MODELLING RUTTING IN FLEXIBLE PAVEMENTS IN HDM-4

Modelling Rutting in Flexible Pavements 1 MODELLING RUTTING IN FLEXIBLE PAVEMENTS IN HDM-4 by Louw Kannemeyer 1. INTRODUCTION 1.1 Background Rutting is defined as the permanent or unrecoverable traffic-associated deformation within pavement layers which, if channelled into wheelpaths, accumulates over time (Paterson, 1987). A primary concern of most pavement structural design procedures is to control rutting. This is achieved by estimating the cover thickness of high quality materials required to protect the natural subgrade against the compressive stresses from traffic, and thus limiting deformation to within acceptable limits over time. This approach has led to the development of various relationships between acceptable rut depth limits and the various measures of material and traffic properties, enabling the design of adequate pavement structures. Due to the deterioration and ageing of road networks, these models (or the principles behind them) were also used for improved management and planning techniques, and for the economic justification of expenditures and standards in the highway sector. Therefore, in modern pavement management systems the routine measurement and prediction of rutting has become an important performance criteria as a result of the influence of rutting on road roughness, dynamic loads and safety (based on the hazard of ponding water), all of which influence the road user costs of vehicle operation and accidents. HDM-III also includes models for the prediction of rut depth over the life of a pavement, based primarily on research conducted in the early 1980 s in Brazil. Since then, the introduction of new technology, both in pavement design and materials, has created the need to review the existing HDM-III rut depth models to consider their continued applicability in light of currently available pavement materials, design and testing. 1.2 Objectives of the study on rut depth The objectives of the study are to use the information on rut depth prediction models obtained from available international literature, as well as actual rut data, to achieve the following: To re-evaluate the applicability of the existing HDM-III rut depth models for the various pavement types, materials and conditions around the world;

2 Modelling Rutting in Flexible Pavements especially the pavement types, materials and conditions not included in the original development of the models. If shown to be beneficial, to replace the existing HDM-III rut depth models with improved models. If such models are not available, to update the existing models to incorporate new technology in pavement materials, design and testing. 1.3 Methodology of study A study of available international literature on rut depth prediction models was conducted. The aim of this study was to identify rut depth prediction models that could be internationally applicable, and to identify parameters of importance in the modelling of rutting. The model or models identified as applicable were then evaluated analytically using rut depth data obtained from international pavement studies and pavement management systems. No additional field observations were conducted at any phase of the study. 1.4 Organisation of the report The report is separated into six parts in which the following topics are addressed: In Chapter 1 the background to the problem is discussed and the objectives of the study and methodology of achieving them are outlined. In Chapter 2 rutting, its causes, parameters affecting it, the existing HDM- III models and other available models are discussed. The methods for processing the additional data obtained for evaluation of the rut depth models are discussed in Chapter 3. Chapter 4 contains the proposed new models for HDM-4. Chapter 5 contains the conclusions. Chapter 6 contains the references. 2. REVIEW OF AVAILABLE RUTTING MODELS 2.1 Introduction During the last two decades many attempts have been made by both researchers and practitioners alike to develop models that could predict the deterioration of a pavement over time, including models for the prediction of rutting. Each model, however, has certain inherent limitations due to the assumptions and data used during the development of the model.

Modelling Rutting in Flexible Pavements 3 This chapter presents the results of the study into mechanisms of rutting, the parameters influencing rutting and finally the models available, including the existing HDM-III models, for predicting rutting. 2.2 Mechanisms of Rutting Traffic-associated permanent deformation, and rutting in particular, results from a rather complex combination of densification and plastic flow mechanisms. Densification, according to Paterson (1987), is the change in the volume of material as a result of the tighter packing of the material particles and sometimes also the degradation of particles into smaller sizes. Rutting due to densification is usually fairly wide and uniform in the longitudinal direction with heaving on the surface seldom occurring, as illustrated in Figure 2.1. The degree of densification depends greatly on the compaction specifications during construction. The density specification should be selected in accordance with the expected loadings and pavement type. Failure to reach the specified compaction during construction will result in an increase of densification under traffic, most of which occurs early in the life of the pavement. It is important to note that for similar rut depth values, the deformation within the pavement may be located within a single weak layer, or more evenly distributed through the depth of the pavement, as illustrated in Figure 2.1. Densification distributed through pavement structure Densification within single pavement layer Surface layer Pavement Layer 1 Pavement Layer 2 Subgrade Figure 2.1: Typical rut profile as result of densification

