SENSITIVITY EVALUATION OF THE MEPDG FOR FLEXIBLE PAVEMENTS TRB Webinar July 25, 2012

Transcription

1 SENSITIVITY EVALUATION OF THE MEPDG FOR FLEXIBLE PAVEMENTS TRB Webinar July 25, 2012 Moderator: Trenton Clark Virginia Asphalt Association Presenter: Charles W. Schwartz University of Maryland

2 2 University of Maryland Charles Schwartz (PI) Rui Li Iowa State University Halil Ceylan (Co-PI) Sung Hwan Kim K. Gopalakrishnan NCHRP Staff Amir Hanna, Program Officer NCHRP 1-47 Project Panel Kevin Hall, U AR (Chair) Michael Ayers, ACPA Imad Basheer, CA DOT Mohamed Elfino, VA DOT Laura Fenley, WI DOT Geoffrey Hall, MD SHA Kent Hansen, NAPA Tommy Nantung, IN DOT Richard Zamora, CO DOT Tom Yu, FHWA Liaison Stephen Maher, TRB Liaison

3 NCHRP 1-47 Final Report 3

4 4 Webinar Outline Study Objectives Sensitivity Analysis Concepts Design Inputs for Study Global Sensitivity Analysis (GSA) Methodology Results from GSA Key Findings and Conclusions Discussion

5 5 Motivation Research is needed to determine the degree of sensitivity of the performance predicted by the MEPDG to input parameter values for specific climatic region and traffic conditions... Users can [then] focus efforts on those input parameters that will greatly influence the pavement design. NCHRP 1-47 RFP Focus on Inputs under Control of Project Designer Definable Refinable HMA Thickness HMA Stiffness PCC Slab Geometry PCC Strength Base Thickness Base Modulus Subgrade Modulus SWCC Parameters Traffic Volume Vehicle Mix Vehicle Speed

6 QUESTION FOR AUDIENCE 6

7 What is the Mechanistic-Empirical Pavement Design Guide (MEPDG)? 7 Mechanistic Empirical OUTPUT INPUT TRAFFIC CLIMATE PAVEMENT Mechanistic Pavement Analysis Models x y z xy z Empirical Pavement Analysis Models Longitudinal Cracking Alligator Cracking Thermal Cracking Asphalt Concr. Rutting Total Rutting Analytic Transfer Functions International Roughness Index

8 MEPDG Software 8 Not available at time of study Same models Same results Version used for study

9 Research Objective 9 h1 h2 volume, speed SSA Va, Vbe, ν, PG, E*(α, δ), thermal properties, etc. M R ν SWCC P 200, D 60, PI M R ν SWCC (Soil Water Characteristic Curve) P 200, D 60, PI Climate Sensitivity is high Sensitivity is Low Groundwater Depth (Flexible Pavement) Determine the sensitivity of the pavement performance predicted by the MEPDG to variability of the material property design input values for flexible pavements.

10 10 Webinar Outline Study Objectives Sensitivity Analysis Concepts Design Inputs for Study Global Sensitivity Analysis (GSA) Methodology Results from GSA Key Findings and Conclusions Discussion

11 Output y Frequency (SI) Types of Sensitivity Analyses Local Sensitivity Analyses Global Sensitivity Analyses 11 Model: y = f (x) (x 0, y 0 ) SI = f (x 0 + Dx) - f (x 0 - Dx) 2Dx Input x Each input varied one-at-a-time Only evaluates sensitivity around the reference value(s) Ignores input correlations, interactions Employed in most past studies SI,s SI Sensitivity Index SI All inputs varied simultaneously Evaluates sensitivity over the entire problem domain Can include input correlations Can quanitify input interactions Extremely computation intensive

12 Model Output Y 12 Sensitivity Metrics: The Past X1 X2 r=1.0 ρ=1.0 Slope=1.0 Model Output Regression Model Output Y j r=1.0 ρ=1.0 Slope=0.1 Average Slope Instantaneous Slope Model Input X k Correlation Coefficients - Pearson r - Spearman ρ Model Input X Multivariate Linear Regression - Regression coefficients (average slopes) Sensitivity ΔY/ΔX ΔOutput/ΔInput!!

13 Selected Sensitivity Metric: Design Limit Normalized Sensitivity Index (NSI) 13 change in predicted distress j corresponding to change of design input the design limit of the distress j Design Limit Normalized Sensitivity Index for Design Input k and Distress j ( ) S DL jk = DY / DL j j DX k / X k ( ) change in design input k about the baseline baseline value of design input k Example: Total Rutting Distress, Design Limit (DL) =0.75 in. Normalized Sensitivity of Total Rutting to Base M R = What is change in Total Rutting if Base M R is increased by ΔX/X = 20%? Change in Total Rutting ΔY = (-0.50)(20%)(0.75 in) = in.

