MECHANISTIC CHARACTERIZATION OF RESILIENT MODULI FOR UNBOUND PAVEMENT LAYER MATERIALS Except where reference is made to the work of others, the work described in this thesis is my own or was done in collaboration with my advisory committee. This thesis does not include proprietary or classified information. Adam Joel Taylor Certificate of Approval: Rod E. Turochy David H. Timm, Chair Associate Professor Associate Professor Civil Engineering Civil Engineering Randy C. West George T. Flowers Director Dean NCAT Graduate School MECHANISTIC CHARACTERIZATION OF RESILIENT MODULI FOR UNBOUND PAVEMENT LAYER MATERIALS Adam Joel Taylor A Thesis Submitted to the Graduate Faculty of Auburn University in Partial Fulfillment of the Requirements for the Degree of Master of Science Auburn, Alabama August 9, 2008 iii MECHANISTIC CHARACTERIZATION OF RESILIENT MODULI FOR UNBOUND PAVEMENT LAYER MATERIALS Adam Joel Taylor Permission is granted to Auburn University to make copies of this thesis at its discretion, upon request of individuals or institutions at their expense. The author reserves all publication rights. Signature of Author Date of Graduation iv VITA Adam Joel Taylor, son of Junior and Patsy Taylor, was born on July 7, 1983 in LaGrange, Georgia. He graduated from Wadley High School in Wadley, Alabama as Valedictorian in 2001. He enrolled in Auburn University to pursue a major in Civil Engineering in the Fall of 2001. In May of 2006, he graduated summa cum laude from Auburn University with a Bachelor of Civil Engineering. As an undergraduate, he also entered into the cooperative eduation program and worked as a co-op student for the National Center for Asphalt Technology (NCAT). He entered into graduate school at Auburn University in May, 2006 where he specialized in the pavements and materials field of Civil Engineering while remaining involved with NCAT through performing research at the NCAT Test Track. v THESIS ABSTRACT MECHANISTIC CHARACTERIZATION OF RESILIENT MODULI FOR UNBOUND PAVEMENT LAYER MATERIALS Adam Joel Taylor Master of Science, August 9, 2008 (B.C.E. Auburn University, 2006) 299 Pages Typed Directed by Dr. David H. Timm In recent years, there has been an industry shift in pavement design methodology from purely empirically based methods (e.g. The AASHTO Guide for the Design of Pavement Structures) to design methods that combine both mechanistic and empirical elements (e.g. the new Mechanistic-Empirical Pavement Design Guide). One of the critical inputs for accurate mechanistic-empirical (M-E) pavement design is accurate characterization of the stiffness of the unbound pavement material layers. This stiffness is quantified as resilient modulus, and this value can be determined either through laboratory testing with the triaxial apparatus or though non-destructive testing in the field with the falling weight deflectometer (FWD). Resilient modulus is typically vi expressed as a function of unbound material stress-state using a non-linear stress- sensitivity model. For this thesis, five unbound materials utilized in the construction of eleven instrumented pavement test sections at the NCAT Test Track were characterized through both triaxial and FWD testing. Additionally, multiple non-linear stress-sensitivity models were evaluated for both testing methods with each material to determine which model provided the best fit to the respective data sets. For the materials tested in the laboratory, stress-sensitivity models that account for both the effects of shear and confining pressure provided a better fit to the triaxial data than models that only accounted for the effects of only one of these variables. The same held true for base layer materials tested with the FWD at the Test Track. Generally, poor agreement was seen between the stress-sensitivity models and moduli generated by different methods for the base layer materials. Reasonable agreement between the data sets was seen for the subgrade material utilized at the Test Track. As a result of this study, the models generated in the laboratory for the various unbound materials were deemed suitable for Level 1 MEPDG pavement design. FWD testing provided a good measure of pavement material variability in the field. The field- calibrated constitutive equations were shown to reasonably predict the backcalculated unbound material moduli at the Test Track. However, it is recommended that these models be made more robust through the addition of additional deflection data and validation at other testing sites containing the specified unbound materials. vii ACKNOWLEDGMENTS The author would like to thank Dr. David H. Timm for his guidance and support in the analysis and writing of this thesis. The National Center for Asphalt Technology Test Track staff, particularly Dr. Buzz Powell, were instrumental in the collection of quality data for use in this project. Burns, Cooley, Dennis, Inc. deserves thanks for conducting the laboratory triaxial testing utilized for this thesis. The author of this paper would like to thank the state departments of transportation from Alabama, Florida, Indiana, Missouri and Oklahoma for their support and cooperation with this research. The Federal Highway Administration also deserves recognition for their support and cooperation. viii STYLE MANUAL AND SOFTWARE USED Style manual used: MLA Handbook for Writers of Research Papers (5 th edition) Computer software used: Microsoft Word, Microsoft Excel, Microsoft Access, EVERCALC v5.0, AUDEF, WESLEA, DATAFIT ix TABLE OF CONTENTS LIST OF TABLES........................................................................................................... xiii LIST OF FIGURES ..........................................................................................................xvi CHAPTER 1 - INTRODUCTION.......................................................................................1 BACKGROUND ..........................................................................................................1 OBJECTIVES OF THESIS .............................................................................................9 SCOPE OF THESIS ......................................................................................................9 ORGANIZATION OF THESIS......................................................................................11 CHAPTER 2 ? LITERATURE REVIEW.........................................................................12 INTRODUCTION .......................................................................................................12 LABORATORY TESTING...........................................................................................15 Sampling of Materials for Laboratory Resilient Modulus Testing............17 Target Moisture Contents and Densities for Laboratory Samples.............19 Stress-States Utilized in Laboratory Testing .............................................20 FIELD MODULUS CHARACTERIZATION ...................................................................21 Field FWD Testing Programs....................................................................24 Load Levels Used for FWD Testing..........................................................25 Spatial Variability......................................................................................26 FWD Deflection Sensor Spacing...............................................................27 Seasonal Variation.....................................................................................29 x Quality Control of Backcalculated Data....................................................32 Software Used in Backcalculation.............................................................35 Determining the Optimal Pavement Cross-Section for Backcalculation..........................................................................................37 Typical Inputs for Backcalculation............................................................44 COMMONLY USED STRESS-SENSITIVITY MODELS ..................................................46 Single Variable Non-Linear Stress-Sensitivity Models.............................47 Multi-Variable Non-Linear Stress-Sensitivity Models..............................51 COMPARISON OF LABORATORY AND FIELD MODULUS VALUES FOR UNBOUND MATERIALS ............................................................................................................56 SUMMARY OF KEY FINDINGS IN LITERATURE.........................................................64 CHAPTER 3 - TESTING FACILITY AND PROGRAM...............................................67 INTRODUCTION .......................................................................................................67 Trafficking .................................................................................................68 THE STRUCTURAL EXPERIMENT .............................................................................69 Unbound Materials used in the Structural Study.......................................71 Section Construction..................................................................................75 Structural Instrumentation .........................................................................83 LABORATORY TESTING...........................................................................................88 FWD TESTING PROGRAM .......................................................................................89 FWD ON GAUGE TESTING ......................................................................................91 SUMMARY REGARDING THE TESTING FACILITY AND TESTING PROGRAM..............93 CHAPTER 4 ? LABORATORY TESTING .....................................................................94 INTRODUCTION .......................................................................................................94 xi Testing Procedure ......................................................................................96 Sample Preparation..................................................................................101 MODELING NON-LINEAR STRESS-SENSITIVITY BEHAVIOR OF TRIAXIAL DATA .................................................................................................................102 UNBOUND MATERIAL COMPARISON AND STRESS-SENSITIVITY............................110 SUMMARY OF KEY FINDINGS FROM LABORATORY TESTING.................................114 CHAPTER 5 ? BACKCALCULATION CROSS-SECTION DETERMINATION.......115 INTRODUCTION .....................................................................................................115 SOFTWARE UTILIZED............................................................................................116 CROSS-SECTION SELECTION METHODOLOGY.......................................................120 N1-N2 CROSS-SECTION INVESTIGATION ..............................................................124 N3-N7 CROSS-SECTION INVESTIGATION ..............................................................143 N8-N9 CROSS-SECTION INVESTIGATION ..............................................................158 N10 CROSS-SECTION INVESTIGATION ..................................................................171 S11 CROSS-SECTION INVESTIGATION ...................................................................188 CONCLUDING REMARKS ON BACKCALCULATION CROSS-SECTION DETERMINATION...................................................................................................202 CHAPTER 6 ? FIELD CHARACTERIZATION OF UNBOUND MATERIALS ........204 INTRODUCTION .....................................................................................................204 MODEL CALIBRATION METHODOLOGY ................................................................205 ADJUSTING STRESSES FOR OVERBURDEN .............................................................208 FIELD-CALIBRATED STRESS-SENSITIVITY MODELS..............................................211 SPATIAL VARIABILITY IN DEFLECTION DATA.......................................................235 MODEL VALIDATION ............................................................................................237 xii HMA CHARACTERIZATION...................................................................................243 SUMMARY OF RECOMMENDED MODELS AND BACKCALCULATED MODULI..........246 KEY FINDINGS FROM FIELD MODULUS CHARACTERIZATION................................248 CHAPTER 7 ? COMPARISON OF LABORATORY AND FIELD-DETERMINED RESILIENT MODULI ....................................................................................................250 INTRODUCTION .....................................................................................................250 METHODOLOGY ....................................................................................................251 LIMEROCK BASE (SECTIONS N1 AND N2).............................................................252 GRANITE BASE (SECTIONS N3 AND N4) ...............................................................256 TRACK SOIL BASE (SECTIONS N8 AND N9) ..........................................................261 TRACK SOIL SUBGRADE .......................................................................................262 POTENTIAL CAUSES OF DISAGREEMENT BETWEEN LAB AND FIELD DATA ...........263 KEY FINDINGS FROM COMPARISON OF LABORATORY AND FIELD RESILIENT MODULI...............................................................................................265 CHAPTER 8 ? CONCLUSIONS AND RECOMMENDATIONS..................................267 SUMMARY OF KEY FINDINGS................................................................................267 Key Findings from Laboratory Testing ...................................................267 Key Findings from Backcalculation Cross-Section Determination.........268 Key Findings from FWD Unbound Material Characterization ...............269 Key Findings from Comparison of Laboratory and Field Resilient Moduli......................................................................................................270 