Research Report

Transcription

1 Development and Limited Implementation of the Validation Process for Parameters Utilized in Optimizing Construction Quality Management of Pavements Research Report Conducted for Texas Department of Transportation P.O. Box 5051 Austin, Texas August 2004 Center for Transportation Infrastructure Systems The University of Texas at El Paso El Paso, TX (915)

2

3 Development and Limited Implementation of the Validation Process for Parameters Utilized in Optimizing Construction Quality Management of Pavements by Imad Abdallah, MSCE Brett Haggerty, BSIE and Soheil Nazarian, Ph.D., PE Research Project TX Impacts of construction Quality on the Life- Cycle Performance of Pavements Using Mechanistic Analysis Conducted for Texas Department of Transportation The Center for Transportation Infrastructure Systems The University of Texas at El Paso El Paso, TX August 2004

4 TECHNICAL REPORT STANDARD TITLE PAGE 1. Report No. TX Government Accession No. 3. Recipient's Catalog No. 4. Title and Subtitle Development and Limited Implementation of the Validation Process for Parameters Utilized in Optimizing Construction Quality Management of Pavements 5. Report Date August Performing Organization Code Impacts of construction Quality on the Life-Cycle Performance of Pavements Using Mechanistic Analysis 7. Author(s) I. Abdallah, B. Haggerty and S. Nazarian 9. Performing Organization Name and Address Center for Transportation Infrastructure Systems The University of Texas at El Paso El Paso, Texas Sponsoring Agency Name and Address Texas Department of Transportation P.O. Box 5051 Austin, Texas Performing Organization Report No. Research Report TX Work Unit No. 11. Contract or Grant No. Study No Type of Report and Period Covered Interim Report Sept. 1, 2002 August 31, Sponsoring Agency Code 15. Supplementary Notes Research Performed in Cooperation with TxDOT and FHWA under project entitled Impacts of construction Quality on the Life-Cycle Performance of Pavements Using Mechanistic Analysis 16. Abstract The implementation of an effective performance-based construction quality management requires a tool for determining impacts of construction quality on the life-cycle performance of pavements. This report present an update on the efforts in the development of a statistical-based algorithm that reconciles the results from several pavement performance models used in the state of practice with systematic process control techniques. The development and limited implementation of the ongoing validation process is presented in this report. 17. Key Words Pavement life-cycle, construction quality, performance-based, impact analysis, probabilistic methods 18. Distribution Statement No restrictions. This document is available to the public through the National Technical Information Service, 5285 Port Royal Road, Springfield, Virginia Security Classif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price Unclassified Unclassified 65

5 The contents of this report reflect the view of the authors, who are responsible for the facts and the accuracy of the data presented herein. The contents do not necessarily reflect the official views or policies of the Texas Department of Transportation or the Federal Highway Administration. This report does not constitute a standard, specification, or regulation. NOT INTENDED FOR CONSTRUCTION, BIDDING, OR PERMIT PURPOSES Imad Abdallah, MSCE Brett Haggerty, BSIE Soheil Nazarian, Ph.D. (69263)

6

7 Acknowledgements The successful progress of this project could not have happened without the help and input of many personnel of TxDOT. The authors acknowledge Mr. Steve Smith, the project PD and David Head, the project PC for facilitating the collaboration and with TxDOT Districts. They have also provided valuable guidance and input. Special thanks to the PAs, Ms. Lisa Lukefahr, and Mr. Tomas Saenz whom were involved in development of the protocol for data collection and have helped in the success of the product being developed under this project. iii

8 iv

9 Abstract The implementation of an effective performance-based construction quality management requires a tool for determining impacts of construction quality on the life-cycle performance of pavements. This report present an update on the efforts in the development of a statistical-based algorithm that reconciles the results from several pavement performance models used in the state of practice with systematic process control techniques. The development and limited implementation of the ongoing validation process is presented in this report. v

10 vi

11 Executive Summary The ability of a flexible or rigid pavement to perform adequately throughout its design life is one of the biggest challenges that transportation agencies face. One factor that has a large impact on the performance of a pavement is the quality of construction. The implementation of an effective performance-based construction quality management program is one way of ensuring that pavements are meeting their expected service life. As a part of that program a tool for determining impact of construction quality on life-cycle performance of pavements is required. TxDOT, amongst other highway agencies, has adopted statistic-based quality assurance/quality control (QA/QC) techniques to improve the quality of roadways. In this report a method of optimizing construction quality management of pavements using mechanistic performance analysis method based on statistical techniques is presented. This method was developed for both flexible and rigid pavements. Ideally, if a pavement section is designed with the same cross section and constructed with the same materials, its performance should be uniform throughout the section. This is not the case in the real world. Almost every constructed road develops distresses randomly in different subsections of the pavement. One reason for the random development of distress is the variability in construction quality. As such the goal in this project is to devise a tool that can be used to identify and minimize variability in material properties that impact the performance of the pavement to ensure a performance period compatible with the expected life of the pavement. With that framework, structural models that predict performance of pavements and material models that relate construction parameters to primary design parameters were identified. Finally, a statistical algorithm that relates the impact of each construction parameter to the performance of a pavement is incorporated into the algorithm. xi

