively involve s canning one side of a plane and are not susceptible to size constraints, many ot her application requi re tight detector configurations in order to s can in between various structur es of a target. Besides electron current limitations, the x-ray generator also imposes limitations on the maximum peak energy spectrum available. This in turn limits the depth into a particular material which can be imaged. Recommendations for Further Developments Detector Configuration With the possibility of new, smaller de tector components (e.g..0.11 inch YSO detectors mounted onto state of the art mini ature photomultiplier tubes ~ 3 in length from Hamamatsu) comes the possibility of modifying the current imaging system to achieve a broader range of applicability. Rath er than the current rigid detector mounting system with fixed detector orientations a nd positions, a new mounting system is proposed to allow 2 more degrees of freedom in dete ctor positioning. That is, rather than the detector to sample geometrical configurations being a pure fu nction of detector to sample spacing and collimation extension, the new sy stem should allow for the detector to be moved both laterally (radially in and out) and ve rtically (up and down) with respect to the X-ray generator and mounting base. Additionall y, the detectors (or at least the detector heads) should be mounted on a pivot so that th eir orientation relative to the backscattered photon field can be varied. Su ch modifications would permit the imaging system to be applicable to a new class of situations w ith tighter, smaller geometries for which the current system, because of its large and rigid detectors, is not practical. Additionally, a

PAGE 173

larger range of detector mobility will allow for more accurate and more efficient selection of backscattered photon field components. Analysis of the backscattered signal components via MCNP5 and supported with experimental observations and analytical calculations has led to a more thorough understanding of the imaging pr ocesses and firmer grasp of the various components of the backscattered photon field and the importan ce of the information th at they carry. As a result of this analysis, a transport model has been developed to detail the mechanics and processes leading to a contrast generated image of various flaw s in a target. Application of this model has further led to the understanding that wh ile the fundamental physics of the imaging process remain the same, the optimal detector and collimat or configuration is often task specific and the more freedom and versatility of the detector configuration the easier and more efficient image acquisition becomes. Detector Materials and Operation Modes Detector material type and operational mode are an integral part of the system and can have a marked effect on the acquire d image quality. Since it is the photon interactions within the sensitive volume of a de tector with lead to a signal and thus and image. Each detector material has its own unique interaction characteristics and thus, to some extent, should produce slightly diffe rent images of a given scenario. Although plastic scintillat or type detectors were consider ed in this investigation, a more thorough investigation into their capabili ties is worth while. In particular, the nature of the plastic detectors, their larg e photon mean free path and fast response time, makes them inclined to be operated in current rather than pulse mode. Since their photon stopping power makes them virtually useless in spectroscopic applicat ions, there is no real value in operating them in pulse mode. A current mode operation would take better

PAGE 174

advantage of their fast response time a nd allow for extremely high count rate and potentially faster scanning times. Additionally plastic scintillators are more rugged, less expensive and can easily be manufactur ed into a variety of shapes. YSO detectors are also worth considering. YSO are more rugged, and have faster response times than NaI. P.I.N. diode de tectors are smaller and can handle extremely high counts. Current state-of-the -art P.I.N. diode components and their properties may be a worthwhile investigation for future applications. MCNP5 simulations to be considered Monte Carlo simulations are a powerful t ool to quickly asse ss the potential of various detector and impinge nt photon beam geometries and to test the range of applicability of the system a nd the delectability of various classes of flaws. Among the more promising configurations that should be simulated and analyzed are those which tilt the detector surface at various degrees w ith respect to the horizon. Currently only simulations with the detector face parallel with the sample surface have been evaluated. The potential benefit of angling the detector f ace, as previously discussed, include faster scanning times and more efficient photon collection. Additionally, Monte Carlo simulations should be run with the photon beam impinging at an angle. This may prove especially useful in analyzing carbon-carbon la minates and other striated type materials. The theory here is that a small delamina tion often does not present significant electron density changes in the transverse direction and possible a photon beam directed along the delamination flaw may experience more signifi cant attenuation differential than a similar beam impingent normally to the flaw length. Other simulations that may be of interest include various detector materials and incident photon beam energies.

PAGE 175

Optimizing geometrical variables Geometrical and configurational variables of the system include detector to sample vertical spacing, detector to source radial separation, coll imation extension and detector surface angle relative to both incident b eam direction and target surface. The perturbations in the backscat tered spectroscopy and photon field has been investigated for the first three variables listed and the effect of the last has been mostly speculation with minimal experimental or analytical data. It is clear that each of these variables affects the properties of the detected back scattered photon field and there by can affect the quality of the image generated by the preferenti al collection of these photons. Current data indicates that these variab les do not function in dependently of one another. That is, the portion of the tota l backscattered photon field observed by the detector is an intimate non-linear function of all of these variables. Thus, while the relative effect of each of th ese variables can be isolated for specific scenarios, the combined net effect of all of them is important to the configuration which results in the optimum balance between the highest quality image and th e fastest image acquisition time. Further investigation into the inte rplay between each of these variables has potential to increase the understa nding of the imaging system and increase the efficiency of focusing optimization routines.

PAGE 176

APPENDIX A SELECTED SABRINA GENERATE D PHOTON TRACK PLOTS The following figures are SABRINA plots with these geometric specifications: 0.6 cm from lead collimator to aluminum sample surface 4.3 cm from NaI to Al. 5.05 cm radius detectors 9 cm center line-to-centerline from the beam. Figure A1. Photons having four or less collisions (void flaw)

PAGE 177

Figure A2. Photons having one collision (void flaw) Figure A3 All photons entering detector (h istory filtered for clarity-void flaw)

PAGE 178

Figure A4 All photons entering detector. Scat ter sites marked with black X (plastic flaw). Figure A5 All collision components, history filtered (plastic flaw)

PAGE 179

. Figure A6 No flaw first thr ough fourth collision components Figure A7. Plastic flaw first thr ough fourth collision components

PAGE 180

Figure A8. No flaw all collisi on components, zoomed in view. Figure A9 Plastic flaw all collis ion components zoomed in view

PAGE 181

Figure A10 Plastic flaw first collision components The following figures are SABRINA plots with these geometric specifications : Lead collimator 2.6 cm from aluminum samp le surface 4.3 cm from NaI to Al. 5.05 cm radius detectors 9 cm center line-to-centerline from the beam.

PAGE 182

Figure A11 Plastic flaw first collision component. Figure A12 Plastic flaw first collision scatter points. Figure A13 Plastic flaw, all scatter component s (note many scatters off the NaI surface and down back into Al)

PAGE 183

Figure A14 First and Second scatter points Figure A15 Multiple scatter sites.

PAGE 184

Figure A16 First, second third scatter points. Figure A17 First, second, thir d, fourth scatter points.

PAGE 185

Figure A18 All scatter points. The following figures are SABRINA plots with these geometric specifications : 4.3 cm to NaI 1 cm to collimator, from Al sample surface. Figure A19 Void flaw first collision components

PAGE 186

Figure A20 Void flaw all collision components. The following figures are SABRINA plots with these geometric specifications : 8.5 cm to NaI 4 cm to collimator, from al uminum sample surface 5.08 cm radius NaI detectors 9 cm from beam Figure A21 First collision components

PAGE 187

Figure A22 Second collision components. Figure A23 Third collision components

PAGE 188

Figure A24 Fourth collision components Figure A25 All collisions. (note photons re flecting off NaI surface and scattering down towards Al sample.

PAGE 189

The following figures are SABRINA plots with these geometric specifications : 8.5 cm to NaI 7 cm to collimator, from al uminum sample surface 5.08 cm radius NaI detectors 9 cm from beam Figure A26 First collision Figure A27 All collisions

PAGE 190

Scenarios from chapter 4 (Figures 4.14 A-L) Figure A28. No flaw. Figure A29. SSL

PAGE 191

Figure A30. SSL zoomed in view of tracks across flaw and aluminum. Figure 31. SSL. Flaw is above CRP

PAGE 192

Figure A32. SSL. Figure A33. ADL flaw is below CRP.

PAGE 193

Figure A24. ADL zoomed in view. Figure A35. VDL. Flaw is below CRP.

PAGE 194

Figure A36 VDL zoomed in view. Figure A37. VSL. Flaw is above CRP

PAGE 195

Figure A38. VDL, flaw shown opaque. Figure A39. VDL.

PAGE 196

Figure A40. VDL, flaw is tran sparent. Notice that photons of all scatter order traverse the flaw directly. Figure A41. VSL zoomed in view.

PAGE 197

Figure A42. VDE. Aluminum and flaw are both set to invisible to emphasize simple photon tracks. Figure A43. ASL. Photon on far right ar e reflecting off NaI, going up into it.

PAGE 198

Figure A44. ASL zoomed in view. Figure A45. ADL

PAGE 199

Figure A46. VDL Figure A47. VDL.

PAGE 200

Figure A48. VSL. Figure A49. SSL.

PAGE 201

Figure A50. SDI zoomed in view of flaw and photon tracks. Aluminum is set to transparent.

PAGE 202

182 APPENDIX B ONE SCATTER SIMPLIFIED PHOTON TRANSPORT MODEL Figure B.1 Typical photon path across a flaw. In the simplified one scatter transport m odel the signal differential leading to an image is a result of a difference in attenuati on experienced by the incident and reflected photon beams caused by the presence of a flaw. In the schematic above, the flaw, region F is represented by the yellow rectangle. The re st of the target material is divided into two regions, the top, T, and the bottom B. The two portions of the photon path, incident y r1 T F B ft fb r2 Y

PAGE 203

and exiting are represented by r1 and r2, respectively. The photon can scatter either above the flaw, within the flaw (unlikely for void type flaws) or below the flaw. The equation detailing the first scatter relative signal intensity returning to the detector from a particular depth in the sample is 2 2 1 10 r re d d e I I where oIis the unattenuated photon intensity, 1 1re and 2 2re are the attenuation functions for the incident portion and exiting portion of the phot on beam, respectively. In this equation, d d is the differential scattering cross-secti on per solid angle per electron as defined by the Klien-Nishiena relationship as : d d = r0 2 2 2 + sin2 ,h= h 1 + h mec2 (1 cos) is the solid angle subtended by the de tector. The solid angle subtended by the detector is simply approximated by the area of the detector divide d by the distance from scatter point to the detector surface, square d. Simple geometric observations reveal the relationship) cos( /1 2 r r, where is the photon scattering angle. This angle is taken as the median between the largest and smallest scattering angles permitted, by geometrical and collimation considerations, to impinge upon the detector. Each of the photon tracks, incident and reflected, is, for eas e of calculations, divi ded into up to three segments, depending upon the scatter point. These segments include top, flaw and bottom and each is handled with its own intera ction characteristics. For most samples the top and bottom segments will be the same materi al and will consequently be treated with the same cross-sections. The simplicity of these approximations is justified by the fact

PAGE 204

that they are applied to both flawed and nonflawed scenarios and since we are concerned with describing a relative c ontrast (ratio) betw een the two, the effect of the rough approximations should, to some extent, cancel out. The point is not to provide a realistic model for photon transport as much as to desc ribe the mechanisms responsible for image contrast acquisition and t hus flaw detection. Thus the calculation is divided into thr ee segments; primary scatter in top region, primary scatters in flaw region and primary s catters in bottom region. The first scenario has a scatter in the top region and the exit ing photon beam will travel only through the top region towards the detector. The second scen ario scatters in the flaw region, this photon beam which travel though the top and part of flaw region before scattering and then through the flaw and t op region upon exiting. The thir d scenario tr avels though all three regions incident and exit ing after being scattered. An uncollimated signal will be composed of all three segments As collimation is increased, each scenario is cut off in order from the first to the third. Thus a well collimated signal will only comprise (to a first approximation as modele d here) photons which have a pr imary scatter in the bottom region of the sample. Additionally, depending upon flaw orientation and detector configuration a photon beam may miss the flaw upon entering or exiting the sample and thus may only traverse it once, rather than twice. (These scenarios are discussed in Chapter 4). The mathematical representation of each of these segments is as follows: For 2 2 1 10r re d d e I I Segment 1 (top): for y [0, ft] r1* 1=y* top

PAGE 205

r2* 2= (r1* top)/ (cos ( )) Segment 2 (top & flaw): for y [ft,fb] r1* 1 = ft* top + (y-ft)* flaw r2* 2= (ft* top)/(cos ( )) +[(y-ft)* flaw]/( cos ( )) Segment 3 (top, flaw & bottom): for y [fb,Y] r1* 1 = ft* top + (fb-ft)* flaw+(y-fb)* bottom r2* 2 = [ft* top + (fb-ft)* flaw+(y-fb)* bottom] / (cos ( )) The total signal strength is the integrated value of segment 1 + segment 2 + segment 3. Or ] [2 2 1 10 0 0 r r Y Y Te d d e I dy I dy I And the contrast is relative signal differential is simply a ratio of the signal strength of flawed to non-flawed areas of a sample. For the calculations presented in table 4.1. the incident photon spectrum was 75 keV peak. The mfps for this photon beam used in these calculations in aluminum and in air were taken form MCNP5 simulations as: 9.75E-01 cm, 6.78E-01 cm and 6.09E+03 cm fro Al top, Al bottom and flaw void, respectively. Different values are taken for Al top and Al bottom to account for the filtering and scattering of photons are they reac h further into the sample. The geometry considered in these calculations is as follows: Two inch thick aluminum plate with a 0. 15 cm flaw 0.65 cm (1 /4 inch) below the surface. The detector is a NaI crystal two inch es in diameter. The detector surface is 2.9 cm above the sample surface and the detector is located nine cm centerline-to-centerline

PAGE 206

from the photon beam. The collimator extens ion is set contingent upon the model being calculated. Figure B.2 Parameters used to determin e optimal collimation length for focusing to a specified depth. A H h 2r D R y C d

PAGE 207

APPENDIX C COLLIMATION DEPENDENT CONTRAST TREND ANALYSIS Figure C.1 Signal intensity as a function of depth and across a void type flaw. 2 1 0 ) (r e d d r e dy y dS ydy dy y dS y S0) ( ) ( Or, roughly Backscattered signal Signal Intensity flaw Incident beam CRP

PAGE 208

y ydy e y S0) (, so let, y y ye dy e A y S0) 1 ( ) ( so that S = A (total signal strength) Percent contrast = 100 *[Sflaw Sno flaw] / Sno flaw *From the figure above it can be seen that all scatters occurri ng above the flawed region are identical for both non-flawed samp les (i.e. above the flaw in the flawed sample is effectively the same as a non-flawed sample). T hus the Absolute difference in signal intensity, for collimation up to the flaw top is a constant. % contrast =d/ Sno flaw where d ~ (Sflaw Sno flaw ) As the collimator extension is increased in length from zero, the critical reference plane (CRP) moves from sample surface down. As the CRP approaches the flaw top, d does not change, and y y noflawdy Ae y S) ( But, y y y y noflawdy Ae dy Ae S0 Thus, percent contrast as a function of y position of the critical reference plane (or synonymously as a function of increa sed collimation extension) is

PAGE 209

y noflaw ydy S d dy d C0[] /( [] / % Or ) /( ] 1 ( [ /( %0 y y y noflawe B d dy e A S d C For d constant and (0 < y < flaw top) which corresponds to (0 < Xcol < Xcrit): Where Xcrit is the optimal collimation length corres ponding to the depth of the flaw top. Figure C.2 Contrast versus collimator extension for collimator from zero to critical (optimal) extension. This optimal co rresponds to collimating to top of flaw. Contrast here increases as a function of increasing collimation since the signal being discriminated against, from sample surface to flaw top, is effectively noise (As it is Xcrit % C o n t r a s t Collimator Ext

PAGE 210

identical between flawed and non-flawed scen arios) and this its inclusion lessens the contrast. As the collimator is extended from Xcr it towards Xbottom (c orresponding to CRP being at flaw bottom) almost no signal is lost as very few photons interact within the flaw (assuming void type flaw) Thus the contrast trend within this region appears as follows: Figure C.3 Contrast versus collimator extension for collimator from critical (optimal) extension to flaw bottom. Xcrit % C o n t r a s t Collimator Ext Xbottom ideal actual

PAGE 211

Xbottom % C o n t r a s t Collimator Ext In reality some photon do interact with in the void and more importantly, some photons which interact deeper and would otherw ise scatter into the detector (as the source has a definite diameter) are lost. This makes th e real trend in this region slightly sloped downward (observed experimentally) in compar ison with the theoretically predicted flat line. As collimation is further increased so that the CRP moves down from the flaw bottom, all signal being discriminated against is useful as theses photons pass through the flaw. The importance of these photons, howev er, decreases exponentially as a function of depth since the incident photon beam inte nsity reaching these areas is exponentially attenuated as a function of depth. This m eans that as you collimat e to deeper within a sample you are cutting out less important signa l per additional unit of depth collimated out. Thus in this region, the contrast appears as follows: flaw top. Figure C.4 Contrast versus collimator extension for collimator extension corresponding to flaw bottom to maximum extension.

PAGE 212

The entire trend of contrast versus collimator extension (or identically contrast versus CRP position) is as follows: Figure C. 5 Contrast versus collimator extension as shown is chapter 4, data form exper iments and MCNP5 simulations agree with this trend. Xbottom Xcrit % C o n t r a s t Collimator Ext Corresponds approximately to flaw height if plotted against CRP rather than Coll Ext.