RECOMMENDATIONS.............................................................................................271 REFERENCES ................................................................................................................273 xiii LIST OF TABLES Table 2-1: Typical Values for Bulk Model Coefficients from AASHTO Design Guide (AASHTO, 1993) ..............................................................................................................48 Table 2-2: Average Regression Coefficients Using Equation 2-12 and Equation 2-14 for Various Soil Types within the LTPP Laboratory Resilient Modulus Database (after Von Quintus and Killingsworth, 1998) .....................................................................................53 Table 2-3: Statistical Summary of MEPDG Models Calibrated to the LTPP Database (after Yau and Von Quintus, 2002) ...................................................................................55 Table 3-1: Axle Weights for Trucking Fleet at NCAT Test Track ...................................69 Table 3-2: Axle Spacings (from Center of Steer Configuration (ft)) for Trucking Fleet at NCAT Test Track ..............................................................................................................69 Table 3-3: Milled Depth for the Structural Sections .........................................................71 Table 3-4: Unbound Material Gradations..........................................................................73 Table 3-5: In-Situ Unbound Material Wet Densities for the Structural Study..................74 Table 3-6: In-Situ Unbound Material Moisture Contents for the Structural Study..................................................................................................................................74 Table 3-7: HMA Densities for Overburden Calculation ...................................................83 Table 3-8: Thermistor Depths for the Structural Sections.................................................88 Table 3-9: Longitudinal Random Locations used for FWD Testing (from end of test section, ft) ..........................................................................................................................90 Table 3-10: FWD Sensor Spacings (Sensor 1 is beneath the load plate) ..........................91 Table 3-11: FWD Drop Heights and Corresponding Loadings.........................................91 Table 3-12: Matrix of Gauges Tested for FWD on Gauge Testing...................................92 xiv Table 4-1: Testing Parameters for Unbound Materials ...................................................102 Table 4-2: Summary of Bulk Stress-Sensitivity Models by Material Type ....................106 Table 4-3: Summary of Deviatoric Stress-Sensitivity Models by Material Type ...........107 Table 4-4: Summary of Universal Stress-Sensitivity Models by Material Type.............108 Table 4-5: Summary of MEPDG Stress-Sensitivity Models by Material Type ..............109 Table 4-6: Representative Stress-States used for Modulus Normalization .....................111 Table 5-1: FWD Testing Dates Used in Cross-Section Determination...........................122 Table 5-2: (1)-5layer Cross-Section Modulus Behavior (Station N1-12, 10/30/06).......128 Table 5-3: (1)-5layer Cross-Section Modulus Behavior (Station N4-12, 08/20/07).......146 Table 5-4: (1)-5layer Cross-Section Modulus Behavior (Station N8-8, 04/23/07).........161 Table 5-5: (1)-5layer Cross-Section Modulus Behavior (Station N10-5, 01/29/07)................................................................................................176 Table 5-6: (2)-4layer Cross-Section Modulus Behavior (Station N10-11, 04/23/07)..............................................................................................176 Table 5-7: (1)-5layer Cross-Section Modulus Behavior (Station S11-3, 11/27/06)........191 Table 6-1: FWD Testing Dates Used in Stress-Sensitivity Model Calibration...............206 Table 6-2: Field-Calibrated Bulk Models (Base Layer, by Section)...............................213 Table 6-3: Field-Calibrated Bulk Models (Base Layer, by Material) .............................213 Table 6-4: Field-Calibrated Deviatoric Models (Base Layer, by Section)......................214 Table 6-5: Field-Calibrated Deviatoric Models (Base Layer, by Material) ....................214 Table 6-6: Field-Calibrated MEPDG Models (Base Layer, by Section).........................215 Table 6-7: Field-Calibrated MEPDG Models (Base Layer, by Material) .......................216 Table 6-8: Field-Calibrated Universal Models (Base Layer, by Section) .......................217 Table 6-9: Field-Calibrated Universal Models (Base Layer, by Material)......................217 xv Table 6-10: Field-Calibrated Bulk Models (Subgrade Layer, by Section)......................226 Table 6-11: Field-Calibrated Bulk Models (Subgrade Layer, by Material) ....................226 Table 6-12: Field-Calibrated Deviatoric Models (Subgrade Layer, by Section).............228 Table 6-13: Field-Calibrated Deviatoric Models (Subgrade Layer, by Material) ...........228 Table 6-14: Field-Calibrated MEPDG Models (Subgrade Layer, by Section) ...............229 Table 6-15: Field-Calibrated MEPDG Models (Subgrade Layer, by Material)..............229 Table 6-16: Field-Calibrated Universal Models (Subgrade Layer, by Section)..............230 Table 6-17: Field-Calibrated Universal Models (Subgrade Layer, by Material) ............231 Table 6-18: Spatial Variability in Field-Calibrated Universal Stress-Sensitivity Models for the Limerock Base Material.......................................................................................236 Table 6-19: FWD Testing Dates used for Model Validation ..........................................238 Table 6-20: Results of HMA Characterization for Structural Sections...........................244 xvi LIST OF FIGURES Figure 1-1: Unbound Material Strains under Repeated Traffic Loading (Huang, 2004) ....3 Figure 1-2: AASHTO Structural Coefficient Correlation Nomograph (AASHTO, 1993) .7 Figure 2-1: M-E Design Flowchart (after Monismith 1992).............................................14 Figure 2-2: Comparison of Resilient Modulus values for Laboratory Samples with Different Moisture Contents (Ping et al., 2001).\ ..............................................................19 Figure 2-3: Comparison of Loading Stress Sequences for Granular Materials (Andrei et al. 2004) .............................................................................................................................21 Figure 2-4: Dynatest 8000 Falling Weight Deflectometer ................................................23 Figure 2-5: Impact of Zone of Influence on Pavement Deflections ..................................28 Figure 2-6: Determination of Bedrock Depth with Deflection Measurements. ................29 Figure 2-7: Seasonal Base Moduli Behavior in Alabama (Parker and Elton 1990)..........30 Figure 2-8: Seasonal Base Moduli Variation at Estillene, Texas LTPP Seasonal Monitoring Site (Briggs and Lukanen 2000).....................................................................32 Figure 2-9: Examples of Quality Deflection Basins (Von Quintus and Killingsworth, 1998)..................................................................................................................................34 Figure 2-10: Examples of Problematic Deflection Basins (Von Quintus and Killingsworth, 1998)..........................................................................................................34 Figure 2-11: Pavement Model used by Parker and Elton in Alabama (Parker and Elton, 1990)..................................................................................................................................40 Figure 2-12: Pavement Model used by Seeds et al. at WESTRACK (Seeds et al., 2000).............................................................................................................41 Figure 2-13: Comparison the Bulk (left) and Universal (right) Model Modulus Predictions for a Given Set of Laboratory Data (Uzan, 1985) ..........................................52 xvii Figure 2-14: Comparison of Laboratory and FWD Moduli in Texas (Nazarian et al., 1998)..................................................................................................................................59 Figure 2-15: Comparison of Laboratory and FWD Moduli for Base Layer Materials at WESTRACK (Seeds et al., 2000)......................................................................................60 Figure 2-16: Comparison of Laboratory and FWD Moduli for Fill and Subgrade Layer Materials at WESTRACK (Seeds et al. 2000)...................................................................61 Figure 2-17: Comparison of Laboratory and FWD Moduli for Base Layer Materials for the Rufus, Oregon Project (Zhou 2000) ............................................................................62 Figure 2-18: Comparison of Laboratory and FWD Moduli for Base Layer Materials for the Rufus, Oregon Project (Zhou 2000) ............................................................................62 Figure 2-19: Comparison of Laboratory and FWD Moduli for Unbound Materials used in Florida (Ping et al. 2001)...............................................................................................64 Figure 3-1: Aerial Photo of NCAT Test Track..................................................................68 Figure 3-2: As-Built Cross-Sections for the Structural Sections at the Test Track...........70 Figure 3-3: Construction Diagram for Section N1 ............................................................76 Figure 3-4: Construction Diagram for Section N2 ............................................................76 Figure 3-5: Construction Diagram for 2003 Structural Sections.......................................78 Figure 3-6: Construction Diagram for Section N8 ............................................................79 Figure 3-7: Construction Diagram for Section N9 ............................................................80 Figure 3-8: Construction Diagram for Section N10 ..........................................................81 Figure 3-9: Construction Diagram for Section S11...........................................................82 Figure 3-10: Average HMA and Base As-Built Thicknesses for All Structural Sections.83 Figure 3-11: Layout of Gauge Array for a Typical Structural Section (after Willis and Timm, 2008) ......................................................................................................................84 Figure 3-12: Gauge Array Prior to Construction...............................................................85 Figure 3-13: CTL ASG-152 Asphalt Strain Gauge Employed at the NCAT Test Track..86 xviii Figure 3-14: Geokon Model 3500 Earth Pressure Cell at the Test Track .........................87 Figure 3-15: Typical Sub-Surface Thermistor Bundle Employed at the NCAT Test Track (after Priest and Timm ?Fatigue,? 2006) ...........................................................................88 Figure 3-16: Random Location Layout for Section N1.....................................................90 Figure 4-1: Triaxial Testing Apparatus (NCHRP, 2004) ..................................................95 Figure 4-2: NCHRP 1-28A Specimen Preparation Flowchart (NCHRP, 2004) ...............97 Figure 4-3: Diagram of Haversine Load Pulse Used in NCHRP 1-28A (NCHRP, 2004) .................................................................................................................99 Figure 4-4: Resilient Modulus versus Bulk Stress ? Limerock Base ? Sample 1 ...........103 Figure 4-5: Resilient Modulus vs. Deviatoric Stress ? Limerock ? Sample 1 ................104 Figure 4-6: Resilient Modulus versus Octahedral Shear Stress ? Limerock ? Sample 1 104 Figure 4-7: R2 Comparison for All Stress-Modulus Models by Material Type..............110 Figure 4-8: Unbound Material Moduli at Representative Stress-States ? MEPDG Model................................................................................................................112 Figure 4-9: Unbound Material Moduli at Representative Stress-States ? Universal Model ..............................................................................................................113 Figure 5-1: Thickness and Material Summary of the Structural Sections at Test Track.120 Figure 5-2: Cross-Section Trials Investigation Flowchart...............................................124 Figure 5-3: Trial Backcalculation Cross-Sections (Sections N1 and N2) .......................126 Figure 5-4: RMS Error Cumulative Distribution Functions (CDF) for the Trial Backcalculation Cross-Sections in Sections N1 and N2 .................................................127 Figure 5-5: Fill Layer Modulus Behavior (Layer 3) (Section N1, All Testing Stations and Dates, 9 kip drops)...........................................................................................................129 Figure 5-6: RMS Error Cumulative Distribution Functions (CDF) for the 3-Layer Trial Backcalculation Cross-Sections in Sections N1 and N2 .................................................130 Figure 5-7: Base Layer Modulus Behavior (Layer 2) (Section N1, (1)-3layer, All Testing Stations and Dates, 9 kip drops) ......................................................................................132 xix Figure 5-8: Base Layer Modulus Behavior (Layer 2) (Section N1, (2)-3layer, All Testing Stations and Dates, 9 kip drops) ......................................................................................133 Figure 5-9: Base Layer Modulus Behavior (Layer 2) (Section N2, (1)-3layer, All Testing Stations and Dates, 9 kip drops) ......................................................................................133 Figure 5-10: Base Layer Modulus Behavior (Layer 2) (Section