12 Implementation Statement At this stage of the project the tools developed are being modified and can be used for limited implementation. The software is undergoing major changes to increase its flexibility and expand its ability to identify and minimize variability in material properties that impact the performance of the pavement to ensure a performance period compatible with the expected life of the pavement. The software is called Rational Estimation of Construction Impact on Pavement Performance (RECIPPE). It can be used to reconcile the results from realistic pavement-performance models used in the state of practice, or those widely accepted by state agencies, with statistical process control techniques and uncertainty analysis methods, to determine project-specific parameters that should be used in construction quality management. xii

13 Table of Contents Chapter 1 - Introduction... 1 Objective... 2 Organization of Report... 2 Chapter 2 - Background... 3 Review of Work Efforts... 3 Progress of Research... 3 Chapter 3 - Develop a Document for Validation of Different Parameters... 5 Overview of the RECIPPE Algorithm... 6 Validation of Probabilistic Process... 8 Appropriateness of Material Models Chapter 4 - Limited Implementation of Validation Process Case Study Chapter 5 - Summary and Conclusions Recommendations References Appendix A - Standards to Measure Construction Parameters Appendix B - Protocol for Collecting Construction Parameters Information for Sites in Texas xiii

14 xiv

15 List of Figures Figure Conceptual Framework of RECIPPE... 6 Figure Flowchart of Process Developed in RECIPPE... 7 Figure Flowchart of the Algorithm Used in RECIPPE for Determining the Impact Value of Construction on Performance... 9 Figure Comparison of Impact Value Results for both Monte Carlo and TPM Methods Figure Results of Impact of Variability of Asphalt Content on Remaining Life Varying AC Modulus Figure Results of Impact of Variability of Asphalt Content on Remaining Life for Four Typical Pavement Sections xv

16 xvi

17 List of Tables Table Advantages and Disadvantages of Probabilistic Techniques... 8 Table Synthetic Data for a Typical Pavement Section Used in Comparing the Two Probabilistic Algorithms Table List of Construction Parameters of each Pavement Layer Table Feasible Ranges and Acceptance Values of Construction Parameters Table Matrix of pavement sections used for validation of RECIPPE Table B1 - Frequency of Measurements for Asphalt/Concrete Layer Table B2 - Frequency of Measurements for Base Layer Table B3 - Frequency of Measurements for Subgrade Layer xvii

18 xviii

19 Chapter 1 Introduction Developing a product that will perform well throughout its expected design life is the objective of any engineering agency. In transportation agencies one of those products is a pavement system. Based on the large number of uncertainty and natural occurrence, the design life of a pavement system is altered at its earliest stages and continues to change through out its life cycle. The beginning of a pavement lifecycle starts at the construction level. The as-design requirements are usually not met due to the uncertainty related to the quality of construction. There are many variables involved in construction practices that allow for deviation from acceptable target design values. For that reason, many transportation agencies agree that a costbased incentive/disincentive mechanism should be a part of the process to implement an effective performance-based construction quality management program. The research in this project attempts to develop a mechanism that identifies quality of construction or meeting a structural-related target variable with minimal variance. In that context, the objective is to device a process that accounts for the variation in pavement performance based on construction practices. This process would be used to maximize construction efforts such as effectiveness of inspection and testing resources during construction and maintaining target-level of pavement performance. The process being developed under this project is a method that ties in parameters related to construction practices for quality assurance/quality control (QA/QC) to pavement performance models based on mechanistic analysis. The process uses statistical base algorithms to optimized construction quality management of pavements. The algorithm considers the fact that relevant construction parameters change with the relative structure of the pavement. Many parameters that are important for a thick pavement designed for an interstate highway may be of secondary significance to a secondary road. The level of acceptable deviations from the target design value for each parameter is established based on quantification of the variability of the construction parameters introduced by: (a) the construction processes, (b) the material properties, (c) the models used to predict pavement performance and those used for data analysis, and (d) the resolution of the procedures used in the field for quality control. 1

20 The software is called Rational Estimation of Construction Impact on Pavement Performance (RECIPPE). It can be used to reconcile the results from realistic pavement-performance models used in the state of practice, or those widely accepted by state agencies. RECIPPE determines project-specific parameters that should be used in construction quality management based on statistical process control techniques and uncertainty analysis methods. Objective This research study is based on three objectives. The first objective is the development of the rational algorithms. The algorithms have been developed and package into a program called RECIPPE. In the first report, details of the algorithms and a case study using the prototype version of the program (Excel version) were discussed. The final version of the program will be included final report. The second objective is to propose field tests that can be used to measure parameters identified in a cost-effective manner. A document that identifies methods for measuring several parameters based on testing standards was developed and is linked into RECIPPE for easy access. A copy of this document is included in this report. The third objective is to establish protocols to validate the algorithms and processes developed during this project. Research work for part of this objective will be included in this report. The remaining research work will be included in the final report. In this report, the efforts and work carried out based on the following tasks are presented: a) develop a document for validation of different parameters, and b) limited implementation of validation process investigate existing techniques and current software packages that relate construction quality to pavement performance, Organization of Report Chapter 2 of this report provides a review of research efforts on previous work documented in the first report. Also summarized in Chapter 2 are the algorithms and methodology using existing models. In chapter 3, the efforts in the validation procedures of different parameters are presented. Chapter 4 includes a limited implementation of the algorithms developed under this project. The last chapter, Chapter 5, includes the summary of the work accomplished and work remaining on this project. Also included is Appendix A, which has a compilation of ways to measure parameters based on existing standards as proposed is task three of the project. Finally, Appendix B has the protocol developed and that will be used to locate, identify and test sites for calibrating and validating material models. 2

