Local Calibration of MEPDG Permanent Deformation Models For Ontarios Flexible Pavements

Presented By:

Afzal Waseem

Supervisor:

Dr. Arnold Yuan

1

Outline
Introduction What is the MEPDG ? What is the Problem ?
The Local Calibration Methodology Reconstructed Sections Rehabilitated Sections Conclusions & Recommendations
2

Introduction
To develop a Local Calibration Methodology Local calibration based on limited PMS database Multiple local optima during local calibration

Local Calibration of Reconstructed and Rehabilitated Flexible Pavement Sections

Pavement Design
AASHTO (1993): Conventionally pavements are design based on AASHTO design guide. Empirical Design Method Only 9 Inputs (ESAL, Mr, Pt, a, m) Long term effects on pavements not considered.

What is MEPDG Method?

MEPDG: Mechanistic-Empirical Pavement Design Guide method

INPUTS

MECHANISTIC

EMPIRICAL

RESULT

1. 2. 3. 4.

AC bottom-up cracking AC top-down cracking AC thermal facture Total Cracking

5. Permanent deformation
6. International Roughness Index

Inputs
Location: Latitude, Longitude & Elevation Climate Information Air Temperature Precipitation Moisture content Sunshine Wind speed Freezing Index
Wind speed

Sunshine

Precipitation Moisture content Air Temperature

Vehicle Inputs Axle spacing Axle configuration Traffic Wander Axle width Dual tire spacing Tire Pressure Operational speed Vehicle class AC Layer Unit weight Air void Effective binder content Poissons Ratio Tensile Strength Thermal conductivity Heat capacity Sieve Passing

Traffic Information AADTT Growth rate Truck distribution Truck traffic classification No. of Lanes Monthly and Hourly Adjustment factor

W/T depth

(Waseem 2013)

Base, Sub-base and subgrade Layers Thickness Poissons Ratio Resilient Modulus Gradation

Permanent Deformation
Permanent deformations can be defined as the combination of transverse undulations, longitudinal depressions and distortions on the surface of the pavement.
a) Surface Rutting

Surface rutting is distributed throughout the layers of pavement

Layer contribution to total surface rutting depends on properties of materials
Asphalt Concrete

Granular Layer Subgrade Layer

b) Trench Study

Rutting Models in DARWin-METM

Three models with 5 local calibration parameters are used to predicts the total rutting in a pavement. = 101 2 3 1. AC Model
= HMA rutting multiplier = Temperature or Climate dependent factor = Load repetitions or Traffic dependent factor

2.

Granular Base / Sub-base Model

= 1

= Granular base rutting multiplier

3.

Subgrade Model = Subgrade rutting multiplier

= 1

Why not Default Models ?

25 20 Observed Rutting (mm) 15 10 5 0 0 5 10 15 20 Predicted Rutting (mm) 25
y = 0.1483x + 4.6263 R = 0.1224 Se=1.786

(Jannat, 2012)

= = = = = 1 Default models are evaluated using pavement sections all over Ontario. Over-prediction of rutting for Ontario. 9

Local Calibration
Calibration is a systematic process to eliminate bias and minimize the residual error between observed and predicted values of a certain distress.

Observed Observed 8 8 7 7 Rutting (mm) 6 6 5 5 4 4 3 3 2 2 1 1 0 0 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Before After Calibration

2007

=
=1

Pavement Pavement Service Service life life (years) (years)

LONGITUDNAL CALIBRATION
Instead of only calibrating the end of cycle distress values, we calibrate the predicted distress curve over time for each pavement sections.

2008

10

Pavement Scenario Analysis & Automation

DARWin-METM official AASHTO software was used for MEPDG analysis of pavements. Local calibration process was automated. DARWin-METM project input file saved with .dgpx extension. Coded in an Extensible Markup Language (XML) the input file is both human-readable and machine-readable. Macros were developed in Microsoft Excel to automate the process of generating desired files and reports. Saves time in local calibration to run hundreds of variations of same project with different set of calibration parameters.
11

Pavement Sections

Locations of the 51 flexible pavement sections on Google Maps

12

Database
1. MEPDG Default values Guide for Ontario recently developed by the MTO staff.
2. MEPDG project inputs used in this research was collected from MTO Pavement Management System (PMS-2) Database.

3. PMS-2 Database is limited to total rutting.

AC Rutting

Granular Rutting

Subgrade Rutting

13

Multiple Local Optima

Extensive computer iterations were analysed. Calibration Coefficients AC1,AC2 & AC3 = AC Layer = Granular base = Subgrade Calibration against total rutting. SG

Multiple local optimums were found.