N2, (2)-3layer, All Testing Stations and Dates, 9 kip drops) .........................................................................134 Figure 5-11: Subgrade Layer Modulus Behavior (Layer 3) (Section N1, (1)-3layer, All Testing Stations and Dates, 9 kip drops) .........................................................................135 Figure 5-12: Subgrade Layer Modulus Behavior (Layer 3) (Section N1, (2)-3layer, All Testing Stations and Dates, 9 kip drops) .........................................................................135 Figure 5-13: Subgrade Layer Modulus Behavior (Layer 3) (Section N2, (1)-3layer, All Testing Stations and Dates, 9 kip drops) .........................................................................136 Figure 5-14: Subgrade Layer Modulus Behavior (Layer 3) (Section N2, (2)-3layer, All Testing Stations and Dates, 9 kip drops) .........................................................................136 Figure 5-15: Backcalculated Modulus Comparison between Trial Cross-Sections (1)-3layer and (2)-3layer ? Sections N1 and N2 ? All Layers ........................................138 Figure 5-16: Measured versus Predicted Strains for Sections N1 and N2 ? (1)-3layer .........................................................................................................................139 Figure 5-17: Measured versus Predicted Strains for Sections N1 and N2 ? (2)-3layer .........................................................................................................................140 Figure 5-18: Measured versus Predicted Base Pressures (N1/N2, (1)-3layer)................142 Figure 5-19: Measured versus Predicted Subgrade Pressures (N1/N2, (1)-3layer..........142 Figure 5-20: Trial Backcalculation Cross-Sections for Sections N3-N7.........................144 Figure 5-21: RMS Error Cumulative Distribution Functions (CDF) for the Trial Backcalculation Cross-Sections in Sections N3-N7........................................................145 Figure 5-22: Fill Layer Modulus Behavior (Layer 3) (Section N3-N7, IWP Testing Stations, All Dates, 9 kip drops, (2)-4layer)....................................................................147 Figure 5-23: RMS Error Cumulative Distribution Functions (CDF) for the 3-Layer Trial Backcalculation Cross-Sections in Sections N3-N7........................................................148 xx Figure 5-24: Base Layer Modulus Behavior (Layer 2) (Sections N3-N7, (1)-3layer, All Testing Dates, All IWP Testing Locations, 9 kip drops).................................................150 Figure 5-25: Base Layer Modulus Behavior (Layer 2) (Sections N3-N7, (2)-3layer, All Testing Dates, All IWP Testing Locations, 9 kip drops).................................................150 Figure 5-26: Subgrade Layer Modulus Behavior (Layer 3) (Sections N3-N7, (1)-3layer, All Testing Dates, All IWP Testing Locations, 9 kip drops)...........................................151 Figure 5-27: Subgrade Layer Modulus Behavior (Layer 3) (Sections N3-N7, (2)-3layer, All Testing Dates, All IWP Testing Locations, 9 kip drops)...........................................152 Figure 5-28: Backcalculated Modulus Comparison between Trial Cross-Sections (1)-3layer and (2)-3layer ? Sections N3 through N7 ? All Layers..................................153 Figure 5-29: Measured versus Predicted Strains for Sections N3-N7 ? (1)-3layer.........154 Figure 5-30: Measured versus Predicted Strains for Sections N3-N7 ? (2)-3layer.........155 Figure 5-31: Measured versus Predicted Base Pressures for Sections N3-N7 ? (1)-3layer .........................................................................................................................157 Figure 5-32: Measured versus Predicted Fill Pressures for Sections N3-N7 ? (1)-3layer .........................................................................................................................157 Figure 5-33: Trial Backcalculation Cross-Sections for Sections N8 and N9 ..................159 Figure 5-34: RMS Error Cumulative Distribution Functions (CDF) for the Trial Backcalculation Cross-Sections in Sections N8-N9........................................................160 Figure 5-35: Seale Layer Modulus Behavior (Layer 3) (Sections N8-N9, All FWD Testing Stations and Dates, 9 kip drops, (1)-4layer) .......................................................162 Figure 5-36: RMS Error Cumulative Distribution Functions (CDF) for the 3-Layer Trial Backcalculation Cross-Sections in Sections N8 and N9 .................................................163 Figure 5-37: Base Layer Modulus Behavior (Layer 2) (Sections N8 and N9, (2)-3layer, All Testing Dates, All Testing Locations, 9 kip drops)...................................................164 Figure 5-38: Base Layer Modulus Behavior (Layer 2) (Sections N8 and N9, (1)-3layer, All Testing Dates, All Testing Locations, 9 kip drops)...................................................165 Figure 5-39: Subgrade Layer Modulus Behavior (Layer 3) (Sections N8 and N9, (2)- 3layer, All Testing Dates, All Testing Locations, 9 kip drops).......................................166 xxi Figure 5-40: Subgrade Layer Modulus Behavior (Layer 3) (Sections N8 and N9, (1)- 3layer, All Testing Dates, All Testing Locations, 9 kip drops).......................................166 Figure 5-41: Sections N8 and N9 ? Cross-Section Modulus Comparison ......................167 Figure 5-42: Measured versus Predicted Strains ? N8/N9 ? (2)-3layer ..........................168 Figure 5-43: Measured versus Predicted Strains ? N8/N9 ? (1)-3layer ..........................169 Figure 5-44: Measured versus Predicted Base Pressures ? Sections N8 and N9 ? (2)-3layer .........................................................................................................................170 Figure 5-45: Measured versus Predicted Subgrade Pressures (Seale) ? Sections N8 and N9 ? (2)-3layer ................................................................................................................171 Figure 5-46: Trial Backcalculation Cross-Sections for Section N10 ..............................173 Figure 5-47: RMS Error Cumulative Distribution Functions (CDF) for the Trial Backcalculation Cross-Sections in Section N10 (3-layer only) ......................................174 Figure 5-48: RMS Error Cumulative Distribution Functions (CDF) for the Trial Backcalculation Cross-Sections in Section N10 (4-layer and 5-layer only) ...................174 Figure 5-49: RMS Error Cumulative Distribution Functions (CDF) for the 3-Layer Trial Backcalculation Cross-Sections in Section N10..............................................................178 Figure 5-50: Base Layer Modulus Behavior (Layer 2) (Section N10, (2)-3layer, All Testing Dates, All Testing Locations, 9 kip drops).........................................................179 Figure 5-51: Base Layer Modulus Behavior (Layer 2) (Section N10, (3)-3layer, All Testing Dates, All Testing Locations, 9 kip drops).........................................................180 Figure 5-52: Fill and Subgrade Layer Modulus Behavior (Layer 3) (Section N10, (2)- 3layer, All Testing Dates, All Testing Locations, 9 kip drops).......................................181 Figure 5-53: Fill and Subgrade Layer Modulus Behavior (Layer 3) (Section N10, (3)- 3layer, All Testing Dates, All Testing Locations, 9 kip drops ........................................181 Figure 5-54: Section N10 ? Cross-Section Modulus Comparison ..................................182 Figure 5-55: Structural Section Rut Depths (7/16/07).....................................................183 Figure 5-56: Measured versus Predicted Strains ? N10 ? (2)-3layer ..............................184 Figure 5-57: Measured versus Predicted Strains ? N10 ? (3)-3layer ..............................185 xxii Figure 5-58: Measured versus Predicted Base Pressures (Surface of Base Layer)- N10 ? (2)-3layer ..............................................................................................................187 Figure 5-59: Measured versus Predicted Fill Pressures (Surface of Fill Layer) ? N10 ? (2)-3layer ..........................................................................................................187 Figure 5-60: S11 Trial Backcalculation Cross-Sections..................................................189 Figure 5-61: RMS Error Cumulative Distribution Functions (CDF) for the Trial Backcalculation Cross-Sections in Section S11 ..............................................................190 Figure 5-62: Fill Layer Modulus Behavior (Layer 3) (Section S11, All Testing Stations and Dates, 9 kip drops) ....................................................................................................192 Figure 5-63: RMS Error Cumulative Distribution Functions (CDF) for the 3-Layer Trial Backcalculation Cross-Sections in Section S11 ..............................................................193 Figure 5-64: Base Layer Modulus Behavior (Layer 2) (Section S11, (3)-3layer, All Testing Dates, All Testing Locations, 9 kip drops).........................................................194 Figure 5-65: Base Layer Modulus Behavior (Layer 2) (Section S11, (2)-3layer, All Testing Dates, All Testing Locations, 9 kip drops).........................................................195 Figure 5-66: Subgrade Layer Modulus Behavior (Layer 3) (Section S11, (3)-3layer, All Testing Dates, All Testing Locations, 9 kip drops ..........................................................196 Figure 5-67: Fill and Subgrade Layer Modulus Behavior (Layer 3) (Section S11, (2)- 3layer, All Testing Dates, All Testing Locations, 9 kip drops).......................................196 Figure 5-68: Section S11 ? Cross-Section Modulus Comparison...................................197 Figure 5-69: Measured versus Predicted Strain Behavior ? (3)-3layer ? Section S11 ......................................................................................................................199 Figure 5-70: Measured versus Predicted Strain Behavior ? (2)-3layer ? Section S11 ......................................................................................................................199 Figure 5-71: Measured versus Predicted Base Pressure Behavior ? (3)-3layer ? Section S11 ...................................................................................................................................201 Figure 5-72: Measured versus Predicted Fill Pressure Behavior ? (3)-3layer ? Section S11 ...................................................................................................................................202 Figure 5-73: Final Selected Backcalculation Cross-Sections..........................................203 xxiii Figure 6-1: Methodology for Adjusting Principal Stresses to Include Overburden under Loading............................................................................................................................210 Figure 6-2: Model R-Squared Summary (Base Layer, by Section).................................219 Figure 6-3: Model R-Squared Summary (Base Layer, by Material) ...............................219 Figure 6-4: Limerock Base Material Universal Model Calibration Data........................221 Figure 6-5: Granite Base Material Universal Model Calibration Data............................222 Figure 6-6: Track Soil Base Material Universal Model Calibration Data.......................223 Figure 6-7: Type 5 Base and Track Fill Material Universal Model Calibration Data...............................................................................................................224 Figure 6-8: Granite Base and Track Fill Material Deviatoric Model Calibration Data...225 Figure 6-9: Model R-Squared Summary (Subgrade Layer, by Section) .........................232 Figure 6-10: Model R-Squared Summary (Subgrade Layer, by Material)......................232 Figure 6-11: Track Subgrade Material MEPDG Model Calibration Data ......................233 Figure 6-12: Seale/Track Subgrade Material Universal Model Calibration Data ...........235 Figure 6-13: Limerock Base Universal Stress-Sensitivity Model Validation .................239 Figure 6-14: Limerock Base Universal Stress-Sensitivity Model Residuals...................239 Figure 6-15: Granite Base Universal Stress-Sensitivity Model Validation.....................240 Figure 6-16: Granite Base Universal Stress-Sensitivity Model Validation (up to 14,000 psi).............................................................................................................241 Figure 6-17: Granite Base and Track Fill Deviatoric Stress-Sensitivity Model Validation.............................................................................................................242 Figure 6-18: Seale and Track Subgrade Universal Stress-Sensitivity Model Validation243 Figure 6-19: Temperature-Stiffness Model Behavior for Section N3.............................245 Figure 7-1: Laboratory versus Field Resilient Moduli Comparison (Limerock Base Material, FWD Testing on 10/30/06) ..............................................................................253 xxiv Figure 7-2: Laboratory versus Field Resilient Moduli Comparison (Limerock Base Material, FWD Testing on 11/27/06) ..............................................................................254 Figure 7-3: Laboratory versus Field Resilient Moduli Comparison (Limerock Base Material, FWD Testing on 1/29/07) ................................................................................255 Figure 7-4: Laboratory versus Field Resilient Moduli Comparison (Limerock Base Material, FWD Testing on 8/20/07) ................................................................................256 Figure 7-5: Laboratory versus Field Resilient Moduli Comparison (Granite Base Material, FWD Testing on 11/27/06) ..............................................................................258 Figure 7-6: Laboratory versus