21 Chapter 2 Background Review of Work Efforts In the first report, analytical tools to determine the impact of construction on pavement performance were presented. The tools will be used to assist the construction engineer in identifying the parameters that most impact the performance of a project under consideration and guide her/him in reducing the variability associated with them. The algorithms developed were based on a systematic statistical process control techniques and uncertainty analysis methods to reconcile the results from existing pavement-performance models used in the state of practice. The algorithms were developed for flexible pavement systems. The outcome so far has been a software prototype developed in Excel that identifies the impact of construction parameters on performance indicators. A combination of two probabilistic techniques, Monte Carlo simulation and Two Point Mass (TPM) methods were used to assess the impact of construction and design parameters on pavement performance. The algorithms for could also be applied to rigid pavements, but the task to further its development were not recommended because the design methodology currently practiced does not warrant the use of the statistical based algorithm used for this project. The focus of this project, based on the agreement with the PMC, is on furthering the research and finalizing tools for flexible pavements. The research efforts for flexible pavement have continued and the tasks completed thus far are presented in this report. Progress of Research The first two years of this project the research efforts were documented in TxDOT Report (Abdallah et al., 2004) and focused on the following tasks: a) Informational search on existing mechanistic models and ways that can be used in developing an algorithm to relate the impact of construction parameters on performance: After a national search, material models were identified and feasible models were selected. Also, popular and well established performance based models were selected. The next step in the research process was the development of the probabilistic algorithms. The 3

22 probabilistic algorithm was developed to combine the material models with the performance model. b) A probabilistic analysis was based on Monte Carlo simulation and TPM: The probabilistic approach differs from a deterministic approach by explicitly accounting for the variability of a parameter. A random parameter can take a range of values and can be represented by different types of probability distributions. In this research project, all input parameters are assumed to be normally distributed. Monte Carlo simulation is a common probabilistic method for simulating and accounting for the variability of a parameter. However, since many parameters are used in the analysis and to accelerate the process for speed the twopoint mass method (Rosenblueth, 1975) was combined with Monte Carlo. The two-pointmass (TPM) method can be used to approximate mean and standard deviation of random variables. The detail of both methods is provided in Chapter 2 of TxDOT Report (Abdallah et al., 2004). c) Develop algorithms to quantify impact of validation on performance: Once the models were selected and the flow of probabilistic algorithm defined, a prototype algorithm was developed. Figure 3.2 in TxDOT Report (Abdallah, et al. 2004) shows the general flow of information used in mechanistic algorithm with the probabilistic methods. The detail and a case study of how to use the program is also provided in the first report. d) Sensitivity analysis to determine and identify important parameters: The mechanistic models selected provided a number of parameters that are used as a measure of construction practices. To optimize the process, a sensitivity analysis was conducted to primarily identify the relative importance of construction parameters on performance indicators. The results of this study were presented the first research report (Abdallah et al. 2004). The results provided an indication of important parameters for pavements with different traffic levels. e) Methods to measure significant parameters: Based on results of the sensitivity study, a search effort to document ways to measure important parameter. The document was developed and to optimize its use it was programmed into RECIPPE. The idea was that users could easily access the different ways to measure parameters. Another advantage to including the document into the program is that when new parameters are added the document could be amended or updated. The document is included in Appendix A of this report. The third year of the project the focus of research was on developing a document for validation using few of construction parameters and to demonstrate the validation process using selected parameters. The details of these tasks are included in this report. 4

23 Chapter 3 Develop a Document for Validation of Different Parameters The validation of the algorithm to quantify the impact of construction parameters on performance is the initial step before being able to utilize RECIPPE. As in any validation process, the ultimate goal is to compare the results from the proposed methodology to known and measurable parameters. On that account, the models in RECIPPE would need to be calibrated. There are three types of models that make up the mechanistic algorithm developed: a) material models, b) the structural model and c) performance models. The material models will be calibrated with information from existing databases and from field data collected at several sites in Texas. The plan for validating and identifying the sites are presented in this report. The efforts of this calibration and validation process will continue throughout the length of this project. However, the structural model and performance models incorporated into the analysis are well developed and established. The structural model is based on a nonlinear model using equivalent linear algorithm. The equivalent linear process was developed under TxDOT Project and the details of the algorithm is provided in TxDOT Research Reports (Abdallah et al., 2003) and (Abdallah et al., 2003). The performance models incorporated into RECIPPE are well established models that are nationally used. At this stage of the project the performance models will be incorporated as since the research and resources to calibrate these models are outside the scope of this project. Additionally, the probabilistic process and the models in the algorithm need to be validated for optimal functionality of the program. In this report, the validity of using the mechanistic analysis using a probabilistic process in RECIPPE to determine impact of variability of construction parameters on performance is presented. This is accomplished with the following: a) Determining the validity of the probabilistic process to identify impact of variability of input parameters on output parameters. b) Determining the appropriateness of using the existing material models used in obtaining the modulus of each pavement layer. c) Determining the accuracy of the life cycle analysis or remaining life models to accurately predict the pavement performance. 5

