Independent Review of the Mechanistic Empirical Pavement Design Guide and Software (2006)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Research Results Digest 307 September 2006 C O N T E N T S PART I, 1 1 Background, 1 2 The Design Process, 3 3 Flexible Pavement Design, 5 3.1 Resilient Modulus for Subgrade, 5 3.2 HMA Complex Modulus, 5 3.3 Fatigue Cracking, 6 3.4 Rutting, 6 3.5 Thermal Cracking, 7 3.6 Other Materials, 7 4 Rigid Pavement Design, 7 4.1 JPCP Design, 7 4.2 CRCP Design, 8 5 Pavement Rehabilitation, 9 5.1 Pavement Evaluation, 10 5.2 Rehabilitation with HMA, 10 5.3 Rehabilitation with Concrete, 11 6 Design Reliability, 11 7 Design Software, 12 8 Recommendations, 13 8.1 Flexible Pavement Design, 13 8.2 Rigid Pavement Design, 13 8.3 Pavement Rehabilitation Design, 14 8.4 Software, 14 8.5 Other Matters, 14 9 Concluding Discussion, 15 PART II, 15 REFERENCES, 30 Appendix A: Brief for Review Teams, 31 Appendix B: Membership of Review Teams, 32 NCHRP Project 1-40A was an inde- pendent, comprehensive review of the Mechanistic-Empirical Pavement Design Guide (MEPDG) and companion soft- ware Version 0.7 delivered under NCHRP Project 1-37A in June 2004. The project was carried out by consultant teams in four areas: new hot-mix asphalt (HMA) pave- ment design; new Portland cement con- crete (PCC) pavement design; composite pavement design and design reliability; and low-volume road pavement design. The review began in August 2004; interim reports were made to the Project 1-40 panel several times in 2004 and 2005, and each team’s findings and conclusions were re- ported at a panel meeting in December 2005. The project was successful and the panel used its interim and final results to direct the development of the new software Versions 0.8 (released November 2005) and 0.900 (released July 2006) by the NCHRP Project 1-40D research team. Part I of this digest summarizes the find- ings, conclusions, and recommendations of the independent review of new HMA pavement design, new PCC pavement de- sign, composite pavement design, and design reliability (the results of the review of low-volume road pavement design will be reported in a future RRD). Part II tabu- lates the responses of the Project 1-40D re- search team to the essential, high-priority recommendations of the independent re- viewers with respect to corrections and improvements to the MEPDG software, which represents the day-to-day imple- mentation of the design guide itself. Most of these recommendations were success- fully incorporated in software Versions 0.8 and 0.900, yielding a stable, robust, and ac- curate tool for pavement design. PART I 1 BACKGROUND Scott Wilson Pavement Engineering, Ltd. (SWPE) was engaged by NCHRP under contract 1-40A, along with two other teams (see Table 1), to conduct an independent engineering review of the new mechanistic-empirical (M-E) pavement de- sign guide (the Guide) and associated soft- ware (designated Version 0.7, July 2004) developed under NCHRP Project 1-37A INDEPENDENT REVIEW OF THE MECHANISTIC-EMPIRICAL PAVEMENT DESIGN GUIDE AND SOFTWARE This digest summarizes key findings from NCHRP Project 1-40A, conducted by three consultant teams headed by Professors Marshall Thompson, Ernest Barenberg, and Stephen Brown. Part I of the digest was prepared by Stephen F. Brown, Scott Wilson Pavement Engineering, Ltd.; Part II was prepared by Michael M. Darter and Gregg Larson, Applied Research Associates, Inc., and Matthew Witczak and Mohamed El-Basyouny, Arizona State University. Subject Area: IIB Pavement Design, Management, and Performance Responsible Senior Program Officer: Edward T. Harrigan NATIONAL COOPERATIVE HIGHWAY RESEARCH PROGRAM

between 1998 and 2004. This was colloquially known as the “AASHTO 2002 Design Guide,” but it was not completed by 2002 and has not, to date, been adopted by AASHTO; it is now termed the Mechanistic- Empirical Pavement Design Guide (MEPDG) and software. These independent third-party reports were conducted at the request of the, then, AASHTO Joint Task Force on Pavements. The detailed brief for the reviewers is shown in Appendix A and the member- ship of each of the review teams in Appendix B. In conducting the review, the engineering basis for each aspect of design and the associated com- putation procedures were assessed. Comments have also been made on the guidance given to potential users of the Guide. Three separate detailed reports, supported by appendixes, have been presented to the NCHRP Panel (1–3) and these form the background to this report, which summarizes the major points and recommendations. The raw material for this review consisted of the following items: • The final report of the 1-37A team (4) and • The Design Guide software (Version 0.7). Additional information was readily made avail- able by the NCHRP Senior Program Officer on re- quest. The approach taken to the various tasks was to study the reports, use the software, conduct dis- cussions among the review teams, and absorb feed- back from the NCHRP panel, following presenta- tions at meetings and in response to interim reports. In addition, experienced pavement engineers in the United States were consulted for their views about the Guide and how their DOTs intend to implement it locally. During the work, the teams became increasingly conscious of the very large parallel research effort— both in progress and being planned—to augment and extend the work done by the 1-37A team. Some of this work was in the form of NCHRP contracts and some was through the FHWA Implementation Group’s activities with the state DOTs. Presenta- tions at panel progress meetings provided some in- sights into this additional work, but, given that the work was incomplete, it was only used as back- ground to the review. However, the initial work re- ported on contract 1-40B, dealing with local cali- bration of the Guide, was influential in identifying both problems with the software and the limitations of the original national calibrations carried out by the 1-37A team. The panel decided, on the basis of this work, to curtail the sensitivity analyses for flex- ible and rehabilitated pavements that were being at- tempted by the review teams using the software. It also became apparent that the available version of the software did not incorporate the latest informa- tion from the 1-37A flexible pavement team. It fol- lows, therefore, that this review was not able to as- sess the software in as much detail as had originally been planned, but observations on the available ver- sion are given in Section 7. The technical developments reported by the 1-37A team are substantial and are presented in detailed reports supported by appendixes (4). The distress prediction procedures bring together for the first time in a single computer program several important aspects of pavement performance that have previously only existed in stand-alone mod- ules. In particular, procedures have combined envi- ronmental prediction with structural analysis and distress computation. Hence, a framework has been created for incorporating future research results, some of which will arise from NCHRP projects. The general practical guidance provided for pave- ment engineers in the 1-37A reports is, overall, highly detailed. However, the documents in their present form are not user-friendly and will require improve- ment if they are to help busy engineers. The docu- mentation and the software must be compatible and consistent, and the same unit of measurement should be used throughout. A further preliminary observation, from the pre- amble to the 1-37A report (4), is that the brief was “. . . development of a design procedure based pri- 2 TABLE 1 Review team leaders Team Leader Key Tasks NCHRP Contract Professor S. F. Brown Pavement rehabilitation, design reliability, executive 1-40A(03) summary report Professor M. R. Thompson New flexible pavement design 1-40A(01) Professor E. J. Barenberg New rigid pavement design 1-40A(02)