Field Resilient Moduli Comparison (Granite Base Material, FWD Testing on 1/29/07) ................................................................................259 Figure 7-7: Laboratory versus Field Resilient Moduli Comparison (Granite Base Material, FWD Testing on 4/23/07) ................................................................................260 Figure 7-8: Laboratory versus Field Resilient Moduli Comparison (Granite Base Material, FWD Testing on 8/20/07) ................................................................................261 1 CHAPTER 1 INTRODUCTION BACKGROUND A fundamental requirement for a successfully constructed structure is a solid foundation. Asphalt pavements are no exception to this requirement. Flexible pavements are constructed in layers with higher quality materials at the surface of the pavements where the loading stresses are highest and lower quality materials deeper in the pavement structure where the loading stresses diminish with depth. The lowermost layers of a flexible pavement structure are often layers of unbound materials (e.g., granular bases or compacted fill) above the existing soil material. These materials are employed to protect the subgrade from stresses capable of causing rutting or pumping of fines (Huang, 2004). Unbound materials are a critical component of the pavement structure, and as such it is important to understand how these materials behave for design. Specifically, accurate stiffness (modulus) characterization of the unbound layers is critical for accurate pavement layer thickness design. For any mechanistically based pavement design, an accurate knowledge of the strains at critical locations in the pavement structure (such as the bottom of the HMA or surface of the subgrade) are necessary to give a quality prediction of the design life of that pavement. Given the fundamental relationship between stress and the multiple of modulus and strain (Hooke?s 2 Law), it is evident that accurate modulus characterization of the unbound pavement layers is necessary to accurately determine the critical strains in the structure based upon the stresses imparted by a design traffic loading. As such, accurate resilient modulus characterization is necessary to model the performance and life span of a given pavement structure. Often, the stiffness of the unbound materials is quantified as the resilient modulus. Equation 1-1 provides the general definition of resilient modulus, which is the ratio of deviatoric stress (from a triaxial compression test) to recoverable strain. This term is utilized since unbound material behavior is not completely elastic; these materials usually experience some permanent deformation after each load application. However, if the load is small compared to the strength of the material and is repeated often, the deformation under the subsequent load repetitions is almost completely recoverable (Huang, 2004). This results in the material behaving in a pseudo-elastic manner under repetitive traffic loads. Figure 1-1 illustrates the concept of recoverable strain under repetitive loading following a small amount of permanent deformation. r d r M ? ? = (1-1) 3 Figure 1-1: Unbound Material Strains under Repeated Traffic Loading (Huang, 2004). Unbound material resilient moduli can be determined either in the laboratory using the triaxial testing apparatus or in the field using a variety of non-destructive testing methods. The laboratory resilient modulus test calculates resilient modulus by directly measuring the load and deformation of the test specimen under repeated loadings that simulate the quick loads imparted on the pavement structure by moving traffic. One of the most common field tests for resilient modulus is done with a falling weight deflectometer (or FWD). This apparatus drops a large loading on the pavement structure that is representative of a heavy vehicle load. After this loading, an array of deflection sensors near the loading measures the resulting deflections of the pavement surface. From these deflection and loading data, the moduli of the pavement layers can be ?backcalculated? by using layered-elastic analyses that relate pavement layer modulus to deformation and loading. Both of these tests determine the resilient moduli of unbound paving material, but achieve that value by very different means. As such, 4 multiple studies have shown poor agreement between laboratory and field-measured resilient moduli (these studies will be discussed in greater detail in Chapter 2). The resilient moduli of unbound paving materials often exhibit non-linear stress- dependent behavior with varying stress-states within the material (Irwin, 2002). This behavior can either be stress-hardening (increasing stiffness with increasing stress) or stress-softening (decreasing stiffness with increasing stress) (Irwin, 2002). Research into unbound material performance through laboratory and field testing has yielded several constitutive relationships relating resilient modulus to stress-state. However, no one relationship is universally used to model resilient modulus data. These models contain a wide range of terms that quantify the stress-state of the unbound materials, such as: bulk stress (?), deviatoric stress (? d ), and octahedral shear stress (? oct ). Equations 1-1, 1-2, and 1-3, respectively, define the bulk stress, deviatoric stress, and octahedral shear stress terms as a function of the axial and confining stresses inducted during resilient modulus testing. Many of these models require non-linear regression analysis techniques to model the behavior of the material resilient modulus as a function of one or more terms that represent the material state of stress. Equations 1-4 and 1-5 are single-variable stress- sensitivity models that relate resilient modulus to bulk stress and deviatoric stress, respectively. These models are commonly specified based on whether the material is a coarse-grained (Equation 1-4) or fine-grained soil (Equation 1-5). Equations 1-6 and 1-7 are multi-variable stress-sensitivity models that model resilient modulus as a function of two stress terms. These models are more universal given they are not constrained to use on a particular soil type. In each model, the first term (the bulk stress) models 5 stress-sensitivity as a function of confining pressure while the second term (either deviatoric or octahedral shear stress) models stress-sensitivity due to shearing stresses. 321 ???? ++= (1-1) 31 ??? ?= d (1-2) (1-3) (1-4) (1-5) (1-6) (1-7) where: M r = Resilient Modulus P a = Atmospheric Pressure (14.7 psi) ? = Bulk Stress ? d = Deviatoric Stress ? oct = Octahedral Shear Stress ? 1 = Axial Stress ? 2 ,? 3 = Confining Stress k 1 , k 2 , k 3 = Regression Coefficients 3 2 1** 1 k a oct k a ar pp pkM ? ? ? ? ? ? + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? = ?? 32 ** 1 k a d k a ar pp pkM ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? = ?? 2 13 2 32 2 21 )()()( 3 1 ??????? ?+?+?= oct 2 * 1 k a r p kM ? ? ? ? ? ? ? ? = ? 2 * 1 k a d r p kM ? ? ? ? ? ? ? ? = ? 6 Recent trends in the pavement design industry have facilitated a need for better understanding of unbound material behavior and its use in pavement design and construction. The pavement industry is currently undergoing a shift from more empirically based design methods (such as that used in the AASHTO Guide for the Design of Pavement Structures (1993)) to mechanistic-empirical pavement design (such as that used by the new mechanistic-empirical pavement design guide (MEPDG) (2004)). These design philosophies utilize very different methods of unbound materials characterization. The current AASHTO design methodology (1993) reduces the resilient modulus value of the unbound paving layers to a single value, the structural coefficient. The structural coefficient was intended to represent the relative strength of the various construction materials used throughout the pavement structure. The AASHTO design methodology uses the resilient modulus to calculate a structural coefficient for a given unbound material. This coefficient is a point estimate that is supposedly representative of the overall layer behavior (under various loading, environmental conditions, etc.). The AASHTO design guide (1993) offers multiple options for obtaining the structural coefficient for unbound materials. The guide emphasizes that the most reliable value will come from laboratory or non-destructive resilient modulus testing in the field. However, the guide also offers multiple correlations to other, less expensive laboratory tests (i.e. California bearing ratio, R-value) that will result in the calculation of a structural coefficient that is given equal weight to one obtained through more precise testing. Figure 1-2 illustrates this concept with a correlation chart from the AASHTO 7 design guide that is used to generate a structural coefficient for granular base materials based on correlations to multiple laboratory unbound material tests. Figure 1-2: AASHTO Structural Coefficient Correlation Nomograph (AASHTO, 1993). The new MEPDG offers a hierarchical approach to the quality of pavement design inputs, with different ?levels? of design input accuracy available based on the needs of a given project. There are three levels of accuracy available for unbound material resilient modulus characterization. The lowest level of accuracy (Level 3) consists merely of a designer estimate or tabulated default value of the material modulus based upon its AASHTO or USCS soil classification. The next highest level of design (Level 2) consists of correlations to other laboratory tests (similar to those found in the 8 AASHTO design guide). The most accurate design level (Level 1) involves comprehensive resilient modulus testing (either in the laboratory or in the field) that result in the generation of a suitable constitutive relationship that models the non-linear stress-sensitivity of those materials. Equation 1-6 is the model recommended by the new MEPDG, but it also allows for the use of other constitutive models to achieve the best possible prediction of resilient modulus behavior. Therefore, this design methodology will take into account material?specific variations of modulus with stress at its most accurate design level. In order to effectively characterize the stiffness behavior of unbound materials, several factors must be considered. First of all, does laboratory or field resilient modulus testing provide a better representation of material behavior? Secondly, is the material stress-sensitive and, if so, which stress-sensitivity model best quantifies the behavior of that material? Answering these questions allow for the most accurate quantification of resilient modulus for pavement design and more accurate modeling of pavement design life. For many agencies, these questions can be resolved through full-scale accelerated pavement testing (APT). One such facility is the National Center for Asphalt Technology (NCAT) Test Track. The Test Track is a 1.7 mile accelerated testing facility containing 46 pavement test sections, each 200 foot long, designed to test the performance of a wide variety of asphalt mixes and structural designs. Each of these test sections are subjected to live traffic loading under a fleet of triple-trailer trucks that are operated by human drivers (a more detailed description of the Test Track is given in Chapter 3). Part of the testing program at the Test Track involves characterization of the 9 various paving layer materials in both the laboratory and the field, making this facility ideal for a study of this nature. OBJECTIVES OF THESIS The overall goal of this study was to mechanistically characterize the unbound paving materials at the NCAT Test Track for effective use in pavement design. Specific objectives included: ? A comparison of laboratory and field derived resilient moduli. ? An evaluation of common non-linear stress-sensitivity models with respect to laboratory and field moduli. ? Developing a recommendation as to the effective use of unbound material moduli in pavement design and analysis. SCOPE OF THESIS Several resources were used to perform this study. First, a review of relevant literature was performed to assess the state of the practice with regards to unbound material characterization. Topics reviewed include: laboratory triaxial resilient modulus testing, non-destructive pavement testing with the FWD, non-linear constitutive relationships relating resilient modulus to unbound material stress state, and agreement between resilient modulus data for a given unbound material collected in both the laboratory and the field. The NCAT Test Track provided an ideal setting for a study of this nature. The five unbound materials utilized for this study were constructed in various capacities 10 within the eleven full-depth instrumented pavement test sections (or structural sections) located at the Test Track. Laboratory triaxial resilient modulus testing on each of these unbound materials was performed by Burns, Cooley, Dennis Inc using test specification NCHRP 1-28A. For in-situ pavement characterization, NCAT operates a falling weight deflectometer (FWD) as part of a regular testing program at the Test Track on each of the eleven structural sections. Four dates worth of deflection data from this testing program were used in the development of field-calibrated stress-sensitivity models for the five unbound materials. An additional four dates worth of deflection data were then used for the model validation process. Additional FWD testing was performed above the embedded pressure and strain instrumentation to assess the measured pavement responses versus the responses predicted with the backcalculation software. EVERCALC v5.0 was the backcalculation software utilized to generate the unbound layer moduli from the FWD data. Microsoft Access was used for managing the database of deflection data and querying the data into a usable format. AUDEF is a file conversion program that was