4 Modelling Rutting in Flexible Pavements Plastic flow involves essentially no volume change, and gives rise to shear displacements in which both depression and heave are usually manifested. Plastic flow occurs when the shear stresses imposed by traffic exceed the inherent strength of the pavement layers (Paterson, 1987). The rutting in this case is usually characterised by heaving on the surface alongside the wheelpaths, as illustrated in Figure 2.2. Plastic flow is controlled through the structural and material design specifications, which are normally based on a measure of the shear strength of the materials used (for example, the California Bearing Ratio (CBR) for soils, and Marshall and Hveem stability for bituminous materials). The best known example of plastic flow is shoving within the asphalt layers, as illustrated in Figure 2.2. Heaving alongside wheelpath Depression within wheelpath Asphalt layer Pavement layer 1 Pavement layer 2 Subgrade Figure 2.2: Typical Rut Profile as Result of Plastic Flow (Shoving) 2.3 Factors Influencing Rutting 2.3.1 Introduction The resistance of pavement structures to rutting is dependent on a number of factors which either relate to applied loads (traffic type and traffic volume), the environment (temperature, rainfall), the pavement structure (materials used and their composition), the construction process, or to a combination of the above. The factors of importance for the various pavement types included within HDM-4 are discussed in this section. The general influence of the factors are based on the findings of laboratory and field observations of numerous experiments.

Modelling Rutting in Flexible Pavements 5 2.3.2 Generalised trends of behaviour In considering general trends in the behaviour of pavements containing different materials, it must be remembered that the state of materials changes with time. Consequently, the general trends in behaviour that are discussed refer to the original asbuilt state of the material. In discussing these trends the pavement types are described in terms of the materials contained in the base of the pavement. The flexible pavement types included in HDM-4 that have distinctly different behaviour patterns are: Granular base (GB) pavements Cement-treated base (SB) pavements Asphalt base (AB) pavements These three pavement types are discussed in more detail, based on information presented by Jordaan (1992). Granular Base Pavements The general trends in deformation of granular base pavements are illustrated in Figure 2.3. The behaviour can be classified into three phases: an initial phase, where some deformation occurs in the wheel tracks; a stable phase, during which little deformation occurs; and finally an increased rate of deformation. The factors influencing the magnitude and duration of various phases are: Construction compaction: The amount of early deformation, also referred to as postconstruction compaction, depends on the densification achieved during the construction of the pavement layers and the quality of the pavement layers. The higher the quality of the layer, the higher the specified level of compaction, and thus the lower the expected initial densification, as seen in Figure 2.3. Material quality: The rate of increase in deformation during the stable phase also depends on the initial quality of the material. Where the initial quality of the material

6 Modelling Rutting in Flexible Pavements W ater ingress stopped Deformation Ingress of water Poor quality material Initial phase Ingress of water Stable phase High quality material Traffic Loading Figure 2.3: Relative behaviour of granular materials (after Freeme, 1983) is very poor and has low densities, high traffic loadings may result in quick shear failure, typical of untreated pavement layers, and the stable phase may be non-existent or very brief, as seen in Figure 2.3. For relatively high quality materials the performance under traffic is much better, and the susceptibility to water ingress is much lower. Moisture content: Over time the pavement surface may crack. The increased moisture content due to ingress of water through a cracked surface layer will result in a decrease in shear strength of granular pavement layers which, when over- stressed by traffic, will result in the shear failure of the layers and thus the increased deformation observed in the final phase. The rate of increase is once again dependent on material quality (high quality materials are less susceptible to ingress of water), the amount of water ingress (rainfall), and traffic loading. Traffic loading: The traffic loading is a combination of the magnitude and volume of the loads; these are combined into the number of standard loads through the fourth power law. Traffic loading is one of the most important factors contributing to rutting. Traffic induces stresses within the pavement structure that have to be withstood, and thus determines the quality of materials required, as well as the behaviour of the pavement in various phases. It is important to note that a few excessive loads or tire pressures for which the pavement was not designed may cause stresses exceeding the shear strength of the material and thus plastic flow, resulting in the premature failure of the layer. Cement-treated base pavements