PAGE 213

193 APPENDIX D IMAGED SAMPLE DESCRIPTIONS Foam Three main SOFI foam type samples were evaluated. The first type is a calibration panel. This is an eight inch thick piece of SOFI mounted on a ten by ten inch aluminum substrate. There are four bored out voids w ithin the SOFI a large and a small (and diameters, respectively) at each substrate level and midplane level (four inches below surface.) The second sample type is a ramp pa nel. This is a large (usually one meter by one meter) contoured piece of aluminum substr ate with various fla nges and bolts with 2 to 8 inches of SOFI foam applied to it. There are often other in serted objects and/or naturally occurring voids and delaminations within the foam as well. The third type of sample are smaller (ten by ten inch with 2 to 4 inch thicknesses) foam blanks. These are blank foam panel not mounted. These were cut and various objects were inserted into them to help develop depth correlations. Aluminum The aluminum samples scanned included fo r the most part the machined aluminum test plates. There are four of these. Two of the plates ha d voids machined into them and tow of the plates had void channels mach ined into them. Of the void non-channeled plates, one was machined with five, hi gh flaws positioned at va rious depths and one had cylindrical voids with various heights. Of the channeled plates, one had three 1 cm

PAGE 214

wide by .1 cm high channels at various depths ( ) and one had 5, 1 cm high .1 cm wide channels at various depths (). Plastics The main plastic sample imaged was a machin ed plastic plate with three, 1 cm wide by 0.1 cm high channel flaws running the leng th of the plate at various depths ( .) Reactor Insulation Panel The Reactor Insulation Panel was supplied by Westinghouse. It was a one meter by one meter by eight inch thick hollow panel of stainless steel. Inside were several layers of corrugated stainless steel foils. Boric acid a nd Boric acid residue was placed on various places of this sample for several images. Concrete & Gypsum The Concrete sample was supplied by LANL. It is a solid concrete block with various flaws (e.g. fishing line, lead point ers, plastic and coppe r wires, steel pipe) inserted at various positions and depths. Ceramic Space Shuttle Thermal Tiles The Space Tiles were provided by NASA. They are small 6 inch by 6 inch by 3 inch ceramic tiles with a ceramic mesh outer binding. Flaws in th e tiles included glue drops and delaminations, both at mesh to ceramic interface.

PAGE 215

195 APPENDIX E SPECTROSCOPIC TRENDS A luminum Flaw plate 1: 1.5 CM to PERP collimator fins. 3 cm to NaI Experimental Backscatter Spectra Flaws A and E 60 & 75 keV peak specs. 0 0.5 1 1.5 2 2.5 3 3.5 050100150200250 channel A75 E75 A60 E60 Figure E1 Spectroscopic trend for fl aw depth by incident energy spectra.

PAGE 216

2 cm to collimator. 3.5 cm to NaI Aluminum flaw plate #1 75keV peak spectrum0 0.5 1 1.5 2 2.5 3 050100150200250 A B C D E Figure E2 Trends for flaw depth. Flaws depth increases from A to E. Current Tally by Collision Componet 5cm to NaI 3. 3 cm Col Ext 2" Al plate w/ 1/4" void .1cm deep -1.00E-05 0.00E+00 1.00E-05 2.00E-05 3.00E-05 4.00E-05 5.00E-05 6.00E-05 0.00E+001.00E-022.00E-023.00E-024.00E-025.00E-026.00E-027.00E-028.00E-029.00E-02 Energy (MeV) 1scat 2scat 3scat 4scat 5scat 6scat 7scat total

PAGE 217

Current Tally by Collision Componet 5cm to NaI 3.3 cm Col Ext 2" Al plate w/ 1/4" void 1.25 cm deep -1.00E-05 0.00E+00 1.00E-05 2.00E-05 3.00E-05 4.00E-05 5.00E-05 6.00E-05 7.00E-05 0.00E+001.00E-022.00E-023.00E-024.00E-025.00E-026.00E-027.00E-028.00E-029.00E-02 1scat 2scat 3scat 4scat 5scat 6scat 7scat total Figure E3 Scatter component break down for spectra. current tally by scatter component 2"Al plate .4cm flaw0.00E+00 5.00E-06 1.00E-05 1.50E-05 2.00E-05 2.50E-05 3.00E-05 0.00E+001.00E-022.00E-023.00E-024.00E-025.00E-026.00E-027.00E-028.00E-029.00E-02 1scat 2scat 3scat 4scat 5scat 6scat 7scat Figure E4. Spectral break down by scatter component.

PAGE 218

Flaw .2cm deep severly overcollimater-1.00E-06 0.00E+00 1.00E-06 2.00E-06 3.00E-06 4.00E-06 5.00E-06 6.00E-06 7.00E-06 8.00E-06 0.00E+001.00E-022.00E-023.00E-024.00E-025.00E-026.00E-027.00E-028.00E-029.00E-02 energy 1scat 2scat 3scat 4scat 5scat 6scat 7scat total Figure E5. Spectral break down by scat ter component. MCNP5 current tally. Normalized Current Tally0.00E+00 5.00E-02 1.00E-01 1.50E-01 2.00E-01 2.50E-01 3.00E-01 3.50E-01 4.00E-01 2.00E-022.50E-023.00E-023.50E-024.00E-024.50E-025.00E-025.50E-026.00E-026.50E-02 .1cm .2cm .4cm .6cm 1.25cm .2cmovercollimated .2cmNoCol .1cmNoCol .4cmOverCol .6NoCol 1.25cmNoNOl Figure E6. Current tall y spectra for various flaw depths and collimation configurations.

PAGE 219

(3.3 cm Col ext) Collision Component0.00E+00 1.00E-01 2.00E-01 3.00E-01 4.00E-01 5.00E-01 6.00E-01 7.00E-01 8.00E-01 9.00E-01 1.00E+00 12345678 Scatters .1cm(3.3) .4cm(3.3) .6cm(3.3) 1.25cm(3.3) no flaw (3.3) Figure E7 Collision component signal contribu tion for various flaw depths in aluminum. No Collimator Collision Components0.00E+00 1.00E-01 2.00E-01 3.00E-01 4.00E-01 5.00E-01 6.00E-01 7.00E-01 8.00E-01 9.00E-01 1.00E+00 12345678 scatters .1cm(0) .2cm(0) .4cm(0) .6cm(0) 1.25cm(0) Figure E8 Collision component contributions fo r various flaw depths with uncollimated detector in aluminum.

PAGE 220

.6cm flaw Collision Comp vs Collimation0.00E+00 2.00E-01 4.00E-01 6.00E-01 8.00E-01 1.00E+00 1.20E+00 12345678 scatters .6cm(0) .6cm(3.3) .6cm(4.4) Figure E9 Collision component contributions to 0.6cm void flaw in aluminum with various collimations configurations. Al Flaw palte #1 :60keV 3cm to collimator 4.5 cm to NaI-0.5 0 0.5 1 1.5 2 2.5 3 3.5 050100150200250 A B C D E Figure E10 Spectroscopic trend with flaw depth. Flaw de pth increases from A to E.

PAGE 221

Ex Set 1 2.3cm ext Varying Detector Heights0 0.5 1 1.5 2 2.5 010203040506070 Energy (keV) 8cm 6cm 4cm 2cm Figure E11. Experimental spectros copic trend with detector height. MCNP set1 2.3 cm ext. Varying Detector Height0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.010.020.030.040.050.060.070.08 Energy 4.3d 2c 6.3d 4c 8.3d 6c 10.3d 8c Figure E12 MCNP5 current tally spectro scopic trend with detector height.

PAGE 222

0 5 10 15 20 25 30 35 40 45 50 1234567 number of scatters MCNP set1 Scatter Component Contribution 4.3d 2c 6.3d 4c 8.3d 6c 10.3d 8c Figure E13 Scatter component contribution for various detector and collimator configurations. MCNP set1 (4.3 cm to NaI)0 0.000000001 0.000000002 0.000000003 0.000000004 0.000000005 0.000000006 0.010.020.030.040.050.060.070.08 Energy (MeV) 1 scat 2 scat 3 scat 4 scat 5 scat 6 scat 7 scat total Figure E14 Spectral breakdown by collision component

PAGE 223

MCNP set1 (6.3 cm to NaI)0.00E+00 2.00E-09 4.00E-09 6.00E-09 8.00E-09 1.00E-08 1.20E-08 1.40E-08 1.60E-08 0.010.020.030.040.050.060.070.08 Energy (MeV) 1 scat 2 scat 3 scat 4 scat 5 scat 6 scat 7 scat total Figure E15 Spectral breakdown by scatter component MCNP set1 (8.3 cm to NaI)-0.000000005 0 0.000000005 0.00000001 0.000000015 0.00000002 0.000000025 0.00000003 0.010.020.030.040.050.060.070.08 Energy (MeV) 1 scat 2 scat 3 scat 4 scat 5 scat 6 scat 7 scat total Figure E16 Spectral breakdown by scatter component

PAGE 224

MCNP set1 (10.3 cm to NaI)0 0.000000005 0.00000001 0.000000015 0.00000002 0.000000025 0.010.020.030.040.050.060.070.08 Energy 1 scat 2 scat 3 scat 4 scat 5 scat 6 scat 7 scat total Figure E17 Spectral breakdown by scatter component Ex Set 2 series1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 1020304050607080 Energy (keV) 0.05cm 0.1cm 0.2cm Figure E18 Experimental tr end with flaw depth.

PAGE 225

Ex set2 series20 0.5 1 1.5 2 2.5 3 1020304050607080 Energy (keV) 1cm 2cm Figure E19 Experimental tr end with flaw depth. MCNP set2 series1 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 1020304050607080 Energy 6.3 d 2c 6.3 d 1c 6.3d .5c Figure E20 MCNP5 trend with collimation extension.

PAGE 226

MCNP set2 series20 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.010.020.030.040.050.060.070.08 Energy 5d 2c 5d 1c Figure E21 Experimental trend with collimation extension. 0 5 10 15 20 25 30 35 1234567 scatter component set2 series 1 scatter component contributions 6.3 d 2c 6.3 d 1c 6.3d .05cFigure E22 Scatter component contributi on trends with vary ing collimation.

PAGE 227

0 5 10 15 20 25 30 35 40 1234567 scatter componentset2 series2 Scatter Component Contribution 5d 1c 5d 2c Figure E23 Scatter component contribution by collimation Flaws 10x10 Al plate. 75keV 2.5 cm to NaI 1 cm Ext Flaws A,B,E at 1/8, 3/16, 3/8 inch depths, respectively0 0.5 1 1.5 2 2.5 3 3.5 01020304050607080 A B E Figure E24 Experimental spectr al trend with flaw depth.

PAGE 228

4.3 cm to NaI 3.3 cm Ext (MCNP5) Flaws at 0, .15 cm, .25cm Depths0 0.05 0.1 0.15 0.2 0.25 0.3 0.010.020.030.040.050.060.070.080.09 43d 1c 15 43d 1c 25 43d 1c 0 Figure E25 MCNP5 current tally sp ectral trend with flaw depth.

PAGE 229

209 APPENDIX F PHOTON CROSS-SECTION DATA All data from X-com32

PAGE 231

211

PAGE 239

219 LIST OF REFERENCES 1. Campbell, S., and Jacobs, A., Detect ion of Buried Land Mines by Compton Backscatter Imaging, Nuclear Science and Engineering, 110, 417-424 (1992). 2. Wehlburg, J., I "Development of a Lateral Migration Radiography Image Generation and Object Recognition System, Ph. D. Dissertation, University of Florida, Gainesville, Florida, May, 1997. 3. Jacobs, J., "Examination of Backscattered Radiation from X Rays Directed into Sand and Cross-Talk Between Adjacent Detector Systems," Masters Research Project, University of Florida, Ga inesville, Florida, December, 1997. 4. Dugan, E., Jacobs, A., Keshavmurthy, S ., and Wehlburg, J.," Lateral Migration Radiography, Research in Nondestructive Evaluation, Vol 10, No. 2,pp 75-108, June, 1998. 5. Jacobs, A., Dugan, E., Howley, J., and Su Z., "Landmine Detection/ Identification Using A New X-Ray Backscatter Imag ing Technique," Th ird Symposium on Technology and the Mine Problem, Monterey, CA, April, 1998. 6. Su, Z, Howley, J., Jacobs, J., Dugan, E., and Jacobs, A., "The Discernibility of Landmines Using Lateral Migration Radi ography," SPIE Proceedings on Detection and Remediation Technologies for Mine s and Minelike Targets III, Vol 3392, pp 878-887, Orlando, Fl, April, 1998. 7. Jacobs, A., Dugan, E., Moore, J., Su, Z ., Wells, C., Ekdahl, D., and Brandy, J., "Imaging Subsurface Defects Using XRay Lateral Migration Radiography/A New Backscatter Technique," Proceedings of ASNT Conference on Real-Time Radioscopy and Digital Imaging, August, 1999. 8. Su, Z., Jacobs, A., Dugan E., Wells, C., Alla rd, A., Caniveau, G., "A Practical Land Mine Detection Confirmation System Based on X-ray Lateral Migration Radiography," Fifth Symposium on Technol ogy and the Mine Problen4 Monterey, CA, April, 2000. 9. Edward T. Dugan, Alan M. Jacobs, Da n Shedlock and Dan Ekdahl Edward T. Dugan*, Alan M. Jacobs, Dan Shedlock and Dan Ekdahl Paper for SPIE 49th Annual Meeting August, 2003 Denver, CO 10. http://www.lockheedmartin.com/wms/f indpage.do?dsp=fec&ci=11322&rsbci=131 82&fti=0&ti=0&sc=400SOFI Lockheed Martin Space Corp. July 16, 2005.

PAGE 240

11. NASA. Return to Flight Focus Area: Ex ternal Tank Thermal Protection System. Marshall Space Flight Center. Huntsville, Al 12. NASA. http://www.ncfi.com/insulation.html April 2005. 13. MCNP5 A General Monte Carlo N-Partic le Transport Code. X-5 Monte Carlo team. Los Alamos National La boratory, New Mexico. 2003 14. Jacobs, A., Dugan, E., Moore, J., Su, Z ., Wells, C., Ekdahl, D., and Brandy, J., "Imaging Subsurface Defects Usi ng XRay Lateral Migration Radiography/A New Backscatter Technique," Proceedi ngs of ASNT Conference on Real-Time Radioscopy and Digital Imaging, August, 1999 15. Dugan, E. Jacobs, A. Lateral Migr ation Radiography Image Signatures for the Detection of Buried Landmines. Gran t Extension Technical Report. ARO Grant Number DAAG-55-98-1-0400. Nu clear and Radiological Engineering Department. University of Florida. 2002 16. Jacqueline MacDonald, J.R. Lockwood, john McFee, Thomas Altshuler, Thomas Broach, Lawrence Carin, Russell Harmon, Carey Rappaport, Waymond Scott, Richard Weaver. Alternatives fo r Landmine Detection. RAND 2003 17. Lawrence Livermore National Laboratory. http://www.llnl.gov/str/Azevedo.html November 1997 18. Garder, Paul D. Landmine Detectio n with Ground Penetrating Radar Using hidden Markov Models. IEEE Transactions on Geoscience and Remote Sensing. Vol.39, No. 6, June 2001. 19. Jacobs. A., Dugan, E., Howley, J., Su, Z ., and Wells, C., "Detection/Identification of Landmines by Lateral Migration Radiography," Proceedings of Second International Conference on The Detecti on of Abandoned Mines, Institution of Electrical Engineers Pub lication No. 458, pp 152-156, Edinburg, UK, October, 1998 20. Su, Z., Concept and Applic ation of a Vehicle-mount ed Land Mine Detecton System Based on Lateral Migration Radi ography, Masters Research Project, University of Florida, Gainesville, FL, August, 1998. 21. Howley, J., "Detailed Study of Lateral Migr ation Radiography Image, 5 Relevant to Landmine Detection," Masters Research Project, University of Florida, Gainesville, Fl, August, 1998. 22. Keshavmurthy, Shyam Prasad Developm ent of Lateral Migration Backscatter Radiography and Associated Image Enhan cement Algorithms Masters Thesis. UF 1996

PAGE 241

23. Salazar, SABRINA Katherine Material Flaw Detection Characteristics of X-Ray Lateral Migration Radiography Masters Thesis. UF 2002. 24. Brygoo, Stephanie X-ray La teral Migration Radiogra phy Non Destructive Flaw Detection Measurements and Simula tions, Masters Thesis. UF 2002 25. Moore, Jeffery D. Simulated Materi al Flaw Detection Using X-ray Lateral Migration Radiography UF 1999 26. Yxlon International Corporation. http://www.yxlon.com/yxlon/yxlon_cms.nsf April 2005. 27. Beiser Arthur. Concepts of Modern Physics 5th Edition. Mcgraw Hill. New York, NY 1995 28. Attix, Frank H Introduction to Radiological Physics and Radiation Dosimetry John Wiley & Sons, Inc New York, NY 1986 29. National Institute of Standards and Technology. http://physics.nist.gov/PhysRe fData/Xcom/Text/XCOM.html August 2005. 30. Uppsala Universitet: Biomed ical Radiation Sciences. http://www.anst.uu.se/hansl und/Med_Tekn/x-ray%20production.pdf May 2005 31. Knoll, Glen E. Radiation Detection and Measurement John Wiley & Sons New York, NY 2000 32. SABRINA 3.54: Three-Dime nsional Geometry Visualization Code System. Radiation Transport Group, Los Alamos National Laboratory, Los Alamos, New Mexico. 33. LabVIEW. National Instruments. Nationa l Instruments Corporation. Austin, Tx.

