Estimation of Latent Pavement Properties Using Condition Survey Data
Final Report - Summary
Samer M. Madanat
Department of Civil and Environmental Engineering
University of California
Berkeley, CA 94720-1720
University of California Transportation Center Year 12 (1999-2000)
1. Background
Pavement maintenance and rehabilitation, and the prediction of future pavement life are heavily dependent on data describing the structure and condition of each pavement segment in the road network. Pavement data for these is typically based on four measures: (1) surface distress and rutting; (2) longitudinal profile (roughness); (3) structural capacity; and (4) skid resistance. These measures were originally developed in conjunction with technologies and surveying systems, which are now over 20 years old. A number of these technologies involve lane closures for detailed pavement measurements, such as visual observation and manual measurement of cracks, and the taking of pavement cores.
Over the past 10-15 years there has been a rapid introduction of continuous automated high-speed pavement condition surveying systems. Commercial systems are now available which can continuously measure and record the following parameters: the type, degree, and intensity of cracking; the width, depth, and profile of rutting; and the pavement layer thickness. Systems for evaluating pavement deflections at driving speeds are under development, and prototype systems are being evaluated. These high-speed surveying systems now provide 100% pavement coverage with a level of detail that was previously unattainable.
For example, several vendors have been heavily involved with the applications of ground penetrating radar for the evaluation of pavement layer thickness. The data are collected at normal driving speed. Similar equipment for measurement of rut depth and roughness at highway speed has been available on the market for 10 years. More recently, equipment and processing techniques have become available for continuous evaluation of surface distress such as different types of pavement cracking. Each type of equipment can produce some measure of pavement condition as a continuous function of distance.
2. Problem Statement
Up until recently, each data source has been used separately to evaluation one aspect of the pavement’s behavior. The goal of the research described herein was to utilize all of the data in a way that provides a more complete picture of the pavement’s current condition and future performance. With such a picture, maintenance and rehabilitation strategies and activities can truly address the real pavement conditions, and can be designed to optimize the remaining life of the pavement at minimal life cycle cost.
The use of high-speed sensor technology by highway agencies has been limited because there is inadequate means for making use of the large volume of data that they generate. The data that is collected is aggregated into traditional condition and performance indices such as the PCI and the IRI which are then used as a basis for pavement and maintenance management decisions. These indices, although simple and widely used in the past, do not take advantage of the extensive capability available from high-speed pavement sensors. Also, the use of separate indices for different aspects of the pavement behavior fails to treat the pavement as a single physical system with a mechanistic rationale for the observed conditions. Therefore, the index approach does not explain the basic causes of the pavement’s behavior, and hence can neither accurately project future behavior nor can it lead to an optimal maintenance and rehabilitation strategy.
With high-speed pavement sensors it is now possible to examine 3 to 4 pavement parameters which are measured independently and continuously along the same length of the pavement. The simultaneous measurement of multiple pavement distresses can provide sufficient information to statistically estimate underlying pavement properties such as layer resistance coefficients. By inferring the values of such variables in-situ, pavement engineers can use them for purposes of deterioration prediction. Furthermore, inferring the causes of the observed deterioration allows pavement engineers to select more effective maintenance strategies.
For example, we can now continuously measure pavement cracking (with video processing and/or laser crack detectors), rutting (with optical, laser, or ultrasonic sensors), and layer thickness (with ground penetrating radar). Pavement cracking is caused by fatigue, which is related to the maximum strain in the pavement structure. The strain in the pavement structure is mechanistically related to the load, the layer coefficients, and the layer thickness. Since loads, thickness, and cracking can now be continuously measured, there is enough information to determine the pavement layer coefficients and the cause of cracking, and to predict its future development.
A pavement represents a physical system characterized by relationships between input (loads, environmental conditions) and observable output (roughness, rutting, cracking, etc.). These relationships involve properties of the pavement system, which can change over time. The pavement maintenance and rehabilitation process involves
à observations of the pavement condition;
à calculating or estimating physical properties related to these observed conditions;
à modeling the remaining life of the pavement based on the estimated pavement properties;
à implementing maintenance and rehabilitation that maximizes the remaining life of the pavement at minimum cost.
3. Summary of Research
In the context of the above framework, the work performed in this research was based on two underlying hypotheses:
1. The key physical properties of the pavement associated with observed pavement conditions can be estimated from the data obtained from a suite of high speed pavement sensors; and
2. These estimates will be sufficiently accurate to lead to more effective maintenance and rehabilitation decisions that are current made with the manner in which the data is currently used.