14 14 Webinar Outline Study Objectives Sensitivity Analysis Concepts Design Inputs for Study Global Sensitivity Analysis (GSA) Methodology Results from GSA Key Findings and Conclusions Discussion

15 Base Cases 15 5 pavement types x 5 climates x 3 traffic levels Total number of base cases = 75 Note: HMA/JPCP on Stiff Foundation is intended to represent both new construction/reconstruction on stabilized foundations or rehabilitation on underlying HMA/PCC layers.

16 Climate Categories 16 Climate Category Location Weather Station Hot-Wet Orlando FL ORLANDO INTERNATIONAL ARPT Hot-Dry Phoenix AZ PHOENIX SKY HARBOR INTL AP Cold-Wet Portland ME PORTLAND INTL JETPORT ARPT Cold-Dry International Falls MN FALLS INTERNATIONAL ARPT Temperate Los Angeles CA LOS ANGELES INTL AIRPORT Months Binder Grade of Data Baseline Range 116 PG PG PG PG PG PG PG PG PG PG PG PG PG PG PG 64-10

17 Binder Grades 17 Climate Baseline PG PGHigh- PGLow+ PGLowTC PGLowTC+ Cold-Dry PG NA PG PG PG Cold-Wet PG NA PG PG PG Hot-Dry PG PG NA NA NA Hot-Wet PG PG NA NA NA Temperate PG PG NA NA NA PGLowTC, PGLowTC+ used for Thermal Cracking sensitivity analyses.

18 Traffic Levels 18 Baseline Inputs Traffic Category AADTT 1 AADTT in Design Lane Est. ESALs (Flexible) 5 Est. ESALs (Rigid) 6 AADTT Range Low 1, M 5M 500-5,000 Medium 7,500 2, M 25M 5,000-10,000 High 25,000 6, M 75M 20,000-30,000 1 Based on MEPDG Interstate Highway TTC4 Level 3 default vehicle distribution. Potential correlations between operating speed and AADTT have been ignored. 2 50% directional split, 2 lanes per direction, 0.75 lane factor for design lane. 3 50% directional split, 3 lanes per direction, 0.55 lane factor for design lane. 4 50% directional split, 3 lanes per direction, 0.50 lane factor for design lane. 5 Based on 15 year design life. 6 Based on 25 year design life.

19 Baseline Flexible Pavement Sections 19 Input Parameter Traffic Level Low Traffic Medium Traffic High Traffic Baseline Min Max Baseline Min Max Baseline Min Max AADTT-Nominal Design Lane HMA Thickness (in) Base Thickness (in)

20 Design Inputs For Evaluation 20 (Nearly) All MEPDG Design Inputs Initial Triage - Literature review - Engineering judgment OAT Local Sensitivity Analyses > Sensitive MEPDG Design Inputs Global Sensitivity Analyses

21 21 Design Inputs Given Special Consideration Unbound Material Properties Gradation, plasticity inputs for soil water characteristic curve (SWCC) correlated to resilient modulus M R HMA Dynamic Modulus Synthetic Level 1 dynamic modulus E* data HMA Low Temperature Properties Creep compliance correlated with stiffness

22 Soil Water Characteristic Curve (SWCC) 22

23 Unbound Material Properties SWCC function of: Percent passing the No. 200 sieve (P 200 ) D 60 gradation parameter Plasticity index (PI) 23 A design input value for M r is selected first and then gradation data giving compatible values of P 200 and D 60 are determined via correlations along with compatible values for PI and LL.

24 Unbound Material Properties 24 P200 D60 %Pass = 100 ( ) e b+g log 10 Size

25 25 HMA Dynamic Modulus Design Input = Synthetic Level 1 Dynamic Modulus Data δ δ+α f η = lower shelf stiffness = upper shelf stiffness = loading frequency (from traffic speed) = binder viscosity (from binder grade, temperature); used to generate synthetic Level 1 G*, ϕ

26 Design Inputs (New HMA) 26

27 27

28 28 Webinar Outline Study Objectives Sensitivity Analysis Concepts Design Inputs for Study Global Sensitivity Analysis (GSA) Methodology Results from GSA Key Findings and Conclusions Discussion

29 29 Global Sensitivity Analyses Procedure Sample the Entire Problem Domain Latin Hypercube MEPDG GSA Simulations Over 40,000 MEPDG runs (Each run takes up to 0.5 hr on a single computer) Respond Surface Modeling (RSM) Continuous response surface model (RSM) for the randomly located GSA simulation results. Sensitivity Index Statistics Based on 10,000 RSM evaluations/base case