PAGE 242

222 BIOGRAPHICAL SKETCH Benjamin Addicott attended the Universi ty of Florida where he received both a B.S. and an M.E. in nuclear engineering.


Permanent Link: http://ufdc.ufl.edu/UFE0013264/00001

Material Information

Title: Characterization and Optimization of Radiography by Selective Detection Backscatter X-ray Imaging Modality
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0013264:00001

Permanent Link: http://ufdc.ufl.edu/UFE0013264/00001

Material Information

Title: Characterization and Optimization of Radiography by Selective Detection Backscatter X-ray Imaging Modality
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0013264:00001


This item has the following downloads:


Full Text












CHARACTERIZATION AND OPTIMIZATION OF RADIOGRAPHY BY
SELECTIVE DETECTION BACKSCATTER X-RAY IMAGING MODALITY













By

BENJAMIN TEICHMAN ADDICOTT


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ENGINEERING

UNIVERSITY OF FLORIDA


2006

































Copyright 2006

by

Benjamin Teichman Addicott



































The Dude















ACKNOWLEDGMENTS

I would like to thank my friends and family, especially my parents and brothers,

my Aunt Deborah, Dr. Alan Jacobs, Dr. Samim Anghaie, Dr. Alireza Haghighat,

Dr. Edward Dugan, The Balta-Cooks', The Dude (and family), and my research group.

Iwould also like to thank NANT, Lockheed-Martin Space Systems Co. and NASA and

the University of Florida Department of Nuclear and Radiological Engineering for

support and funding of this project.
















TABLE OF CONTENTS



A C K N O W L E D G M E N T S ................................................................................................. iv

LIST OF TA BLE S ......... .... ..................................... ............ .. viii

LIST OF FIGURES ......... ......................... ...... ........ ............ ix

ABSTRACT .............. ............................................ xix

CHAPTER

1 IN TR OD U CTION ............................................... .. ......................... ..

New Foam Imaging Backscatter Modality.....................................................
M otiv atio n ........................................................................ 1
Spray O n Foam Insulation .......................................................... ............... 2
Purpose of Investigation ...... .......................................... ........... ............................ ..3
Radiography by Selective D election ........................................ ........................ 4

2 BA CK GROUND ........................................................ .......... ............. ....

P rem ise .............................. ......... .......................................... .... 5
New Backscatter Radiography System ......................................................................7
System D description and Configuration................................. ....................... ........... 7

3 TH EOR Y ............................................................... ..... ...... ........ 10

Photons ......................................................... .10
G general P hysics of P hotons ..................................................................... ... ... 10
Photon Interactions ..................................................... ........ .............. .. 11
P hotoelectric effect........... ...... .......................................... .. ........... 13
C om pton scattering .............................................. .......................... 14
R ay leig h scatterin g ........................... ................................. .................... 16
Photon Production and Energy Distribution Spectrum ................... ...............17
A tte n u atio n ................................................................. ................................. 1 9
D election M odalities........... ................................................................. ....... ..... .. 20
D etector C om ponents ................................................. ............................. 20
Scintillator.................................. ...................................... 20
Photo m ultiplier................................................. 2 1



v









Preamplifier ......... ................ ........... ........ 21
M odes of O operation ... ........... .............................................. .. ....... .. ......2 1
P u lse m o d e ............................................................................................. 2 1
Integrating m ode ................... ................ ................... ......... 22
N a l ................................................................................................................. 2 2
Plastic Scintillator.................. ..................................... .. ........ .... 23
D etector Com parison........................................................................... 23
Com pton Backscatter Radiography ........................................ ........................ 25
Radiography by Selective D election ........................................ ....... ............... 26
M onte Carlo M ethods (M CNP5) ...................................... ........................... ....... 27
SABRINA Supplemental Track Plotter................................. ........................ 28

4 RADIOGRAPHY BY SELECTIVE DETECTION....... ....................................29

R S D ..........................29........ ................. .........
P hoton T transport M odel ..................................................................... ..................48
Introduction and C concept ........................................................ ............... 48
Support /Evidence and Characteristics.....................................................51
Im age contrast (bright vs. dark im ages).....................................................53
Collim ation trends (optim ization) ..................................... ............... 64
Im age pixel shifts and shadow s......................................... ............... 73
S ig n al inten sity ...................................... ............ ........ ............... 7 5
Applications and Limitations...................... ...... .............................. 76

5 BACKSCATTER FIELD DISTRIBUTION AND DETECTOR PLACEMENT ....77

B ackscattered X -ray Signal Profile ................................................. .....................77
D etector Placem ent Considerations.................................... ..................................... 80

6 RSD OPTIMIZATION AND IMAGING CHARACTERISTICS .............................84

O ptim ization P rin cip les ................................................................... .....................84
S O F I F o am ................................................................8 5
Voids in foam ......................................... ..... ......... ... ............89
High density absorber and scattering type flaws............. ................94
S h a d o w s .................................................... ................ 9 7
Aluminum .............. ...... .......... ..................... ..... ......... 101
P plastic .............. .. ...... .... .......... ......................................1 12
Concrete and Gypsum ................................ ......... ..... ............... 114
R actor Insulation ............ ..... .... ...... .. .... ...... ............ .. .......... .... 118
Space Shuttle Thermal Protection Shield (TPS) Insulation Tiles ...................121

7 SP E C T R O SC O P Y ........................................................................ ......................123

Flaw Type and O orientation ......................................................... .............. 124
Flaw M material ..................................... .............. ........ .. ........ .... 124
F law D e p th .................................................................................................. 1 3 0
T arg et M material ............................................................... 134









D etector M material ........................ .. .. ........................ ...... ............... 140
Detector and Collimation Configuration ...................................... ............... 142
D etector to sam ple spacing........................................... .......... ............... 142
C ollim ation extension................................................ ............................ 144
M onte C arlo V erification................................................. ............................. 149
A applications ........................................................................................................ 149

8 CONCLU SION S .................................. .. .......... .. .............151

Applications ................. ............................. ..... 151
Recommendations for Further Developments............................ 152
D detector Configuration ......... .................... ..................... .....................152
Detector Materials and Operation Modes ............. .................................... 153
M CNP5 Simulations to be Considered................................... ............... 154
Optimizing Geom etrical Variables......................................... ............... 155

APPENDIX

A SELECTED SABRINA GENERATED PHOTON TRACK PLOTS......................156

B ONE SCATTER SIMPLIFIED PHOTON TRANSPORT MODEL........................182

C COLLIMATION DEPENDENT CONTRAST TREND ANALYSIS.....................187

D IM AGED SAM PLE DESCRIPTION S ........................................ .....................193

E SPECTROSCOPIC TRENDS ............................................................................ 195

F PHOTON CROSS-SECTION DATA...................................................................209

LIST OF REFEREN CES ........................................................... .. ............... 219

BIOGRAPH ICAL SKETCH ...................................................... 222
















LIST OF TABLES


Table page

4.1 Relative contrasts as calculated and observed experimentally .................................64

7.1 Percent contrast for various flaw types in aluminum ......................... ............129
















LIST OF FIGURES


Figure page

1.1 Mean free path of various peak energy photon beams in SOFI and in aluminum.........3

2.1 Sim ple schem atic of landm ine detection system ........................................ ................6

2.2 Tiled sample landmine scans. VS-1.6 antipersonnel mine, 2.5 cm depth-of-burial,
15 m m resolu tion .............................. ................................................. ............... .. 6

2.3 Schematic of detector components. Gray arrow represents photon beam direction.
Center of detector is 9 cm from photon beam center. ..........................................8

2.4 System signal flow chart for image acquisition......................... ...................

2.5 Yxlon x-ray tube head and four Nal collimated detectors ..........................9

3.1 Oscillating electric (E) and magnetic (M) fields about a propagating photon............ 10

3.2 Dominant interaction type as a function of energy and material Z number ................13

3.3 Photoelectric cross-section for aluminum. Log / Log scale..................................14

3.4 Compton interaction between photon and stationary electron............................... 15

3.5 Relative Klein-Nieshena Compton collision cross-section as a function of energy
and scattering angle............. ................................. ................... 16

3.6 Typical bremsstrahlung x-ray energy spectrum .................................. ............... 18

3.7 Bremsstrahlung spectrum with characteristic x-rays...............................................18

3.8 Scatter components of Compton backscatter radiography signal...............................26

3.9 Scatter component contributions to collimated RSD backscatter imaging signal......27

4.1 First collision components in SOFI reaching the detector with collimator
ex te n sio n ...................................................... ................ 3 2

4.2 Multiple collision components in SOFI reaching the detector with collimator
ex te n sio n ..................................................................... 3 2









4.3 First collision components in aluminum.................................... ....................... 33

4.4 M multiple collision components in aluminum.. .................................... ...............34

4.5 First collision com ponents............................................................. 35

4.6 Second collisions com ponents. ...... ....................................................................... 35

4.7 Third collisions com ponents. ........................................ .......................................36

4.8 Fourth collisions com ponents ............................................. ............................. 36

4.9 Higher order (fifth and greater) collisions components.................... ..................37

4.10 Scatter component contribution to detector current tally for geometry of
F ig u re s 4 .5 to 4 .9 ................................................................... 3 8

4.11 Relative contrast by scatter order........................ ..................................39

4.12 Directional distribution of tally components by scatter order. ................................40

4.13 Percent signal, relative contrast and total contrast contribution by scatter
components for 1 cm collimator extension. .................................. .................41

4.14 Percent signal, relative contrast and total contrast contribution by scatter
components for 1.32 cm collimator extension ............................... ............... .42

4.15 Percent signal, relative contrast and total contrast contribution by scatter
components for 1.5 cm collim ator extension ................................. ............... 42

4.16 Percent signal, relative contrast and total contrast contribution by scatter
components for 2.32 cm collimator extension ............................... ............... .43

4.17 Schematic of CRP. Photons at and below CRP can pass under collimator and
enter detector. ................................................................. ......... 45

4.16 Severely under-collimated. first scatters. 5.08 cm radius Nal 1.14 cm from SOFI.
Dark line indicates CRP above which no first scatters can enter the detector.........46

4.19 First Scatters. Nal 6.14 cm from SOFI 5 cm collimator extension.........................47

4.20 First Scatters. Nal 11.14 cm from SOFI 10 cm collimator extension. ...................47

4.21 First Scatters. Nal 14.15cm from SOFI 14cm collimator extension. ...................48

4.22 One scatter photon transport model. Double lines indicate upper and lower
bounds from im portant scatters. ......................................................................... 49









4.23 SABRINA generated photon track plot. Aluminum plate with shallow scattering
type flaw .......................................................................................................... ...... 52

4.24 SABRINA generated photon track plot. Aluminum plate with shallow scattering
type flaw .............................................................................53

4.25 Difference is important photon exiting paths caused by a focusing collimator
e x te n sio n ............................................................................................................. 5 4

4.26 V ShL : V oid shallow long. .............................................. ............................. 56

4.27 V ShI: V oid shallow incident........................................................... .... ...........57

4.28 V ShE : V oid shallow exit. ............................................... .............................. 57

4 .29 V D L : V oid deep long ........................................................................ ..................58

4 .30 V D I:V oid deep incident......................................... .............................................58

4.31 VDE: Void deep exit..................... ................................59

4.32 ScShL : Scatterer shallow long ............................................ ........... ............... 59

4.33 ScShI: Scatter shallow incident. ........................................ ......................... 60

4.34 ScShE : Scatter shallow exit ............................................. ............................. 60

4.35 ScDL: Scatter deep long. ................................. ......................................61

4.36 ScD I: Scatter deep incident. ........................................... ............................ 61

4.37 ScD E : Scatterer deep exit. .............................................. ............................. 62

4.38 MCNP data plots of contrast vs. collimation length....................... ................65

4.39 Experimental data points of contrast vs. collimation length.............................. 65

4.40 Analytically derived contrast trend as CRP is moved from sample surface to
flaw top ................. ......... .......... .......................................67

4.41 Analytically derived contrast trend as CRP moves below flaw bottom ...................68

4.42 Contrast vs collimator extension trend. MCNP5 simulations and 1st scatter
m odel. ................................................................................69

4.43 Contrast vs collimation extension. MCNP5 simulation data and 1st scatter
m odel approxim ation.......... .......................................................... ...... .... ..... 70









4.44 Contrast vs collimation extension. MCNP5 simulation data and 1st scatter
m odel approx im action ........................................................................ ..................70

4.45 Geometrical considerations for shadow pixel shifting relative to detector
p o sition .............................................................................. 74

4.46 Bright shadow cast by void in Al seen by detector in lower right hand corner........75

4.47 Bright shadow cast by void in Al seen by detector in lower left hand corner..........75

5.1 Klien-Nishiena differential scattering cross-section for 55 keV photon ..................78

5.2 Backscattered photon flux across a plane parallel to SOFI sample surface ................79

5.3 Percent difference in signal due to void flaw in SOFI as a function of scatter field
com ponent. ......................................... ............................. 80

5.4 RSD focusing. Each collimation configuration A, B, C selects for photons
originating at and below each specific depth A, B, C, respectively.......................82

6.1 Two important paths of a backscattered photon .................................. ............... 86

6.2 MCNP5 simulated geometry. Four, two inch thick layers of SOFI on aluminum
substrate ............................................................... ..... ..... ......... 88

6.3 Detector contribution by cell as a function of collimation. 60 keV incident
spectrum ............................................................................88

6.4 Detector contribution by cell as a function of collimation. 75 keV incident
sp e ctru m .......................................................................... 8 9

6.5 Void-type flaw in SOFI. CRP is optimally set to be just above flaw........................91

6.6 Images of foam calibration panel with varying degrees of collimation. Collimation
increases clockwise from lower left. ........................................ ...... ............... 93

6.7 Void-type flaw in SOFI. Thin arrow demonstrates the path difference induced by
the lack of scatter at the void site. ........................................ ........................ 94

6.8 Two mechanisms (1-lack of scatter, 2- increased attenuation) for generating low
intensity signal from an absorber-type flaw in SOFI. ............................................95

6.9 Scan of depth staggered aluminum inserts in SOFI. Bright inserts are above CRP
and dark inserts are below CRP. ........................................ .......................... 97

6.10 Mechanism for shadow image generation. Dashed line represents true flaw
position, solid arrow indicated shadow image detection position............................98

6.11 SOFI panel with contoured aluminum substrate. B&W ..........................................99









6.12 SOFI ramp panel with contoured aluminum substrate. .......... ...............................100

6.13 Bolt of flange panel under SOFI.............................................. ......................... 101

6.14 Clockwise from upper left, detectors 1,2,3,4. ................. .............................. 103

6.15 Photon exit paths across a void channel. ..................................... ............... 104

6.16 Shadow shifting with detector position.................................. ....... ...... ............ 105

6.17 Bright shadow images of aluminum flaw plate. 5 cylindrical void flaws at
various depths are imaged as bright with severe over-collimation ......................106

6.18 Schematic of flaw shadow and detector orientation relationship ..........................107

6.19 Correlated (processed) image of sample aluminum plate ............. ... ..................108

6.20 Uncollimated image of aluminum flaw plate..................................................... 109

6.21 Collimation set to discriminate just above shallowest flaw. Flaw depth increases
from lower right, counterclockwise to center....................................................... 110

6.22 Over-collimated image of aluminum sample plate................................. ............111

6.23 Over-collimated image of small channel aluminum plate ..................................... 112

6.24 Under-collimated image of small channel aluminum plate ...................................112

6.25 Plastic flawed plate #1, uncollimated on left, collimated on right...........................113

6.26 Correlated image of LANL block...................................................... ............... 115

6.27 LAN L block color im age ................. ........................................... ............... 115

6.28 Clock radio, glass tube, wire, and acrylic rod inside cinder block. .......................116

6.29 Various objects behind 1 inch of gypsum............................ .........117

6.30 Miniature stereo, glass, fiber optic cable, copper wire, behind 1 inch of gypsum
(dryw all). ................................... ................................. ........... 117

6.31 Reactor insulation panel image showing steel nameplate and shadow also
corrugated interior foil structure evident ............................................ .................118

6.32 Steel reactor insulation panel with boric acid residue. ...........................................119

6.33 Color image of insulation panel, boric acid on far side clearly evident. ..............19









6.34 Reactor insulation panel correlation image. Plastic bag with boric acid on far
side of panel. Interior structure of foil also apparent. .........................................120

6.35 Side view of reactor insulation panel, showing several layers of corrugated foil. .120

6.36 Internal structure of corrugated foil inside reactor insulation panel.....................120

6.37 Space shuttle insulation tile. Density variation in surface adhesive evident. Dark
spots are glue conglomerations between ceramic tile and laminate covering........121

6.38 Space shuttle insulation tile, under-collimated. Bright smears are glue below
CRP, dark circles on left are drilled holes.................. ...............122

7.1 MCNP5 generated backscatter spectra of various flaw materials in Al, high
collimation .............. ..... ...................... ................... ........ 126