GB
AC AC1=0.3, AC2=1, AC3=1

14

Multiple Local Optima

At least 17 local optima were found
12 Total rut depth (mm) 10

8
6 4 2 0 1996

Observed (0.6, 1.0, 0.2) - 3.89 (0.3, 0.9, 0.3) - 3.46 (0.4, 0.3, 0.4) - 3.46 (0.35, 0.6, 0.35) - 3.45 (0.2, 0.1, 0.5) - 3.45
1998 2000 2002 Year 2004 2006 2008 2010

5 local optima showed similar rut predictions along time Calibration coefficients differ from each other Individual layer calibration is needed
15

Layers Contribution to Rutting

Layer contribution to the total surface rut depth, How much percentage of the total rutting comes from the AC surface layer, base layer, sub-base layer, and the subgrade soil?

Reduce the inherent indeterminacy of the permanent deformation models.

Unique determination of the five local calibration parameters can be expected.

16

Layers Contribution to Rutting

Layer Contribution in Ontario has not been found in Literature. 1. Empirical Studies 2. Findings from DARWin-METM Global Models 3. Layers Contribution to Rutting from Software Packages
80 Percentage Rutting (%) 70 60 50 40

32 %

58 %

9%

Reconstructed AASHO Sections

ALF-FHWA (Thin) ALF-FHWA (Thik) SPS-1 ALF-TxMLS

30
20 10 0 AC Base

Global Models Subgrade

AC Layer =50 %

Granular Layer =50 %

Subgrade Layer = 0 %

Rehabilitated Sections

17

Local Calibration Methodology

Rutting layer contribution studies were used to calculate observed rutting in AC, Granular and Subgrade layer.
Granular ( ) and Subgrade ( ) layer calibration by adjusting initial point obtained from: where = = Measured layer rut i = Calculated layer rut from the

Rutting model for Granular ( ) and Subgrade ( ) layer are calibrated.

AC layer layer calibration by adjusting initial point obtained in a similar manner. Initial Iterate for Calibration AC layer , = , = of AC Model

default global model.

18

Local Calibration Methodology

Calibration of AC Rutting Model (3 calibration factors)
AC1 = multiplier coefficient AC2 & AC3 = exponential coefficients

1. Full Factorial Sampling of 3 factors in 3 levels

DARWin-METM

2. Response Surface Model used as surrogate model.

full quadratic polynomial function
AC1 , AC2 , AC3 = 1 + 2 AC1 + 3 2 + 4 3 + 5 1 2 + 6 2 2 + 7 3 2 + 8 1 2 + 9 2 3 + 10 1 3

MATLAB

3. Minimization of unconstrained multivariable function

min ()

4. Termination Criteria
19

Local Calibration
1. Section-by-section Calibration
o Calibration section was individually calibrated for the available rut points throughout the service life of the pavement which is called the section-by-section longitudinal calibration.

2. Pooled Calibration
o The rutting RSS of all sections are pooled together to calculate the total RSS which is then minimized to get a single set of calibration parameters. o For comparison purpose, a calibration was also performed. pooled longitudinal
20

Local Calibration Results: Local Calibration of Reconstructed Sections

RECONSTRUCTED SECTIONS
Ten reconstructed flexible pavement sections.

Section-by-section Longitudinal Calibration Parameters:

Rutting Model I.D.
9 43 191 376 1049 1053 1188 1189 1200 1311 0.14 0.16 0.29 0.31 0.2 0.26 0.37 0.47 0.37 0.34 AC 1.738 1.091 0.975 1.2 0.991 1.262 1.04 1.13 1.48 1.182 0.229 0.92 1.105 0.835 0.985 0.719 0.97 0.87 0.58 0.892 Granular GB 3.06 3.17 1.57 0.41 0.96 0.63 1.11 1.27 1.61 0.66 Subgrade SB 0.033 0.033 0.039 0.025 0.037 0.016 0.04 0.041 0.05 0.022 AC 0.063 0.098 0.022 0.309 0.257 0.097 1.064 0.383 0.454 1.105

RSS
Granular 0.2378 0.2281 0.2128 1.0798 0.7121 0.3104 0.4038 0.2239 1.3401 2.8296 Subgrade 0.0074 0.0051 0.0047 0.0243 0.0163 0.007 0.0082 0.0055 0.0363 0.063

Total Bias
0.12 -0.075 -0.002 0.038 -0.018 0.055 -0.737 -0.303 -0.319 -0.158

Total RSS
0.6601 0.5649 0.4363 2.6932 1.9728 0.8067 2.9352 1.0645 3.4449 8.2359

Comparison with Global Calibration Statistics:

Parameters Number of data points Standard deviation (mm) 2 Section-by-section local calibration 68 0.58 0.887 Pooled local calibration 68 2.22 0.016 Global calibration (NCHRP 2004) 387 3.07 0.399 Global calibration (AASHTO 2008) 334 2.72 0.577

21

Local Calibration Results:

11 10 9 8 Observed Rutting [mm] 7 6 5 4 3 2 1 Observed Rutting (mm) I.D. # 9 I.D. # 43 I.D. # 191 I.D. # 376 I.D. # 1049 I.D. # 1053 I.D. # 1188 I.D. # 1189 I.D. # 1200 I.D. # 1311 0 1 2 3 4 5 6 7 8 Predicted Rutting [mm] 9 10 11 11

RECONSTRUCTED SECTIONS
(a)
R2 = 0.8824 N = 68 Se = 0.583 10 9 8 7 6 I.D. # 9 I.D. # 43 I.D. # 191 I.D. # 376 I.D. # 1049 I.D. # 1053 I.D. # 1188 I.D. # 1189 I.D. # 1200 I.D. # 1311 0 1 2 3 4 5 6 7 8 Predicted Rutting [mm] 9 10 11 R2 = 0.016 N = 68 Se = 2.217

(b)

5
4 3 2 1 0

After-calibration prediction versus observation plots. (a) section-by-section calibration; (b) pooled calibration. 22

Local Calibration Results:

Observed Section-by-Section Pooled 6 5 Rutting [mm] 4

RECONSTRUCTED SECTIONS
2
Observed Section-by-section Pooled

1.5 Rutting [mm]

3
2

0.5

(a) AC Layer Rutting

0

(b) Granular Layer Rutting

2008 0

2004

1997

1998

1999

2000

2001

2002

2003

2005

2006

2007

2001

1997

1998

1999

2000

2002

2003

2004

2005

2006

2007 2007

Pavement Service life [years] 0.6 0.5 Rutting [mm] 0.4 0.3 0.2 0.1 Rutting [mm] Observed Section-by-Section Pooled

Pavement Service life [years] Observed 8 7 6 5 4 3 Section-by-section Pooled

2
1

(c) Subgrade Layer Rutting

2004
1997 1998 1999 2000 2001 2002 2003 2005 2006 2007 2008 Pavement Service life [years]

(d) Total Rutting

1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2008 Pavement Service life [years]

2008

23

Local Calibration Results:

Pre-overlay Rut:

REHABILITATED SECTIONS

The pre-overlay rut is an important input parameter which has significant influence on the calculation of total surface permanent deformation of rehabilitated sections. 1. Frequency Distribution of the pre-overlay rut of 1175 flexible pavement sections
gave 7 mm average value.

2. Regression Analysis of pre-overlay rut showed poor correlation with DMI and RCI

3. Scenario Analyses of rehabilitated pavement sections confirmed a pre-overlay rut

value of between 7 mm and 8 mm is optimum.

Pre-overlay rut value of 7mm was used for rehabilitated pavement sections whose pre-overlay rut value was unavailable.
24

Local Calibration Results:

S. No.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

REHABILITATED SECTIONS
I.D.
139 217 347 348 349 350 353 356 357 358 361 377 378 379 386 803 811 951 981 1.01 0.9 1.01 0.276 1.05 1.1 0.49 0.27 0.98 0.225 0.366 0.44 0.3 1.01 0.295 0.356 0.4 0.287 1.25

Rutting Model AC
1.23 1.23 1.23 1.601 1.255 1.155 1.235 0.935 1.055 1.447 1.408 1.175 1.08 1.23 1.166 1.115 1 1.157 1.17

Granular
1.09 1.274 1.86 1.124 1.297 1.308 1.303 0.713 1.337 0.726 0.614 0.853 0.985 1.336 0.804 1.14 2.245 1.237 0.641

Layers RSS AC
0.982 0.299 0.444 0.22 0.238 0.467 0.242 0.324 0.109 0.257 0.181 1.067 1.071 0.571 0.296 0.245 0.085 0.567 1.878

0.77 0.77 0.77 0.581 0.745 0.85 0.865 1.185 0.945 0.774 0.721 0.925 1.13 0.77 0.989 0.935 1 0.983 0.82

Granular
0.52 0.215 0.241 0.362 0.419 0.962 1.463 0.757 0.525 0.538 0.277 2.064 6.596 1.078 0.762 0.112 0.112 0.391 2.731