marily on existing technology.” This is an important point in presenting this evaluation. Since the 1-37A work began in 1998, there had been substantial progress in pavement engineering research, the find- ings of which were not considered for incorporation in the Guide. The review team also noted the remarks made by the 1-37A authors in the preamble to their report, in which the authors present their views about the strengths and weaknesses of their work and the further research and development activity that they consider desirable. The 1-37A authors have been en- gaged by NCHRP to conduct some of this additional work in parallel with this review. The key issues to be addressed are • Difficulties experienced with calibration of the design models against actual pavement behavior in the field, because of limitations with the Long-Term Pavement Performance (LTPP) database and the related difficulties in dealing with design reliability; • Problems encountered in developing the software; • The approximate nature of some of the distress prediction models, which have been included as “place holders” for improved versions to be introduced in future; • The need for enhanced application of finite element analysis to deal with the non-linear properties of the lower layers in flexible pave- ments; and • The need for procedures for designing con- crete pavements and overlays of less than 6-in. slab thickness. 2 THE DESIGN PROCESS Figure 1, taken from the 1-37A report (4), sum- marizes the design process used by the Guide soft- ware. The design process has three stages: 1. Evaluation (the data input stage), 2. Analysis (the complex distress prediction stage), and 3. Strategy selection (the design decision stage). A new philosophy has been adopted: applying mechanistic principles to carry out very detailed dis- tress development computations over the design life of the pavement and incorporation of these proce- dures within a comprehensive piece of computer software. In addition, empirical predictions of riding quality (smoothness) are also carried out using the International Roughness Index (IRI) as the quanti- tative parameter. The user must run the software repeatedly for a single design. Hence, the previous familiar, well-defined procedures, involving the di- rect determination of required layer thicknesses for a pavement to accommodate specified traffic, materi- als, environmental conditions, and performance re- quirements over a particular design life have been changed. For an engineer to design a pavement, it is necessary to assume an initial structure, compute the distress development over the required design life using the software, adjust the initial design if this ex- ceeds acceptable levels, re-compute the distresses, and continue in this way until a satisfactory design is achieved. Three levels of design are available; the user can select the one that best suits the level of input detail available. Level 1 requires extensive, detailed input, including data from the laboratory testing of materials, while Level 3 relies on empirical rela- tionships between easily obtainable parameters and 3 Figure 1 Flow Diagram of Design Procedure (after ARA, 4)

those required for the design computations. Level 2 requirements fall between these two extremes. For new pavement design, the details required for Lev- els 1 and 2 are not generally available at the design stage of a project, given that the actual materials may not be identified until construction is about to commence. These comments apply to the imported materials, but not the subgrade soils. Even so, the properties of soils and granular materials are diffi- cult to determine in the laboratory in a way that re- produces the field situation. By contrast, for an ex- isting pavement, field testing, particularly using the Falling Weight Deflectometer (FWD), can be carried out to determine the in situ properties of materials. In addition, for bound layers, cores can be cut to pro- vide appropriate laboratory test specimens. The first stage of the design process involves entering all the input data, which, for pavement re- habilitation design, includes all the field and labo- ratory testing associated with structural evaluation of the existing pavement. The first stage also em- braces input for traffic loading and for the Extended Integrated Climatic Model (EICM), both of which represent major advances relative to previous design guides. A key change in the Guide relative to earlier versions is the move from the concept of Equivalent Single Axle Loads (ESALs). This was based on very old data from the AASHO Road Test (5), the conditions of which differed significantly from those on modern highways, and application of the “4th Power Law” to determine the relative damag- ing potential of different wheel loads. The avail- ability of detailed traffic load spectra and the ability to perform numerous pavement structural analyses has led to this change. Consequently, more accurate distress predictions can be obtained using the actual traffic load spectra. The EICM, developed from earlier work (6–8), has been fully incorporated in the Guide software but its detailed application, which is central to the distress prediction computations, requires many pre- viously unconsidered material properties, many of which would be difficult to measure in a practical situation, so extensive reliance has to be placed on empirical methods of estimation. Use of the EICM allows the prediction of temperature and moisture conditions at all depths in the pavement throughout its design life. Given that the mechanical properties of materials and the development of distress are strongly influenced by these two variables, the soft- ware offers clear benefits. The definition of design reliability has changed since the previous AASHTO Guide (9) to reflect the use of distress prediction techniques. Reliabil- ity used to be defined as the probability that the ac- tual number of ESALs to a terminal serviceability would be less than the predicted number. In the new Guide, reliability is defined as the probability that a particular distress, such as fatigue cracking, will be less than a selected critical value. The Guide software incorporates structural analy- sis subroutines as an essential part of the design process in Stage 2. Finite element analysis has been used for rigid pavements and the process has been effectively sped up through the application of neural networks. Linear elastic layered system analysis is used and the program (JULEA) has been incorpo- rated for flexible pavements. Finite element analysis is available for dealing with the non-linear resilient properties of soils and granular materials at Level 1. Given the vital role of back-analysis in pavement evaluation for rehabilitation design, which requires that non-linearity be considered, it is difficult to see why the program does include a suitable subroutine. Indeed, very little is said about this key aspect of re- habilitation design. A particular level of reliability can be specified for each distress mode and for IRI, which has re- placed the Present Serviceability Index (PSI) as the measure of pavement smoothness. The inclusion of IRI, as a measure of riding quality in the Guide, does not jibe with the analytical (mechanistic) approach to design. It is not possible to compute IRI using theo- retical analysis and, indeed, the Guide deals with its prediction empirically, albeit by incorporating vari- ous predicted levels of distress. It can be argued that IRI has no place in a pavement design guide because it is, essentially, a tool for pavement management. Furthermore, if correct design is applied in order to limit cracking (and rutting, in the case of flexible pavements), global moisture-related movements are restricted, and good construction practice is fol- lowed, then good riding quality will result. Experi- ence among the users of the Guide suggests that the predicted values of IRI are rarely critical to assess a design. Agencies interested in assessing the value of IRI for a design that satisfactorily addresses rutting and cracking could still use empirical relationships to determine a value, but these relationships should not be regarded as a structural design parameter. The M-E design approach taken in developing the Guide involved incorporating theoretical con- 4