7.2 MCNP5 generated backscatter spectra of various flaw materials in Al, low
collimation .............. ..... ...................... ................... ........ 127

7.3 Unresolved experimental spectral trends. Experimentally acquired data for the
spectral shift observed for void type flaw at various depths.............................132

7.5 Unresolved simulated spectral trends. MCNP5 simulations for spectral shifts
observed from void flaws at various indicated depths. ................... ...............133

7.6 Resolved simulated spectral trends. MCNP5 simulations for spectral shifts
observed from void flaws at various indicated depths. ................... ...............134

7.7 Normalized (first moment) experimental data 75 keV backscatter spectra from a
plastic targ et. ...................................................................... 136

7.8 Normalized (first moment) experimental data 75 keV backscatter spectra from an
alum inum target .................. ................................. .. .... .............. .. 136

7.9 Normalized (first moment) experimental data 75 keV backscatter spectra from a
steel target. .......................................... ............................ 137

7.10 Normalized (first moment) experimental data 75 keV backscatter spectra. No
collimator extension, 6 cm from Nal to Sample surfaces. .....................................137

7.11 Normalized (first moment) experimental data 75 keV backscatter spectra. ...........138

7.12 Photoelectric cross-sections for lead ............. ............................................. 139

7.13 Photoelectric cross-sections for aluminum. ................................... ............... 140

7.15 Normalized experimental spectral shifts for various detector to sample distances
ov er alu m inu m .............................................................................. ............... 14 3









7.16 Collim ator extension and CRP......................................... ............................ 146

7.17 Normalized experimental spectral shifts as a function of collimation extension. ...148

A l Photons having four or less collisions (void flaw).....................................................156

A2 Photons having one collision (void flaw) ...................................... ............... 157

A3 All photons entering detector (history filtered for clarity-void flaw) ........................157

A4 All photons entering detector. Scatter sites marked with black X (plastic flaw).......158

A5 All collision components, history filtered (plastic flaw).........................................158

A6 No flaw first through fourth collision components..............................................159

A7 Plastic flaw first through fourth collision components ........................... ..........159

A8 No flaw all collision components, zoomed in view................................................160

A9 Plastic flaw all collision components zoomed in view ......................... ...........160

A 10 Plastic flaw first collision components ........................................ ............... 161

All Plastic flaw first collision component................................... ...............162

A12 Plastic flaw first collision scatter points. ...................................... ............... 162

A13 Plastic flaw, all scatter components (note many scatters off the Nal surface and
dow n back into A l) ........... .... .......... .................. .... .... .... .. .. ............ 162

A14 First and second scatter points......................... ................... ............. .. 163

A 15 M multiple scatter sites ....................... .................. ... ...................... 163

A 16 First, second, third scatter points. ........................................ ........................ 164

A17 First, second, third, fourth scatter points.........................................164

A 18 A ll scatter points. ....................... ...................... ...........................165

A19 Void flaw first collision components..................................................165

A20 Void flaw all collision components. ............................................. ............... 166

A 21 First collision com ponents ............................................... ............................ 166

A 22 Second collision com ponents........................................................ ............... 167

A 23 Third collision com ponents............................................. ............................. 167










A 24 Fourth collision com ponents ........................................................... ............... 168

A25 All collisions..................................................168

A 26 First collision ............................................. 169

A 27 A ll collision s ....................................................... 169

A 2 8 N o fla w ............................................................................................................... 1 7 0

A 2 9 S S L ..................................................................................1 7 0

A30 SSL zoomed in view of tracks across flaw and aluminum ..............................171

A31 SSL. Flaw is above CRP ................ ........ ............171

A 3 2 S S L ................................................................17 2

A33 ADL flaw is below CRP. ................................................. ........172

A 24 A D L zoom ed in view ......... ........... ........... ........................ 173

A35 VDL. Flaw is below CRP. ........................................ ........ ........ 173

A 36 V D L zoom ed in view .......................................................... ........ 174

A 37 V SL. Flaw is above CRP ................................................ .............................. 174

A38 VDL, flaw shown opaque. ............................................... .............................. 175

A 39 V D L ........................................................................... 175

A40 VDL, flaw is transparent ................................ .......176

A 41 V SL zoom ed in view ......... .................. ........ ...................... ............... 176

A42 VDE. Aluminum and flaw are both set to invisible to emphasize simple photon
track s. ................ ............................. ................................ .... 177

A43 ASL. Photon on far right are reflecting off Nal, going up into it. ........................177

A 44 A SL zoom ed in view ..................................................................................178

A 4 5 A D L .......................... .... ....... ........................................................ ......... 17 8

A 4 6 V D L .............................................................................................................. 1 7 9

A 47 V D L. ....................................................... .. ............... 179

A 4 8 V SL ............. ........ .............................. ............................... ........ 180


xvi









A 4 9 S S L ............................................................. ................ 18 0

A50 SDI zoomed in view of flaw and photon tracks. Aluminum is set to transparent. 181

B.1 Typical photon path across a flaw .................................. .............. 182

B.2 Parameters used to determine optimal collimation length for focusing to a
sp ecified d ep th ................................................. ................ 18 6

C. 1 Signal intensity as a function of depth and across a void type flaw.......................... 187

C.2 Contrast versus collimator extension for collimator from zero to critical (optimal)
extension. This optimal corresponds to collimating to top of flaw....................189

C.3 Contrast versus collimator extension for collimator from critical (optimal)
extension to flaw bottom ...... ........................... .......................................190

C.4 Contrast versus collimator extension for collimator extension corresponding to
flaw bottom to maximum extension.............. ............. ............. ..............191

C. 5 Contrast versus collimator extension ...................................................................192

El Spectroscopic trend for flaw depth by incident energy spectra ..............................195

E2 Trends for flaw depth. Flaws depth increases from A to E............... .................196

E3 Scatter component break down for spectra. .................................. ............... 197

E4 Spectral break down by scatter component. ....................................................... 197

E5 Spectral break down by scatter component. MCNP5 current tally .........................198

E6 Current tally spectra for various flaw depths and collimation configurations. .........198

E7 Collision component signal contribution for various flaw depths in aluminum.......199

E8 Collision component contributions for various flaw depths with uncollimated
detector in alum inum ............ .............................................................. ........... 199

E9 Collision component contributions to 0.6cm void flaw in aluminum with various
collim nations configurations. ...... ......................................................................200

E10 Spectroscopic trend with flaw depth. Flaw depth increases from A to E..............200

El Experimental spectroscopic trend with detector height..................................... .....201

E12 MCNP5 current tally spectroscopic trend with detector height ............................201


xvii









E13 Scatter component contribution for various detector and collimator
configurations .................. ..................................... .......... ........ 202

E14 Spectral breakdown by collision component.............................................. 202

El5 Spectral breakdown by scatter component.................... ........... .... ........... 203

E16 Spectral breakdown by scatter component.......... ................. ..............203

El7 Spectral breakdown by scatter component................... ........... .... ........... 204

E18 Experimental trend with flaw depth. ............................................. ............... 204

E19 Experimental trend with flaw depth. ............................................. ............... 205

E20 M CNP5 trend with collimation extension.................................... ............... 205

E21 Experimental trend with collimation extension ............................. ......... .......206

E22 Scatter component contribution trends with varying collimation ..........................206

E23 Scatter component contribution by collimation .............................................207

E24 Experim ental spectral trend with flaw depth..........................................................207


xviii















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Engineering

CHARACTERIZATION AND OPTIMIZATION OF RADIOGRAPHY BY SELECTIVE
DETECTION BACKSCATTER X-RAY IMAGING MODALITY


By

Benjamin Addicott

May 2006

Chair: Edward Dugan
Cochair: Alan Jacobs
Major Department: Nuclear and Radiological Engineering

Backscatter x-ray imaging techniques have been developed at the University of

Florida for applications ranging from the detection of buried landmines to the

non-destructive-examination (NDE) of various industrial materials such as aluminum

and carbon-carbon composites. Recently, a new backscatter x-ray imaging system with

geometries and components different than that employed in previous systems has been

developed under contract with Lockheed Martin.

The primary purpose of this new system is the NDE of the foam thermal insulation

material used by NASA on the space shuttle external fuel tank. The imaging modality

has also been applied to evaluate other objects such as samples made from aluminum,

plastics, steel, concrete, gypsum and titanium as well as to reactor vessel head steel

insulation panels. Detailed subsurface images were acquired on each of these various

materials indicating the wide range of applicability of this imaging modality.









This investigation aims at describing the physics and phenomena associated with

this new imaging modality. The photon transport and interaction processes leading to an

acquired image are explored. Both Monte Carlo simulations and analytical calculations

are used, in conjunction with experimental results, to develop an analytical model

detailing the mechanics of the photon transport process governing the imaging modality.

Trends in detector response and spectroscopic profiles due to variations in

parameters such as detector collimation length, detector to target-object spacing, flaw

type (i.e., void, scattering or absorbing type flaws) and orientation and media type is

cataloged and explained in terms of this model. System variations such as detector type

and detector mode of operation are also investigated.

Consideration of these trends as well as the simple analytical model developed is

then applied, in conjunction with experiments and previously collected data, to identify

key parameters in detection efficiency and apply them towards system optimization and

an overall assessment of the applicability and limitation of the imaging modality.














CHAPTER 1
INTRODUCTION

This research endeavor details the design and optimization of a unique Compton

backscatter type non-destructive x-ray imaging modality developed over the past several

years at the University of Florida .1-4 The origin of the system dates back to systems

initially designed for the detection of buried landmines. 58 Success of this imaging

modality at acquiring definitive images of subsurface landmines had led to its application

in the NDE (non-destructive examination) of various industrial materials. Currently, a

new state-of-the-art system has been developed at the University of Florida designed

specifically for the NDE of the foam thermal insulation on the space shuttle external

tank 9, but applicable, with modification, to a wide range of industrial type materials.

New Foam Imaging Backscatter Modality

Motivation

The new system is essentially a derivative of the original landmine oriented system.

The present system, funded largely by a contract from Lockheed Martin, 10 is designed

for and applied primarily to the imaging of the special foam like material used for

thermal insulation of the space shuttle fuel tanks. This foam insulation, or spray-on-foam

insulation (SOFI),11' 12 because of its extremely low density, represents a unique imaging

challenge.

The integrity of the insulating SOFI barrier must be ensured before shuttle launch is

permitted. Imaging this material and optimizing the system for detecting certain classes









of flaws in this material has consequently become a priority and much of the optimization

methods and investigations have been conducted towards this end.

Spray On Foam Insulation

The SOFI material used by Lockheed Martin and NASA roughly comprises a

mixture of carbon, nitrogen, chlorine, fluorine, oxygen, and hydrogen. The exact

chemical composition was deemed proprietary and therefore had to be estimated 13 from

the composition of similar industrial foam insulation materials. The density of the

material is given as 0.03g/ cm3, about two orders of magnitude lower than that of

previously imaged materials such as aluminum (2.7 g/ cm3), plastics and composites

(~1 g/ cm3).

Due to the extremely low density of the foam, a photon, on average, will have

significantly fewer interactions, or travel a greater distance in the medium, before it

scatters and returns to the detector. The mean free path in this material (SOFI), roughly

inversely proportional to the density, is orders of magnitude larger than that of any

previously investigated materials. It is therefore reasonable to expect the physical

processes resulting in the imaging of this material to be somewhat different than those for

other higher density materials, such as aluminum or landmines. Figure 1.1 displays the

average mean free paths obtained from MCNP5 14 Monte Carlo calculations of several

incident photon spectra in both aluminum and SOFI.











Avg MFP (cm) vs Peak Energy


60 mfp (cm)
Foam
~,, OAl





fo0...
60 KE-I

Emax
75 keV



Figure 1.1 Mean free path of various peak energy photon beams in SOFI and in
aluminum

Purpose of Investigation

This discussion aims at describing the physics and phenomena which govern the

new imaging modality. Detector response and spectral features of this imaging modality

and the parameters which affect them will be investigated and their mechanisms detailed.

Also covered will be the methods employed to investigate these phenomena and how

they are applied to the optimization of the present system as well as suggested

modifications for further development of future generation backscatter type imaging

modalities.

Careful observations and experimentation, complemented by MCNP5 Monte Carlo

simulations have afforded a comprehensive understanding of the photon interactions and

detector responses indicated by particular images. This knowledge has allowed simple









algorithms to be established to optimize image quality and increase system efficiency.

Methods have also been developed to systematically and accurately focus the detector

collimators to target a specific depth in material as well as to distinguish between types of

flaws in various materials and, in some situations, to describe flaw depth and three-

dimensional orientations.

Radiography by Selective Detection

Detector response and image acquisition observed throughout this investigation are

described by a concept referred to as Radiography by Selective Detection (RSD). The

theory of RSD is that by preferentially selecting specific components of a scattered

photon field, information relating to specific locations and properties of an imaged

sample can be extracted. That is, sensitizing the detector towards selected photons

within a specific volume will result in larger relative signal fluctuations caused by flaws

and features within that volume.

In this imaging modality, collimators are used in conjunction with the detectors to

discriminate against certain components of the reflected scatter field. This effectively

enhances the relative contribution and, thus, the contrast resulting from more important

and often deeper regions of a sample where a flaw or region of interest lies. In this study,

a simplified analytical model of the RSD imaging modality is developed and supported.

Detector responses and imaging characteristics of the system are then described in terms

of this proposed model.














CHAPTER 2
BACKGROUND

Premise

Backscatter radiography methods were developed at the University of Florida 15 as

a means to effectively detect and positively identify buried landmines. The method was

developed in response to problems arising from other land mine detection systems

available at the time. 16-18 Most significant among these problems was that of false

positive identification of buried landmines. Previous methods identified landmines only

as subsurface irregularities making it difficult and often impossible to distinguish

between true landmines and landmine sized objects (e.g. rocks, debris, etc.).19

In the system designed at the University of Florida, positive landmine

identification via realistic image acquisition was accomplished by placing some of the

detectors behind a series of collimators. 20 The collimators were positioned so that the

majority of the shallow first scattered photons would be prevented, geometrically, from

reaching the detectors. This essentially provided the detectors with a less cluttered view

of the subsurface features of the landmines and other buried objects. The landmine

detection system configuration and a demonstration of its imaging capabilities is

presented below in Figures 2.1 21 and 2.2. 22





















Uncolllmated
Detectors


Lead Collimator


I cm
4


Mine






Figure 2.1 Simple schematic of landmine detection system


i saa-
15 J3 E

B "Fj3BB-
I E
U 15 -
S1B B B
29 JS a.


T NC


Raster Scan (ixels)
Front


Raster Scan (p ixe)
Rear


U ncollimated D etecto r Images


511


Raster Scan (piels)

Front


Raster Scan (p ixes)

Rear


Collimated Detector Images


Figure 2.2 Tiled sample landmine scans. VS-1.6 antipersonnel mine, 2.5 cm depth-of-
burial, 15 mm resolution


31020



2 15JS-


N x tK-









Previous research endeavors centered upon understanding this process and

optimizing this system initially for landmine detection and, later, for other industrial non-

destructive examination applications have been carried out over the past several year.21-

28 Analysis and evaluation of these along with the inevitable progress of technology have

led to a gradual and steady advancement of this imaging method and the understanding of

the mechanisms which govern it.

New Backscatter Radiography System

Recently, the backscatter x-ray system has been acutely modified and redesigned to

meet the specific needs of imaging SOFI and other drastically different materials than

the original system was designed for. The new system features different detector and

collimator configurations, different photon beam energies as well as new electronic

components and detector geometries. These differences, combined with the fresh

perspective afforded by optimizing and testing a new system configuration on a new and

unique material (i.e., SOFI), have led to new imaging approaches and concepts as well as

a comprehensive analytical model which accurately describes the photon transport

process and resultant detector responses.

System Description and Configuration

The system used in this series of investigations consists of four sodium iodide

[Nal (Tl)] scintillation detectors and an Yxlon MCG41 x-ray generator 29 mounted onto a

scanning table with X Y scan motion capabilities. The detectors are positioned at the

corners of an eighteen by eighteen centimeter square, centered on the x-ray beam. Each

detector comprises a two inch diameter by two inch thick Nal scintillation crystal

mounted onto a photomultipler tube (PMT) and a fast preamplifier specifically designed

to handle high count rates. The customized, ultra-low-noise, high count rate preamps









have a maximum noise level of about 5 mV for a 1 volt pulse output, while operating in

close proximity to a strong electro-magnetic field (x-ray generator tube). The preamp

pulses have a typical rise time of 100 nano-seconds and fall time of about 1000 nano-

seconds, yielding a total pulse width of about 1.1 micro-second. This specific pulse

width (1.1 micro-seconds) allows sufficient light and charge collection time from the Nal

and PMT (about five time constants), while allowing the detectors to measure backscatter

fields up to 700,000 counts per second, without experiencing statistically significant

pulse pile-up. A schematic of the RSD detector components and their configurations is

presented below in Figure 2.3, followed by a flow chart of the entire image acquisition

process from detection to display in Figure 2.4. Figure 2.5 is a picture of the x-ray

generator tube head and collimated detectors.








fast pre-amp
PMT


Nal
lead collimator









Figure 2.3 Schematic of detector components. Gray arrow represents photon beam
direction. Center of detector is 9 cm from photon beam center.
