24 The progress and results on the first two tasks are discussed in this report. As the algorithms become optimized and finalized and as field data becomes available, the remainder of the validation will be presented and discussed in future reports. The third task is accomplished in an independent study. As better performance models become available or get developed under accelerated pavement testing programs, they can be incorporated into RECIPPE. In the meantime these performance models are commonly used throughout the transportation industry. Overview of the RECIPPE Algorithm An overview of the algorithm used in RECIPPE is fitting before discussing the validation process. The detail of the preliminary algorithm developed is presented in TxDOT Research Report (Abdallah et al., 2004). However, a summary of the process is presented here and the latest modifications to the algorithm are also included. As presented in Chapter 3 of first research report, the methodology is divided into three phases. Figure 3.1 from the report is presented below as Figure 3.1 for clarity. The three phases start from construction parameters (represented in the inner circle of the figure. The second part is the material characteristics models (represented by the middle circle in the figure), and it is the link between the construction and pavement performance (which is represented by the outer circle of the figure). These three phases form the mechanistic process and are used in RECIPPE with the probabilistic algorithm to estimate the impact of construction variability on performance. Once the parameters and models for the three phases are established the probabilistic algorithms allows pavement engineers to maximize effectiveness of inspection and testing resources during construction process. Construction Parameters compaction, moisture content & dry density Layer Moduli Fatigue Cracking & Rutting Figure Conceptual Framework of RECIPPE 6

25 To demonstrate the conceptual use of the algorithm, lets assume that the construction parameters are defined and are used as input to the material models. Also, lets assume that the performance models are selected and the results of material models are used in mechanistic analysis to calculate pavement performance. With these assumptions in mind, Figure 3.2 illustrates the role of RECIPPE in construction quality management. QC/QA STEP 1 Analysis STEP 2 Pavement Performance Damage Time STEP 3 Perform analysis until performance vari ability is minimized and Ins pection and testing resources during construction can be maxi mized? Impact of Construction Parameters on Performance 5% 6% 12% 10% 48% 21% P200 Percent Air Voids Viscosity Percent asphalt content Dry density of base Specific gravity of subgrade Determine Impact of construction variability on Pavement performance Figure Flowchart of Process Developed in RECIPPE As a new pavement is being constructed, samples of construction parameters for each layer are collected, in situ and appropriate laboratory tests are performed to obtain values and statistics that can them be used as input into the program (indicated in Step 1 of Figure 3.1). These measurements are ones made for QA/QC and are currently based on tests from the TxDOT Guide Schedule. The initial sampling rate can also be adopted from the guide schedule. Once the measurements are made, the values are transferred to RECIPPE for analysis. The algorithm uses the probabilistic process to obtain the variability of performance values based on mechanistic analysis as depicted in Step 2. In Step 3 of the figure, RECIPPE produces an impact chart that identifies construction parameters that impact pavement performance. This impact chart varies based on performance models selected. If the impact of construction variability shows that the pavement performance is critical or lower than expected, the next step is to identify construction parameter that impacts performance most and reduce the variability of that critical parameter. The analysis is repeated until the condition is satisfied. With each new 7

26 iteration, a different parameter could be identified as critical in impacting performance. The reason is, the variability for critical parameters are reduced or modified with each run. Once the performance variability is satisfied, inspectors and testing resources can be utilized and thus maximizing resources and ensuring quality and performance of pavement. The role of RECIPPE, as illustrated in the figure, can be very useful in monitoring impact of construction variability on performance. Validation of Probabilistic Process First, investigating and analyzing the effectiveness of the probabilistic process incorporated in RECIPPE is discussed. The probabilistic approach differs from a deterministic approach by explicitly accounting for the variability of a parameter. The first report of this project compares the differences between the two approaches. The algorithm in RECIPPE assumes that the variability and or uncertainty in a parameter is defined by assuming a normal distribution represented with a mean and a coefficient of variation (COV). As explained in the previous report the COV values of the inputs and outputs are used to develop the impact values. Also presented in the report are the two techniques that are used in simulating each input parameter. Table 3.1 below summarizes the advantages and disadvantages of each technique. If processing time is of no concern then Monte Carlo simulation is advisable. Monte Carlo simulation requires a large number of simulated values to simulate each parameter however, the results of the analysis can be used to determine higher order statistics other than the mean and COV value. On the other hand, in instances where processing time is a factor, Monte Carlo simulation is not advantageous since the as the number of parameter increases, the processing time increases. Table Advantages and Disadvantages of Probabilistic Techniques Technique Advantages Disadvantages Monte Carlo Two Point Mass Simulate several parameters simultaneously Simulate one parameter with two values Require NS*n simulation to simulate n parameter one at a time Require 2 n simulation of several parameters simultaneously For example if the number of simulation to represent each parameter as a normal distribution is 500 and there are 20 parameters involved, then the total number of simulations that are process through the analysis to determine the impact of each of the parameters is ten thousand. If however, the interest is in determining the impact of all 20 parameters simultaneously the total number of runs is 500 as indicated in Table 3.1. As for the Two Point Mass technique, the opposite is true. Only two simulated values are required to describe the variability of each input parameter. For 20 parameters, only 40 runs are required. However, to investigate the effect of all twenty parameters simultaneously, the number of runs is 2 n. For 20 parameters the number of runs would be over one million. To take advantage of both techniques, the RECIPPE algorithm utilizes both at different stages in the analysis. To determine the impact of each parameter independently, Two Point Mass 8

