The Importance of Local Calibration of the MEPDG

World of AsphaltAsphalt ConferenceCharlotte, NCMarch 13, 2012

Kevin D. Hall, Ph.D., P.E.University of Arkansaskdhall@uark.edu

Primary Objectives

• Verify the applicability of the MEPDG for local conditions

• Calibrate to local conditions

• Reduce bias and standard deviation for distress prediction models

Calibration of MEPDG

(adapted from FHWA)

Concept of Model Calibration

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4

Pred

ictio

n (in

.)

Measurement (in.)

Before Calibration

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4Pr

edic

tion

(in.)

Measurement (in.)

After Calibration

Bias and Error Concept

“Actual” Distress

“Pre

dict

ed”

Dis

tres

s

BIASERROR

National Calibration

(LTPP sections used in the national calibration for new flexible pavement)(from MEPDG Users’ Guide,2004)

Recommendations(from National Calibration)

• Available data include LTPP and PMS.

• Some distress data formats may need to be converted.

• Flexible pavement longitudinal cracking and transverse cracking are questionable.

• JPCP is good for use.

• Calibration with more data is recommended.

Across the U.S…

Stage of Implementation StatesNot considering at allIn planning stagesIn process Arkansas, California, Florida,

Mississippi, New MexicoFinished some models North Carolina, Washington, Montana,

Minnesota, Texas, Kentucky, OhioAlready finished all modelsMEPDG is in use Missouri, Indiana

(As of July 2010)

Local Data Sources

• Long Term Pavement Performance (LTPP)– General Pavement Studies (GPS)– Specific Pavement Studies (SPS)

• Pavement Management System (PMS)– TOP25 for flexible pavements

LTPP sites in Arkansas

GPS

SPS

TOP25 from AHTD

Validation and Calibration of New Flexible Pavement

Pavement Sections38 Sections• 18 LTPP• 20 TOP25

Experimental DesignBase Type

Hot-Mix Asphalt (HMA) Thickness

No. of sectionsThin

(≤4 in.) Intermediate Thick (≥8 in.)

Unbound base

0113,0114,0804, 070079G,070079A, 070079P

R20149G,R20149P,090001G, 090001A,090001P,090048G, 090048A,090048P,070018G,

070018A,070018P

17

ATB 0803

0115,0116,0117,0118,01190120,0121,0122,0123,0124,

R80065G,R80065A, R80065P,R50067G, R50067A,R50067P

17

CTB 2042,3048, 3058,3071 4No. of

sections 1 26 11 38

Underlined sections are randomly selected for validation; G is Good; A is Average; P is PoorATB: Asphalt Treated Base; CTB: Cement Treated Base.

Measured Performance

Predicted Performance

Design Criteria

Comparison (uncalibrated)

Comparison (calibrated)

Measured Performance

Predicted Performance

Design Criteria

Alligator Cracking

0

5

10

15

20

25

30

35

40

0 10 20 30 40

Pred

icte

d al

ligat

or c

rack

ing,

%

Measured alligator cracking, %

LTPPPMSEqualityRegression

y = 0.1190 xR2 = 0.09

0

5

10

15

20

25

30

35

40

0 10 20 30 40Pr

edic

ted

allig

ator

cra

ckin

g, %

Measured alligator cracking, %

LTPPPMSEqualityRegression

y = 0.2172 xR2 = 0.22

Determined in accordancewith LTPP Distress IdentificationManual

Alligator Cracking

0102030405060708090

100

5 10 15 20 25 More

Freq

uenc

y, %

Alligator Cracking, %

Histogram of alligator cracking (Validation)

MeasuredPredicted

0102030405060708090

100

5 10 15 20 25 MoreFr

eque

ncy,

%Alligator Cracking, %

Histogram of alligator cracking (Calibrated)

MeasuredPredicted

Longitudinal cracking

0102030405060708090

100

1000 2000 3000 4000 5000 More

Freq

uenc

y, %

Longitudinal cracking (ft/mi)

Histogram of Longitudinal Cracking (Validation)