Figure 2.4 System signal flow chart for image acquisition


Figure 2.5 Yxlon x-ray tube head and four NaI collimated detectors














CHAPTER 3
THEORY

Photons

General Physics of Photons

Photons are a form of electromagnetic radiation. Like all forms of electromagnetic

radiation, photons travel in waves and are accompanied by oscillating electric and

magnetic fields. These fields are perpendicular to each other as well as to the direction of

photon propagation and rotate about the axis of travel. A qualitative representation of

propagation of electromagnetic waves and their associated electric and magnetic fields is

presented in Figure 3.1.




E E




hv
M b


E







Figure 3.1: Oscillating electric (E) and magnetic (M) fields about a propagating photon.









Although photons are necessarily accompanied by an electric field, they have no

net electric charge and, to a first approximation, are not influenced by the electric charges

and fields of neighboring particles or waves.30

Photon Interactions

Although photons, like all electromagnetic radiation, are most completely described

by their quantum mechanical wave functions and probabilities, for our purposes,

considering the energies (45-100 keV) and the processes that we are interested in

(Compton scattering and photoelectric absorption), they can be well represented by

discrete particles with energies dictated by a classical and definite wavelength.

Photon interactions, then, may be accurately and relatively completely described by

five major interaction types. These are, photoelectric effect, Compton Scattering, pair

production, Rayleigh (coherent) scattering, and photonuclear interactions. Each of these

interaction types is described by a quantum mechanical or empirical microscopic cross-

section detailing its probability of occurrence for a photon in a particular phase space.

These cross-sections, modified by certain atom and energy dependent incoherent

scattering functions and form factors, give the probability of each particular type of

interaction per atom in a sample. The sum total of these, or the total cross-section,

represents the total interaction probability per atom for a photon in a particular phase

space. Multiplying these microscopic cross- sections by atom density produces the

macroscopic cross-sections which represent the probability per unit path length of

interaction of a particular type. Typically, microscopic cross-sections are given in units

of c cm2 / atom or in barns (10 24 m 2 / atom ) per atom, depending upon whether they









have been appropriately modified. Macroscopic cross-sections are typically given in

units of Y c p/M or(cm2 /g)(g/cm3) (atoms /g) = cm1

Once these cross-sections have been defined, another parameter, the mean free path

(mpf) can then be introduced. This is a typical attenuation length or free flight travel

length given as 1/ E = cm. This is the length that the uncollided portion of a beam of

photons will be attenuated by a factor ofe' (2.718). The MFP can also be interpreted as

the distance in a material that a photon of certain energy would have a probability e 1 of

traveling through without interaction. 31

Each of these photon cross-sections is a function of at least the energy of the

photon and Z of the material. For the typical materials (low Z < 15) and photon energies,

<- 100 keV, the dominant interaction types are Compton scattering and photoelectric

effect. For lower energies, < 30 keV coherent scattering of photons can also become

important. Figure 3.2, 31 below, plots the three important cross-section, i.e. photoelectric,

Compton, and pair production, as a function of both energy and material Z number.

Clearly, the first two interaction mechanisms, photoelectric and Compton scattering, are

the dominate interaction mechanisms for the typical Z range and energy range

[highlighted in the (red) circle] that we are concerned with. The rest of our treatment of

photons in this discussion will be for finite wave/particles with definite energies and

knowable pre and post collision velocities.
















S80- dominant dominant

60-
o0 Compton effect -
N 4 0 d o m in a n t ; -




0.01 0.05 0.5 1 5 10 50 100
hv in MeV




Figure 3.2 Dominant interaction type as a function of energy and material Z number



Photoelectric effect

Photoelectric interaction is an interaction between a photon and a bound electron.

In these interactions, a photon is completely absorbed by a bound atomic electron. This

electron is then ejected from the atom. It is therefore necessary for the incident photon to

carry at least the energy required to free the electron. This is effectively the electrons'

binding energy. If the photon imparts less than the binding energy to the electron, the

electron will merely be excited and consequently release a photon upon de-excitation

rather than being ejected form the atom. A typical photoelectric cross-section, shown in

Figure 3.332 for aluminum, is a strong function of energy. In fact it varies roughly as

(1/E3) .32 This results in a very large absorption cross-section at low energies.







14




1 04











102






10 10-
Photon Energy (MeV)
S Photoelectric Absorption


Figure 3.3: Photoelectric cross-section for aluminum.32 Log /Log scale.

Compton scattering

Compton scattering is best conceptualized as an elastic billiard ball type collision

between a photon 'particle' and a free electron. In this process, a photon with known

direction and energy collides with an assumed stationary (unbound) atomic electron.

After the collision, both the photon and electron velocity vectors (energy and position

vectors) can be accurately described via classically considered conservation laws of

energy and momentum.















hv'





hv
e-



Figure 3.4: Compton interaction between photon (thick, red) and stationary electron
(thin, blue)

It is, for the most part, these types of collision which concern us, as it is these

collisions which predominately result in photons being backscattered from the target to

the detector.

The Compton Scattering probability is described in terms of a differential cross-

section which relates the probability in cm2 / steradian per electron for a photon of a

specific energy to scatter into a specific solid angle about a specified angle. This

differential probability distribution is the Klein-Nieshena differential cross-section. The

Klein-Nieshena cross-section uniquely relates initial energy, scattering angle and final

do- v' v' v
energy of a photon. This relationship is given as: = -( sin2 0) where,
dQ v v v

v' 1 ho
S= is the ratio of scattered to initially photon energy and a = Is
v 1 +a(1- cosO) moc2

the reduced initial photon energy. As this relationship dictates, the scattering

probabilities will be peaked (have local maximums) in the directly forward (0) and

directly backward (180) directions. The scattering will also follow a forward peaking










trend as the energy is increased. The Klein-Nieshena relative scattering probabilities for

several incident photon energies are plotted below in Figure 3.5 for 0 to 180 degrees

scatters.




2-
Incident photon
energy


1.6-


1.4-


1.2-


1-


0. 8-



0 0.5 1 1.5 2 2.5 3
eta

Figure 3.5: Relative Klein-Nieshena Compton collision cross-section as a function of
energy and scattering angle, eta (in radians).

Rayleigh scattering

Rayleigh scattering, pronounced mainly at low energies (hv < 30 keV) and high Z

materials, is the highly forward elastic scattering which occurs between an incident

photon and a bound electron. The important difference between coherent (Rayleigh) and

incoherent (Compton) scattering is that in Compton scattering, the collision is between a

photon and an essentially unbound electron, whereas in coherent scattering the collision

is between a photon and the target atom as a whole. This difference allows for the vast

majority of the momentum and energy to be retained by the incident photon while the









target atom receives only the minimum amount of momentum to ensure conservation.

Due to the relatively huge mass of a target atom relative to a single electron (as in

Compton scattering), the photon does not transfer much energy to the atom. This is very

similar to bouncing a ball off of a wall. In this case, they wall, being fixed, does not

absorb a significant amount of energy or momentum and the ball rebounds with

essentially the same amount of energy as it had before the collision. Coherent scattering

is highly forward peaked. Similar to incoherent (Compton) scattering this peaking is

accentuated at higher energies.

Photon Production and Energy Distribution Spectrum

Photons utilized in research are produced via an x-ray generator. X-ray generators

operate on the principle that high energy electrons incident upon a tungsten target

produce a spectrum of bremsstralung 33 photons. The initial production of these photons

is more or less isotropic. The energy is distributed, initially, according to a typical

bremstraulung spectrum as shown in Figure 3.6. The addition of aluminum of copper

filters hardens the primary bremstraulung spectrum. Figure 3.7 shows target

characteristic x-rays superimposed on the continuous bremstrahlung spectrum. For a

tungsten target, the generator voltage has to exceed at least 69.5 keV, the K-shell binding

energy, for the characteristic x-rays to be generated. These are the type of spectra used in

the investigations presented in Chapters Five through Seven.

































Figure 3.6 Typical bremsstrahlung x-ray energy spectrum.33


Figure 3.7. Bremsstrahlung spectrum with characteristic x-rays.









Important parameters describing this spectrum are the most probable energy, the

average energy, and the peak energy.

The spectral analysis completed as part of this investigation makes use of the

scattering-to-absorption ratio. It is this ratio, which includes both coherent and

incoherent scatters, that governs the spectral shifts and trends realized in our

measurements and simulations. This can be visualized by considering a uniform

spectrum of photons traveling through and being attenuated by a particular material.

Since for lower energies the attenuation coefficients are higher, due to the photoelectric

effect, it is logical to predict that the emergent beam spectra will be shifted towards

higher energies. This is a result of the lower energy photons being preferentially

absorbed. Other shifts in energy can likewise be explained.

For our particular situation where we are interested mainly in backscattered

photons, an appropriate down shift in energy, due to energy lost in scattering can be

observed. This downshift is theoretically dependent upon the initial photon energy, the

scattering angle, and the number of scatters encountered before reaching the detector.

Attenuation

Linear attenuation coefficients are described for monoenergetic uncollided photons

beams as the sum of all interaction cross-sections multiplied by the appropriate atom

density. This coefficient, given in terms of 1/cm is then multiplied by the photon path in

cm and the exponential function of this product, given as e-Nx then represent the

fractional uncollided photon intensity as a function of distance traveled in a medium.

This simple equation can easily be expanded into a series with appropriate

coefficients and weighting parameters to provide a description of the uncollided photon

intensity of a particular photon energy distribution.











Detection Modalities



Detector Components

Scintillator

Scintillation detectors function by turning radiation energy into visible light,

which is subsequently collected and converted into an electrical signal. The process by

which visible light is produced from incident radiation takes place in the scintillation

material, the portion of the detector which interacts with the radiation, and the process is

referred to as fluorescence. This process involves the absorption of some portion of the

incident radiation's energy by an electron. The electron is then elevated from its normal

energy state into an excited state. The excited state is necessarily less stable that the

original ground state of the electron and thus de-excites back to this more stable ground

state. With this de-excitation comes a photon of light of wavelength determined by the

energy gap that the initial electron traversed in its excitation.34

In inorganic scintillation crystals, such as the sodium iodide (Nal) used in most of

these investigations, electrons are excited from the valence band to a conducting band

across an energy gap called the "forbidden band." In this energy gap, no sublevels are

found so that there is no real probability of finding an electron between the valence and

conduction bands. In pure inorganic crystals, the de-excitation of an electron to the

valance band with the proper photon emission is not realistically efficient for practical

detection requirements. To compensate for this, and to increase the probability of the

resultant photon being in the useful visible range of the electromagnetic spectrum, small

amounts of impurities, referred to as activators, are added to the crystal. These impurities









have an energy band within the forbidden band gap of the pure crystal. With the

appropriate activator added, an electron-hole pair will migrate though the crystal until

they reach an activator impurity site where they will quickly de-excite with the release of

a useful photon of visible light. This light is then funneled into a photomultiplier tube

where it is turned into a measurable electronic signal.34

Photo multiplier

Photomultiplier tubes collect the visible scintillation photons and convert them

into a measurable electric signal. The total process progresses in three distinct stages.

Photons from the scintillation crystal impinge upon the photocathode region of the tube

where they are converted to electrons. These electrons are then channeled down the

electron multiplier where they are proportionately multiplied by several (typically 5-7)

orders of magnitude. After this multiplication process, the electrons are then collected at

the anode end of the tube where they have effectively become a now measurable electric

signal proportionate to the incident scintillation photons and thus, to the original radiant

energy deposition.34

Preamplifier

The preamplifier serves as an intermediate signal amplification step between the

detector and the analytical circuit used to process the detected signal. The circuit

components and time constant of the preamplifier have important implication on the

detector behavior as a whole.

Modes of Operation

Pulse mode

The detection of signal energy distribution (spectroscopy) or true count rates

requires the detector to be used in pulse mode. In this mode, each incident quantum of









radiation produces a pulse. Each pulse in turn is individually collected by the detector

and processed as a count. The voltage height of this pulse is proportional to the energy

deposited. This is the most common mode of operation and for most of the experiments

conducted in this investigations, pulse mode operation is used.34

Integrating mode

In situations where the detected count rate is high so that pulse pile up occurs,

current (or integral) mode of operation can be used to mitigate such detector saturation

problems. In integral mode, the total charge generated over a set time is collected. This

set time, known as the detector response time, is large compared to the time between

individual events, and thus the association between charge created and individual

interactions is lost. The benefit, however, is that the detector does not need a recovery

time between individual pulses and thus is capable of handling much higher count rates.

One important property of integral mode of operation is that the current measured

or detected is not exactly equivalent or necessarily strictly proportional to the true count

rate. The reason for this is that it is the total charge deposited in the detector over a set

time which is measured. This charge is dependent upon the number of particles

interacting, the type of interaction, as well as the energy of the interacting particles. In

other words one particle which deposits most of its energy will be detected the same as

two particles each of which deposit half as much energy.34

Nal

The NaI is the standard by which the other materials are measured. The system

was originally designed with NaI detectors and most of the experiments and images

acquired on the system have been with this detector type. The advantage of NaI is its fast

response time and large photoelectric cross-section. This property allows NaI to be used









as thin crystals and still collect full energy deposition. Additionally, because most of the

photons which impinge upon Nal deposit their full energy, it has good energy resolution

and is appropriate for spectroscopic investigations. Disadvantages of Nal include that it

is brittle and hygroscopic.

Plastic Scintillator

The advantage of the plastic scintillation detectors is that they are more resilient,

cheaper, not hygroscopic and can be produced in almost any desired shape. The plastic

detectors are also capable of handling higher count rates. The mean free path (mfp) of a

photon in plastic is on the order or 2 cm, much larger than that of Nal crystals(-mm).

However, the relative photoelectric cross-section for plastic type scintillators is smaller

than that of Nal (Appendix F). Consequently, many photons do not deposit their full

energy within the plastic scintillation material. For this reason, energy resolution is poor

in plastic type scintillators and they are of almost no use in spectroscopic analysis.

Plastics are more conducive to integration mode operation at high count rates.

Detector Comparison

Both the Nal and the plastic detectors may be operated in both pulse and

integrating mode. The concept is that integral mode is not susceptible to pulse pileup

problems encountered at high count rates. Since pulse mode counts each individual

photon by collecting the appropriate charge deposited in the detector, pulse mode

detectors are limited, in count rate, by an associated dead time which is the time required

to collect and dissipate the charge of an individual photon. This dead time is ultimately

limited by the charge collection time of the detector material itself which in Nal is about

0.23 is, indicating a maximum ideal (assuming cps could be limited only by this time

constant) count rate in pulse mode of about 800,000 cps (5 time constants 35 ) before









significant pulse pile up and dead time effects result. In reality, however, the maximum

count rate is governed by the time constant of the entire detector circuit, including the

pre-amplifier. Current or integrating mode, however, collects the charge produced by

multiple photons incident over a predetermined integration period. In integral mode, it is

the total energy deposited over a period of time and not the energy deposited per

interaction that is important. Because current mode operates on a voltage produced by

the incident photons and the number of electron-hole pairs they generate, greater weight

is given to a higher energy photon since they generally deposit more energy and thus

create more electron-hole pairs in the detector material. This results in images which

show biasing towards high energy photon detection. This effect is more pronounced in

Nal than it is in the plastic scintillators. This is due to the fact that more energy is

deposited in Nal than in the plastic material (per unit distance traveled by a photon). A

photon impingent on a Nal crystal is likely to deposit all or most of its energy within that

crystal resulting in a proportional amount of electron-hole pairs being produced. That

same photon incident upon a plastic scintillation material (of comparable thickness) is

less likely to deposit the majority of its energy in the detector. A lower energy photon,

however, will deposit a larger portion of its energy since its MFP is smaller. The result

of this phenomenon is that there is essentially a maximum energy deposition limit above

which no single photon will usually deposit the rest of its energy. Consequently, unlike

Nal, most photons, regardless of their energy deposit about the same amount of energy

and therefore produce about the same number of electron-hole pairs in a relatively thin

plastic scintillation material.









A problem with this feature of the integrating mode (primarily in the Nal) is that

the change in signal strength resulting from photoelectric attenuation, which is dominant

at low energies, is not observed as well. Often this differential photoelectric signal is a

significant portion of the overall contrast and discounting it results in image degradation.

Another problem with neglecting the low energy portion of the signal is that often the

high energy portion of the incident spectrum does not interact as intimately with the

medium and consequently contributes to noise rather than signal.

Compton Backscatter Radiography

Compton Backscatter Imaging (CBI) Radiography is a non destructive imaging

modality used to image objects when transmission type methods are not feasible. The

important feature of CBI is that access to only one side of an object is required. For

transmission type radiography a detector is placed on the opposite side of an object

relative to the source and is sensitive to photons which traveled through the object; CBI

techniques place the detector and source on the same side of the object and the detector is

then sensitive to photons which interacted in the object and scattered backwards into the

detector.

Many traditional Compton methods employ a large unobstructed detector which is

sensitive to the entire distribution of backscattered photons (at least those scattering into

the relevant solid angle subtended by the detector). Since the majority of backscattered

photons will suffer a scatter near the surface (within half of a mean free path) of the

substrate and then return to the detector, the signal and the image generated directly

reflects electron density variation within this surface and shallow subsurface region of the

sample. An example of this configuration is illustrated in Figure 3.5 below.












detector










target






Figure 3.8 Scatter components of Compton backscatter radiography signal.