cepts for various aspects of the design process and modifying them through field calibration based, almost entirely, on data from the LTPP GP sections. This approach to deriving material “models” gener- ally involves the use of multiple regression analyses to derive material constitutive models from labora- tory or field data—an application of a philosophy that has been used extensively in the United States for many years. This approach is a poor substitute for applying sound theoretical concepts. Several of the design models, such as that for reflection crack- ing in asphalt overlays, are entirely empirical and lack the advantage of being derived from a theoret- ical base. A major shortcoming of the Guide, in its present form, is a lack of balance in the level of detail and accuracy that is combined in the distress computa- tions. For instance, it is possible to base a design on a combination of detailed, measured, complex mod- ulus master curves for HMA, but with simple esti- mates of subgrade and granular layer resilient mod- ulus derived from soil classification or values of the California Bearing Ratio (CBR). 3 FLEXIBLE PAVEMENT DESIGN The M-E pavement design approach is based on an assumption that load-induced pavement structural responses (e.g., stresses, strains, and deflections) can be used to predict the development of pavement dis- tress, in the form of rutting or cracking, through the use of Transfer Functions. Thus, the key factors of interest considered in the flexible pavement design review were the procedures used to characterize the elastic moduli of the various paving materials and the subgrade soils, because these form the most im- portant input to the structural response calcula- tions. The veracity of the transfer functions was also closely studied. A detailed review of flexible pavement design is given by Thompson et al. (1) and further obser- vations are included in Brown et al. (3). 3.1 Resilient Modulus for Subgrade The Guide uses CBR for soils and granular ma- terials as a basis to estimate resilient modulus (Mr). However, research has shown that the relationship between these parameters is not reliable for either material type (10, 11). This unreliability results from the non-linear stress-strain relationships involved and the fact that CBR is, at best, a measure of undrained shear strength, which does not relate closely to re- silient properties at relatively low stresses, but may relate to permanent deformation resistance. This relationship was derived from research at TRL in the United Kingdom. Croney (12) has noted that the relationship is based on wave propagation measure- ments for in situ Mr, which were at very low stress levels, and field CBR tests, the results from which were not very reliable. Given that the typical default values of Mr at OMC values provided in Table 2.2.51 of the Guide are based on the CBR relationship, these values should be reviewed. In addition, some guidance en- abling the user to select Mr values that consider non- linearity for granular materials and soils would be helpful, particularly for Level 2. Guidance for esti- mating values for fine-grained soils from parameters such as the clay content, Plasticity Index, and water content, could also be provided. For candidate gran- ular materials, a database of laboratory properties could be developed for local areas with an empha- sis on resilient modulus and resistance to permanent deformation. The Guide does not generally reflect develop- ments in soil mechanics of relevance to pavement design, with the exception of the way in which soil suction-water content characteristics have been used to adjust resilient modulus values. However, there are errors in the equation used by the Guide soft- ware (2.3.5) to compute the change in Mr as a con- sequence of changes in water content. In addition, no allowance is made for surface infiltration in flex- ible pavement design. 3.2 HMA Complex Modulus A prediction model for the complex modulus of HMA is used in the Guide for situations when exper- imental results are not available and for the lower lev- els of design. The so-called Witczak model is used, but others are available such as the Hirsch model (13) and consideration should be given to offering the op- tion of using either. The latest version of the Witczak model (14) should be incorporated. It would be helpful for the complex modulus computations to be available as stand-alone options in the software. Whichever method is used, the design computa- tions require that a complex modulus master curve be available that relates this parameter to reduced loading frequency. The frequency for a particular 5

situation is derived from an estimation of load pulse duration in the HMA sublayer under consideration. Given that this assumption is inherent in the Witczak model, the conversion of loading time (t) to frequency ( f ) is based on the relationship: When dealing with experimental data from sinu- soidal loading of materials, an alternative conver- sion has been used in other circumstances, notably in the classic Van de Poel approach to bitumen stiff- ness estimation (15): However, in the context of the Guide, Equation 1 is considered appropriate. In its present form, the Guide procedure for esti- mating complex modulus in the HMA layers results in unrealistic values, particularly for thick layers, showing a decrease in predicted modulus with depth in hot weather that is counterintuitive. This quandary results from the loading time/frequency effect over- riding the temperature effect. Selecting a design fre- quency related to vehicle operating speed is recom- mended as an alternative to the present method. Effective in situ loading time needs to be inves- tigated further, not least because its definition for use in design is complex. Brown (16) pointed out that, even for a particular vehicle speed, there is no unique load pulse length for the layer because the vertical stress pulse at a particular depth differs from the horizontal pulses and both also vary with depth. The procedure in the Guide uses the vertical stress pulse, which is a simplification that needs to be justified. Brown suggested that the appropriate value should be determined by averaging the verti- cal and the two horizontal pulse lengths and also av- eraging them through the HMA layer or sublayer being considered. 