The Development of Models to Predict Pavement Performance in Frost Conditions JOINT CSHRP/QUEBEC BAYESIAN APPLICATION Recent analysis of the performance data of the Quebec pavement network inventory system show that some 20% of the 30,000 km network is experiencing distresses associated with frost action. It is also observed that the rate of deterioration for these pavements, in terms of change in roughness with time (dIRI/year) is, on average, twice that of pavements unaffected by frost. This translates into a multimillion dollar problem in additional annual maintenance and rehabilitation costs. Additional costs related to construction of frost resistant pavements and indirect costs such as the lost of productivity due to spring load restrictions or increased vehicle damage associated with increased roughness also have to be considered. It is generally felt that current design methods can not properly address severe frost conditions and do not allow for optimization of structural design. The Quebec ministry of transportation has therefore undertaken a major study with the following two objectives: 1) Revisit the whole approach to pavement design in frost conditions and develop a rational method which specifically deals with freezing and thawing of pavement structures. 2) Review and assess the performance of existing mitigation methods and evaluate the potential of using alternative insulation materials such as plastic, rubber and wood residues. The development of a rational design procedure involves the assessment of the cost benefit ratio of different design options. Deterioration models are thus required to allow for the evaluation of the expected performance of each option. 2. JOINT CSHRP/AGENCY APPLICATION PROGRAM The main research activity of the Canadian Strategic Highway Research Program (CSHRP) is the Long Term Pavement Performance (LTPP) program. The general objectives of CLTPP are as follows: To evaluate Canadian practice in the rehabilitation of flexible pavements and to subsequently develop improved methodologies and strategies. To identify ways to increase pavement life through the development of cost effective rehabilitation procedures based upon the systematic observation of in service pavement performance. In order to support these objectives, the ten Canadian provinces have undertaken the systematic observation of 65 inservice test sections located across Canada. The pavement types under study include asphalt concrete constructed over granular base courses which were overlaid in 1989 or 1990 using either virgin or recycled asphalt concrete. The experiment is scheduled to extend over at least ten years (1999) and possibly 15 years. In essence, CLTPP is a full scale field experiment designed to provide real time data which will be used to examine the trends relating the input factors of design, construction and service condition to the measured field performance. One of the important deliverables of CLTPP is the development of a Bayesian modeling methodology and supporting software (BSTAT and XLBAYES). The Joint CSHRP/Agency Bayesian Application Program is aimed at transferring the Bayesian methodology to the provincial highway agencies and providing the agencies with an analytical tool to solve specific problems in their jurisdiction. The joint application program involves CSHRP, eight participating provincial agencies and a consulting team. CSHRP is responsible for the overall project management, for the contracting arrangements with the consultant, for supplying the CLTPP data and for final report publication. Each provincial highway agency is responsible for the definition of the problem, for the supply of the necessary resources for the analysis and for the preparation of the final report. Finally, the consultant (VEMAX/DFI) is responsible for technical support on Bayesian statistics and model development. A training session was provided by CSHRP and by consultant team to the interested agencies during which a model development template was proposed. Participating agencies then undertook the definition of their specific problem and model development. A midterm workshop was also held to provide a forum for the agencies to present their modeling application, exchange ideas and experience with the others participants and receive feedback from the consultant. Projects are scheduled to be completed by the end of July 1995 and the results of some of the applications will be presented at a special session of the annual Conference of the Transportation Association of Canada in Victoria, B.C. in October of 1995. 3. PROBLEM STATEMENT 3.1 Frost Action on Pavements During winter, frost penetrates into the pavement system and generally reaches the subgrade soil. In a frost susceptible layer, water is sucked through the frozen fringe and forms an ice lens. As the frost front progresses, several ice lenses may form. The accumulation of segregation ice (lenses) causes a volumetric increase of the soilpavement system which translates into heaving of the surface (Figure 1). The phenomena is rarely uniform; therefore heaving is generally uneven which causes distortion of the surface of the pavement. The main distresses occurring at the surface of the pavement and associated with the heaving phenomena are roughness and longitudinal as well as meandering cracking. During spring, the accumulated segregation ice melts, thus increasing pore water pressures. Under poor drainage conditions, bearing capacity of subgrade soils can be drastically reduced, thus aggravating fatigue related distresses such as alligator cracking and structural rutting. 