Here, the shallow first scattered component of the backscattered field, represented

by the thick (red) arrow is the dominant signal contribution and clearly overwhelms the

contribution of the deeper penetrating and/or multiple-collided photons which are usually

considered as noise. This is intuitive if the exponential attenuation of photons is

considered, necessitating that each additional distance traveled into the sample by the

photon beam results in exponentially less photons available to scatter back into the

detector. Additionally, the contributions of multiple-scattered photons are further

reduced since these photons must both have and survive multiple collisions without being

absorbed.

Radiography by Selective Detection

Radiography by Selective Detection (RSD) techniques are similar to CBI methods

in that they rely on Compton (mostly) backscattered photons to generate an image of the

investigated object. The difference is that RSD techniques employ collimators and









calculated detector positioning to select for specific regions of a backscattered photon

field. The effect of this is to enhance the sensitivity to the detector response function to

variations of specific components of the backscatter field. A simple illustration of this

principle is shown in Figure 3.9 below.



detector

collimator















Figure 3.9. Scatter component contributions to collimated RSD backscatter imaging
signal.

Here, the shallow first scattered field, represented by the thick (red) arrow is

effectively discriminated against by the collimator (lower (pink) cylinder). The resultant

image is thus generated largely form deeper penetrating photons. In RSD methods

detector and collimator geometries and orientation are governing parameters in selecting

the portion of the detected backscattered photon field, and hence the region of the sample,

viewed by the detector.

Monte Carlo Methods (MCNP5)

Throughout this investigation Monte Carlo methods, implemented via MCNP5, are

used to simulate experimental setups and detector responses. Monte Carlo methods are a









means of solving a problem through statistical sampling of probabilities and are used

when deterministic methods are not desirable. Effectively, Monte Carlo methods arrive

at a particular solution by tracking particles and tallying individual events until enough

information has been obtained to infer a reasonable answer. Each event is determined by

sampling from a pool of random numbers distributed according to the appropriate

interaction probabilities.

SABRINA Supplemental Track Plotter

SABRINA36 is an application code which, in conjunction with MCNP,

graphically displays the simulated geometry and/or the photon tracks and interactions

mechanics. It utilizes the MCNP geometry input deck and a special PTRAC card which

causes MCNP to generate a file in which selected history data (location, interactions, and

velocity components) of photons run in the MCNP simulation are recorded.














CHAPTER 4
RADIOGRAPHY BY SELECTIVE DETECTION

RSD

Radiography by Selective Detection (RSD) produces images via a signal

differential due to a linear attenuation difference experienced by single and multiple

scattered photons as they traverse various regions of a sample. The photons essentially

travel in a simple reflection path between the source and the detector (much like optical

photons), with an appropriate backwards scattering occurring in the target object. This

approximation is roughened by photons which interact more than once in the sample

before scattering into the detector. Photons having more than one scatter in the target

material deviate, to varying degrees, from those having only one collision. For the

examined configurations, analysis and experiments indicate that these multiple collided

photons behave essentially as single scattered photons (for the purpose of providing

contrast in imaging modalities) in that they transverse flaws directly, as the once scattered

photons do, on the way from their last scatter to the detector. That is, the mechanism for

generating flaw contrast is essentially the same for single and multiple scattered photons.

The contrast, regardless of the number of scatters, is a function of the attenuation

difference afforded by the flaw as the photons impinge upon it and exit, after scattering,

towards the detector.

The effect of these multiple scattered photons, to a first approximation, is

tantamount to a broadening of the initial impingent photon beam. Photons having more

than one scatter lose some degree of their original incident directionality. Thus,









depending upon how many scatters the photons have suffered, their backscattered field

distribution is skewed from the primary once scattered backscatter photon field.

Additionally their final scattering points are necessarily displaced from the incident

photon beam axis. The contrast observed by these photons, however, is generated by the

same mechanism responsible for single scatter photon contrast. That is, the attenuation

differential afforded by a flaw as a photon traverses it. For photons having many

collisions (usually 4 or more) this attenuation differential can become negligible in

comparison to the total photon path length in the target material and thus the contrast for

these very high order photons is often much lower than for the primary and secondary

scattered photons. This concept of effective beam spreading can be observed in Figures

4.1 and 4.2, below. Each of these figures is a SABRINA generated photon track plot

from an MCNP5 simulation. The simulation models a one inch aluminum substrate

below eight inches of SOFI foam. The detector is 2 inch diameter Nal and is located 9

cm, centerline-to-centerline, from the impingent beam (2mm in diameter) and 5.14 cm

above the foam surface. The collimator is extended 4 cm past the Nal surface, or 1.14 cm

from the SOFI surface. Figure 4.1 shows first collision components of the backscattered

radiation field that reach the detector.

Intuitively, all these collisions occur along the axis of the impingent beam, shown

by the black arrow. Figure 4.2 is the SABRINA generated plot of the same MCNP5

simulation displaying multiple (second order and higher) scattered photons. As

demonstrated in the figure, except for a few outliers, the effect of multiple scattered

photons can be approximated by an effective broadening of the impingent photon beam.

That is, the result of total scatters from a narrow beam can be approximated well by









considering only single scatters from a broader, diverging beam. This concept is meant

to augment the understanding of the contrast generating mechanism involved in the

imaging modality rather than to be used as a quantitative model for system optimizations.

The multiple scattered photons make up an effective source distribution within the target

material which has a wider distribution than the effective source distribution of the once

scattered photons. Thus, intuitively, the last scatter in the target material, before photon

detection, of the multiple scattered photons does not occur along the impingent beam axis

as the first scatter does. If the initial beam were diverging, however, then the first scatter

site distribution for the diverging beam would be similar to the multiple scatter sites of

the line source which is actually impingent. The dark arrows in Figure 4.2 demonstrate

the concept of effective impingent beam widening which would account for the effect of

multiple scatters. Many of the scatters occur within this area and upon their final scatter

are directed towards the detector. These arrows are meant to indicate approximately the

effective multiple scattered photon source distribution and are not a quantitative

representation of an actual beam divergence. (Note: the right half of the area i/ i/hin the

arrows does not display scatters because the SABRINA plot was filtered to only include

photons scattered into the detector shown.)











detector

c I ....- ..
coL'x toi
bLrFoam


Figure 4.1 First collision components in SOFI reaching the detector with collimator
extension.


- detector

collimator



foam


Figure 4.2 Multiple collision components in SOFI reaching the detector with collimator
extension. Diverging arrows demonstrate approximate area of multiple
scattered source distribution.

The concept of effective beam broadening is further demonstrated, in aluminum

targets, by the following figures, 4.3 and 4.4. Figures 4.3 and 4.4 are SABRINA plots of

an aluminum target with a 75 keV impingent photon beam. The detectors are 5.08 cm









radius Nal positioned 9 cm from the photon beam and are 5 cm from the aluminum

surface with a 4.4 cm collimator extension. Figures 4.3 and 4.4 show the first collision

and multiple collision backscattered components which reach the Nal detectors,

respectively. As again illustrated by the black arrows, the first collision components all

originate along the impingent beam axis while the higher order components can be

modeled as originating from a radial axis of a broader, diverging beam. The validity of

this approximation is based upon the fact that, upon suffering a scattering collision, a

photon is necessarily deflected at some angle away from the initial photon beam. In

order for the photon to be detected, it must either scatter directly into the Nal, or suffer

another event that scatters it into the detector. The photons which do not scatter directly

into the Nal, as the figures illustrate, effectively make up a distributed source within the

target itself. This effective distributed source of multiple scatters is very similar in effect

to the primary scatters of an initially broader, diverging, beam having a spread

approximated by the black arrows in Figure 4.4.

detector collimator









fla/
'*a. .....

fIa /
flaw~


Figure 4.3 First collision components in aluminum














detector collimator
It~


flaw


1A1


Figure 4.4 Multiple collision components in aluminum. Arrows indicate approximate
multiple scattered photon source distribution.


The following set of figures, 4.5 4.9, show the effective beam spreading as a


function of scatter components. These plots demonstrate that the dominant mechanism


for image contrast generation regardless of the number of collisions is the attenuation


difference provided by the flaw as the photons directly traverse it. The figures are again


generated by the SABRINA application using MCNP5 simulation data. Each figure


models a 10 x 10 x 1 inch aluminum plate with a void type flaw running along the axis


from beam to detector. The flaw is 0.4 cm in height and 1 cm wide and 1 cm below the


aluminum surface. The 2.54 cm radius detector is positioned at 9 cm from the beam


center and 2.9 cm above the aluminum surface with a 1.5 cm lead collimator extension.


This collimation is configured so that the CRP (critical reference plane, see Figure 4.17)


is located just below the flaw channel bottom. In each figure the black arrows again


roughly indicate the effective single scatter beam divergence that would approximate the


multiple scattered photons. As each of these demonstrate the vast majority of photons


directly and linearly traverse the flaws and consequently the resulting acquired image


............. .......
.. .. ... .. .. .. .. ... .. .. ... .. ... .. .. ... ..
... .. ..............
.. .. .. ... ... .. .. ... ..
.. ... .. .. ... ..
...............
.......... .............
.. .. ... .. .. .. ... .. .. ... ... .. .. ........ ......
.... ... ... ......... .......
...... .. .. .. .. ......... .....
........ . . . . . .









contrast can be directly related to the difference in attenuation properties induced by the

flaw. This is the same effect we would observe if the signal comprised all first scattered

photons originating from a wider range than the initially impingent beam. That is the

dominant mode for image generation is the same regardless of the scatter order of the

photon.


det


collimator_ -




Alfaw


flaw


Figure 4.5. First collision components. Black arrow indicates impingent beam axis and
line of scatter origination. Note that photons directly traverse the flaw.


collimator


detector


flaw


Figure 4.6 Second collisions components. Black arrows indicate approximate effective
beam divergence. Note that photons directly traverse the flaw










detector


colimator


flaw


Figure 4.7 Third collisions components. Black arrows indicate approximate effective
beam divergence. Note that photons directly traverse the flaw


detector


collimator


PAl


faw


Figure 4.8 Fourth collisions components. Black arrows indicate approximate effective
beam divergence Note that photons directly traverse the flaw


~B~h~~


pL










collimator


detector








flaw



Figure 4.9 Higher order (fifth and greater) collisions components. Black arrows indicate
approximate effective beam divergence. Note that photons directly traverse
the flaw

The degree to which the effective beam divergence is observed is a function of the

relative scatter component contribution to the detector, as dictated geometrically by the

collimation configuration, as well as the mean free path of the target material. For

relatively high density materials such as aluminum, shown in the figures above, most of

the second and third order scatters occur close enough to the initial beam so that

neglecting them in an approximation is valid. The relative importance of the first seven

scatter components for this simulation, given as percent contribution to the detector

current tally, for the simulations depicted above in Figures 4.5 4.9, are plotted below in

Figure 4.10. Even for this highly collimated situation, contributions of the first three

scattering components make up over 70% of the total signal. Furthermore, comparison of

this data against a similar MCNP5 run without the flaw channel reveals that the majority

of the contrast contribution and thus the important part of the signal comprised mostly

first and second scatters as shown below in figure 4.11. This is indicative of the proposed












contrast providing mechanism. As scatter order increases so does the total path length


of a photon in the target material. As path length increases (with scatter order) the


relative attenuation differential afforded by the flaw (since it remains the same size)


decreases. Thus higher order scatter components have lower relative contrasts even


though they may represent larger portions of the total signal.



Relative Scatter Component Contribution to detector Tally


1 ",',),

.,,,,,


: u u

-,) ,),


1st 2nd 3rd 4th 5th
scatter component


6th 7th


total


Figure 4.10. Scatter component contribution to detector current tally for geometry of
Figures 4.5-4.9.










Percent Contrast by Scatter Components












0









1st 2nd 3rd 4th 5th 6th 7th total
scatter order


Figure 4.11. Relative contrast by scatter order. This is the contrast that would be
observed if each scatter component could be isolated and individually
considered.

Additionally, a directional distribution of the tally collision component breakdown

(for the simulation discussed above) reveals that for each scatter component considered,

up to seventh, the vast majority of the detected photons impinge upon the detector at an

angle of thirty degrees or less to the horizontal, just as the once scattered photons do.

This indicates that even high order scatter events do not bring the photon significantly far

away (geometrically: i.e. the scattering angles do not vary by more than a few degrees)

from the impingent beam axis. If photons in general, were to scatter farther before being

deflected into the detector, we would observe a more significant deviation in angular

direction components as scatter order increased. There is, as expected, a noticeable

increase in detected photons entering further away from the horizontal with increasing










scatter order. However, the majority of photons enter at relatively the same range of

angles as the once scattered photons, implying that they traverse the flaw at a similar

angle and are thus similarly attenuated by the flaw. The important photons, as defined

above and in Figure 4.10 and Figure 4.11, are shown in Figure 4.12 to be composed of

more that 90% photons entering within thirty degrees of the horizontal.


Scatter Component Detector Entry Angle


094 0984
cosine angle from horizontal (mu)


Figure 4.12. Directional distribution of tally components by scatter order.

The following four figures, 4.13 4.16, are plots of data taken from MCNP5

simulations. Each shows the percent signal, relative contrast, and contrast contribution of

each scatter component (up to seventh) of the signal. The four plots are taken from four

separate simulations each of a 40 x 40 x 5.08 cm aluminum target with a 0.08 cm high

and 1 cm wide flaw channel 0.1 cm below the aluminum surface. Each simulation was

modeled with a 5.08 cm diameter Nal detector 2.9 cm from the target aluminum surface











and offset from the impingent beam (center-to-center) by 9 cm. The collimation in each


of these runs was varied as 1 cm, 1.32 cm, 1.5 cm and 2.32 cm extension past the Nal


surface. In these plots, the percent signal of each component is calculated by dividing the


signal strength of that component by the total signal strength. The relative contrast is


calculated by dividing the difference between the nth components of flawed versus non-


flawed scenarios by the appropriate component of the non-flawed scenario. This relative


contrast represents the contrast that would be observed if the detector was only sensitive


to the nth scatter component of the backscattered field. The contrast contribution in these


plots is calculated by dividing the signal difference of the nth scatter component by the


total signal difference. This represents the contribution to the contrast generated by each


scatter component.


40X40X5.08 AL PLATE 9 CM TO DET CENTER 1 CM COL EXT 2.9 CM TO DET SURFACE
0.08cm high flaw 1 cm wide 0.1 cm below surface

12000%


100 00%


8000%
*II-- II 1E .:.:._1 1 ;I
D :-: l : T
6000% i i:


40 00%


20 00%


0 00% ----
1ST 2ND 3RD 4TH 5TH 6TH 7TH TOTAL
Scatter Component


Figure 4.13. Percent signal, relative contrast and total contrast contribution by scatter
components for 1 cm collimator extension.












40X40X5.08 AL PLATE 9 CM TO DET CENTER 1.32 CM COL EXT 2.9 CM TO DET SURFACE
0.08 cm high flaw 1cm wide 0.1 cm below surface


* ;rll-i
m l-ii_ i i i i i i: T
0rh-,-,l iTl: T ,-,-, il: lli.-.Il,, i


1ST 2ND 3RD 4TH 5TH
Scatter Component


6TH 7TH TOTAL


Figure 4.14 Percent signal, relative contrast and total contrast contribution by scatter
components for 1.32 cm collimator extension


40X40X5.08 AL PLATE 9 CM TO DET CENTER 1.5 CM COL EXT 2.9 CM TO DET SURFACE
0.08 cm high flaw 1cm wide 0.1 cm below surface


120 00%


10000%


80 00%


6000%


4000%


20 00%


0 00%


-20 00%


scatter component



Figure 4.15. Percent signal, relative contrast and total contrast contribution by scatter
components for 1.5 cm collimator extension


120 00%



100 00%



80 00%



60 00%



40 00%



20 00%


0 00% I-


Sl 1 1- I I I I
m l-lD l i i ,_ 1 ii ii Ti
r ,,- _,i ii : T ,_ _,i ll -n ,_ l _,i I


IhL,


IV


I I


"I I_ JIH M I


i" 1 I











40X40X5.08 AL PLATE 9 CM TO DET CENTER 2.32 CM COL EXT 2.9 CM TO DET SURFACE
0.08cm high flaw 1 cm wide 0.1 cm below surface

12000%

100 00%

8000%
Ii-- Ii_ I I l :.:,Ii -iii I i

60 00%


40 00%


20 00%


000% I oil
--T IIC IC. -TH 1TH T..T-L

-2000%

-40 00%
Scatter Component


Figure 4.16. Percent signal, relative contrast and total contrast contribution by scatter
components for 2.32 cm collimator extension

As these plots demonstrate, the relative contrast of the first collision component is

usually (with exceptions for extremely over-collimated and under-collimated cases) the

largest. This is because for these cases, the first collision path length is the shortest and

thus the flaw represents the largest relative attenuation difference. As scatter order

increases, relative contrast percentages generally decrease. This is due to the relatively

longer path length of multiple scattered photons in the target material and consequently

lessened effect of the attenuation difference caused by the flaw. In fact, as the figures

also indicate, for scatter components on the order of 5 or more, the resultant signal is

considered noise and can detract from the desired contrast. The above plots also

indicate, as expected, that increased collimation increases the signal contribution of

higher order scatters. This is accomplished mostly by eliminating shallow low order









scatter components geometrically from entering the detector. The contrast contribution

is a function of both the relative contrast and the signal contribution of each scatter

component of the signal. Consequently, even though a particular scatter component may

have the highest relative contrast, it may not represent the dominant contrast contribution

if it does not compose a significant percentage of the total signal. Similarly, the fact that

a particular scatter component dominates the signal or even the contrast does not imply

that it necessarily produces the largest relative contrast. The justification for including

higher order (second, third, and fourth) scatter components even though they may have

lower relative contrasts than the first scatter component is that the advantage of decreased

scanning time provided by the stronger signal outweighs the disadvantage of lower

contrast.