used to convert the deflection files from an Access database format to a deflection file that could be used by EVERCALC. This software also converted the backcalculation output files into usable EXCEL spreadsheets. Finally, DATAFIT non-linear regression modeling software was used to generate non- linear stress-sensitivity models for the various resilient modulus data. The availability of multiple data sets regarding the unbound material behavior was especially useful in performing a complete study regarding unbound material characterization. Equations 1-4 through 1-7 were then evaluated to determine which stress-sensitivity model provided the best fit to both the laboratory and backcalculated 11 resilient moduli. Finally, comparisons were made between the laboratory and the field- measured resilient moduli. ORGANIZATION OF THESIS A review of relevant literature concerning unbound material characterization is presented in Chapter 2. Chapter 3 gives an overview of the NCAT Test Track, as well as information regarding the in-situ unbound material properties and construction of the eleven instrumented pavement test sections. This chapter also outlines the FWD testing program utilized at the Test Track and provides details concerning the FWD testing performed atop the embedded pavement response instrumentation. A detailed look at the laboratory resilient modulus testing procedure is given in Chapter 4, along with the results of this testing on the unbound materials utilized in the structural sections at the Test Track. Chapter 5 outlines the procedure utilized to select the optimum cross-section for backcalculation for each of the eleven structural sections, as well as the results of this investigation. Chapter 6 presents the results of the field modulus characterization using the FWD on the structural sections at the Test Track. This chapter includes the calibration and validation process used to develop constitutive relationships for the various unbound materials based on field testing. Chapter 7 gives a comparison between the unbound material behavior in the lab and in the field, along with an assessment on the degree of agreement. Chapter 8 is the final chapter in which the conclusions and recommendations of this study are presented. 12 CHAPTER 2 LITERATURE REVIEW INTRODUCTION The mechanistic-empirical (M-E) pavement design process is based on the mechanics of materials that relates an input, such as a wheel load, to a particular pavement response, such as a stress or a strain at a structurally critical location within the pavement structure (Huang, 2004). The M-E design process utilizes detailed knowledge of a trial pavement structure regarding the material composition, expected traffic loading, and environmental conditions to more accurately model the pavement structure and project its design life (Priest and Timm ?Fatigue,? 2006). Monismith (1992) states that ?the intent of the analysis and design process is to simulate, in advance, the expected performance of the asphalt pavement so that the optimum thicknesses of the various components can be selected and the available materials used effectively.? Given accurate knowledge of the materials to be used in the pavement structure and the environmental and traffic conditions, the pavement responses at critical locations in the pavement structure (bottom of the HMA for fatigue cracking, surface of the subgrade for rutting, etc.) can be simulated using various methods, such as layered- elastic analysis (Huang, 2004). With these calculated critical responses, the cycles to failure for the pavement under loading of a critical design vehicle can be calculated by 13 utilizing calibrated transfer functions. These equations typically take the form of Equation 2-1 for fatigue and Equation 2-2 for rutting (Priest and Timm ?Fatigue,? 2006). Therefore, this method of pavement design can generate reliable optimal thicknesses that have basis in mechanistic theory and are not based purely on empirical methods. Figure 2-1 shows a flow chart outlining the conceptual framework of M-E design and analysis. 2 1 * 1 k t f kN ? ? ? ? ? ? ? ? = ? (2-1) 4 1 * 3 k v r kN ? ? ? ? ? ? ? ? = ? (2-2) where: N f = Number of cycles to failure via fatigue cracking ? t = Critical horizontal tensile strain at the bottom of the HMA layer N r = Number of cycles to failure via structural rutting ? v = Critical vertical compressive strain at the surface of the subgrade soil k n = empirical constants 14 Figure 2-1: M-E Design Flowchart (after Monismith 1992). One of the critical inputs for M-E pavement design is accurate knowledge of the stiffness characteristics of the unbound materials to be used in the pavement structure, specifically the resilient modulus of these layers. The new MEPDG (2004) offers three different levels of accuracy in determining the resilient modulus of the unbound layer Traffic Environment Pavement Material Characteristics Trial Pavement Layer Thicknesses and Material Combinations Degree of Risk Analyses for Specific Distress Modes Fatigue Cracking Permanent Deformation (Rutting) Thermal Cracking Distress Criteria INPUTS DISTRESS ANALYSES Selection of Optimal Thicknesses and Material Requirements DESIGN DECISIONS 15 materials. A level 3 design value is the least accurate and consists of a designer estimate or typical values of the resilient modulus at the optimum moisture content for various AASHTO and USCS soil classification. A level 2 design value consists of correlations between resilient modulus and other material properties (such as California bearing ratio and dynamic cone penetrometer penetration data). These correlations are very similar to those found in the AASHTO Guide for the Design of Pavement Structures (1993). Level 1 design is the most accurate design value and involves using either laboratory triaxial testing or field non-destructive testing to generate a constitutive relationship between material stress-state and resilient modulus. Characterization of resilient modulus in this manner is the focus of this study and of this literature review. The focus of this literature review is to assess the state of the practice with regards to resilient modulus testing and data analysis methods. First of all, resilient modulus testing in the laboratory using a triaxial chamber and in the field using a falling weight deflectometer (FWD) will be discussed. Next, the various constitutive relationships that relate unbound material stresses to resilient modulus will be discussed, along with examples of their use in practice. Finally, studies that have compared the results of laboratory and field resilient modulus (using the triaxial chamber and FWD, respectively) will be presented and discussed. LABORATORY TESTING The most common laboratory test utilized to determine the resilient modulus of granular base materials and subgrade soils for flexible pavement construction is the triaxial resilient modulus test. According to the specification outlined by NCHRP 1-28A 16 (2004), ?The resilient modulus test simulates the conditions in a pavement due to the application of moving wheel loadings. As a result, the test provides an excellent means for comparing the behavior of pavement construction materials under a variety of conditions and stress-states.? The primary advantage of the triaxial test is that it allows the unbound material to be tested in a controlled environment over a variety of stresses, making it ideal for the development of the stress-sensitivity model relating stress-state to resilient modulus (Nazarian et al., 1998). Though testing is conducted in a controlled environment, there are several disadvantages to this mode of testing as well. First, the laboratory resilient modulus is not completely representative of in-situ conditions due to several factors: sample disturbance, differences in aggregate orientation, differences in water content, and differences in level of compaction (Seeds et al., 2000; Nazarian et al., 1998). Imperfect instrumentation in the triaxial apparatus creates difficulties in reproducing the in-situ state of stress and accurately measuring axial deformation in laboratory samples (Seeds et al., 2000). Additionally, multiple test specifications and variable equipment calibration and verification procedures tend to make triaxial results variable both within and between laboratories (Seeds et al. 2000). The laboratory specimens represent the properties of a material from a very small location and not necessarily the larger mass of material that would respond to the pass of a typical truck axle (Seeds et al., 2000). Finally, laboratory resilient modulus testing tends to be very time consuming and expensive to perform in order to generate accurate results (Nazarian et al., 1998). Throughout the literature concerning laboratory resilient modulus testing, multiple specifications by which this test was conducted were encountered. Zhou (2000) 17 utilized AASHTO T274-82 for their testing. Ping et al. (2001) utilized AASHTO T292- 91 for their triaxial testing. Yau and Von Quintus (2002) performed a synthesis on resilient modulus data generated within the LTPP database that had been tested according to LTPP Protocol P46. The MEPDG (2004) recommends using one of two specifications for laboratory resilient modulus testing, either NCHRP 1-28A or AASHTO T307. According to Andrei et al. (2004), the NCHRP 1-28A procedure was developed to harmonize the existing resilient modulus test protocols and develop a single test method utilizing the best features of the existing specifications. This particular testing specification will be discussed in greater detail in Chapter 4 (laboratory testing). Sampling of Materials for Laboratory Resilient Modulus Testing As previously noted, one of the primary disadvantages of laboratory resilient modulus testing has to do with preparing samples that are representative of the unbound samples that are compacted in the field. NCHRP 1-28A (2004) recommends compacting samples to the in-situ density and moisture content, but allows samples to be compacted to 95 percent of maximum laboratory dry density and the optimum moisture content in the absence of these data. To generate accurate laboratory modulus values, the material must be sampled and then recompacted in such a manner that yields representative results. Nazarian et al. (1998) conducted a study in which the laboratory determined moduli of unbound materials utilized in various pavement structures were compared for the materials obtained from the quarry and for materials excavated from the existing 18 pavement structure. The authors concluded that there is a large difference in material properties based on the location from which these materials were obtained. The quarry and in-situ unbound materials were found to have varying gradation curves, Atterberg limits, and optimum moisture contents of the quarry materials were found to vary from the in-situ moisture content by an average of 2 percent. The authors also concluded that that quarry materials compacted to Proctor maximum dry densities and moisture contents yielded notably different constitutive model regression coefficients than the in- situ samples compacted to the field densities and moisture contents. Yau and Von Quintus (2002) conducted a study in which the LTPP resilient modulus database was analyzed to determine the effect of sampling technique on testing results. The authors concluded that sampling technique (boring with an auger versus trenching in the pavement structure) had an effect on the resilient modulus test results for the uncrushed gravel and crushed stone base and subbase materials. This is logical given the ability of a boring to alter the gradation of a coarse-grained material. For the subgrade soils in the database, whether the sample was obtained through disturbed or undisturbed sample proved to impact at least one of the regression coefficients in the constitutive model for each of the soil types. Ping et al. (2001) conducted a study in which disturbed samples of soils taken from various excavated pavement sites throughout Florida were tested at both the in-situ moisture content and the optimum moisture content. The results of this comparison showed that the average laboratory resilient modulus at optimum compacted conditions was about 1.1 times higher than the average laboratory resilient modulus at in-situ conditions. The data for this comparison are shown in Figure 2-2. It was also noted that 19 the average dry density at optimum compacted conditions was higher than the field- measured in-situ dry density. Therefore, sample preparation did have an impact on the results of resilient modulus testing. Figure 2-2: Comparison of Resilient Modulus values for Laboratory Samples with Different Moisture Contents (Ping et al., 2001). Target Moisture Contents and Densities for Laboratory Samples Several authors make various recommendations about whether the triaxial samples should be compacted to optimum or in-situ conditions. Based on the results of the study outlined above, Ping et al. (2001) recommended the use of the laboratory determined optimal conditions as a standard representation of unbound material conditions for design. Parker and Elton (1990) also utilized the optimum moisture 20 content and maximum laboratory dry density for testing, citing that these conditions would reasonably approximate field conditions. Zhou (2000) utilized data from two field FWD testing sites, and the data from one site was compacted to the maximum dry density and optimum moisture content, while the other was compacted to field conditions. Buchanan (2007) states that samples for laboratory resilient modulus testing should be compacted as close as possible to in-situ conditions for testing. Hence, there is no general consensus in literature as to whether triaxial resilient modulus samples should be compacted to optimum laboratory or in-situ measured conditions. Stress-States Utilized in Laboratory Testing According to the MEPDG (2004), the stress-states at which the laboratory triaxial samples are tested should be representative of the stress-states that are to be expected in the field. Both NCHRP 1-28A (2004) and AASHTO T307 (2003) define a range of stresses to be tested based on whether the material tested is located in the base or the subgrade (since varying depths in the pavement structure correspond to different stresses under loading). The resilient modulus test is designed to characterize the unbound layer moduli under loading conditions that will not result in the failure of the material (Buchanan, 2007). These testing methods have varying treatments of the stress- sequences that the material endures. The original NCHRP 1-28 laboratory specification procedure holds the confining pressure constant while increasing the deviatoric stress, causing the material to rapidly approach the Mohr-Coulomb failure envelope on a plot of shear stress versus normal stress (Andrei et al., 2004). A similar method of load application is applied in the AASHTO T307 (2003) procedure. The NCHRP 1-28A 21 procedure takes a different approach by holding the principal stress ratio constant (? 