Total Total RSS Bias

-0.28 -0.152 -0.318 0.029 -0.143 0.044 -0.133 -0.015 0.145 -0.053 0.073 0.296 0.429 -0.186 0.114 -0.072 -0.059 -0.272 0.012 2.804 0.83 0.794 0.383 0.437 1.624 1.515 1.914 0.86 0.762 0.644 5.735 9.764 2.386 1.863 0.426 0.325 0.951 6.331

25

Local Calibration Results:

13 12 11 10 Observed Rutting (mm) 9 8 7 6 5 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Predicted Rutting (mm)

REHABILITATED SECTIONS
I.D. # 139 I.D. # 217 I.D. # 347 I.D. # 348 I.D. # 349 I.D. # 350 I.D. # 353 I.D. # 356 I.D. # 357 I.D. # 358 I.D. # 361 I.D. # 377 I.D. # 378 I.D. # 379 I.D. # 386 I.D. # 803 I.D. # 811 I.D. # 951 I.D. # 981 13 12 11 10 Observed Rutting (mm) 9 8 7

R2 = 0.967 N = 150 Se = 0.520

R2 = 0.339 N = 150 Se = 1.756

6
5 4 3 2 1 0 0 1 2 3 4 5 6 7 8 9 Predicted Rutting (mm) 10 11 12

I.D. # 139 I.D. # 217 I.D. # 347 I.D. # 348 I.D. # 349 I.D. # 350 I.D. # 353 I.D. # 356 I.D. # 357 I.D. # 358 I.D. # 361 I.D. # 377 I.D. # 378 I.D. # 379 I.D. # 386 I.D. # 803 I.D. # 811 I.D. # 951 I.D. # 981

(a) Section-by-section Calibration

(b) Pooled calibration

26

Summary
1. Developed two stage layer-by-layer, section-by-section local calibration methodology. o Multiple Local Optimums o Layer Contribution studies
2. Automated methodology using DARWin-METM coupled with Excel Macros. 3. Local Calibration of 9 Reconstructed Sections. 4. Local Calibration of 19 Rehabilitated Sections. o Pre-overlay study
27

Conclusions
1. For local calibration with PMS data, layer-by-layer calibration has to be done, otherwise there would be multiple local optima. 2. Based on the layer contribution of rutting from AASHO study, section-by-section calibration has higher accuracy than pooled calibration but pooled calibration is still better than the results from global calibration. 3. The pooled calibration parameters ( = . , = . , = . , = . and = ) of rehabilitated pavement sections has better R2 (0.464) and standard error (1.608 mm) than that obtained during calibration of global rutting models. 4. The level-3 pre-overlay rut input value of 7 mm is proposed.
28

Recommendations
1. Layer contribution study for flexible pavement sections in Ontario and its sensitivity to major input parameters is desperately needed. 2. New pavement sections should be included in this database before recalibration of MEPDG rutting model for Ontario. Similarly information for more reconstructed sections should be collected to enhance the local calibration study. 3. It is recommended that MTO should collect more rehabilitated pavement sections whose performance history and pre-overlay rut information is available.

29

References
Waseem, A. (2013). "Methodology Development and Local Calibration of MEPDG Permanent Deformation Models for Ontario's Flexible Pavements." MASc Civil Engineering, Ryerson University, Toronto. Waseem, A., Yuan, X.-X., and Shehata, M. (2013). "Evaluation of Pre-overlay Rut with MEPDG Rutting Model for Flexible Pavements using Ontario's Long Term PMS Data." CSCE 3rd Specialty Conference on Engineering Mechanics and Materials, Montreal, Quebec Waseem, A., and Yuan, A. (2013). "Longitudinal Local Calibration of MEPDG Permanent Deformation Models for Reconstructed Flexible Pavements Using PMS Data." 8th International Conference on Road and Airfield Pavement Technology Taipei, Taiwan. http://www.scribd.com/doc/146792015/Met hodology-Development-and-LocalCalibration-of-MEPDG-PermanentDeformation-Models-for-Ontario-s-FlexiblePavements-Thesis-by-Afzal-Waseem http://www.scribd.com/doc/149970728/Ev aluation-of-Pre-overlay-Rut-Depth-forLocal-Calibration-of-the-MEPDG-RuttingModel-in-Ontario-Pavements

http://www.scribd.com/doc/149984444/Lo ngitudinal-Local-Calibration-of-MEPDGPermanent-Deformation-Models-forReconstructed-Flexible-Pavements-UsingPMS-Data
30

Questions

32