The scattered photons viewed by an RSD configuration are those specific photons

which interacted in or traveled through a specific region of interest. In radiography by

selective detection, collimators and detector placement are coordinated so that only

certain components of a backscattered signal are detected. In many cases, this amounts to

using the collimators to discriminate against all interactions occurring above a specific

region of interest. This allows a signal originating from deeper within a sample to be

collected and relative differences caused by small or deep flaws to become observable.

In the RSD imaging modality, this collimation-induced specificity for signal components

is referred to as focusing. By focusing to a specific depth, the modality effectively

discriminates against all photons having scatters above this depth. The depth to which a

RSD configuration is focused is described by a critical reference scattering plane (CRP)

which is an imaginary plane located at the depth at which the first significant primary









scatter contribution to the detector occurs. The concept of a CRP is demonstrated by the

schematic presented in Figure 4.17. In this figure, below, the CRP is shown as a dark

horizontal line. The effect of the collimator on once scattered photons originating from

above this plane and below this plane is shown.




Detector



Scollimator



CRP __

target




Figure 4.17 Schematic of CRP. Photons at and below CRP can pass under collimator
and enter detector. Photons scattering from above CRP are blocked by the
collimator.

Figures 4.18 4.21 demonstrate how a particular collimation configuration focuses

to a specific depth by discriminating against shallower components of the returning

scatter field. These figures are again SABRINA photon track plots ofMCNP5

simulations. The target is eight inches (20.32 cm) of SOFI foam on an aluminum

substrate. Each figure shows the CRP location by a dark horizontal line at the site of the

first important scatter event. Each scenario simulated has a collimator sleeve to sample

separation of 1.14 cm. Thus a 5 cm collimator extension implies a distance of 6.14 cm

from detector (Nal) surface to sample surface. In Figure 4.18, the collimator is fully

withdrawn so that photons scattering from all depths in the target may impinge upon the









detector. In Figure 4.19, the collimator is extended 5 cm past the Nal. This forces the

CRP to a depth of 5.2 cm below the foam surface. In this figure it is evident that no

scatters occurring above the CRP are reaching the detector. Figure 4.20, shows the effect

of further increasing the collimator extension to 10cm past the detector surface. Simple

geometrical calculations reveal that the CRP is now 11.6 cm below the SOFI surface and,

as indicated in the figure, this is the minimum depth that photons must penetrate before

being able to directly scatter into the detector. Figure 4.21, features still further

collimation as the collimator is extended 14 cm past the Nal surface. Here the CRP is

moved to a depth of 16.7 cm and, as the SABRINA track plot demonstrates, no primary

scatter events occur above this plane and enter the detector.


detector

collimator
CRP



foam a

void








Figure 4.18. Severely under-collimated. first scatters. 5.08 cm radius NaI 1.14 cm from
SOFI. Dark line indicates CRP above which no first scatters can enter the
detector.










detector

collimator


CRP

foam








Figure 4.19. First Scatters. Nal 6.14 cm from SOFI 5 cm collimator extension. Dark line
indicates CRP above which no first scatters are tallied.


---- detector

-collimator


fo am


CRP


Figure 4.20. First Scatters. Nal 11.14 cm from SOFI 10 cm collimator extension. Dark
line indicates CRP above which no first scatters are tallied.













collimator





foam


CRP




Figure 4.21. First Scatters. Nal 14.15cm from SOFI 14cm collimator extension. Dark
line indicates CRP above which no first scatters are tallied.

RSD modalities can often be modeled as one scatter phenomena in that the photons

behave much like the once scattered photons in a traditional Compton imaging system.

Essentially the physics of RSD can be considered similar to traditional CBI, except that

the photons having scatters above a selected depth are discriminated against. In this

idealization, RSD methods effectively remove a specified amount of material from the

surface of a sample and thereby view the lower layers, below the CRP.

Photon Transport Model

Introduction and Concept

A rough analytical model has been developed under consideration of experimental

observations, MCNP simulation, and photon transport physics. The purpose of this

model is to facilitate visualization and understanding of the phenomena leading to an

image as well as to approximate detector responses and system optimization parameters









for various types of materials and flaws. The model used to describe the current system

is a simplified "one scatter" model, shown below in Figure 4.22.

In this model, each potential first scatter site along the length of the impingent

beam axis is considered and the integrated relevant path length leading from each

element to the detector is then calculated. The difference between the attenuation of this

path for flaw versus no flaw conditions is then taken to approximate a contrast ratio. This

model facilitates quick calculations and serves as a good analytical model upon which to

base optimization parameters.



1st significant scatter that
passes under collimator a Detector
determined by CRP
Collimator




Flaw
Deepest scatter that
reaches detector,
determined by
photon penetration.
Figure 4.22 One scatter photon transport model. Double lines indicate upper and lower

bounds from important scatters.

Here the signal difference between flawed and non flawed regions of a sample can

be visualized as a difference in attenuation. The photon beam potentially crosses a flaw

twice, as shown above, once impingent and once after scattering. The difference in

attenuation of the photon beam over its entire path (incident and scattered) between

flawed and non-flawed regions is the primary contributor to the signal difference

observed in the detector. In this model the lack of scattering within a volume due to a









low density flaw can be shown to be equivalent to an increase in attenuation of the

photon beam that otherwise would have scattered within that volume and returned to the

detector. That is, the photons which do not interact in the low density flaw continue to

penetrate through the sample where eventually they will interact. When they do interact,

it will necessarily be deeper within the sample and they will necessarily have a longer

distance to travel though the material to the detector. Consequently these photons will

have a greater probability of being attenuated on the way to the detector than their

counterparts which interacted in the shallower volume of a non-flawed sample. [This is,

however, contingent upon flaw orientation and detector and collimation configurations.

If the system is configured such that the selectedportion of the back scattered field

traverses the flaw again on the way out, the overall effect on attenuation may be an

increase or decrease, depending upon flaw height and material interaction

characteristics, as discussed later in this chapter].

Analytical calculations can easily be performed based upon the simplified one

scatter model. In this one significant scatter model, the attenuated photon intensity at

each point along the incident beam path (shown as the thick, red arrow) is scattered and

further attenuated towards the detector. The integral of this path over all scattering points

then results in a final once scattered intensity at the detector. The detected signal intensity

then has the following form: AeaebB(y). Here A is the original photon intensity and the

exponential terms, a and b, account for the attenuation between the source and the first

scatter and between the first scatter and the detector, respectively and B(y) is a function

which accounts for the angular cross-section of the scatter and the solid angle subtended

by the detector. The percent difference between this integrated intensity for a flawed









versus a non-flawed region in then taken to be the percent contrast. More mathematical

detail of this treatment is in Appendix B.

While in reality a detected signal is composed largely of multiple scattered ( 2nd

order and higher) photons, the usefulness of the simplified once scattered transport model

is that it provides a means of understanding and visualizing contrast generating

mechanisms and thus augments our ability to predict and understand trends, optimization

and relevant image features such as shifts and shadows.

Support /Evidence and Characteristics

This photon transport model of the RSD modality is based largely upon

experimental observations and supported with MCNP5 simulations, SABRINA track

plots, and analytical calculations. Trends observed in detector response arising from

parameter variations such as flaw type and orientation and collimation configuration

strongly suggest a scatter model of this type. Specifically, the image contrast, positive

and negative (depending upon flaw type and orientation) as well as collimation induced

contrast trends and pixel shifts observed in acquired images all lead to and are well

described by the proposed transport model. Figure 4.23 and Figure 4.24 are SABRINA

generated track plots of MCNP5 simulations. The simple linear tracks shown here

support the premise of the once scattered transport model, i.e. that contrast is primarily

generated by a differential in attenuation experienced by photons as they directly traverse

a given flaw. Figure 4.23 is a plot of a highly collimated detector over an aluminum plate

with a scattering-type flaw channel (modeled as C2Hs02 plastic). As the plot shows,

many of the detected photons suffer multiple scatters. However, for the purpose of

generating image contrast, even these high order scatter photons behave essentially as

primary or first scatter photons. That is, the majority of the photons, regardless of scatter









order, traverses the flaw in approximately the same manner with approximately the same

angle and consequently results in similar contrast. Figure 4.24 is a similarly produced

SABRINA track plot except that the collimation is much less pronounced. In this

scenario the photons similarly traverse the flaw directly and thus generate an image

contrast based upon the differential in scattering and attenuation characteristics provided

by the flaw. As the Figures 4.23 and 4.24 demonstrate, the attenuation process and

contrast mechanics are essentially the same for all important photons, regardless of their

scatter angle and order.







collimator



flaw











Figure 4.23. SABRINA generated photon track plot. Aluminum plate with shallow
scattering type flaw. Severe collimation.











detector


collimator

flaw -


**these photons are reflected ofth and
are traveling down not up into the detector



Figure 4.24. SABRINA generated photon track plot. Aluminum plate with shallow
scattering type flaw. Less collimated.

Image contrast (bright vs. dark images)

Images generated with the RSD imaging modality display flaws as either bright

(high intensity) or dark (low intensity) regions of a sample. Bright and dark regions are a

result of relatively more or relatively less photons, respectively, reaching the detector due

to the presence of a flaw. This, in turn, is usually caused by a lack of or increase in

photon beam attenuation resulting from the presence of a flaw. In the proposed transport

model, this differential attenuation is a function of both flaw type and orientation as well

as detector and collimator configurations. That is, the intensity of a detected signal is not

uniquely a function of electron density of the region of the sample where the incident

photon beam impinges. The explanation for this is that, RSD techniques are sensitive to

the photons' exit paths though the media, which depending upon system geometry and

flaw depth, can be the source of significant attenuation. The contrast between exit paths









of important contributing photons due to collimation in RSD imaging modalities is

illustrated in Figure 4.25



Uncollimated Collimated



















Figure 4.25 Difference is important photon exiting paths caused by a focusing collimator
extension.

The additional distance traveled in the material by collimated RSD contributing

photons, highlighted by the (red) circle in the figure above, provides additional influences

on the contrast presented by collimated RSD images.

Considering relevant parameters in flaw type and orientation and collimation

configurations, most relevant imaging scenarios can be idealized by eighteen simple

models. The parameters are critical reference scattering plane (a function of collimation

configurations), flaw type, and flaw orientation (length). The collimator can be set so

that the flaw lies either above or below the critical scattering reference plane. (This

imaginary plane, again, represents the depth at which the first important scatter occurs

that can pass under the collimator and enter the detector.) The flaw, relative to the object









media, can be low density (void), high density scatterer or absorber. The orientation of

the flaw can be such that both the incident and exiting photon field pass through it, only

the incident field passes though it, or only the exiting field passes though it. These eight

characterizing parameters henceforth referred to by the following letter designations:

D deep, the flaw is below the critical reference plane

Sh-shallow, the flaw is above the critical reference plane

V- void, the flaw is a void

Sc- scatter, the flaw is a scattered (higher density)

*low density scatterers are treated as voids

A-absorber, the flaw is an absorber (higher density)

*qualitatively, low density absorbers are either treated as weak absorbers

or as voids and treatment is contingent upon flaw dimensions and relative

attenuation characteristics between flaw and target medium. This scenario

was not considered in detail.

L-long, the flaw is oriented so that both incident and reflected photon beams pass

though it

I-incident, the flaw is oriented so that only the incident photon beam passes through

it

E-exit, the flaw is oriented so that only the reflected beam passes through it on its

way to the detector.

These parameters can be organized into eighteen permutations relative to a control,

non flawed sample. They are specified by three letter designations such that the first

letter indicates flaw type (V/A/Sc), the second letter indicates the collimation









configurations (D/Sh) and the third letter indicates the flaw orientation (L/I/E). The

permutations are thus:

VShL, VShI, VShE, VDL, VDI, VDE, ScShL, ScShI, ScShE, ScDL, ScDI, ScDE,

AShL, AShI, AShE, ADL, ADI, ADE

They are depicted in the following Figures 4.26 4.37:


Figure 4.26. VShL: Void Shallow Long. More photons reach a depth at which they can
scatter to the detector (below reference plane) due reduced attenuation on the
way down. More photons reach the detector on the way out due to lack of
attenuation on the way out. BRIGHT


E........ mEnu



























Figure 4.27. VShI: Void Shallow Incident. More photons reach a depth at which they
can scatter to the detector (below reference plane) due to reduced attenuation
on the way down. BRIGHT


Figure 4.28. VShE: Void Shallow Exit. More photons reach the detector on the way out
due to reduced attenuation on the way out. This is also the cause of shifts and
shadows: BRIGHT


~


















m...............


Figure 4.29. VDL: Void Deep Long. This scenario can either be viewed as bright or
dark and is contingent upon the combination of two main mechanisms. The
first mechanism is the lack of scattering at the flaw site due to the lack of
material. This mechanism is similar to traditional Compton backscatter
mechanism with a specified amount of material removed form the top by
preferential discrimination of the collimator. This mechanism tends to lead to
a dark image as expected by a void type flaw in traditional Compton
backscatter radiography. The other major mechanism is the increase signal
intensity due to the decreased attenuation of the exiting photons due to the
flaw. In this mechanism, the incident photons that traverse the flaw scatter
below it and experience a lessened attenuation on the way out due to the
presence of the flaw. This mechanism tends to produce a bright image. The
interplay between these two mechanisms and thus the overall contrast (bright
or dark) of the image is very sensitive to parameters such as flaw height and
depth, detector and collimator configurations and the attenuation properties
(scattering to absorption ratio and over mfp) of the target
material


Figure 4.30. VDI: Void Deep Incident Almost same effect as VDL but dark because
photons are effectively transported deeper by the flaw and now travel further
out through the material. Again similar to traditional CBI: DARK


.... U. ... .. .. .E



























Figure 4.31 VDE: Void Deep Exit. More photons reach the detector on the way out due
to reduced attenuation on the way out: BRIGHT


Figure 4.32. ScShL: Scatterer Shallow Long. Less photons reach a depth at which they
can scatter to the detector (below reference plane) due to additional
attenuation on the way down. Less photons reach the detector on the way out
due to additional attenuation on the way out. DARK (note: if scatterer is pure
scatterer or very low density, the opposite effect may be observed and these
flaws are treated as 'voids'.)


imEm m ..... .. .


















HMMMMMMMM


Figure 4.33. ScShI: Scatter Shallow Incident. Less photons reach a depth at which they
can scatter to the detector (below reference plane) due to additional
attenuation on the way down. DARK


Figure 4.34. ScShE: Scatter Shallow Exit. Less photons reach the detector on the way
out due to additional attenuation on the way out: DARK























Figure 4.35. ScDL: Scatter Deep Long. More photons are scattered at a shallower
depth: BRIGHT


Figure 4.36. ScDI: Scatter Deep Incident. More photons are scattered at a shallower
depth: BRIGHT


IMMEM OMN. L ..............

























Figure 4.37. ScDE: Scatterer Deep Exit. Additional attenuation from denser material on
the way out: DARK

The six scenarios considered with an absorber type flaw result in photon paths

identical to those for the six scatter type flaws depicted above and are, for the sake of

reducing unnecessary redundancy, not shown. Flaws of dense absorber material will

always produce a dark image because whether the dominant process is transmission or

reflection dense absorber materials always decrease intensity relative to no flaw.

As the figures above indicate, the relative intensities, high or low, are a primary

result of the differential in attenuation provided by a flaw. The mechanisms by which a

flaw can perturb the degree of attenuation, as illustrated above, include both primary

attenuation differences resulting from the portion of the flaw traversed by the photon

field (incident and/or exit) as well as secondary differences resulting from the effective

translation in scattering depth, either deeper (void) or shallower scattererr), resulting from

the presence of a flaw. Besides changing the path-length (and consequently the degree

of attenuation) of an exiting photon beam, translating the scattering position (in depth)

changes the solid angle subtended by the detector and thus produces second order effects

on the photon intensity reaching the detector.









The scenarios presented above are first scatter simplifications meant to aid in the

understanding of important contrast generating mechanisms and physical processes

occurring within the imaging modality. In real world applications, results deviate to

varying degrees from these models due to a number or second order effects. Namely,

geometries which select for higher order collision components and geometries which are

more accurately described by a combination of two or more of the scenarios described

above rather than just one. In these situations, bright and dark images as well as expected

absolute contrast become very sensitive to flaw, target, photon beam, and detector

characteristics and dimensions. Thus, while these scenarios accurately describe a wide

range of true applications, there are many which are not perfectly described by this very

simplified one scatter linear model.