30 Sampling with Complete Coverage x 2 30 Monte Carlo n=100 x 2?? x 1 x 2 Latin Hypercube n=25 x 1 Monte Carlo n=25 x 1

31 31 AutoIt Windows Scripting Automates manual entry of input values to MEPDG. Extracts results from MEPDG output spreadsheets lines of codes. Cloud Storage

32 Parallel Processing 32 UMD Pavement Materials Lab ISU Jelling Lab 40,000+ MEPDG Runs

33 Predicted Distresses Span Wide Range 33 New HMA Cold-Wet Climate 1300 Simulations

34 34 Respond Surface Modeling for Derivatives y x Continuous response surface model (RSM) fit to the randomly located GSA simulation results Enables computation of sensitivity index derivatives

35 Respond Surface Modeling Artificial Neural Networks p 1 p 2 p 3 p 23 0,1 iw 1,1 0,1 iw 23,5 Σ Σ Σ Σ Σ b 1 1 b 2 1 b 3 1 b 4 1 b 5 1 n 1 1 n 2 1 n 3 1 n 4 1 n 5 1 f 1 f 1 f 1 f 1 f 1 a 1 1 a 2 1 a 3 1 a 4 1 a 5 1 1,2 iw 1,1 1,2 iw 5,1 Σ f 2 b 1 2 n 1 2 a Climate Type, Traffic Speed, Pavement structure, etc Rutting Depth Input Layer Hidden Layer Output Layer Outputs

36 RSM Predicted Excellent Performance of ANN RSMs 36 Long Cracking Alligator Cracking AC Rutting Total Rutting IRI CD CW T HW HD MEPDG Predicted

37 Spreadsheet for Quick MEPDG Calculation 37

38 38 Webinar Outline Study Objectives Sensitivity Analysis Concepts Design Inputs for Study Global Sensitivity Analysis (GSA) Methodology Results from GSA Key Findings and Conclusions Discussion

39 QUESTION FOR AUDIENCE 39

40 GSA Sensitivity Index Distributions 40 Sensitivity Metric for Ranking: NSI μ+2σ =Mean SI + 2 Standard Deviations New HMA AC Rut Depth All Climate Zones, Traffic Levels 10,000 Simulations w/ Censoring

41 41 Design Input Maximum NSI m+2s Values (ANN RSMs) Long. Alligator Thermal AC Rut Total Rut Crack Crack Crack Depth Depth IRI Max HMA E* Alpha Parameter HMA E* Delta Parameter HMA Thickness HMA Creep Compliance m Exponent N.A. N.A N.A. N.A. N.A Base Resilient Modulus Surface Shortwave Absorptivity HMA Air Voids HMA Poisson s Ratio Traffic Volume (AADTT) HMA Effective Binder Volume Subgrade Resilient Modulus Base Thickness Subgrade Percent Passing No HMA Tensile Strength at 14 o F N.A. N.A N.A. N.A. N.A Operational Speed HMA Creep Compliance D Parameter N.A. N.A N.A. N.A. N.A HMA Unit Weight Base Poisson s Ratio HMA Heat Capacity Subgrade Liquid Limit Binder Low Temperature PG HMA Thermal Conductivity Binder High Temperature PG Subgrade Poisson s Ratio Groundwater Depth Subgrade Plasticity Index Aggregate Coef. Of Thermal Contraction N.A. N.A N.A. N.A. N.A Hypersensitive Very Sensitive Sensitive Non-Sensitive

42 42 Design Input Maximum NSI m+2s Values (ANN RSMs) Long. Alligator Thermal AC Rut Total Rut Crack Crack Crack Depth Depth IRI Max HMA E* Alpha Parameter HMA E* Delta Parameter HMA Thickness HMA Creep Compliance m Exponent N.A. N.A N.A. N.A. N.A Base Resilient Modulus Surface Shortwave Absorptivity HMA Air Voids HMA Poisson s Ratio Traffic Volume (AADTT) HMA Effective Binder Volume Subgrade Resilient Modulus Base Thickness Subgrade Percent Passing No HMA Tensile Strength at 14 o F N.A. N.A N.A. N.A. N.A Operational Speed HMA Creep Compliance D Parameter N.A. N.A N.A. N.A. N.A HMA Unit Weight Base Poisson s Ratio HMA Heat Capacity Subgrade Liquid Limit Binder Low Temperature PG HMA Thermal Conductivity Binder High Temperature PG Subgrade Poisson s Ratio Groundwater Depth Subgrade Plasticity Index Aggregate Coef. Of Thermal Contraction N.A. N.A N.A. N.A. N.A Hypersensitive Very Sensitive Sensitive Non-Sensitive