27 is used and when the impact of all parameters is needed Monte Carlo is used. More detail of both techniques is provided in the first report of this project. The final algorithms will contain a combination of the two techniques and the results will be included in the final report. Figure 3.3 below provides an illustrative flow of the use of either technique in RECIPPE. As indicated in the figure, the construction parameters that are identified as part of the analysis require the results of the sampling statistics (Mean and COV) to simulate the cases based on both Monte Carlo and Two Point Mass to evaluate impact of variability. The simulated cases are used to determine the modulus of each pavement layer. The process continues and remaining life values for each case is estimated. The statistics from all cases are determined and the impact value is obtained. Asphalt Content Dry Density Compaction Etc Sample and Testing Statistics ( µ DD, COV ) Monte Carlo Simulation or 2Pt. Estimate Models relating construction to section properties Modulus=f(construction) Mechanistic analysis Based on probabilistic method E 1,t 1 E 2,t 2 Etc Variability of Performance Fatigue Cracking / Rutting -σ +σ -2σ +2σ µ, COV Impact Chart 25% 1% 8% 1% 9% Figure Flowchart of the Algorithm Used in RECIPPE for Determining the Impact Value of Construction on Performance 57% 9

28 The impact value is as follows: COV output COV input I = 3.1 where COV output is variability of performance and COV input is the variability of construction parameter. The impact of each parameter is then normalized as follows: I i NIV i = i = 1,2,..., n n i = 1 ( I ) i 3.2 The normalized impact values are further compiled to develop an impact chart that assists in identifying the level of impact of each parameter on performance as illustrated in the last part of the flow diagram in Figure 3.3. To determine the validity of the process several simulation were performed to identify the accurate functionality of the process. First step in the process was to determine the appropriate number of simulation required using Monte Carlo technique for a satisfactory representation of a parameter. The results of this analysis were presented in Figure 3.6 of TxDOT Research Report (Abdallah et al., 2004). The results show that the number of simulation based on Monte Carlo is 500. Next was to show the appropriateness of using TPM in the analysis. To check this process, a comparison of the impact of one parameter using both Monte Carlo and TPM. The following example is provided to illustrate and compare the two probabilistic methods. The data in Table 3.2 is based on a typical pavement section showing both the design and material model values that are used as input in RECIPPE. For purposes of this example only the material model for the AC layer is activated and the base and subgrade material models are not used. The modulus value of the base and subgrade layer is assumed constant and set to the design value. The mean and COV values for all parameters are set as assigned in the table and only the COV value of the asphalt content is varied. The program is executed five times varying the COV value from 10% to 50% by 10% increments. For each COV value the program is executed using each probabilistic method. The results are presented in Figure 3.4. The figure compares the impact value based on varying the COV value for both Monte Carlo and TPM. As shown in the figure, the results of both analyses are identical when the COV value of the impacting parameter is low. For higher COV values the impact values of the two methods are very comparable with little significance in their difference of impact values. The results of this example show the interchangeability of the two methods without affecting to final output of RECIPPE. These results are independent of construction parameter and pavement performance model used. Another factor to examine in the figure is change in variability of the output based on the variability of the input. The impact due to variability might change for every parameter and the shape of the curve in the figure will not be the same for all parameter. That is the effectiveness of this algorithm to account for impact of variability independent of the analysis process used. 10

29 What is also shown in the figure is the direction of change in the analysis. The change is monotonic, meaning the impact value increases as the variability of impact value increases. As the COV value of the input changes the impact value on the performance parameter changes. This demonstrates the usefulness of the probabilistic algorithm to identify the impact of variability of construction parameters on variability of performance. The final version of RECIPPE will be optimized by balancing both accuracy and processing time. Table Synthetic Data for a Typical Pavement Section Used in Comparing the Two Probabilistic Algorithms. Data Type Input Values Parameter Mean COV AC Thickness 5in. 1% Base Thickness 12in. 0% Design AC Modulus 500 ksi. N/A Base Modulus 50 ksi. N/A Subgrade Modulus 10 ksi. N/A Aggregate Passing No % 5% AC Mix Air Void 1.50% 10% Material Model Asphalt Viscosity poise 3% Asphalt Content 3.25% Vary Loading Frequency 18 Hz N/A Temperature 78.5 degrees F N/A Impact Value 94% 92% 90% 88% 86% 84% 82% 80% 0% 5% 10% 15% 20% 25% 30% 35% Asphalt Content COV TPM Monte Carlo Figure Comparison of Impact Value Results for both Monte Carlo and TPM Methods. 11