3.2 Current Design Procedures There are two types of design approaches typically used for the mitigation of frost action on pavements. The first approach puts the emphasis on the limitation of damages due to frost heaving. Most of the currently used methods of this type are empirically based. They typically prescribe the total pavement thickness required to achieve full or partial (ex.:50% of frost penetration) pavement protection. Calculation methods of various levels of complexity and precision are then used to predict frost penetration. Other methods use mechanistic based models to predict frost penetration and the resulting response (heaving) of the pavement system. The response is then compared to allowable limits, generally derived from experience. The second approach is based on the adjustment of the structural design in order to compensate for the loss of bearing capacity during spring thaw. Methods of this type typically use reduced spring or equivalent annual mechanical properties of soils and materials to account for the spring thaw effect in standard structural design procedures. Few of these approaches allows for the assessment of the expected performance of a pavement structure in given climatic and traffic conditions, nor do they allow for the assessment of the consequences of actions taken (or not taken) to mitigate frost related problems. It is felt that current methods have severe limitations. Some parameters, such as the longitudinal and transversal variation of frost susceptibility of the subgrade soil which are likely to have a significant impact on pavement performance, are not taken into consideration in current design procedures. Further, most methods are insensitive to important site specific conditions (geomorphology, drainage, etc.) affecting temperature and water regimes. Moreover, current methods do not associate factors related to frost action with long term performance. 3.3 Proposed Design Procedure The proposed design procedure is meant to be complementary to existing design methods. As shown on Figure 2, preliminary design, focusing mainly on mechanical aspects, would be done using existing design procedures. The output of this preliminary design, in addition with information on factors related to frost action, design objectives and constraints is then used as input for the new design model. The proposed model is a fourstep iterative process. The first step uses stateoftheart mechanistic models to predict mechanical and thermal responses of the pavement system in given conditions. This predicted pavement response is then used to estimate pavement performance through distress specific models. The third step is to verify if design objectives are met within the specified constraints. If the design is not optimal, improvements will be proposed in the fourth step of the design model through an expert system (or decision trees). Subsequent iterations will allow for the validation of improved structures until a satisfactory solution is reached. The unique feature of this procedure is it's capacity to predict long term performance of pavement structures subject to frost action. The development of the performance models which will link response to performance is the main challenge in developing this design procedure. 4. THE DEVELOPMENT OF PERFORMANCE MODELS 4.1 Model type Haas et al.[1] have grouped prediction models into four basic types: 1) Mechanistic models, 2) mechanisticempirical, 3) empirical or regression and 4) subjective models. Purely mechanistic models predict fundamental response of the pavement system (stress, strain and displacement) for given conditions. These models are not practical unless the response is associated with performance using deterioration functions which are usually derived by regression analysis. If this approach is taken, these models are typically referred to as mechanisticempirical models. Empirical models are obtained by directly linking site characteristics (structure, climate, traffic, etc.) with functional and/or structural deterioration indicators (dependent variables) This is generally accomplished using statistical regression analysis. If the link is built using engineer's experience exclusively, the models are then referred to as subjective models. In order to accommodate the assessment of various design features, design models need to be flexible and specific to factors that can be controlled by the designer. Hence, as discussed earlier, the design model will ultimately be mechanisticempirical based. However, it was believed that the development of first generation empirical models would provide a good opportunity to learn the process while at the same time create useful products for simple design or management applications. Figure 3 illustrates the concept of empirical and mechanisticempirical modeling for mechanical action as well as for frost action on pavements. Models are needed to link site characteristics and long term performance (empirical path), and to link site response to performance (mechanistic empirical path). 4.2 Performance indicators As summarized previously, the two dominant aspects of frost action are heaving during winter and loss of bearing capacity during spring. It is believed that the major distresses associated with heaving are distortion and longitudinal (including meandering) cracking. The loss of bearing capacity during thawing will in turn amplify fatigue type distresses such as fatigue (alligator) cracking and structural rutting. It is postulated that pavement deterioration due to frost action is adequately represented by these four distress types; therefore, they will be used as dependent variables in four distress specific performance models. The work completed thus far in the project entails the development of the two models associated with frost heave, namely: distortion and longitudinal cracking. The remaining two models (fatigue cracking and structural rutting) will be dealt with in subsequent stages of the overall project. 