Measured

Predicted

0

1000

2000

3000

4000

5000

6000

7000

0 2000 4000 6000

Pred

icte

d lo

ng. c

rack

ing,

ft/m

i

Measured longitudinal cracking, ft/mi

LTPPPMSEquality

Longitudinal cracking recorded in wheelpath only

Transverse Cracking

0102030405060708090

100

500 1000 1500 More

Freq

uenc

y, %

Transverse Cracking, ft/mi

Histogram of Transverse Cracking (Validation)

Measured

Predicted

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 500 1000 1500 2000 2500 3000

Pred

icte

d tr

ansv

erse

cra

ckin

g, ft

/mi

Measured transverse cracking, ft/mi

LTPP PMS

No determination maderegarding ‘cause’ ofcracking

Total Rutting

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.0 0.1 0.2 0.3 0.4 0.5 0.6

Pred

icte

d ru

ttin

g, in

ch

Measured rutting, inch

LTPPPMSEqualityRegression

y=1.2433 xR2=0.87

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.0 0.1 0.2 0.3 0.4 0.5 0.6

Pred

icte

d ru

ttin

g, in

chMeasured rutting, inch

LTPPPMSEqualityRegression

LTPP: dipstick or photographic

Ark: straightedge

Smoothness (IRI)

0

10

20

30

40

50

60

40 60 80 100 120 140 More

Freq

uenc

y, %

IRI, in/mi

Histogram of IRI (Validation)

Measured

Predicted

40

60

80

100

120

140

40 60 80 100 120 140

Pred

icte

d IR

I, in

/mi

Measured IRI, in/mi

LTPP PMSEquality Regression

y=1.2606 x R2=0.93

In MEPDG, IRI is a function of the otherpavement distresses;

MEPDG default Initial IRI: 63 in/mileArkansas initial IRI (LTPP and Top 25): 70 in/mile

Results(Calibration Coefficients)

Calibration Factor Default CalibratedAlligator cracking

C1 1.0 0.688

C2 1.0 0.294C3 6000 6000

AC ruttingβr1 1.0 1.2βr2 1.0 1.0βr3 1.0 0.8

Base ruttingβs1 1.0 1.0

Subgrade ruttingβs2 1.0 0.5

Validation(Calibrated models)

• 20% of the 38 sites (8 sites)

0

5

10

15

20

25

30

35

40

0 10 20 30 40

Pred

icte

d A

lliga

tor

Cra

ckin

g (%

)

Measured Alligator Cracking (%)

LTPP

PMS

Equality

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.0 0.1 0.2 0.3 0.4 0.5 0.6

Pred

icte

d ru

ttin

g, in

ch

Measured rutting, inch

LTPP

PMS

Equality

Statistical Analysis

Alligator cracking Total rutting

MeasuredBefore

calibrationAfter

calibration MeasuredBefore

calibrationAfter

calibrationAverage 2.0688 0.5512 2.0070 0.1945 0.2586 0.1852Standard deviation 4.9029 2.0661 1.4456 0.0670 0.0885 0.0628p-value(t-test) N/A 0.0000

0.8160(>0.05) N/A 0.0000

0.0552(>0.05)

p-value(F-test) N/A 0.0000 0.0000 N/A 0.0000

0.2176(>0.05)

N 371 371 371 363 363 363

Alligator cracking model and rutting models are improved.

Points to Ponder

• Start at the bottom: get “Level 3” established before $$ for Level 1.

• Although BIAS and/or ERROR are reduced, calibrated models will not be ‘perfect’.

• Calibration/Validation is an ONGOING effort.– Major emphasis in the Pvmt Mgmt section– Not all data is quality data: be intentional!

• Understand the SENSITIVITY of the models– New models are out there…

Acknowledgements• Mark Evans• Mark Greenwood• David Ross• Jacqueline Hou• Bobbie Jordan• Elizabeth Mayfield-Hart • Thomas Greg Nation• Sarah Tamayo• Tymli Frierson• Andrew Warren• Elisha Wright-Kehner• Claude Klinck

Dr. Robert P. Elliott Danny X. Xiao Daniel Byram Dr. Vu Nyguen Dr. Qiang Li Dr. Yangsheng Jiang Wenting Luo Lin Li