1 /? 3 ) and simultaneously increasing the deviatoric and confining stresses, minimizing the probability of premature sample failure (Andrei et al., 2004). This contrast in loading philosophies is illustrated in Figure 2-3. Figure 2-3: Comparison of Loading Stress Sequences for Granular Materials (Andrei et al. 2004). FIELD MODULUS CHARACTERIZATION An alternative method of determining pavement material moduli comes from the use of non-destructive pavement testing equipment on the existing pavement structure. Non-destructive testing (NDT) typically applies some form of loading to the in-situ pavement structure so that pavement deflections can be measured at various distances from the loading point. This form of testing is necessary to perform the process of 22 backcalculation to determine the respective pavement layer moduli at a particular testing location. Irwin (2002) provided the following definition of backcalculation: ?The procedure to determine Young?s modulus of elasticity for pavement materials using measured surface deflections by working elastic layer theory ?backwards? is generally called backcalculation.? Perhaps the earliest form of NDT was called the Benkelman Beam. This apparatus was used to measure the pavement deflection basin between two rear truck tires. Other forms of NDT testing include seismic pavement analyzers and falling weight deflectometers. Seismic pavement analyzers utilize a small seismic source and can determine pavement layer moduli by measuring and analyzing the generated seismic waves (Nazarian et al., 1998). Falling weight deflectometers (FWD) are commonly used NDT equipment that apply a pulse loading to the pavement. The pavement deflection basin resulting from this load is measured by a series of velocity transducers at various distances from the loading apparatus (Irwin, 2002). This equipment uses a pulse loading to more closely simulate the pavement loading applied by a moving wheel load (Irwin, 2002). A Dynatest 8000 Model FWD is shown in Figure 2-4. The focus of this literature review will be FWD testing, since this form of NDT was available for this research. 23 Figure 2-4: Dynatest 8000 Falling Weight Deflectometer. FWD testing has multiple advantages. First, it allows testing the in-situ condition of the pavement without damaging the pavement structure by trenching or coring. Secondly, it allows for the determination of the structural capacity of a pavement. This is critical for determining optimum overlay thicknesses and potentially identifying structural weaknesses in a given pavement (Von Quintus and Killingsworth, 1998). Typically, the overall expense associated with non-destructive testing is less than with laboratory testing (Seeds et al., 2000). The FWD loadings can typically be adjusted to approximate the critical wheel loadings experienced by the pavement structure (Nazarian et al., 1998). Finally, the sampling frequency with non-destructive testing can be increased much more economically than with laboratory testing. Therefore, a better 24 representation of the natural and random variability within the pavement properties can be obtained (Seeds et al., 2000). Despite the appeal of being able to accurately characterize paving materials in the field, there are multiple disadvantages to FWD testing as well. One disadvantage of the FWD is that generating accurate pavement layer moduli is not an exact science, and accurate modulus determination requires good quality control of raw data and precision in analysis (Von Quintus and Killingsworth, 1998; Nazarian et al., 1998). It is possible for backcalculation software to produce multiple combinations of pavement layer moduli from the same deflection basin. This non-unique solution dilemma therefore requires some interpretation of the most logical combination of layer moduli (Seeds et al., 2000). Most backcalculation programs assume pavements are loaded in a static manner, when in reality wheel loads are dynamic in nature (Seeds et al., 2000). Additionally, most backcalculation programs assume purely linear-elastic behavior, when HMA typically behaves as an elastic visco-plastic material and unbound materials behave in a non-linear elastic fashion (Seeds et al., 2000). Finally, extremely accurate pavement cross-section information (thicknesses and layer composition) is required to generate reasonable backcalculated results (Seeds et al., 2000). Field FWD Testing Programs To accurately assess the in-situ pavement conditions, an FWD testing program must be able to accurately quantify the variability inherent to the pavement structure. FWD moduli represent both the pavement conditions at the time of measurement and the stress conditions induced by the applied loading (Parker, 1991). Therefore, multiple 25 tests are useful to quickly gain a quality data representation of the overall pavement structure. There are several types of variability inherent to most pavements: loading variability, spatial and construction variability, and seasonal variability. These sources of variability can readily be accounted for with a quality FWD testing program. Load Levels Used for FWD Testing A common practice in FWD testing is to test the pavement at load levels that are representative of the load magnitudes to be placed on the pavement structure by live traffic (Parker, 1991; Nazarian et al., 1998). Timm and Priest (?Material Properties,? 2006) utilized 2 repetitions of a 9,000 lb load at each testing location. The 9,000 lb load is useful because it represents the loading placed on the pavement structure by one-half of a standard 18,000 lb axle load (Parker and Elton, 1990). Sometimes multiple loadings can be used at each testing location to account for varying traffic levels and to measure the stress-sensitivity of the underlying pavement layers (Parker and Elton, 1990; Von Quintus and Killingsworth, 1998). Von Quintus and Killingsworth (1998) performed analysis on Long Term Pavement Performance (LTPP) database deflection data that contained 6 kip, 9 kip, 12 kip, and 16 kip loadings. Parker and Elton (1990) utilized 9 kip, 12 kip, and 15 kip loadings in their FWD testing program. Zhou (2000) utilized FWD loadings between 13 kN and 67 kN (approximately 3 kip and 15 kip). Sebally et al. (2000) state that the state of Nevada uses 3 to 4 drops of the FWD at each testing location with increasing load levels. These load levels range between 27 kN and 67 kN (approximately 6 kip to 15 kip) for standard pavements and loads between 45 kN and 90 kN (approximately 10 kip to 20 kip) for thicker pavement structures. Seeds et al. (2000) 26 used increasing load levels of 27, 40, 53, and 71 kN (6, 9, 12, and 16 kip) at the WESTRACK accelerated testing facility. Spatial Variability As mentioned previously, one of the primary advantages of FWD testing is the ability to test multiple pavement locations in a short period of time. This is important because pavements are generally not homogenously constructed structures, and contain several possible sources of spatial variability. Some of these sources of variability include: layer thickness variability throughout the longitudinal profile of the pavement, density variation due to construction and traffic, varying depth to bedrock or the water table, and varying density and moisture content of the unbound materials (Irwin, 2002). Irwin (2002) recommends testing at various points across the transverse profile, due to variations in deflection testing results within and between the wheelpaths. Priest and Timm (2006) performed testing at multiple longitudinal locations throughout the pavement structure in both the inside and outside wheelpath to quantify spatial variability. Nazarian et al. (1998) utilized 11 longitudinal FWD drop locations at their sites to be tested. Parker (1991) utilized testing at 10 longitudinal locations spaced 200 feet apart at the FWD testing sites used for seasonal materials characterization. The state of Nevada tests at 0.15 km (0.09 mile) intervals for highway projects in each direction of travel (Sebally et al., 2000). Testing was performed at 10 meter (33 foot) intervals within each of the pavement test sections at WESTRACK (Seeds et al. 2000). Therefore, spatial variability can be quantified by FWD testing at multiple longitudinal and transverse pavement locations. 27 FWD Deflection Sensor Spacing Another critical component of the FWD is the configuration of the deflection sensors used to measure the deflection of the pavement surface immediately after loading. A sufficient number of these sensors must be present to yield an accurate deflection basin for backcalculation. Zhou (2000) utilized a KUAB model FWD with 6 deflection sensors, three of which were adjustable depending on the project requirements. Rwebangira et al. (1987) utilized a KUAB FWD with four sensors spaced 11.8 inches apart with one sensor beneath the loading plate. Timm and Priest (?Material Properties,? 2006) utilized an FWD with 7 sensors spaced 12 inches apart with one sensor beneath the loading plate. Nevada utilizes an FWD using 7 sensors that have variable spacing depending on the thickness of pavement for a given project. The sensors are spaced at 0, 203, 305, 457, 915, 1219, and 1524 mm (approximately 0, 8, 12, 18, 36, 48, and 60 inches) away from the loading plate for standard projects. The sensors are spaced at 0, 305, 610, 864, 1219, 1524, and 1829 mm (approximately 0, 12, 24, 34, 48, 60, and 72 inches) away from the loading plate for Interstate Highway projects with thicker pavements. The logic behind the variable sensor spacing is that closely spaced sensors near the load are useful for the accurate delineation of pavement surface conditions while the sensors farther from the load are useful for the accurate determination of subgrade layer moduli (Sebally et al., 2000). The FWD at WESTRACK utilized a similar sensor setup, with seven sensors spaced at 0, 305, 457, 610, 914, 1219, and 1524 mm (approximately 0, 12, 18, 24, 36, 48, and 60 inches) away from the load plate. 28 The sensor configuration is critical in determining how accurately the moduli of the layers can be determined. The presence of sensors far away from the loading is critical to accurately determining the moduli of the deeper subgrade layers. This is because at larger distances from the loading, only the deeper pavement layers are influenced by the loading and thereby deflecting (Irwin, 2002). This concept is illustrated graphically in Figure 2-5 below. Figure 2-5: Impact of Zone of Influence on Pavement Deflections. The presence of outer sensors can also be useful in determining whether a rigid layer exists at a given testing site. By plotting the deflections of the outer sensors against the normalized radial distance of these sensors from the loading (dividing the load plate radius by the radial distance), this curve can be extrapolated to ascertain an approximate depth to the stiff layer. This is based on the assumption that the depth to the stiff layer is approximately where the deflections will be equal to zero. If the curve FWD Sensors HMA Granular Base Soil Load Zone of Influence Boundary 29 intersects the x-axis at a positive value, this value gives a good approximation of the depth to the stiff layer (Irwin, 2002). This concept is illustrated in Figure 2-6. Figure 2-6: Determination of Bedrock Depth with Deflection Measurements. Seasonal Variation A critical component of any FWD testing program is ensuring that it captures the performance variability associated with the changing seasonal conditions at the test site. Several studies have been performed that document the effects of seasonal changes on the various pavement layers. Parker and Elton (1990) conducted a study in which FWD testing was performed at several seasonal sites throughout the state of Alabama at approximately 2 month intervals over the course of three years to quantify the seasonal modulus variability. Deflection versus Normalized Radial Distance (a/r) 0 1 2 3 4 5 6 7 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 a/r Sen s o r De fle c t ion ( m illi- inche s) Testing Location with No Stiff Layer Effects - Intersection at the Origin Testing Location with Stiff Layer Effect on Deflection - Intersection at Positive Radial Distance 30 Figure 2-7 shows the behavior of the backcalculated base layer moduli with time as well as the variations in temperature in rainfall as percentages of their respective maximum monthly averages. In the absence of frost action, temperature and moisture were the major seasonal factors to consider in the state of Alabama. The figure shows the general trend that an increase in temperatures combined with low rainfall levels corresponded to an increase in base moduli. A decrease in temperature combined with increased rainfall levels would serve to decrease the base layer moduli. This was believed to occur since high moisture levels in the absence of higher temperatures would lead to a moisture build-up in the unbound layers, and a subsequent reduction in strength. Figure 2-7: Seasonal Base Moduli Behavior in Alabama (Parker and Elton 1990). 