4.3 Methodology Empirical as well as mechanisticempirical models require a large number of observations in order to provide a good coverage of all factors involved over a reasonable inference space. Since the performance models are fundamentally a function of time, data need to be available over a reasonable time span for pavement being observed (10 to 20 years). This usually leads to major problems in pavement performance analysis due to the lack of good quality lifecycle data. Even large research programs such as SHRP, which includes the intensive observation of some 2000 pavement test sections, have experienced major problems in achieving statistical significance in early performance analysis. This problem was recognized early by the Canadian Strategic Highway Research Program (CSHRP) which was facing a small sample size problem while being committed to deliver early performance analysis results. This led CSHRP's technical steering committee to adopt an alternative modeling strategy known as Bayesian analysis). Bayesian statistics were developed specifically to cope with small sample size by providing a structured way to introduce prior information (prior) such as old databases or expert judgment into the regression analysis. The Bayesian algorithm balances prior information with new field observations as a function of the relative degrees of freedom between the two data sets and the relative fit of the prior and data models to the specified functional form. The fit of the models is in turn a function of the sample size and it's distribution. Figure 4 illustrates the three phases of Bayesian analysis needed to develop performance models within this project. A major product of CSHRP is the Bayesian modeling methodology and the supporting software (BSTAT and XLBayes) to assist in the development of Bayesian pavement performance models. The methodology was successfully used in the early analysis of the Canadian Long Term Pavement Performance (CLTPP) project data, and most recently in the joint CSHRP/Agency application projects by eight provincial highway agencies. The development of performance models for pavements subjected to frost conditions is part of the joint applications efforts. 5. LITERATURE REVIEW A literature review was conducted focusing on pavement performance modeling and more specifically on work related to quantifying the effect of frost action on pavements. Several documents were identified in the literature search. We first noted that there was much confusion around the definition of pavement performance. For some authors, performance was related to the response of the pavement structure (heaving, deflection, strain, ...) under a given solicitation by traffic and climatic agents. In other cases, performance was defined as a rate of deterioration. This confusion has been noted by Haas, Hudson and Zaniewski [1] who propose the use of the terms (damage) or (deterioration) to describe long term performance of pavements. Two working papers written by Kajner et al.(1994) [2] and Kajner and Sparks (1994) [3] describe the methodology proposed to develop deterioration models using Bayesian statistics. The first paper deals with the development of a roughness model for asphalt concrete (AC) overlays on AC pavements. In this paper, relevant literature is reviewed and summarized. Recommendations are made on the selection of dependent variable, independent variables and functional form for the regression model. Two development strategies are also proposed. The second paper has a similar content but deals with reflection cracking in asphalt concrete overlays on flexible pavements. The two papers provide good summary of the stateoftheart with respect to deterioration models for these specific distresses. General characteristics of the roughness models developed by Rauhut et al.(1993)[4], Queiroz et al.(1987)[5], Watanada et al.(87)[6], Paterson and AttahOkine (1992)[7] and Bein et al.(1989)[8] as described by Kajner et al. are summarized in table 1. The literature search did not identify a large body of research relating pavement deterioration to frost action and except for one document [12], work reported [9][10][11] deals exclusively with the effect of spring thaw. White and Coree (1990) [9] have revisited the AASHTO road test results focusing specifically on the effect of spring thaw. Using survivor analysis, they made interesting observations on the relationship between the structure's total thickness, the number of freeze thaw cycles, and loads. Based on their findings, it is possible to estimate the probability of a structure to survive (n) freezethaw cycles or to assess the load restriction needed for the structure to survive spring thaw. Esch et al. [10] studied the performance of 120 paved highway sections in an attempt to identify factors associated with good performance. Alligator cracking, rut depth and spring peak deflection were used as performance indicators and attempts were made to relate them with 175 material, dimensional and environmental factors. The most significant factor affecting pavement performance was found to be the percentage of material in the base and the subbase layers passing 75 and 20 um sieve. In spite of multiple regression analysis, no empirical equation could be derived to predict performance with reasonable accuracy. Allen et al. [11] reported on the development of a mechanistic design method for use in seasonal frost areas. The method is structured around five computer programs which compute pavement moisture and temperature conditions, resilient modulus and Poisson's ratio, stresses and strain in the pavement system, and cumulative damage. these models were tested to assess cumulative damage. In each case, the model predicts the number of loads required to induce failure in the pavement system. The definition of failure, as well as the fatigue criteria, vary from one model to the other. The only model identified which does not specifically deal with spring thaw is described in the 1986 AASHTO design guide [12]. The proposed method relates heave rate, drainage condition, and percentage of the surface affected by frost action to loss in pavement serviceability index (PSI). The origin of the (empirical) model is not specified in the guide. The method prescribes the use of an effective modulus to account for varying climatic effects. It is generally accepted that seasonal variation in material properties and pavement behavior have a significant impact on long term performance, but that the causal relationship between these variables is still not well understood. The Strategic Highway Research Program (SHRP) has therefore undertaken a large monitoring program to provide the data needed to attain a fundamental understanding of the magnitude and impact of temporal variations in pavement response and performance (Richter, 1991 [13]). Programs such as the SHRP Seasonal Variation and the Minnesota Road Research Project (MN/Roads) have evolved from the need to better understand pavement performance and should provide a sound basis for model development. |