Prediction of Rutting in

Alternative AsphaltConcrete

Overlay Methods

JOINT C­SHRP/NEW BRUNSWICK BAYESIAN APPLICATION
Prepared for: Canadian Strategic Highway Research Program (C­SHRP)

Prepared by:
Research Section
Engineering Services Division
New Brunswick Department of
Transportation

October, 1995

EXECUTIVE SUMMARY

This project is a joint Canadian Strategic Highways Research Program (C­SHRP) and a New Brunswick Department of Transportation (NBDOT) application of Bayesian statistical modeling methodology. The objective of this project was to demonstrate to NBDOT an application for Bayesian modeling and not specifically to produce definitive predictive performance models ready for design application.

Originally three models were considered to predict the rutting performance of asphalt concrete (AC) overlay methods used for AC pavement rehabilitation. The overlay methods modeled were thin overlay, thick overlay with padding, and thick overlay with milling. An important aspect of using Bayesian Statistics in this project was to supplement actual data with expert judgement. Judgement on the rutting performance of each type of overlay was solicited from pavement experts within NBDOT. Field data specific to each model was collected and databases were developed. This field data supplemented with the expert judgement were then analyzed using the XLBayes software developed through the Canadian­Long­Term Pavement Performance (C­LTPP) Project of C­SHRP to give first generation models.

After analyzing the first generation models, a second iteration was performed using just two models, thick and thin overlays. The thick overlay with milling and the thick overlay with padding models were combined to give just a thick overlay model.

The following are the resultant model equations for the thin and thick overlay models for the second iteration where V = % Air Voids; R = % Retained on the 4.75mm sieve; A = Age of the overlay in years; C = % Crushed particles; T1 = annual log Traffic measured in KESAL/year; and T2 = cummulative log Traffic measured in KESAL/year

For the Thin overlay
Thin Overlay Equation

For the Thick overlay:
Thick Overlay Equation

It was concluded that Bayesian methodology is a viable tool and has application for NBDOT. It demonstrated that it has the potential to assist in optimizing rehabilitation design of overlays.

Several recommendations for future modeling came out of the study, both in reference to further development of the rutting model and in other areas where expert judgement could be combined with data to get earlier results.

TABLE OF CONTENTS

EXECUTIVE SUMMARY

1.0 INTRODUCTION

2.0 TEAM MEMBERS

3.0 METHODOLOGY

3.1 Step 1­ Decide What You Want to Model

3.2 Step 2 ­ Select Dependent Variable

3.3 Step 3 ­ Select the Model Type

3.4 Step 4 ­ Select Independent Variables

3.5 Step 5 ­ Postulate Functional Form

3.6 Step 6 ­ Assemble Information

3.6.1 Compiling Actual Sample Data Base ­ "Data"

3.6.2 Encoding Expert Judgement to Calculate the "Prior"

3.7 Step 7 ­10 Perform Bayes, Use Models to Predict Performance, Evaluate Model and Iterations

4.0 ITERATIONS

4.1 ITERATION 1

4.1.1 Definition of Model

4.1.2 Dependent Variable

4.1.3 Independent Variables

4.1.4 Model Type and Functional Form

4.1.5 Model Inputs

4.1.5.1 Actual Data ­ "Data"

4.1.5.2 Encoding Expert Judgement Data ­"Prior"

4.1.6 Analysis of Data ­ "Posterior"

4.1.7 Model Runs

4.1.7.1 First Model Run

4.1.7.2 Second Run

4.1.7.3 Third Run

4.1.7.4 Fourth Run

4.1.8 Sensitivity Analysis

4.1.8.1 Building Predictive Cases

4.1.8.1.1 Thin Model

4.1.8.1.2 Thick Overlay With Milling Model

4.1.8.1.3 Thick Overlay with Padding Model

4.1.8.2 Sensitivity of Input Assumptions

4.1.8.2.1 Thin Overlay Model

4.1.8.2.2 Thick Overlay with Milling

4.1.8.2.3 Thick Overlay with Padding

4.1.9. Inference from Analysis of Iteration 1

4.2 ITERATION 2

4.2.1 Definition of Model

4.2.2 Dependent Variable

4.2.3 Independent Variables

4.2.4 Model Type and Functional Form

4.2.5 Model Inputs

4.2.5.1 Actual Data ­ "Data"

4.2.5.2 Encoding ExpertJudgement­ "Prior"

4.2.6 Analysis of Data

4.2.7 Model Runs

4.2.7.1 Thin Model

4.2.7.1.1 First Model Run

4.2.7.1.2 Second Model Run

4.2.7.1.3 Third Model Run

4.2.7.1.4 Fourth Model Run

4.2.7.1.5 Fifth Model Run

4.2.7.1.6 Sixth Model Run

4.2.7.1.7 Seventh Model Run

4.2.7.2 Thick Model

4.2.7.2.1 First Model Run

4.2.7.2.2 Second Model Run

4.2.7.2.3 Third Model Run

4.2.7.2.4 Fourth Model Run

4.2.7.2.5 Fifth Model Run

4.2.7.2.6 Sixth Model Run

4.2.7.2.7 Seventh Model Run

4.2.7.2.8 Conclusion

4.2.8 Sensitivity Analysis

4.2.8.1 Building Predictive Cases

4.2.8.1.1 Thin Overlay Model

4.2.8.1.2 Thick Overlay Model

4.2.8.2 Sensitivity of Input Assumptions

4.2.8.2.1 Thin Overlay Model

4.2.8.2.2 Thick Overlay Model

4.2.9 Inference from Analysis of Iteration 2

4.2.9.1 Thin Model

4.2.9.2 Thick Model

5.0 RESULTS

5.1 ITERATION 1

5.1.1 Thin Overlay

5.1.2 Thick Overlay with Milling

5.1.3 Thick Overlay with Padding

5.2 ITERATION 2

5.2.1 Thin Overlay

5.2.2 Thick Overlay

6.0 DISCUSSION AND RECOMMENDATIONS

REFERENCES

APPENDICES

(Continue)

Return to Main Page