3.3 Fatigue Cracking Fatigue cracking of HMA is dealt with by pro- cedures for the traditional, bottom-up, cracking mode and by introducing, for the first time, a similar approach for top-down cracking. The procedures follow well-established lines, except that the labora- tory models have been heavily modified through a field calibration process, despite very large variabil- f t= 1 2 2π ( ) f t= 1 1( ) ity. No mention is made in the Guide of the possi- bility of an endurance limit in fatigue for thick as- phalt construction, which has been widely discussed in recent years (17). The following generic HMA fatigue algorithm should be introduced to give users an alternative option: in which AC =Maximum tensile strain in the HMA EAC =Stiffness of HMA layer K1, K2, K3 =Material constants Top-down cracking is assumed to be longitudinal without any reason being given. Top-down cracking has been a subject of much research in recent years and is a more complex problem than assumed in the procedure proposed in the Guide. The standard error of estimate is very large, making use of the model unreliable and impractical. It would be better at this stage to omit the prediction of this distress type, pending further research to give a sounder basis for design. 3.4 Rutting The Guide has abandoned the traditional vertical resilient subgrade strain criterion for rutting. This is welcomed, but the alternative method proposed for prediction of rut depth analytically, involving per- manent strain prediction models for each layer, is not adequate for practical use. The development of permanent strain in pavement materials is very dif- ficult to predict from basic material characteristics because of the many variables involved. For instance, in HMA, this particular mechanical property, even more than dynamic stiffness or fatigue cracking re- sistance, requires a laboratory test to be conducted on the material under consideration. This is because the aggregate structure has a fundamental influence on the result. Consequently, details such as particle sur- face characteristics, shape, and grading, together with packing and orientation after compaction, are all in- fluential and cannot reliably be predicted using an empirical model. The philosophy used in the Guide assumes that most rutting is caused by volume change in the HMA following compaction under traffic, whereas much N Kf AC K AC KE= ( ) ( )1 2 31 1 3 ( ) 6

field evidence shows that rutting is principally a shear- ing phenomenon, at least for well-constructed HMA layers, and a major output from the SHRP research supported this (18). The permanent deformation model used in the Guide takes the same general form for HMA, un- bound materials, and soils, and expresses permanent strain as directly proportional to resilient strain. There is an empirical relationship between these two param- eters for HMA, but no fundamental reason why this should be so. Permanent strain is essentially con- trolled by the level of repeated applied stress ex- pressed as a ratio between shear and normal com- ponents. Ongoing work in this field under NCHRP Project 9-30A will provide an opportunity to iden- tify improved techniques for modeling permanent deformation in HMA. Major assumptions have been made in the Guide procedure based, in some cases, on little evidence. One example of this is the relative contribution to rutting from the various layers, which was derived from minimal field data. For the present, it would be better to rely on good mixture design and testing to limit rutting from the HMA layers and to use allowable stress criteria to deal with the lower layers. Stress criteria could be based on accumulated research knowledge from re- peated load triaxial testing, which has identified “Threshold Stress” limits for many materials (17). The similar “Shakedown Limit” concept (19) also holds promise for future application. 3.5 Thermal Cracking The Guide uses a complex procedure to predict thermal cracking in HMA. However, the properties of the HMA layer are assumed to be constant with depth, which could present a problem for pavements with thick HMA construction. Difficulties could arise in the event of an unusually cold winter unless the database in the EICM is extended to accommo- date a longer history of records. 3.6 Other Materials The Guide procedures for design of pavements in- corporating chemically stabilized materials should be brought up to date to reflect recent advances, notably in connection with the use of lime stabilization. The Guide could be improved by including more advice on sustainable construction (e.g., issues such as recycling and the use of materials that would not comply with current specifications, such as marginal aggregates and industrial byproducts). Cold-mixed asphalt-treated granular materials and techniques, such as foamed bitumen, fall into this general cate- gory, but are not considered in the Guide. These var- ious materials will become increasingly important as demands for more sustainable construction are made by highway authorities and end-product, performance-based specifications are introduced. 4 RIGID PAVEMENT DESIGN 4.1 JPCP Design The design process for JPCP represents a signif- icant step forward. It is well formulated and presents the state of knowledge available to the 1-37A team at the time that their work commenced. The mecha- nistic concepts, incorporation of many parameters not included in previous design procedures (e.g., cli- matic effects, transfer functions converting pave- ment responses to pavement distresses, and incor- poration of traffic stream and axle load distribution concepts) are all to be commended. Typical designs for JPCP carried out by the re- view team (2) appear reasonable and agree with ex- perience. It is, therefore, considered that the Guide procedures for JPCP can be implemented when the detailed issues set out below have been resolved. 4.1.1 Issues to be Resolved The exact role that curling and warping plays during hardening of the concrete and during daily/ monthly temperature and moisture cycles is not clear from the documentation and models presented in the Guide. At issue is the arbitrary approach to estab- lishing equivalent temperature gradients to account for the built-in curling/warping gradient and the mois- ture gradient within the slab. Further experimental and modeling work is required to accurately deter- mine the effect on incremental damage. Curling and warping produce static, long-term stresses of a significant magnitude. It is not clear that these stresses can be simply added to those caused by transient traffic loading when using a linear fa- tigue damage accumulation model such as Miner’s Hypothesis. There is little discussion of how negative tem- perature gradients cause curling stresses to produce top-down cracking. It is not clear how these effects 7