30 Appropriateness of Material Models The next stage in the validation is to investigate adequacy of material models selected based on the literature search. The models were developed based on studies conducted on different pavement layers. These models were gathered based on a nation wide search. The material models selected were for AC, base and subgrade layers that relate construction parameters to modulus of pavement layers. Originally, the expectation of the material model search was to produce a pool of models that would be evaluated and models that were most appropriate product of this research will be selected. However, as presented in the previous report, only one model was selected for each layer of a flexible pavement. The material model for the AC layer is an equation that relates the dynamic modulus of the AC to parameters such as temperature, asphalt content and air voids content. A popular model known as the Witczak model serves this purpose (Ayres et al., 1998). The equation that is: log E AC P = f ( log f ) V ( log f ) η t P t p P 0.5 ac 1.1 f p V 0.5 ac f where E AC is dynamic modulus of AC mix (in 10 5 psi), η is bitumen viscosity (in 10 6 poise), f is the load frequency (in Hz), V v is percent air voids in the mix by volume, P ac is percent effective bitumen content by volume, and P 200 is percent passing No. 200 sieve by total aggregate weight. The advantage of using the Witczak model was that a version of it, would be part of the AASHTO 2002 design guide. The base and subgrade models were based on the constitutive model. This material model accounts for granular and cohesive materials are needed. The general form of the universal equation is M R k2 = k θ σ 1 k3 d 3.4 where θ = σ 1 + σ 2 + σ 3 = bulk stress ; σ d = σ 1 σ 3 = deviator stress (σ 1, σ 2 and σ 3 are the three principal stresses). Parameters k 1, k 2 and k 3 are material regression constants statistically obtained from laboratory tests. Santha (1994) developed regression equations for cohesive and granular materials that estimate parameters k from various construction parameters such as the moisture content, compaction, and percent saturation. The equations developed if that study were adopted for this project. The equations are expanded on in the previous report. The material models were combined into one algorithm and the list of parameters used in the material models is provided Table 3.3 according to their corresponding pavement layer. A sensitivity study reported in the previous report identifies significant parameters. The sensitivity study was a preliminary study performed based on four typical pavement sections that are designed in Texas. The study considered all parameters for the three material models as equally 12

31 important. The range of feasible parameters based on the current TxDOT specification is listed in Table 3.4. The table shows the range of values for each parameter and the typical acceptance values used in construction practices currently practiced. These values were obtained from TxDOT specification and will be provided as default coefficient of variation values (COV) to guide users in the analyses. Table List of Construction Parameters of each Pavement Layer Layer Parameter Thickness Modulus Poisson Ratio Asphalt Content Asphalt-concrete AC Mix Air Void Asphalt Viscosity Aggregate Passing Sieve No.200 Loading Frequency Temperature Thickness Modulus Poisson Ratio Maximum Dry Density Degree of Compaction Base Aggregate Passing No.40 Percent of Clay Percent of Silt Optimum Moisture Content Moisture Content Thickness Modulus Poisson Ratio Maximum Dry Density Optimum Moisture Content Moisture Content Subgrade Degree of Compaction Aggregate Passing No.40 Percent of Clay Percent of Silt Liquid Limit Plastic Index 13

32 Table Feasible Ranges and Acceptance Values of Construction Parameters Layer Parameter Range Acceptance % Passing Sieve #200 2% - 7% ± 5% Asphaltconcrete Base Subgrade Thickness - ± 0.75 Viscosity ± 20% AC Mix Air Voids 5% - 9% ± 2% Asphalt Content - ± 1% Thickness - ± 1 Compaction 100% Density - ± 3 % Passing Sieve #40 50% - 85% ± 5% Plasticity Index - ± 2% Compaction, PI 20 98% Compaction, 20<PI <35 105% Compaction, PI < 35 95% - 100% Moisture Content, PI 35 Optimum Moisture Content Density - ± 3 % Passing Sieve #40 - ± 5% In order to test the algorithm and the model s validity using field data, a database search was conducted. The field data was to be used to check appropriateness of the models and to calibrate the models for Texas road conditions and construction practices. The database search effort was conducted both locally in Texas and nationally but was not successful. Most databases contained partial information. Unable to locate a comprehensive database, the search pattern focused on identifying data for parameters of each material model separately. On that consideration, the Long-Term Pavement Performance (LTPP) database was found to contain data that can be used for the asphalt-concrete layer. The LTPP database is online and a user interface with a definition guide is available to extract data. The Long-Term Pavement Performance (LTPP) database has been populated with data since 1987 with more than 2,400 test sections on in-service highways at over 900 locations throughout North America. The extensive data collection effort includes inventory, material testing, pavement performance monitoring, climatic, traffic, maintenance, rehabilitation, and seasonal testing modules. The data are housed in an information management system that is the world's largest pavement performance database. The online LTPP database offers a user-friendly format for exploring, extracting, and organizing the extensive LTPP data from all over the nation. In addition, the website provides the capability to identify LTPP database tables and fields for data extraction. 14