Further support of this model is provided by the agreement, between predictions

based upon rough analytical calculations, experiments, and MCNP5 simulations, of the

percent contrast for the simplified scenarios described above. The percent contrast

(relative to a control no flaw scenario) for each experimental, analytical and MCNP5

simulations for a few of the scenarios is listed in Table 4.1 below. Several important

differences between the calculations, MCNP5 models and experimental results should be

noted however. The transport model, upon which the calculations are based are 2-D

geometry and only roughly approximate the real world 3-D scenarios. Additionally, for

ease of calculations, the impingent photon beam is treated as monoenergetic. The

experimental results also include real world uncertainties and efficiencies that neither

MCNP5 the calculated results consider.









Table 4.1. Relative contrasts as calculated and observed experimentally


Configuration 1 scatter MCNP5 Experiment Bright/Dark

Agreement

VDL* 33.8% 22.3% Bright Yes

VShL 54.2% 29.7% Bright Yes

VDE 15.9% 17.1% Bright Yes

VDI -22.9% -21.1% dark Yes

2.54 cm radius Nal detector 2.9 cm from sample surface, 9 cm centerline-to-

centerline form impingent photon beam. 0.15 cm high void channel flaw 0.65 cm deep,

1.5 cm collimator extension.

** MCNP5 data has error of less than 1%.

Collimation trends (optimization)

Both experiments and Monte Carlo simulations reveal inverse parabolic plots of

contrast versus collimator extension for a wide range of flaw types and system

configurations. The trend shows contrast to increase exponentially as collimation

extension is increased from no extension up to a critical optimal extension length. The

contrast then remains roughly constant as the collimation is increased slightly and then

begins to decrease exponentially as the collimator extension is further increased. Figures

4.3826 and 4.3928 are plots from previous publications of MCNP and experimental data,

respectively, and both reveal this trend.













Contrast vs. Collimator Length


25.00
S20.00
15.00
10.00
o 5.00
0.00


-ED


--Poly.
(Current)


1.75 1.85 1.95 2.05 2.15 2.25

Collimator Length (cm)


Figure 4.38 MCNP data plots of contrast vs. collimation length. This data is taken from
an MCNP run with a 2 inch square detector located 5.96 cm center-to-center
from the impingent beam (radius 0.5 cm). The detector height is 6 cm and the
void flaw channel is 0.3 cm below the aluminum surface. Analytical
calculations based upon first scatter geometry predict an optimum collimation
length of 1.98 cm which matches well with the MCNP derived current tally
optimum.




Contrast versus Collimator Length for Various Flaw Depths and Detector
Aperture Sizes (A) using a 75 kVp X-ray Generator Voltage


1.50 1.75 2.00 2.25 2.50
Collimator Length (cm)


Flaw Depth, Aperture size
_X--_ --- 3 mm, A = 1 cm
---5 mm, A = 1 cm
ir-* : 3i -----7 mm, A = 1 cm
---x-.- 3 mm, A = 2 cm
---x--- 5 mm, A = 2 cm
2.75 3.00 -*-- -7 mm, A = 2 cm


Figure 4.39 Experimental data points of contrast vs. collimation length. This data was
taken experimentally and the results have not been verified in this study. The
trend, however, assumes the predicted shape for contrast vs collimation
curves.


i,




I _____~I_









This trend also supports and is well described by the simple direct scatter transport

model. To illustrate, consider the simple, ideal case of a regular void type flaw of

arbitrary dimensions positioned anywhere within a sample. (Note that as the collimator

is extended, for any particular geometry, the critical reference scattering plane is moved

accordingly deeper and the combination will henceforth be referred to as a movement of

the reference plane.)

As the reference plane is lowered from the surface of the object to the top of the

flaw, image contrast should increase since the scatters in the material above the flaw

contribute only to noise and thus reduce the relative contribution of the important

scatters. This increase in attenuation should be roughly logarithmic (inversely

exponential) since the photon is attenuated exponentially and thus each additional unit of

depth into the sample contributes exponentially less photons to the signal. The trend

then, as the reference plane moves from the sample surface to the top of the flaw should

look approximately like Figure 4.40. This trend is derived analytically, based upon the

first scatter model, and is further detailed in Appendix C.

























Flaw top




S II I I I I I I I I I I I I I
0 1 2 3 4

CRP
Figure 4.40 Analytically derived Contrast trend as CRP is moved from sample surface to
flaw top.

As the reference plane moves from the top of the flaw towards the bottom of the

flaw, the contrast should remain about constant since no scatters occur within the void

region of the flaw and consequently no photons are eliminated by varying the reference

plane within this region. The trend will not be exactly flat in this region however,

because increasing the collimator extension within this region does affect the contribution

of important multiple scattered photons and thus we expect, and observe, a slight

downward slope within this region. Also, of note is that if the flaw is anything other than

a void, i.e. a scatterer or an absorber, this downward slope will be more exaggerated since

important scatters will occur within the flaw height region.

Once the reference plane reaches the flaw (void) bottom and is moved towards the

bottom of the sample or to the maximum penetration depth of the incident photon beam,









we expect the contrast to decrease. This decrease is due to the fact that the photons

below the sample, which are now being discriminated against by the collimator, are

important photons in that they traverse the flaw and contribute to the signal differential.

Since a relatively constant (assuming little variance in the K-N cross-section and solid

angle) portion of these photons which interact below the flaw contribute to the signal, we

expect and observe the contrast to taper off approximately as the signal does, i.e. -

exponentially. This trend is shown below in Figure 4.41 as the reference plane moves

from the flaw bottom towards the maximum photon penetration depth.





C
0





a
S
t




CRP


Figure 4.41 Analytically derived contrast trend as CRP moves below flaw bottom

Combining the three segments described above results in a similarly shaped curve

with similar inflection points observed in both experiments and MCNP5 simulations.

Furthermore, it can be demonstrated that the peak in the curve, the optimum collimation








69



length, corresponds to the depth of the top on the flaw, and that the width of the plateau


in the trend corresponds approximately to the height of the flaw.



Contrast vs Collimator Extension
40X40 cm Al Plate 9 cm to Det Radially 2.9 cm to Det Surface
0.08cm high flaw void channel 1cm wide .1cm below surface


45 00%


40 00%


35 00%

30 00%


S25 00%


u 20 00%


1500%


1000%


5 00%


0 00%
0 05 1 15 2
col ext (cm)


--MCNP5 cont
-- 1scat cont
-- 1st MCNP5


Figure 4.42 Contrast vs Collimator extension trend. MCNP5 simulations and 1st scatter
model.















Contrast vs Collimator Extension
40X40 cm Al Plate 9 cm to Det Radially 2.9 cm to Det Surface
0.18 cm high flaw void channel 1 cm wide .3 cm below surface


12000%



10000%



80 00%



60 00%



40 00%



20 00%



0 00%



-20 00%


-- MCNP5 cont
-- 1scat cont
1st MCNP5


collimator ext (cm)


Figure 4.43 Contrast vs collimation extension. MCNP5 simulation data and 1st scatter

model approximation.



40X40X5.08 AL PLATE 4 CM TO DET CENTER 2.9CM TO DET SURF
0.08cm high void flaw 1cm wide .1cm below surface


50 00%



40 00%



3000%


3 --- MCNP5 cont
20 --1scat cont
2000%st MCNP5
l 1st MCNP5


10 00%



0 00%



-1000%


collimator ext (cm)


Figure 4.44 Contrast vs collimation extension. MCNP5 simulation data and 1st scatter

model approximation.









Figures 4.42 4.44 reveal this trend for two flaw geometries for both MCNP5 data

and for lst scatter analytical calculations as described in Appendix B. Data for Figure

4.42 was acquired for a 40 x 40 cm aluminum plate, 5.08 cm thick. The detector is a Nal

crystal located 9 am centerline-to-centerline from the beam and 2.9 cm above the

aluminum surface. The flaw was a .08 cm tall void channel 1 cm wide and located at 0.1

cm below the surface. The collimator extension for the trend was varied from 1 cm to

2.32 cm. The MCNP5 data plotted is from a current tally crossing the detector surface.

The 1st scatter data is calculated analytically as described in appendix B. The mfp used

was taken from the average mfp as calculated by MCNP5. Figure 4.43 was similarly

derived except that the flaw channel was changed to a height of 0.18 cm at a depth of 0.3

cm. Geometry for data shown in Figure 4.44 is identical to that of Figure 4.42 except that

detector to photon beam centerline-to-centerline spacing was changed from 9 cm in

Figure 4.42 to 4 cm in Figure 4.44.

As the plots demonstrate, the trend, including inflection points (as described in

appendix B) and optimal collimation length assumes the predicted shape for both the

analytical calculations and theMCNP5 simulations. Further the trends, MCNP5 and

calculated display identical optimal collimation length, indicating that the contrast

generating mechanism is indeed a first scatter type phenomena. That is, the contrast is

generated by the attenuation differences as the flaw is traversed directly by the photons.

The calculated data and the MCNP5 acquired data, while displaying the same trend, do

not match up identically. This is expected. The discrepancy between the first scatter

analytical model and the MCNP5 data is attributed to two main factors. The first being

that while MCNP5 uses an impingent x-ray beam with the true energy distribution the









analytical calculations assume an appropriate mono-energetic photon beam. Thus all

properties and attenuation characteristics, which are treated as continuous in MCNP5, are

treated as a one group approximation in the calculations. The other reason is that

MCNP5 data is for total detected signal while the analytical model considers only 1st

scattered photons. Thus, as previously demonstrated and thus expected, the analytical

calculations display higher contrasts that the MCNP5 simulation because the analytical

model only considers 1st scattered photons which typically have the highest contrast.

Also of note is that for Figure 4.38 and Figures 4.42, 4.43 and 4.44 (Figure 4.39 is

excluded because it is used only to demonstrate the trend and the exact experimental data

points have not been reproduced or confirmed in this study) the optimal overall contrast

is achieved at the same collimation configuration dictated by the optimal first scatter

configuration. This is true even for situations when the true first scatter contribution is

below 10% of the signal (e.g. for data in Figure 4.44 1st scatter component at the

optimum collimation length represents only 1.2% of the total signal). This supports the

two main premises of the proposed photon transport mechanism. The first scatter

component of a signal has the highest relative contrast and thus contributes a larger than

expected (from overall signal fraction) contrast contribution. The significant higher order

components of the signal behave essentially the same as first scatter components for the

purposes of providing detectable image contrast.

A more comprehensive analytic explanation of this trend in contrast versus

collimation, as handled by the transport model is included as Appendix C.









Image pixel shifts and shadows

Pixel shifts observed in the images between various detectors serve as an important

proof of principle tool for validating the transport model. A clear angle and thus a

projection is realized between the detector and the investigated object. The model

predicts that, much like optics, a relative shift will be observed between images taken

from different view points, i.e. different detectors. The phenomena responsible for both

pixel shifts and shadows, or more appropriately, pixel shifts of shadows, are

demonstrated in the following schematic labeled Figure 4.45. As the illustration indicates

there are four important variables which determine the pixel shift observed by a particular

detector. These include the thickness of the sample (or the penetration depth of the

effective impingent photon beam), the depth of the flaw within the sample, the position of

the detector (both x and y components) and the collimation extension past the detector

surface. In Figure 4.45 these variables are labeled as t, d, R, and c, respectively, and the

shift in shadow relative to the true flaw location is given as s.










s/(t-d)=R/(h+t)


d=t-s(h+t)/R


t


Figure 4.45 Geometrical considerations for shadow pixel shifting relative to detector
position.

The following set of images, Figures 4.46 and 4.47, demonstrate, according to the

above schematic, the shadow pixel shifts as observed by the various detectors. In each

image, the arrow point in the direction that the effective shadow is cast. This is away

from the position of the detector. That is, the shadow image is formed when the flaw is

between the incident beam and the detector, thus the image appears on the opposite side

of the flaw as the detector. Thus the arrow points from the detector position towards the

beam.










Detector 2


Figure 4.46 Bright shadow cast by void in Al seen by detector in lower right hand
corner.


Detector 3


-649903


0 10 :'0 4Cb 60 0 C: 9r 100 110io 10


Figure 4.47 Bright shadow cast by void in Al seen by detector in lower left hand corner.

Signal intensity

Total signal intensity in the RSD imaging modality is observed to decrease


markedly with increased collimation. Considering the exponential attenuation of the


incident signal this should be expected based upon the same arguments presented above









in the collimation extension section. As collimation is increased and the critical

scattering reference plane is consequently moved deeper into the sample, more of the

scattered signal is being discriminated against. As the scattered signal varies as

interaction rates which vary with incident photon flux, the characteristic exponential

attenuation is also observed in the signal intensity decrease with collimation extension.

Applications and Limitations

To date, RSD imaging modalities have been successfully applied to various

materials including, SOFI foam, aluminum, plastics, landmines buried in soil, concrete,

drywall, and reactor insulation. Each of these materials possesses unique interaction and

attenuation properties and presents specific important flaw types. Therefore, each

specific imaging task represents its own unique problems and inherent limitations.














CHAPTER 5
BACKSCATTER FIELD DISTRIBUTION AND DETECTOR PLACEMENT


Backscattered X-ray Signal Profile

The backscattered photon field is distributed according to the appropriate

differential scattering cross-section. Thus, the backscattered photon distribution should

be symmetric about the axis of the impingent beam. The signal profile should also be

peaked at 180 degrees (direct backscatter) and sinusoidally taper off towards a minimum

at 90 degrees. The sharpness of the peak and speed of the tapering are functions of the

incident photon energy, with the scatter profile becoming more isotropic as energy is

decreased. The unmodified (i.e. without consideration of atomic form factor) Klien-

Nishiena cross-section for a 55 keV photon beam is presented below (Figure 5.1) as a

function of scattering angle. The backscattered photon flux through a plane parallel with

the surface of foam target (w/ Al substrate) is presented below as Figure 5.2. This plot is

taken from data acquired via MCNP5 simulations. The foam target in Figure 5.2 is eight

inches thick with a one inch aluminum substrate. The tally plane is 1 cm above the

surface of the foam. The impingent photon beam is 75 keV peak spectra with a 0.5 cm

diameter. In Figure 5.2, as expected from the shape of the differential scattering cross-

section, the highest backscattered photon flux is directly above the source (180 degrees

scattering angle). This plot, however, represents the flux across a surface and is thus

more exaggerated than the Klien-Nishiena cross-section because the flux is dependent

upon both the relative fraction of photons scattered into a particular solid angle as well as









the relative orientation of the reference plane to the incident photons. Since photons

scattered at smaller angles (to the horizon), i.e. closer to 90 degrees than to 180 degrees

across the reference plane at more severe angles (further away from perpendicular), these

surfaces have less effective area and thus will display lower fluxes than the scattering

cross-sections would otherwise dictate. Additionally, since the tally is taken across a

plane rather than a sphere, the photons reaching the outer mesh voxels are geometrically

attenuated by the 1/rA2 law and thus the fluxes are further reduced.


Odeg ttr
V' ward scatte-r


180 deg
backscatter


90 deg
scatter


U1111111111111 11111111111111111
0 0.5 1 1.5 2 2.5 3
ibeta relative to lcident beam

Figure 5.1 Klien-Nishiena differential scattering cross-section for 55 keV photon


















0 00 MEN 0




x




Figure 5.2 Backscattered photon flux across a plane parallel to SOFI sample surface

As the transport model indicates (chapter 4), the primary cause of a detected signal

differential (i.e. image contrast) is a change in the scattering/attenuation characteristics

of the photon field as a result of a 'flaw' in the imaged target material. Since to a first

approximation, the presence or absence of a flaw changes only relative interaction rates

within a specific volume (about the flaw) and has no effect on the directional distribution

of the scattered photon field, we would expect that the same signal (i.e. same relative

contrast) could be detected from any particular solid angle component of the

backscattered photon field. In fact, for a perfectly symmetrical flaw in a uniform

medium, this is indeed the case as illustrated in Figure 3. Figure 3 is a plot of relative

differences (% contrast) in photon fluxes across a meshed tally plane oriented parallel to

a target sample surface. The target, as in Figure 5.1, is a SOFI foam material on an

aluminum substrate. The data for the plot is taken from 2 MCNP5 simulations. The first

simulation utilizes a 15 x 15 cm mesh tally taken over an eight inch thick sample of SOFI

on a one inch aluminum substrate. The second simulation is identical except that a one










cm diameter spherical void flaw is placed two and a half cm below the sample surface.

In both cases the incident photon beam is 55 keV peak and the mesh tally plane is located

1 cm above the surface of the foam. The backscattered photon flux though the mesh tally

for simulation 1 is shown above as Figure 5.2 and the backscattered flux for simulation 2

(with the flaw) would have an identical relative distribution. The percent difference

between these two simulations is presented below as Figure 5.3. As the plot shows,

relative differences in the signals are essentially the same across all voxel elements of the

mesh plane. The high and low peaks observed towards the outer edges of the plot are

statistically insignificant and are a result of the relatively few number of photons crossing

the outer mesh elements.








0.2r" 1


-025C h

-1
-15 -10 -5 1 51 0r I




Figure 5.3 Percent difference in signal due to void flaw in SOFI as a function of scatter
field component. Notice that the contrast is evenly distributed across the
entire scatter field.

Detector Placement Considerations

The ideal scenario described above in which the same relative contrast is obtained

regardless of scatter component selected does not indicate that detector placement and

orientation are irrelevant for image acquisition. On the contrary, in real situations,

detector orientation and appropriate selection of scatter-field components are of