are accounted for in the neural network computation environment. More documentation on top-down ver- sus bottom-up cracking and further validation of both phenomena are needed. There must be a separation between permanent warping that occurs during concrete hardening, and the warping that results from climatic changes during the pavement service life. How these effects com- bine to produce a critical tensile stress and the influ- ence of creep during the initial hardening stage should be considered. The Guide recommends −10°F as the effective temperature to determine permanent curl/warp. How- ever, this value will be affected by time of placement, joint spacing, load transfer at joints and base/slab interface conditions, some of which cannot be pre- dicted at the design stage. There is insufficient jus- tification for use of this single value. The model assumes shrinkage warping can be accounted for by use of an equivalent negative temperature profile that produces a concave upward curling of the slab. However, shrinkage warping is not always an addi- tive effect and might counteract downward curling. There is hardly any discussion of modeling for expansion and contraction of pavement slabs result- ing from the change of average slab temperature on a daily or seasonal basis. The effects on load trans- fer efficiency (LTE) and of compressive force in the pavement during summer on fatigue calculations for transverse cracking need to be considered. Local calibration/validation of the design proce- dure is required using local experience. State DOTs are encouraged to develop catalogs of designs for their conditions to ensure that the designs are ratio- nally developed and incorporate agency experience. The omission of longitudinal cracking as a dis- tress mechanism is questionable. It is a particularly damaging form of distress because such cracks can propagate indefinitely and are known to be a serious form of distress in the Western United States. In ad- dition, corner cracking is not included as a quanti- fied distress mechanism, although it is discussed in the Guide. 4.1.2 Input Data A major issue is the amount and complexity of input data called for in the design procedure. Some data are highly critical; others have relatively little effect on design. A procedure to classify these data into three categories is recommended. Type 1 data could incorporate information that is absolutely vital such as • Traffic data; • Slab thickness; • Concrete strength, shrinkage and thermal properties; • Subgrade properties; • Base type and properties; and • Climatic conditions. Type 2 data could embrace information not used in the various design models or for which the design procedure is not sensitive. Type 3 data could incor- porate parameters to which the design procedure is sensitive but the appropriate data would not normally be available at the time of design. Inclusion of values that are tentative at best could lead to misleading con- clusions with regard to the sensitivity of the design to these parameters. The Guide recommends that the design value for Modulus of Subgrade Reaction (k) be obtained from an assumed value of resilient modulus that is, in turn, obtained from a correlation with CBR. This approach is regarded as very unreliable. A dynamic k value is assumed, whereas CBR is obtained from a static test. This issue is discussed in more detail in Section 2.2 in connection with flexible pavements. Different soil types will have significantly different relationships between dynamic and static test results. Estimated resilient modulus values given in the Guide for the base and subbase layers seem too high, particularly when they are estimated from soil classification. 4.1.3 Design Sensitivity The software for rigid pavement design was shown to work well and run much faster than that for flexible pavements. Consequently, the Review Team carried out some work on sensitivity issues that re- vealed some anomalies. In one particular study, it was found that subgrade type and strength had very little effect on slab cracking, even though the subgrade moduli values ranged from less than 1,000 psi to nearly 30,000 psi. In many cases, pavements on stiffer soil performed worse than those on softer subgrades. In an equal number of cases, the opposite was true. 4.2 CRCP Design The design procedure for CRCP is a major step forward in understanding factors that might affect the 8

behavior and performance of pavement. The analy- sis procedures provide valuable guidance on the im- portance of key parameters affecting the performance of CRCP. The algorithms incorporated to compute structural response and estimate damage are logically sound and instructive. However, the models used to predict punchouts, the principal failure mechanism, are extremely complicated and only one possible mechanism has been considered. The review team considers that the models do not correctly analyze the response and failure mechanisms of CRCP due to loads and climatic conditions, because the equa- tions appear to be more research-oriented than use- ful as design tools. The procedure requires numer- ous inputs to solve more than 70 equations. Even though many of the input variables are computed or assumed by the software, many of them are not familiar to most design engineers. For the above reasons set out in detail in Barenberg et al. (2), the review team believes that major issues must be resolved before the CRCP design procedure can be implemented. 4.2.1 Issues to be Resolved One of the primary issues to be resolved relates to determining the thicknesses of PCC required to provide a desired level of reliability. Using the Guide, the thickness of CRCP is equal to or somewhat greater than that for JPCP, which is contrary to the de- sign procedures used in many states (for which sim- ilar thicknesses are apparent). The excessive CRCP thickness is probably caused by the Guide’s proce- dures for estimating factors that affect punchouts. A basic assumption made in the design proce- dure is that there is a good correlation between crack spacing and crack width. Field observations reveal that this is not the case—crack width depends on the time when the crack occurs because time affects the shrinkage drying that initiates cracking. Another reason is that crack spacing varies and crack width depends on the spacing of the two cracks it dissects. Principally for these reasons, the review team questions the accuracy of the entire algorithm for punchout prediction. A major concern is the apparent confusion be- tween bond and friction between the base and slab. Bond prevents one layer from moving relative to the other, whereas friction indicates a force needed to cause relative horizontal movement. Bond will pre- vent one layer from lifting from the other whereas friction has no resistance to vertical separation. For CRCP design, no account is normally taken of bond but slab/base friction is considered in order to pro- vide for changes in slab length. Another concern is the effect of crack spacing on punchouts. The descriptions in the 1-37A report (4) imply that the probability of punchouts increases as the transverse crack spacing becomes smaller. At the same time, there are several statements to the effect that smaller crack spacing will result in smaller crack width with better load transfer efficiency. In some cases the number of punchouts was found to increase with the base stiffness, which is counter- intuitive; the exact reason for this phenomenon is not known and must be resolved. 4.2.2 Design Sensitivity Punchout distress was found to be sensitive to co- efficient of thermal expansion, curling and warping, slab thickness, percent steel, and concrete strength, but not to aggregate type or to the erodability of the base, which is contrary to the experience of most DOTs and to the impression left by the 1-37A report. Base/slab friction has no effect on performance. Although curing effectiveness influences crack devel- opment and spacing and, hence, punchouts, the com- putations are insensitive to curing method. Designs are also insensitive to coarse aggregate type, which is contrary to the experience of many state DOT engineers. 5 PAVEMENT REHABILITATION Detailed observations on the Guide’s coverage of rehabilitation design are given by Brown et al. (3). Given that the same philosophy and distress models are used for the design of pavement rehabilitation as for new pavements, the observations in Sections 2 and 3 of this report also largely apply to rehabilita- tion design. The principal difference is at Stage 1 (Figure 1), in which the input data accommodate the results of material and structural evaluation in- vestigations carried out on the existing pavements. The Guide should emphasize that rehabilitation de- sign will become increasingly important because the highway system in the United States has largely been built. Also, rehabilitation design can be con- ducted to a greater level of reliability than new pave- ment design, because the highway already exists and measurements can be made of key parameters 9