33 A filtration process is used to extract specific data that is necessary by users to minimize unnecessary information that are contained in the multiple tables that make up the database. This fact is necessary since many of the parameters required for the calibration of the material model are not located in different tables, and the testing sample varies for any given parameter. The research efforts was first focused on SPS sites and later expanded to other Texas sections. The information required that needs to be extracted from the LTPP database are: Asphalt content Viscosity Percent of aggregate passing sieve #200 Percent air voids Elastic Modulus of the AC To illustrate the filtration process, the example of extracting data for the asphalt content is discussed. The asphalt content can be found in the Inventory IMS module which contains tables with listings of plant mixed asphalt data. One of the tables, Plant-mixed asphalt bound layers original mixture properties, contains information such as the following. STATE_CODE, LAYER_NO, SHRP_ID, SAMPLE_TYPEHVEEM_COHESIOMETE R, BULK_SPEC_GRAVITY_MEAN, MAX_SPEC_GRAVITY, NUMBER_BLOW, PCT_AIR_VOIDS_MAX, PCT_AIR_VOIDS_MIN, PCT_AIR_VOIDS_MEAN, ASPHALT_CONTENT_STD_DEV, RECORD_STATUS, ANTISTRIP_AGENT_TYPE, ASPHALT_PLANT_TYPE_OTHER, HVEEM_STABILITY, MARSHALL_FLOW, RETAINED_STRENGTH_INDEX, PERCENT_STRIPPED, HVEEM_STABILITY_NO, TENSILE_STRENGTH_RATIO, EFFECTIVE_ASPHALT_CONTENT, NO_SAMP_ASPHALT_CONTENT, ASPHALT_PLANT_TYPE, CONSTRUCTION_NO, ASPHALT_CONTENT_MAX, ASPHALT_CONTENT_MIN, ASPHALT_CONTENT_MEAN, BULK_SPEC_GRAVITY_STD_DEV, BULK_SPEC_GRAVITY_MIN, MARSHALL_STABILITY, BULK_SPEC_GRAVITY_MAX, VOIDS_MINERAL_AGGR, NO_SAMP_PCT_AIR_VOIDS, PCT_AIR_VOIDS_STD_DEV, MOISTURE_SUSCEPT_TEST, ANTISTRIP_AGENT_AMOUNT, ANTISTRIP_AGENT_CODE, ANTISTRIP_AGENT_TYPE_OTHER, MOISTURE_SUSCEPT_TEST_OTHER, NO_TESTS_BULK_SPEC_GRAVITY This list shows a considerably large amount of information to filter through in order to obtain the asphalt content. To link several of the databases together, identification keys are used to link the variety of tables together thereby allowing us to match data collection. Crucial identification keys such as SHRP ID, the road identification key, are used to extract, link, and match the appropriate data that is required. Few of the other identification keys that were used are: SHRP_ID, STATE_CODE, LAYER_NO, RECORD_STATUS, ASPHALT_GRADE, CONSTRUCTION_NO, and ASPHALT_SPECIFIC_GRAVITY. 15

34 The next set of extraction was the backcalculated modulus from FWD data. Also extracted were the pavement property information and the raw FWD data. After the data is extracted from the data base a data mining process was used to ensure the best set of data possible was used in the calibration of the AC material model. Things such as consistency in drop heights for FWD data, number of backcalculated layer, and RMS error of backcalculation were consideration in ensuring data reliability. Compilation of field data for base and subgrade material models was not as successful. As presented in Table 3.3, the number of construction parameters required is sizeable (9 parameters for base and 11 parameters for subgrade) and no comprehensive database was made available and or found of Texas pavement sections. Therefore, as part of the research, a test plan was developed to help locate several on going project in Texas where data would be collected. The initial attempt to locate sites was based on trying to identifying new construction projects where parameters for subgrade, base, and AC could be collected and therefore creating a small database that could be used for validation and calibration. After several month of attempting to locate and test sites, the processes were not very successful. Only limited and partial data was collected. The failure could be attributed to several reasons mainly due to schedule and miscommunication between the researchers and District personnel. The project meeting committee in a day long meeting outlined a new strategy to search for sites. As an outcome the researchers developed a protocol that would identify each parameter that needs to be obtained and the number of samples and tests that would be required. This was important to clarify to the Districts the efforts involved in the data collection process. Also, due to time limitations all projects, independent of the stage of construction, will be targeted. This allowed for the pool of available project to be expanded. The protocol was sent out to several District personnel that are in a position to identify and help locate potential sites. At least six total sites from different corners of Texas should be located. The protocol that was developed and sent out to the Districts is provided in Appendix B. The protocol provides a simple explanation of the purpose of the data collection followed by a listing of all parameters required separated for each pavement layer. For each parameter a designated testing process is identified and the frequency of testing is provided. Any parameter that is not feasible to be tested by TxDOT due to lack of resources, only needs to be sampled. The research team at UTEP will perform the necessary laboratory testing. The research team would work with TxDOT field personnel to collect the data. With immense help of the Project Director (PD), several research sites were located and are being tested, and necessary data is being collected. The data collection process and progress of the calibration process will be included in the next report. 16