43 43 GSA Results: New HMA Max NSI Values for Longitudinal Cracking

44 44 GSA Results: New HMA Max NSI Values for Alligator Cracking

45 45 GSA Results: New HMA Max NSI Values for AC Rutting

46 46 GSA Results: New HMA Max NSI Values for Total Rutting

47 47 GSA Results: New HMA Max NSI Values for IRI

48 Most Sensitive Inputs by Property Category: New HMA 48

49 49 Webinar Outline Study Objectives Sensitivity Analysis Concepts Design Inputs for Study Global Sensitivity Analysis (GSA) Methodology Results from GSA Key Findings and Conclusions Discussion

50 Key Findings: Methodology 50 Mean + 2 standard deviations NSI is the best and most robust ranking metric Relates change in design input to corresponding change in predicted distress relative to its design limit Captures both the average sensitivity and its variability Categories: Hypersensitive (HS): NSI μ+2σ > 5 Very Sensitive (VS): 1 < NSI μ+2σ < 5 Sensitive (S): 0.1 < NSI μ+2σ < 1 Nonsensitive (NS): NSI μ+2σ < 0.1 Artificial Neural Network response surface models provide robust and accurate relationships between design inputs and outputs

51 51 Key Findings: General Most NSI frequency distributions showed welldefined single peaks Indicates that NSI did not vary significantly over problem domain For all pavements, design inputs for the bound layers (i.e., HMA) exhibited the highest sensitivities Sensitivity values did not vary substantially or systematically across climate zones

52 52 Key Findings for Designers: Flexible Only HMA properties were most consistently in the highest sensitivity categories: HMA dynamic modulus lower (δ), upper (δ+α) shelves HMA thickness Surface shortwave absorptivity Poisson s ratio Sensitivity values were consistently higher for longitudinal cracking, alligator cracking, and AC rutting than for IRI and thermal cracking Little or no thermal cracking was predicted when correct binder grade used

53 Asphalt Binder Asphalt Mix Asphalt General 53 GSA of MEPDG using Level 3 dynamic modulus data. Level 1 Input Dynamic Modulus at different combination of Temperature and Frequency Level 3 Input Cumulative % Retained 3/4 inch sieve: Cumulative % Retained 3/8 inch sieve: Cumulative % Retained #4 sieve: % Passing #200 sieve: Shear Modulus and Phase Angle at 10 rad/sec at different temperatures Superpave binder grading Effective binder content (%) Air voids (%) Total unit weight(pcf) Poisson's Ratio Thermal conductivity asphalt (BTU/hrft-F) Heat capacity asphalt (BTU/lb-F) Effective binder content (%) Air voids (%) Total unit weight(pcf) Poisson's Ratio Thermal conductivity asphalt (BTU/hrft-F) Heat capacity asphalt (BTU/lb-F)

54 NSI Values (Mean +2 StDev) Level 1 vs. Level 3 HMA E* Level 1 Level

55 55 Additional Key Findings: HMA Over Stiff Thermal conductivity and heat capacity of stabilized base are additional sensitive inputs Not inputs for non-stabilized base in flexible pavements Increased sensitivity of longitudinal cracking to operating speed

56 56 Issues for Further Study Base, subgrade design inputs had significant sensitivities in only a small number of cases realistic? Realism of unexpectedly high sensitivities for some inputs: E* lower (δ), upper (δ+α) shelves Poisson s ratio for HMA Poisson s ratio for base, subgrade HMA unit weight (especially for longitudinal cracking) Replace/eliminate low sensitivity inputs with internal default values?

57 Rutting Depth (in) Sensitivity Analysis in Terms of Service Life 57 Permanent Deformation: Rutting AC Rutting Design Value = 0.25 Total Rutting Design Limit = 0.75 y = x R² = Distress at end of analysis period Pavement Age (month) Pavement Service Life SubTotalAC SubTotalBase SubTotalSG Total Rutting TotalRutReliability Total Rutting Design Limit Power (Total Rutting)

58 58 Overall Conclusions Rigorous sensitivity analysis methodology is essential Very few surprises in results This is a good thing! There were some surprises in results Identify MEDPG areas that may merit further investigation Results provide practical guidance for pavement designer Importance of Level 1 characterization for HMA Results suggest areas where MEPDG/DARWin-ME could be simplified Eliminate some inputs/replace with internal defaults

59 59 Webinar Outline Study Objectives Sensitivity Analysis Concepts Design Inputs for Study Global Sensitivity Analysis (GSA) Methodology Results from GSA Key Findings and Conclusions Discussion

60 Contact Info: Dr. Charles W. Schwartz University of Maryland