(e.g., traffic, material properties, and structural in- tegrity) through field and laboratory testing. Equi- librium water conditions will have been established in the subgrade and the level of distress that exists at the time of evaluation can be quantified. 5.1 Pavement Evaluation An advantage of rehabilitation design is that field testing, particularly using the Falling Weight Deflectometer (FWD), can be carried out to deter- mine the in situ properties of materials. The pave- ment survey procedure in the Guide specifies very wide spacing (typically 500 ft for FWD) between test points, even for Level 1 analysis. Such spacing may be appropriate for network analysis, but for detailed evaluation of a site, from which rehabili- tation measures are to be developed, a spacing of around 60 ft is desirable. Other field testing tech- niques, including the Dynamic Cone Penetrometer and Ground-Penetrating Radar can also be effec- tively deployed for existing pavements. In addition, for bound layers, cores can be cut to provide appro- priate laboratory test specimens. The application of good structural back-analysis of FWD deflection data provides vital information on the effective stiffness of each of the pavement layers. This information can be used to assess both the level of distress (particularly through cracking) and to provide input for overlay or other rehabili- tation design measures. One of the most important aspects of back-analysis is the ability it gives the designer to fit the theory to the actual pavement by ensuring that the measured resilient response of the pavement matches the value determined from structural analysis. This is a situation that cannot be achieved when analysing a pavement that has yet to be constructed. In the Guide, the characterization of existing HMA stiffness when dealing with HMA overlays and with PCC overlays is inconsistent. For the HMA overlays, FWD back-analyzed values can be used, but not for the PCC overlays, which are considered illogical. Use of the FWD for evaluation of concrete pavements is less well covered in the Guide than that for flexible construction. In particular, the procedure for determining a Modulus of Subgrade Reaction (k) from back-analysis needs to be clarified. In addition to its use for determination of in situ effective layer stiffnesses, the FWD can be applied to the measurement of other important pavement characteristics such as interlayer bond, concrete joint properties, and the detection of incipient cracks in cement treated bases. Use of the FWD can either be part of a rather simple evaluation process based on measured deflection parameters, an intermediate process using simple two-layer back-analysis, or a more complex procedure for experienced pavement engineers (3, 19). 5.2 Rehabilitation with HMA Relative to the advice given in the Guide, more reliance should be placed on data from field testing and less on laboratory testing of cores. The stiffness of cores tested in the laboratory is likely to differ from FWD-derived values, even after allowance for possible loading time and temperature differences. This difference occurs because cores essentially give point values for coherent material, whereas FWD- derived values represent the effective stiffness in situ, accounting for cracking, de-bonding, or voided asphalt materials. Improved advice on the effect of age hardening and traffic damage on HMA should be given in the Guide, in order to better understand the in situ material characteristics for rehabilita- tion design. The Guide requires that the “as new” material properties for an existing pavement be determined. Given that the pavement may have been in service for many years, making this determination is diffi- cult. It is also illogical because the pavement param- eters that influence rehabilitation design should be the existing ones, not those that may have occurred at the time of construction. This feature of the Guide results from the same approach being adopted for rehabilitation as for new construction. The Guide uses equivalent laboratory values of resilient modulus for unbound material derived from the FWD values using simple ratios based on earlier work. This approach does not seem well-founded. Design stiffness for chemically stabilized materi- als is mainly related to laboratory values in the Guide. However, weaker stabilized materials may well have deteriorated to a granular condition after many years of trafficking in an existing pavement. Because this condition can be quantified using back-analysis of FWD data, such analysis is recommended for all lev- els of design. Interface conditions between various pavement layers are considered in the Guide. A value between 0 (no bond) and 1 (full bond), apparently on a non- 10

linear scale, can be incorporated in the analysis for calculating load-induced stresses and strains. The concept is believed to be based on assumptions in the JULEA program. However, advice on the sen- sitivity of the design thickness to this parameter and the recommended values for various cases are not described. The designer commonly assumes that full bond exists between all pavement layers. However, field studies have shown that this is often not the case and the power of FWD back-analysis could be har- nessed to quantify an interlayer stiffness (20). Reflection cracking is noted in many sections of the Guide as a distress mechanism commonly asso- ciated with HMA overlays to both rigid and flexible pavements. However, reflection cracking is treated simplistically compared with other aspects of perfor- mance, and prediction is based on a simple empirical formula. The empirical reflection crack model pro- duces reasonable results, but needs to be expanded to cover a wider range of pavement conditions in terms of thickness, stiffness, and joint characteristics. Some suggestions are given by Brown et al. (3). The use of a fabric placed below the HMA over- lay to prevent or delay reflection cracking is one of the techniques included in the Guide. For design purposes, it is assumed, without justification, that a properly installed fabric can reduce the HMA overlay by 2 in., but this value is an entirely em- pirical allowance. The Guide makes no specific ref- erence to the use of grid reinforcement in asphalt. Materials of this type provide genuine reinforcing elements, the best ones having the ability to inter- lock with the HMA aggregate. They can be made of high-tensile polymeric material, glass fiber, or steel, and some are supplied with a geotextile back- ing to aid installation. When correctly installed, some of these materials have been shown to be effective at reducing reflection cracking, particularly from thermal movements. A recent report provides useful information on this subject (21). Advice on optimizing strengthening recommen- dations and the available options, other than HMA overlay, should be given in the Guide. These can in- clude inlays, combinations of inlay and overlay, par- tial or complete reconstruction, and recycling options. 