31 Timm and Priest (?Material Properties,? 2006) conducted a study at the NCAT Test Track in which FWD testing was performed at approximately one month intervals as a means of quantifying the seasonal behavior of the pavement layer moduli. The findings from this study were that the HMA modulus was greatly affected by seasonal changes and those seasonal effects seemed to decline lower in the pavement structure. Both the backcalculated base moduli and subgrade moduli seemed to experience a slight reduction in modulus with increasing temperature. This was opposite of the expected result since hotter pavement temperatures reduce asphalt modulus and allow for higher stresses to be imparted on the unbound layers, causing an increase in modulus for a stress-hardening material. The authors concluded that this seasonal effect might be an artifact of the backcalculation process, and that an annual average unbound layer modulus would be most representative of those materials. Briggs and Lukanen (2000) conducted a study in which the seasonal pavement layer moduli for 25 different LTPP seasonal monitoring sites across North America were analyzed to quantify seasonal effects. The data for this study were collected between the fall of 1993 and the spring of 1995 on a monthly basis. The authors concluded that seasonal temperature variation affects the modulus of all the pavement layers, not just the surface layer. The study showed that the temperature-induced modulus variations of the surface layer induce changes in the moduli of the underlying unbound layers. Figure 2-8 illustrates this concept by showing the base modulus variation for a seasonal site near Estillene, Texas, a location that subject to warmer summer temperatures and not subject to frost action. The authors noted that this temperature influence was more notable on thicker asphalt pavements as opposed to thinner ones. It was also evident that 32 the moduli of the unbound layers were most affected by a freezing cycle, and that precipitation had relatively little impact on the unbound layer moduli. Figure 2-8: Seasonal Base Moduli Variation at Estillene, Texas LTPP Seasonal Monitoring Site (Briggs and Lukanen 2000). The general consensus of the literature on seasonal variation of backcalculated moduli is that there can be a significant fluctuation in modulus values with changing conditions over time. Therefore, a quality FWD testing program should take this into account and test the pavement structure at regular intervals across multiple seasons to best quantify seasonal pavement layer moduli.. Quality Control of Backcalculated Data To get accurate backcalculated modulus values from FWD testing, it is necessary to regularly calibrate the FWD deflection sensors to run good quality control (QC) of the 33 output deflection basins (Von Quintus and Killingsworth, 1998; Irwin 2002). Irwin (2002) states ?It is a basic principle that a mismatch between the theoretical assumptions and the actual data will almost inevitably result in an error, sometimes large and sometimes small, in the calculated modulus.? Rwebangira et al. (1987) performed a study in which they compared the sensitivity of multiple backcalculation programs to various inputs. The authors concluded that every backcalculation program they experimented with was very sensitive to the deflection measurements. Von Quintus and Killingsworth (1998) point out that layered-elastic analysis methods are not necessarily applicable to all deflection basins, and moduli backcalculated from these basins may not be representative of the layer moduli. In their analysis of the LTPP deflection basins, Von Quintus and Killingsworth eliminated several types of deflection basins as being unsuitable for analysis. Typically, these basins exhibited an increase in deflection farther away from the loading point or a large magnitude of deflection increase or decrease between sensors. Figure 2-9 below shows an example of quality deflection basins that exhibit a reasonable decrease in deflection as distance from the load increases, while Figure 2-10 illustrates ?problem? deflection basins where there is an increase in deflection farther away from the load. 34 Figure 2-9: Examples of Quality Deflection Basins (Von Quintus and Killingsworth, 1998). Figure 2-10: Examples of Problematic Deflection Basins (Von Quintus and Killingsworth, 1998). . A common representation of the overall difference between the measured deflection basin and the layered-elastic predicted deflections is the root mean squared 35 (RMS) error. This term, shown as equation (2-3) below, represents the overall percentage error between the calculated and measured deflections (Sivaneswaran et al., 2001). Commonly, minimization of this error term is desirable to ensure reasonable backcalculated moduli. Von Quintus and Killingsworth (1998) utilized the practice of eliminating deflection basins with an RMS error term above a specific cut-off value. For their data set, this value was set at 2.5 percent. Nevada also uses a cut-off value of 2.5 percent RMS error for their deflection basins (Sebally et al., 2000). For the FWD testing at the WESTRACK experiment, RMS error values were ranked on a relative scale, with RMS error less than 1 being considered ?excellent?, values between 1 and 4 percent RMS error being ?very good?, and values between 5 and 7 percent RMS error being ?good? (Seeds et al., 2000). N d dd ErrorRMS N i m cm 2 1 *100(%) ? = ? ? ? ? ? ? ? ? ? = (2-3) where: N = Number of Sensors d m = measured deflections d c = calculated deflections Software Used in Backcalculation Several different types of backcalculation software were encountered through various literature involving FWD testing. Some of the more commonly used programs include ELMOD (developed by Dynatest), EVERCALC (developed by the Washington State DOT), MODCOMP (developed by Cornell University), PADAL (developed by the University of Nottingham), and WESDEF (developed by the U.S. Army Waterways 36 Experiment Station) (Irwin, 2002). These different backcalculation programs are equipped with a wide range of features to assist in analysis. Most backcalculation programs treat all the layers as linear-elastic, and ignore the stress-dependency of the unbound materials. However, both BOUSDEF (Zhou, 2000) and EVERCALC (Sivaneswaran et al., 2001) are capable of modeling the non- linear behavior of the base and subgrade material if multiple FWD loadings are utilized. These programs are able to perform this analysis due to having built-in layered-elastic analysis software to compute comprehensive stress-states at critical locations within the pavement structure after any FWD loading. For programs not containing this feature, a separate layered-elastic analysis would have to be performed to generate these critical stresses. ELMOD is the software designed to be compatible with Dynatest FWD data and is capable of adjusting the HMA modulus based on temperature and adjusting the base and subgrade moduli for seasonal variations (Parker, 1991). Some FWD programs are capable of utilizing the outer deflections to calculate the depth to any existing stiff layer within the pavement structure, while other programs require the entry of a fixed stiff layer depth. Both ELMOD and EVERCALC can utilize the outer deflection sensors to predict the depth to bedrock if indicated to do so by the user, while CHEVDEF and ELSDEF require the location of the stiff layer to be specified by the user (Parker and Elton, 1990; Sivaneswaran et al., 2001). Though many different backcalculation programs exist, studies have shown that these programs can often yield varying results. Von Quintus and Killingsworth (1998) state that layer moduli generated with different backcalculation programs should not be used interchangeably due to differences in the calculation schemes within these 37 programs. Parker and Elton (1990) evaluated the serviceability of four different backcalculation programs for use in their study. They concluded that while there were differences between the output values generated by each program on identical deflection basins, no one program provided obviously superior results. Rwebangira et al. (1987) also concluded that even with identical input values, the three backcalculation programs evaluated (BISDEF, MODCOMP2, and SEARCH) would generate very different layer moduli. The authors attribute these differences to MODCOMP2 not having the option of including a stiff layer for analysis and SEARCH treating stress distribution differently than the other two linear-elastic based programs. Therefore, it is necessary to be consistent with a given backcalculation procedure and ensure that those results are reasonable for any given project. Determining the Optimal Pavement Cross-Section for Backcalculation A critical component in generating accurate backcalculated data is the determination of the backcalculation cross-section (or pavement model). This cross- section defines the pavement layers within the backcalculation software. Prior to developing this cross-section, accurate knowledge of the pavement structure to be tested must be obtained. Accurate layer thicknesses are critical for backcalculation to ensure accuracy in material characterization. Von Quintus and Killingsworth (1998) report that a 10% change in layer thickness can easily result in more than a 20% change in layer modulus. Rwebangira et al. (1987) state that the predicted moduli are very sensitive to changes in thickness for both the base and surface layers. The authors reported that a one inch change in surface layer thickness could alter both the base and surface layer 38 moduli by over 60% each. As a result, coring, trenching, or surveyed construction thicknesses are recommended to obtain the site-specific layer thicknesses. A critical decision involving the cross-section selection involves how many layers to use in the model and whether or not to include a stiff layer for analysis. Rarely is the as-built pavement cross-section used as the pavement model for backcalculation. It is often necessary to combine and subdivide various pavement layers to achieve reasonable results. In their study regarding the evaluation of pavement layer moduli in Alabama, Parker and Elton (1990) were required to develop a suitable pavement model for backcalculation at each of the testing sites. For this model, all the lifts of hot mix asphalt were combined for the surface layer. For the second layer, the granular base and subbase layers were combined for each site. Additionally, several of the sites included the processed subgrade layer (fill) in the second layer. The authors concluded that the in-situ moisture and density measurements were excellent indicators as to whether to include this fill layer with the base/subbase layer or with the deep subgrade. The results of their analysis showed that this layer should be included with the base/subbase layer if it has moisture and density values similar to those measured in the base/subbase (and vice versa should the moisture and density values be more similar to the deep subgrade). The inclusion of the processed subgrade in the base/subbase layer typically had little effect on the HMA layer modulus, minor impact on the deep subgrade modulus, and significant impact on the base/subgrade layer modulus. In cases where the moisture and density values of the processed subgrade and base/subbase were similar, the composite base layer including the processed subgrade exhibited more consistent and reasonable 39 base layer modulus values than a base layer not including the processed subgrade for backcalculation. Beneath the composite base layer, the deep subgrade was modeled as the third layer in the pavement model. Below this subgrade, a rigid layer was set at 20 feet beneath the pavement surface. The authors utilized a stiff layer in the pavement model for several reasons: due to the existence of bedrock at a finite depth at many testing locations, due to the natural stiffening of most soils with depth due to increasing confinement leading to an over-prediction of subgrade moduli in the pavement model, and to avoid the calculation of infinitely large subgrade moduli for soils with non-linear stress-strain behavior. Therefore, the stiff layer was utilized to generate more reasonable pavement moduli at each of the test locations. Figure 2-11 shows the final pavement model used by Parker and Elton for their study. 40 Figure 2-11: Pavement Model used by Parker and Elton in Alabama (Parker and Elton, 1990). Seeds et al. (2000) proposed a three-layer backcalculation cross-section for analysis of the deflection data at the WESTRACK experiment. The contrast between the 41 actual pavement section and the cross-section for backcalculation are shown in Figure 2- 12 below. For these sections, the total thickness of HMA, granular base, and fill material were placed in multiple lifts (two lifts each for both the HMA and granular base, three for the engineered fill). In the final backcalculation cross-section, the top two layers were grouped by material type, using one layer each for combined HMA and granular base thickness. The engineered fill was also combined with the natural subgrade soil due to the backcalculation software (EVERCALC) producing similar results for these two layers. Additionally, a rigid layer was added to the program to account for the stiff layer effects imposed by the water table (typically 3 to 4 meters below the pavement surface). Figure 2-12: Pavement Model used by Seeds et al. at WESTRACK (Seeds et al., 2000). 42 Timm and Priest (?Material Properties,? 