Evaluation of Rutting in

Nova Scotia's Special "B"

Asphalt Concrete Overlays

JOINT C­SHRP/NOVA SCOTIA BAYESIAN APPLICATION
Prepared for: Canadian Strategic Highway Research Program (C­SHRP)

Prepared by:

Anthony P. Ramia, P.Eng.

Graduate Student, Technical University of Nova Scotia

Project Engineer, Maritime Testing Limited, Dartmouth, NS

 

Nouman Ali Ph.D., P.Eng.

Associate Professor, Technical University of Nova Scotia

 

Kent Speiran, M.Eng., P.Eng.

Planning Division, Nova Scotia Department of Transportation and Communications

 

October, 1995

 

EXECUTIVE SUMMARY

The increase of truck traffic on Nova Scotia's 100 series highways has resulted in an increase of wheel path rutting. To remedy this problem, in 1989 the Nova Scotia Department of Transportation and Communications (NSDOT&C) introduced a high friction Type "B" asphalt concrete (AC) mix, more commonly referred to as Special "B". In this paper an empirical model is developed to evaluate the performance of Special "B" as a rut resistant AC mix. The model is developed using the Bayesian regression software and methods developed by the Canadian Strategic Highway Research Program (C­SHRP).

In addition to developing this technique, C­SHRP has also successfully utilized Bayesian regression methodology to develop several pavement distress models, including one for rutting. The Bayesian methodology allows for the inclusion of subjective data, referred to as prior data, from experienced engineers, as well as field data in its modelling approach. The resultant model from the data is referred to as the "posterior".

For this project a team consisting of members of NSDOT&C, the Technical University of Nova Scotia (TUNS) and consultant from private industry was formed to transfer the technology of the Bayesian C­SHRP project in order to develop a rutting model for Special "B". The field data was gathered from a data base of C­SHRP test sections in Nova Scotia. Five expert pavement engineers from both NSDOT&C and private industry were solicited to provide the subjective data. The "XLBAYES" software program, developed for the Bayesian C­SHRP project was utilized for all analyses. A linear and linear­log functional form were investigated for this project. Each expert predictions were analysed individually for each functional form. Subsequently, the encoded judgements of the individual experts were combined into a "group prior" and a single posterior regression model was developed for each of the functional forms implemented in this modelling project.

This report is the result of NSDOT&C's participation in the "Joint C­SHRP/Agency Bayesian Applications" project in which eight of the provincial highway agencies attempted to solve a pavement ­ related problem within the province using the C­SHRP developed Bayesian modelling methodology and software.

ACKNOWLEDGMENTS

The authors of this report would like to express their gratitude and appreciation to Gerard Lee, Romeo Poirier and Eric Theriault for their time and valuable contribution to the development of this project. A special thanks to John Archibald, Stuart Clare, Jim Edwards, Frank Gervais, and Jim Talbot for their time and participation in this project, and for their valuable input as pavement engineers. Finally, we would like to thank Luc Frechette of C­SHRP for initiating this joint application project and Lyle Kajner and Mark Nickeson of Vemax Management Inc. for their training and consultation throughout this project.

TABLE OF CONTENTS

1.0 INTRODUCTION

1.1 Project Objectives

1.2 Bayesian Methodology

2.0 PROJECT TEAM

2.1 Model Selection

2.2 Dependent Variable

2.3 Independent Variables

2.4 Model Type and Functional Form

3.0 C­SHRP TEST SECTIONS

3.1 Test Site Cross Sections

3.2 Nova Scotia's Special "B" AC Mix

3.3 Expert Judgement

3.4 MODEL INPUTS

3.4.1 NEW DATA

3.4.2 Data Quality Assurance

3.4.3 Prior Data

4.0 MODEL RUNS

4.1 Linear Model

4.1.1 Sensitivity Analyses ­ Model 1

4.2 Combined Linear Model :

4.2.1 Sensitivity Analysis ­ Model 1A

4.3 Linear­Log Model

4.3.1 Sensitivity Analysis ­ Model 2

4.4 Combined Linear­Log Model

4.4.1 Sensitivity Analysis ­ Model 2A

4.5 Comparison of Models

5.0 SUMMARY

5.1 Discussion and Conclusions

 

REFERENCES

LIST OF TABLES

Table 1 Preliminary List of Independent Variables

Table 2 Layer Description of Pavement Structure

At Highway 102 C­SHRP Test Site9

Table 3 Layer Description of Pavement Structure At

Highway 103 C­SHRP Test Site

Table 4 Rehabilitation Strategies for Nova Scotia's C­SHRP Test Sections

 

Table 5 NSDOT&C Specifications on Gradation of CombinedAggregates for Types "B" and "B­HF"

Table 6 Summary of Model 1 for All Experts

Table 7 Model 1 ­ Comparison of Regression Coefficients Across Experts and Data Set

Table 8 Comparison of Model 1 Statistics

 

Table 9 Sensitivity of Predictions ­ Posterior Model 1

 

Table 10 Summary of Combined Model 1A

 

Table 11 Summary of Regression Statistics ­ Model 1A

 

Table 12 Sensitivity of Predictions ­ Posterior Model 1A

 

Table 13 Summary of Model 2 for All Experts

 

Table 14 Comparison of Regression Coefficients

Table 15 Summary of Model Statistics

 

Table 16 Summary of Prediction Cases ­ Posterior Model 2

 

Table 17 Combined Linear­Log Model 2A

 

Table 18 Summary of Regression Coefficients ­ Model 2A

Table 19 Sensitivity of Predictions ­ Posterior Model 2A

Table 20 Summary of Developed Combined Models

 

Table 21 Summary of Statistics for Combined Models

Table 22 Wheel Path Rutting Severity Guide

LIST OF FIGURES

Figure 1 Flow Chart of the Bayesian Methodology

Figure 2 Pavement Structure Cross Section At Hwy 102, Nova Scotia, Prior to C­LTPP Rehabilitation Strategy

Figure 3 Pavement Structure Cross Section At Hwy 103 Nova Scotia, Prior to C­LTPP Rehabilitation Strategy

 

Figure 4 Sensitivity Analysis of Predictions ­ Prior Model 1

 

Figure 5 Sensitivity Analysis of Predictions ­ Posterior Model 1

Figure 6 Sensitivity of Posterior Standard Error Prior DOF Estimate ­ Model 1 .

 

Figure 7 Sensitivity of Posterior Standard Error to Prior Standard Error Estimate ­ Model 1

 

Figure 8 Sensitivity Analysis of Predictions ­ Model 1A

Figure 9 Sensitivity of Posterior Residual Variance to Prior Residual Variance Estimate

Figure 10 Sensitivity of the Posterior Residual Variance to Prior DOF Estimate ­ Model 1A

 

Figure 11 Sensitivity Analysis of Prior Predictions ­ Model 2 34

Figure 12 Sensitivity Analysis of Predictions ­ Posterior Model 2

Figure 13 Sensitivity Analysis of Posterior Standard Error to Prior DOF Estimate ­ Model 2

Figure 14 Sensitivity Analysis of Posterior Error to Prior Standard Error Estimate ­ Model 2

 

Figure 15 Sensitivity of Predictions ­ Model 2A

Figure 16 Sensitivity of Posterior Residual Variance to Prior DOF Estimate ­ Model 2A

Figure 17 Sensitivity of Posterior Residual Variance to Prior Residual Error Estimate ­ Model 2A

 

(Continue)

Return to Main Page