5.3 Rehabilitation with Concrete For the determination of flexural strength in existing concrete, consideration should be given to specifying an indirect tensile test and deriving the flexural strength from this test. Indirect tensile tests are much easier to perform than flexural tests because they can be carried out on cores, rather than on beams. In addition, the relationship between in- direct tensile and flexural strengths is believed to be much more certain than that between compressive and flexural strengths. The effect of load transfer efficiency (LTE) on computation of stresses in PCC and in the prediction of joint faulting is highly significant and strongly affects the occurrence of traffic-related reflection cracks. Therefore, it is appropriate that considerable effort has been made in the Guide to express LTE as correctly as possible. However, the details relating to effects of aggregate interlock and dowel LTE ap- pear to require some corrections (3). Faulting is most sensitive to a quantity described as “differential energy of subgrade deformation,” calculated in the Guide from the deflection under load and various joint parameters. Although this cal- culation is entirely logical (because it is the factor that takes account of the effects of traffic loading), the calculation is surprisingly insensitive to subgrade fines content and the number of wet days per year. Calibration work for concrete pavement restora- tion should be carried out using data from sites cov- ering a wider range of conditions than previously, and more data are needed on the performance of CRCP overlays to existing PCC pavements for heavily trafficked sites. The prediction of transverse cracking and joint faulting in JPCP overlays does not appear to match the available evidence particularly well. The issue should be investigated further. 6 DESIGN RELIABILITY Pavement design and rehabilitation are complex and involve many uncertainties, variability, and ap- proximations. Reliability is critical for estimating the effect of construction and material variability on pavement design and performance. The same vari- ance and reliability models have been used for de- sign of both new and rehabilitated pavements. Design reliability is defined as the probability that each of the distress types will be less than a selected criti- cal level over the design period. Average inputs are used in the design guide and average responses and distress are predicted. The various distress mecha- nisms are taken to have normal (Gaussian) proba- bility distributions. 11

Variability and, hence, reliability has been esti- mated by comparing predictions with measured val- ues from LTPP sites. Depending on the level of the input data, variability will include factors such as measurement errors and errors associated with ma- terial characterization parameters, traffic and envi- ronmental conditions, and errors associated with the model prediction algorithms. For the deterioration modes, there was more calibration data available at low levels of pavement distress than for more seri- ously damaged pavements. This means that the vari- ability determined for the higher levels of distress is likely to be less reliable because it is based on fewer data points. Apart from thermal cracking, the same standard deviations have been used for all design Levels. This implies that there is no reduction in vari- ability, and hence improvement in reliability, when more accurate input parameters are used. The use of techniques such as a Monte Carlo sim- ulation should be investigated so that the variability of input parameters can be used directly to estimate the variability in predicted performance and, hence, the reliability of design. Reliance on LTPP data as a basis for model cal- ibration and establishing the reliability of designs is questionable. A recent report by Hajek et al. (22) con- cerning the traffic data for LTPP sections raises seri- ous questions about the reliability of the data. A major concern with numerous calibration con- stants used in the Guide is their application at so many levels. It would be preferable if the procedures could progress as far as possible using only analyti- cally based concepts (without calibration constants) and apply a single adjustment or correction factor at the end, if it is required. To form judgments about this, it would be necessary to have high-quality field performance data available. Such data are generated from Accelerated Full-Scale Pavement (APT) test- ing such as that conducted at WesTrack (23), MnRoad (24) and, more recently, at the National Center for Asphalt Technology (25). 7 DESIGN SOFTWARE The version of software made available for this review (dated June 2004) did not incorporate the lat- est revisions for flexible pavement and rehabilitation design from the 1-37A contractors. Nonetheless, it is clear that the software has many useful characteris- tics that can be of great help to pavement engineers. However, this version had many errors, which re- sulted in failed runs, and numerous problems were encountered by the many users who worked with it. When used for rigid pavement design, the software performed much more efficiently and quickly than for flexible pavements. However, nearly all reviewers re- ported some problems and, in a number of instances, calculations and results did not appear to agree with what was written in the Guide. Specific issues that arose from the review team’s use of the software are set out in the detailed reports (1–3). The Guide’s very detailed analytical computa- tions pertaining to distress prediction procedures, which are also incorporated in the software, result in excessive computing times, particularly for flex- ible pavements. Given that the user has to make manual adjustments to an initially unsatisfactory design before re-running the software and has to continue to do this until the design is acceptable, the overall design process requires considerable time. The software provides a Microsoft Windows envi- ronment. Users might expect the iterative compu- tations to be incorporated in the software so that various “what if” scenarios (such as alternative ma- terials) could be explored with the outputs given in terms of required layer thicknesses. The input data required for design are extremely comprehensive. A tabular summary with critical observations has been provided by Brown et al. (3). Much of the data input seems too detailed and un- realistic or unnecessary, such as the requirement to specify cement content for PCC when mechanical properties are also required. Very extensive data are required on climatic conditions, requested on an hourly basis. This relates to the very detailed analy- ses of distress, the results of which are presented on a monthly basis. For rehabilitation, the assump- tion that construction records will be available and provide a reliable source of input for evaluation and design is naïve. It is far better to measure the charac- teristics required by the software, preferably through field testing or, alternatively, by laboratory testing of field samples. Given the desirability to input detailed traffic loading data, the software developed under NCHRP Project 1-39 must be made compatible with the Guide software. Compatibility would allow the detailed data collected by DOTs to be directly used in design. In rehabilitation design, the input data required for existing HMA and PCC are similar to those for new materials. Parameters such as cement type and the curing technique for concrete or rheological char- 12