2006) utilized a three-layer cross-section for analysis of the deflection data from the instrumented sections at the NCAT Test Track. In that study, four trial cross-sections were evaluated on the basis of RMS Error, reasonableness of backcalculated moduli for the different layer materials, and measured (with embedded instrumentation) versus predicted (with layered-elastic analysis) pavement response from FWD testing directly above embedded instrumentation. In their analysis, a stiff layer was deemed inappropriate for use in analysis because the trial cross-section using a stiff layer generated unreasonably high RMS error values and unreasonably low HMA stiffnesses. For the final backcalculation cross-section, a combined HMA layer was used above a composite base layer (consisting of the granular base material combined with the improved subgrade material) above the deep subgrade layer. A common theme in the literature is to ensure that the site conditions are well known for the purposes of backcalculation. Von Quintus and Killingsworth (1998) indicated that a layer with different conditions (material type, moisture content, density, etc.) should most often be modeled as a separate layer. Irwin (2002) also suggests that water content and density be used as a differentiating factor between two layers for backcalculation, even if these layers consist of the same material type. Both Ping et al. (2001) and Nazarian et al. (1998) recommend that test pits be excavated near the testing site to accurately measure the pavement layer thicknesses as well as to determine the moisture contents and densities of the pavement layers in-situ. Nevada recommends 43 coring near the testing site to determine HMA thicknesses while utilizing construction information to acquire the base and fill layer properties (Sebally et al., 2000). The literature points out several pavement conditions that could prove potentially problematic in establishing a viable cross-section for backcalculation. Typically, layers with relatively small thickness within the pavement structure are combined with similar layers for the purposes of backcalculation. If this layer is too thin to have much impact on the surface deflections, then these deflections cannot be used to determine the modulus of that layer (Irwin 2002). Additionally, unbound layers with significantly lower moduli than their supporting layers tend to present problems for the backcalculation process. These issues should be taken into account when deciding upon a backcalculation cross-section for analysis. When evaluating trial cross-sections, several factors should be considered. First, the match between measured and predicted deflections should be reasonably low (i.e. low RMS error for the solution) (Irwin, 2002). Secondly, good deflection matches does not necessarily mean that the backcalculated moduli are reasonable (Seeds et al., 2000). The modulus values for the individual layers should be compared to ensure that there are no fluctuating moduli or compensating layer effects present (Timm and Priest ?Material Properties,? 2006). Additionally, the solution should be checked to ensure that the variability within the backcalculated solution can be attributed to variability within the pavement structure (spatial or seasonal) and is not an artifact of the backcalculated solution (Irwin, 2002). 44 Typical Inputs for Backcalculation Most backcalculation programs require multiple inputs in order to accurately determine pavement layer moduli. In addition to the deflection measurements, some of these inputs include: the pavement layer thicknesses (site-specific), the FWD sensor spacing and load plate radius (FWD specific), a modulus range for solution iterations in each layer, a seed (or initial) modulus value for each layer, the number of allowable solution iterations, Poisson ratios for each layer, and RMS error and modulus tolerances for each pavement layer. These inputs must be carefully selected to generate a viable backcalculation solution. An iteration range and seed moduli are typically required for each pavement layer in backcalculation to create a boundary for a reasonable solution. Rwebangira et al. (1987) conducted a sensitivity analysis on various backcalculation programs to their required inputs. The authors concluded that the range of moduli used in BISDEF could impact the accuracy of the backcalculated solution. The best results were achieved when the modulus ranges were on the same order of magnitude as the typical modulus values of the layer materials and that narrow modulus ranges could cause the solution to rest on the upper and lower modulus boundaries. The seed moduli selected within the modulus iteration range seemed to have little effect on the backcalculated moduli (provided reasonable seed moduli were used). The authors used a range of 250 to 850 ksi with a seed value of 375 ksi for HMA, a range of 10 to 50 ksi with a seed value of 30 ksi for base layer materials, and a range of 3 to 23 ksi with a seed value of 14.5 ksi for subgrade materials. Both Zhou (2000) and Parker and Elton (1990) recommend using seed moduli that are close to the typical estimated moduli for that layer material. Parker and 45 Elton (1990) also suggest that wide iteration boundaries be set to minimize the constraint on solution iterations. Sebally et al. (2000) recommend that reasonable modulus boundaries be set for each layer with the seed moduli being set mid-range. Considerable variance was found in literature concerning the number of iterations the program should use along with the RMS error and modulus tolerance used to determine when the measured and backcalculated deflection basins have sufficiently converged. Parker and Elton (1990) used 30 iterations with ELMOD to get deflection and modulus tolerances within 10 percent. This contrasts with the findings of Bush and Alexander (1985) who determined that 3 iterations were most often sufficient to get deflection matching within 3 percent. Rwebangira et al. (1987) determined that a number of iterations greater than 2 typically don?t have much effect on the backcalculated moduli, though 10 iterations were used for their study. The authors also determined that using a modulus tolerance below 0.2 percent did not improve the results, while modulus tolerances above 5 percent generate erratic solution behavior. The EVERCALC software (Sivaneswaran et al., 2001) recommends using 5 iterations with a 1 percent tolerance for both RMS error and backcalculated layer moduli. For each pavement layer to be backcalculated, a Poisson ratio must be specified as a general input. Parker and Elton (1990) noted that ELMOD uses a Poisson Ratio of 0.35 for all materials. For the layered-elastic computation of stresses beneath the FWD load, Parker and Elton (1990) used Poisson ratios of 0.35, 0.40, and 0.45 for the HMA, granular base, and subgrade materials, respectively. Zhou (2000) used Poisson Ratios of 0.35, 0.40, and 0.45 for the HMA, granular base, and subgrade materials, respectively. The MEPDG (2004) does not recommend testing of pavement materials for Poisson?s 46 ratio and contains a table of standard values for use with various unbound material types. Irwin (2002) claims that a quality estimate of Poisson?s ratio is important, but ?the consequences of being slightly incorrect are not very significant.? Typically, the stiffness of the rigid layer is fixed within the program and the bedrock modulus is required as a program input if a stiff layer is to be used. Parker and Elton (1990) noted that ELMOD automatically locates a stiff layer with fixed properties using the measurements from the outer deflection sensors. In their analysis of other backcalculation software (CHEVDEF and ELSDEF) the authors used a bedrock stiffness of 1,000 ksi and a Poisson Ratio of 0.5. The MEPDG (2004) recommends using bedrock stiffness between 750 and 2000 ksi and a Poisson?s ratio between 0.1 and 0.25. Rwebangira et al. (1987) concluded that both the HMA and base layer moduli were very sensitive to the depth to the stiff layer, making an accurate determination of this depth very important. COMMONLY USED STRESS-SENSITIVITY MODELS Given its definition as the ratio of deviatoric stress to recoverable strain, it stands to reason that the state of stress is the primary factor that affects the resilient modulus of a given unbound material. A common issue in resilient modulus testing is the selection of the most appropriate constitutive model relating material stress-state to resilient modulus. There are multiple constitutive equations in literature that relate resilient modulus to a combination of a variety of different stress-states. Some of these stress- states include: the confining pressure (? 3 ), the difference in principal stresses (or 47 deviatoric stress) (? d ), the sum of the principal stresses (or bulk stress) (?), and the octahedral shear stress (? oct ). Single Variable Non-Linear Stress Sensitivity Models The most basic non-linear stress-sensitivity models found in literature relate resilient modulus to a single stress term. All of these models are typically recommended for use with either a coarse-grained (typically non-cohesive) or fine-grained (typically cohesive) soil based on the stress-states used for the regression analysis. Therein exists a fundamental drawback with the use of these models: no one model is suitable for use with all soil types. One of the most common models in practice relates the resilient modulus to the sum of the principal stresses (or bulk stress). This model accounts for the stress- dependency of coarse-grained soils, which is typically seen with an increase in resilient modulus at increasing confining pressures (Andrei et al., 2004). This model is shown as Equation 2-4, and will be referred to as the ?bulk? model throughout this thesis. The atmospheric pressure term (p a ) is typically used to eliminate the influence of units of pressure on the calculated resilient modulus (Papagiannakis and Masad, 2008). A form of this model that utilizes only the confining stress term instead of the sum of the principal stresses is shown as Equation 2-5 (Papagiannakis and Masad, 2008). (2-4) 2 * 1 k a r p kM ? ? ? ? ? ? ? ? = ? 48 (2-5) The model form shown in Equation 2-4 has been widely employed for modeling the stress-dependency of coarse-grained granular materials (Parker and Elton, 1990; Zhou, 2000; Ping et al., 2001; Irwin, 2004; AASHTO, 1993). This model is the recommended constitutive relationship for triaxial testing in the AASHTO Guide for the Design of Pavement Structures (1993). The AASHTO guide states that for base materials the value of the k 1 coefficient should be between 3,000 and 8,000 psi and the value of k 2 should be between 0.5 and 0.7. Table 2-1 shows the typical coefficients recommended for the bulk model in the AASHTO guide for varying material properties and moisture contents. Table 2-1: Typical Values for Bulk Model Coefficients from AASHTO Design Guide (AASHTO, 1993) 2 3 1 * k a r p kM ? ? ? ? ? ? ? ? = ? 49 Parker and Elton (1990) conducted triaxial testing on soils collected from various FWD testing sites throughout Alabama and reported the results using the bulk model. The authors reported that the average k 1 value for dampened base materials in Alabama was 6.09 ksi, and the average k 2 value was 0.43. These coefficients were in reasonable agreement with those presented in the AASHTO guide concerning damp base materials (4 to 6 ksi for k 1 and 0.5 to 0.7 for k 2 ). The authors attribute the higher k 1 values to the significant cohesion exhibited by natural aggregate materials in Alabama. These materials also tend to have lower frictional resistance between the aggregate particles, resulting in a lower k 2 value compared to the AASHTO recommendations. Zhou (2000) conducted a study in which FWD testing was performed on two different pavement structures at multiple load levels. The use of multiple load levels, combined with utilizing backcalculation program with built in layered-elastic analysis software, allowed the author to use the bulk model to model the stress-sensitivity of the base layer materials in-situ. For each FWD loading, the stress-state at the mid-depth of the base layer was simulated using LEA and regressed against the backcalculated modulus. The k 1 coefficients generated from the data set at each job site fall between approximately 9 and 40 MPa (approximately 1.3 and 5.8 ksi). The k 2 coefficients range between 0.29 and 0.72 (calibrated to metric units). These coefficients are in reasonable agreement with those presented in the AASHTO guide, especially considering the considerable variability inherent to field testing. Typically, the resilient modulus of the fine-grained or cohesive soils is modeled as a function of the deviatoric stress, or the octahedral shear stress. The octahedral shear stress is a scalar invariant of the principal stresses which represents the state of shear 50 within the material, shown in Equation 2-6 (Irwin, 2002). When the confining pressures are assumed to be equal on all sides of the test specimen (as with a typical triaxial test), this term can be expressed as a coefficient multiple of the deviatoric stress (shown in Equation 2-7) (Papagiannakis and Masad, 2008). The most common single-term model using the octahedral shear follows the same form as the bulk model, and is shown in Equation 2-8. (2-6) doct ?? * 3 2 = (2-7) (2-8) The more commonly used model for the stress-sensitivity of fine-grained soil moduli is a power model using the deviatoric stress, shown as Equation 2-9. This model form is recommended by several authors for use in modeling cohesive soils (Ping et al., 2001; Parker and Elton, 1990). A bi-linear variation of this model (shown in Equations 2-10 and 2-11) is utilized by the ILLI-PAVE and KENLAYER layered-elastic analysis programs (Andrei et al., 2004; Seeds et al., 2000). The power-law form of the model shown in Equation 2-9 will be referred to as the ?deviatoric? model for the purposes of this thesis. (2-9) 2 * 1 k a d r p kM ? ? ? ? ? ? ? ? = ? 2 13 2 32 2 21 )()()( 3 1 ??????? ?+?+?= oct 2 * 1 k a oct r p kM ? ? ? ? ? ? ? ? = ? 51 ddr kwherekkkM ?? ??+= 1132 )( (2-10) ddr kwherekkkM ??