35 Chapter 4 Limited Implementation of Validation Process The validation of the process being develop under this project of optimizing construction requires field data from actual construction sites. The performance models as indicated in Chapter 3 are the current state of practice in determining remaining life of flexible pavements. However, at this stage of the project the material models have not been fully calibrated. Currently, the AC material model is being calibrated based on LTPP database. As for the base and subgrade material models the field and laboratory testing efforts are on their way and once actual data becomes available, the validation process will be undertaken and produced in a final report. In this chapter a limited implementation of the validation process is presented to illustrate the consequence of variability in construction. This process will provide a guideline for the validation of the calibrated models. Case Study The case study presented in this project will demonstrate the use of RECIPPE to identify the impact of construction variability on variability of pavement performance. Four typical pavement sections varying from thin to thick pavement sections are used in this study. The pavement sections have varying thickness from 3 in. to 6 in. and from 6 in. to 12 in. for AC and base respectively. For each pavement section the modulus of AC layer is varied from 100 ksi, to 500 ksi to 900 ksi with the base and subgrade modulus values fixed at 50 ksi and 10 ksi respectively. The matrix in Figure 4.1 illustrates the experiments that are analyzed. Base and Subgrade Modulus (ksi) Base = 50 and Subgrade = 10 Table Matrix of pavement sections used for validation of RECIPPE Pavement Sections AC Modulus (ksi) Thin-Thin Thin-Thick Thick-Thin Thick-Thick 3in.-6in. 3in.-12in. 6in.-3in. 6in.-12in. 100 ksi 500ksi 900ksi 17

36 To perform the analysis for the sections presented above using RECIPPE, the material models for the base and subgrade layers were bypassed with the modulus values set constant and only the AC model included in the analysis. Based on Equation 3.3, the model is based on the following parameters: viscosity, load frequency, percent air voids, percent asphalt content, and percent passing No. 200 sieve. In this study the parameter that was selected to illustrate its impact on performance is the asphalt content. The remaining parameters were held constant. The performance models selected were fatigue cracking, subgrade rutting and AC rutting. Figure 4.1 presents the results of the first set of experiments, where the thickness was for thin-thin pavements and the AC modulus varied. In all three graphs the variability calculated for the three performance indicators are plotted on the y-axis against the input variability of the asphalt content (plotted on the x-axis). The results show similar trends for all three cases of thin-thin pavements. As expected the variability of performance indicators increase as variability of input (in this case it s the asphalt content) increases. The most impacted performance indicator is subgrade rutting where the variability varied from 3% to 10% as the variability of the asphalt content varied from 10% to 50%. The other to performance indicators, although were not impacted as subgrade rutting, also varied as the variability of the asphalt content varied. This trend is similar for the remaining pavement sections where the modulus of AC is varied. However, if the four pavement sections are compared the results tell a different story. Figure 4.2 shows the results of the middle row of experiments in Table 4.1. This set presented is for the AC modulus of 500 ksi, base modulus of 50 ksi, and subgrade modulus of 10ksi. In this case the figure shows a comparison of different pavement sections. Although as expected the variability increases, the change from one pavement section to another varies. In cases such as a thin-thick section the impact is less significant compared to a thick-thin section. The varying trend due to variability of the input shows the nonlinear effect inherent in the analysis and further justifies the need to incorporate a probabilistic analysis in RECIPPE. Comparing the top two graphs to the bottom two graphs in Figure 4.2 suggests that the impact of asphalt content increases for thicker AC layers as expected. The subgrade rutting is impacted highest for thick AC and thin base. Also, the fatigue cracking performance indicator becomes a factor as the AC layer thickness is increase. Finally, the AC rutting performance indicator does not seem as sensitive as the other two parameters for the asphalt content. The overall results of these experiments illustrate the versatility of the algorithm in RECIPPE to analyze different flexible pavement sections. This case study is for varying one parameter. However, for actual field data, the number of varying construction parameters is more, and the combining effect of their variability increases the complexity of the analysis. The algorithm incorporated in RECIPPE is able to handle such complexities. Also, as can be concluded from the results of the case study, RECIPPE s algorithm can be used to analyze pavements ranging from a project level to a network level construction. The next step, after the preliminary validation process presented in this chapter, is to test the flexibility of each of the material models after the calibration with field data is completed. Also, as the algorithm progresses and is optimized, the validation process will continue and will be documented in future reports. As more data becomes available more case studies will be presented to illustrate the use and advantages of utilizing RECIPPE. 18