acteristics of HMA binders in their “as-new” state after short-term aging, are impossible to determine for an in-service pavement. The various changes proposed for the design process, which have been outlined in this report, should be incorporated into the software. In addition, it is recommended that stand-alone versions of the structural analysis and EICM subroutines should be accessible to the user so that particular outputs, such as values of stress or strain, could be accessed. Appendix D of the 1-37A report is the software User Guide that, helpfully, takes the reader through some typical designs with views of the relevant screens provided to illustrate what can be expected as a design case is being worked through. However, inconsistencies in terminology between this docu- ment and the main report will need to be corrected in the context of publishing a user-friendly manual for the whole design process. Serious work remains to be done on the software. A professionally conducted beta testing exercise is recommended before a final version is made freely available for the use of design engineers. It would be very unfortunate if new versions contained the level of significant errors apparent in the review version. 8 RECOMMENDATIONS The three detailed reports, which form the back- drop to this digest (1–3), have set out numerous es- sential and desirable actions that the review team considers necessary before the Guide is acceptable for use in engineering practice. The sections below set out the most important of these changes and at- tempt to capture the essential features of the review’s overall recommendations. 8.1 Flexible Pavement Design 8.1.1 Essential 1. Incorporate the new prediction model (Bari and Witczak, 14) for HMA complex modulus and allow users the option of using the Hirsch model (15). 2. Improve the application of soil mechanics principles to the characterization of soils and unbound materials and, in particular, omit the use of CBR in design. Provide improved guidance on the estima- tion of Mr from basic soil characteristics. Correct the adjustment procedure for Mr resulting from changes in water content. 3. Resolve queries about the correct determina- tion of in situ loading time for an HMA layer. 4. Omit the top-down cracking distress predic- tion computations and await the results of ongoing research on this subject. 5. Replace the rut depth prediction procedures with good HMA mixture design combined with a permanent deformation test and the introduction of allowable stress levels for the granular layers and subgrade. 6. Provide a more flexible approach to the mod- elling of asphalt fatigue and introduce the concept of an endurance limit, particularly for heavy-duty pavements. 8.1.2 Desirable 1. Omit the use of IRI as part of the structural design procedure, because it is essentially a pave- ment management tool. 2. Emphasize local calibration of the models and predictions rather than adjusting national calibrations. 3. Improve the guidance given to use of chemi- cally stabilized materials and to other cold-mix and recycled materials. 4. Include surface water infiltration as a factor to be accounted for in design. 8.2 Rigid Pavement Design 8.2.1 Essential 1. Simplify the design process, particularly with regard to the required inputs. 2. Conduct state-level calibration/validation for JPCP. 3. Improve the treatment of curling and warping for JPCP. 4. Address the incompatibility between CRCP design thicknesses generated by the Guide and cur- rent practice. 5. Improve the way in which the “zero stress” condition is defined with respect to the time of con- struction for CRCP. 6. Consider other mechanisms for CRCP pun- chout failure, in addition to the one used in the Guide. 8.2.2 Desirable 1. Include longitudinal and corner cracking as distress mechanisms and provide improved advice 13

with regard to raveling. Review the permanent warp assumptions when a granular base is used. 2. Improve the way in which the Guide deals with climatic effect for JPCP design. 3. Address the issues of crack width and crack spacing as part of the validation process for CRCP. 4. Reconsider the significance of curing condi- tion and base erodability for CRCP. 5. Resolve the effect of stabilized bases for CRCP and the relationship between friction and bond. 8.3 Pavement Rehabilitation Design 8.3.1 Essential 1. Use in situ material properties obtained from pavement evaluation as input parameters for reha- bilitation design. 2. Give better advice on HMA stiffness predic- tion for existing pavements. 3. Give advice on other uses of the FWD, in addition to the determination of pavement layer stiffnesses. 4. Specify closer spacing for FWD testing, cor- ing, and DCP testing for the various design levels. 5. Investigate and carry out more research of laboratory-resilient modulus predictions of unbound materials from field values determined from FWD data using various conversion factors. 6. Improve the procedures for structural evalu- ation of concrete pavements. 7. Improve the determination of LTE between slabs and across cracks. 8. Check and correct, as appropriate, the detail concerning base erodability, upward curl, and over- burden on subgrade in relation to the computations for faulting in concrete slabs. 8.3.2 Desirable 1. Give recommendations on the effect of inter- layer bond condition on pavement evaluation, life prediction, and recommended treatment. 8.4 Software 8.4.1 Essential 1. Correct all the errors in the software and up- date it in the light of developments. 2. Conduct a professional beta testing exercise before issuing the software for general use. 8.4.2 Desirable 1. Allow the HMA complex modulus prediction subroutine(s), together with JULEA, EICM, and the finite element analysis programs to be used indepen- dently of the design process. 2. Provide for critical response parameters in terms of load-induced stresses and strains to be avail- able as part of the software output. 8.5 Other Matters 8.5.1 Essential 1. Address the shortcomings in the calibration procedures for rehabilitation models used in rigid pavement rehabilitation. 2. Re-address the whole issue of reliability in light of the work dealing with local calibration under way in Project 9-30. 3. Investigate the application of Monte Carlo simulation or alternative methods in order to im- prove the understanding of variability in design and the effects of variability in input data. 4. Improve the guidance to users of the Guide within a more succinct, user-friendly version of the document. Ensure that the terminology in the soft- ware and in the Guide is consistent. 5. Revise the Guide for use in the short term by offering a more balanced approach in which the various elements of design at each level are approx- imately of the same complexity. 6. Expand the advice on recycling options. 7. Introduce the concept of long-life pavements. 8.5.2 Desirable 1. Omit IRI from the design process because it is a pavement management tool and cannot be pre- dicted analytically. 2. Given the problems encountered with the use of LTPP data for calibration of many elements in the Guide, reconsider adoption of the national calibra- tion, at least for flexible pavements. 3. Over the long term, reduce the dependence on empiricism in the Guide and increase the basis of sound theory, making use of the wide range of research conducted since work on the Guide began in 1998. 4. Introduce simple design charts based on local calibration for various states. 5. Propose a better model for reflection cracking. 14