The long-term objective of the present work is to develop a hardware/software system for automated prediction of pavement life and for selection of the optimum maintenance and rehabilitation. The near-term objective of this research was to develop prototype software, test the software on actual field data, and develop software and hardware specifications for a final system. The prototype software was to, at a minimum, produce estimates pavement layer coefficients based on pavement input data (loads and environmental exposure) and on measurements obtained from high-speed pavement sensors. Achievement of this objective was based on development of a methodology for determining these underlying pavement properties from field observations. An additional near term objective was to use the estimated values of the pavement layer coefficients to predict future deterioration and remaining life. This was achieved by using these estimates as input values (causal variables) in improved pavement deterioration models.
This research performed included the development of a methodology for the evaluation of pavement properties from data collected by high-speed sensors, and the development of pavement performance models that form the basis of the methodology. These include a pavement rutting progression model, a pavement cracking initiation and prediction model system.
The research also included the development of an analytical procedure for inferring pavement resistance by using measurements obtained from high-speed sensors. The procedure is based on Bayesian updating principles, and combines the predictions from the relevant pavement performance model and the measurements obtained from the sensors. The methodology was tested on two different types of field data: a data set obtained from the Mn/Road site, a set of specially built pavement sections, and condition survey data obtained from various sites in the U.S. The results of the tests indicated that the use of the measurements obtained from high-speed sensors in the analytical procedure jointly with the performance prediction models helped improve the precision of pavement rutting progression.
The research included writing complete documentation of the software that includes the methodology developed in this research, as well as hardware specifications for its future implementation.
4. Major Findings
The work carried out in this project demonstrated how newly formulated models for pavement rutting and cracking can be implemented using data from current high speed pavement survey technology. The empirical work showed how data from conventional high speed rutting, cracking, and pavement thickness (GPR) sensors can be integrated into the models to predict future performance of pavement sections. The demonstration also showed how compensation could be made for the absence of historical condition data (as is often the case) by using spatial variability in structurally homogeneous sections, as identified through the GPR thickness data.
Our analytical approach for using the predictions of the pavement-rutting model, together with measurements of pavement rutting for predicting future rutting provided good results. From the tests performed using the field data from MnRoads, two conclusions were made:
1. The Bayesian analytical procedure outperforms existing methods for predicting pavement rutting;
2. Increasing the number of data points improves the rut depth prediction precision, irrespective of the prediction method used.
Both conclusions are consistent with statistical theory and with engineering intuition.
Software was developed for implementation and simulations were conducted to predict future rutting using current condition data from an in-service pavement.
1. Archilla R. and Madanat S., “Development of a Pavement Rutting Model from Experimental Data,” ASCE Journal of Transportation Engineering, Vol. 126, No. 4, 2000.
2. Prozzi J. and Madanat S., “Analysis of Experimental Pavement Failure Data using Stochastic Duration Models”, Transportation Research Record 1699, Transportation Research Board, National Research Council, 2000.
3. Archilla R. and Madanat S., “Estimation of Highway Pavement Deterioration Models by Combining Experimental Data from Different Sources”, ASCE Journal of Transportation Engineering, Vol. 127, No. 5, 2001.
4. Archilla R. and Madanat S., “A Statistical Model of Pavement Rutting in Asphalt Concrete Mixes”, Transportation Research Record 1764, Transportation Research Board, National Research Council, 2001.
Technology Transfer Activities
1. The research proposal and results were published in Access Magazine, in the following article: Madanat S. “The Opportunities and Challenges of New Technologies in Pavement Management”, Access, No. 17, University of California Transportation Center, University of California, Berkeley, CA, Fall 2000.
2. Presentations were made at the following conferences: Annual Meetings of the Transportation Research Board, 2000, 2001 and 2002.
Other Accomplishments
1. Ricardo Archilla was awarded the Council of University Transportation Centers’ Milton Pikarski Award for best dissertation in Transportation Engineering in 2000. Dr. Archilla is now Assistant Professor in the department of Civil Engineering at the University of Hawaii at Manoa.
2. Jorge Prozzi has completed his dissertation in 2001. He is now Assistant Professor in the department of Civil Engineering at the University of Texas at Austin.
Percent Complete: 100%
Direct Cost: $42,223