Modelling Rutting in Flexible Pavements 7 The general deformation trends of cement-treated base pavements are illustrated in Figure 2.4. For these pavements, the expected initial increase in deformation due to postconstruction compaction is much lower and even negligible in most instances. This is followed by a stable phase during which little or no deformation occurs, and finally a phase during which the rate of deformation increases. This final phase often only occurs after moisture ingress through secondary cracking causes fines to be pumped out from under the cement-treated base of the pavement. Furthermore, it is most often only during this phase that the difference in behaviour between high and poor quality materials becomes evident. In general, the resistance to deformation of cement-treated base pavements is similar to or even better than the deformations expected from very high quality (crushed stone) granular materials. Furthermore, the susceptibility to moisture of the cement-treated material is also less than that of the granular material used to construct the cement-treated layer. The factors influencing the magnitude and duration of various phases are similar to those discussed for granular materials. It is important to note that for cement-treated base pavements, most of the relative strength of the pavement is usually concentrated within these layers, and as such construction quality has a considerable influence on the performance of the layer. Deformation Poor quality material Water ingress stopped Ingress of water Traffic Loading High quality Figure 2.4: Relative behaviour of cement-treated materials (after Freeme, 1983) Asphalt-Treated Base Pavements The general deformation behaviour of asphalt base pavements is illustrated in Figure 2.5. The behaviour during the various phases is similar to that of granular base pavements. Three phases are once again distinguished: an initial phase, where some

8 Modelling Rutting in Flexible Pavements deformation occurs in the wheel tracks, followed by a phase of gradual decrease in rut rate to a constant rate, and finally a phase with an increased rate of deformation. The main difference in behaviour between asphalt layers and granular layers occurs in the final phase, where asphalt layers are far more water-resistant than granular layers, but as a result of their visco-elastic behaviour they are more temperature susceptible. The factors influencing the magnitude and duration of various phases are construction compaction, material quality (which, for asphalt layers, refers to the mix properties of binder content, air voids and aggregate type), and traffic loading. As a result of the nature of asphalt layers, the influence of moisture content is replaced by that of temperature. The influence of the asphalt s characteristics is discussed below (Verhaeghe, 1993): Binder content: The selection of a suitable binder content for a given grading of aggregate is one of the main problems in the design of a bituminous mixture. From the point of view of deformation, asphalt mixes should contain just enough binder to give cohesion and to enable adequate compaction to be achieved, without undue risk to plastic deformation under the prevailing conditions of traffic and temperature. Too much binder will lubricate the mix to such an extent that the mixture will lack internal friction and become unstable. Air voids content: The percent of air voids within an asphalt mix also influences the behaviour of the mix. The higher the percent of air voids, the more resistant the mix is to deformation. But due to the increased permeability to air, an increased rate of hardening of the binder will occur, reducing the fatigue life of the asphalt. If the air voids content is too low, the asphalt mix will become unstable, resulting in plastic flow of the layer under heavy trafficking, slow moving loads or high maximum temperature. According to Road Note 31 (TRL, 1994), numerous studies indicate that the minimum air voids after trafficking should always exceed 3 percent to avoid potential plastic flow, but should be less than 5 percent to keep hardening of the binder (under tropical conditions) to a minimum. Aggregate type and quantity: The resistance to permanent deformation of an asphalt mix is also dependent upon the interaction between particles of the coarse aggregate to form a mechanical interlocking structure; the higher the particle to particle contact within the mix, the more resistant the mix will be to deformation. Thus both the shape and texture of coarse aggregate is of importance. It also has been found that the higher the stone content the lower the deformation, but the more difficult it is to achieve the required compaction. Temperature: The dependence of the flow properties of bituminous mixtures on temperature is due to changes in the rheological properties of the binder, the dominant factor being the great dependence of viscosity on temperature. From simulation tests and general experience it is well known that the resistance to deformation of bituminous materials decreases rapidly as temperature increases, especially if the ambient temperature approaches or exceeds the softening points of the binders used in such mixes.

Modelling Rutting in Flexible Pavements 9 Poor mix design Deformation Increase in temperature of asphalt material Poor creep resistant mix Good creep resistant mix Traffic Loading Figure 2.5: Relative behaviour of asphalt materials (Freeme, 1983) 2.4 Prediction of rutting 2.4.1 Background Two approaches exist to predict rutting as a result of densification and plastic flow. The first approach is mostly used in pavement design procedures and limits deformation to below a specified failure limit; these models are not useful for performance modelling because of the need to predict not the limit, but the trend of rutting during the life of the pavement (Paterson, 1987). The second approach predicts the trend of rutting during the life of a pavement, identifying the response of a pavement to actions of traffic, environment and maintenance. As such, this second approach is useful for pavement performance predictions. The deformation trends observed for pavements of various base types discussed previously are summarised in Figure 2.6, highlighting the difference in influence of the two mechanisms of permanent deformation. Curve A in Figure 2.6 represents a pavement with generally adequate structural and material design in which the deformation occurs primarily through densification. The curve follows a concave trend and tends asymptotically towards a limit which is likely to be a function of the compaction specifications relative to the level of applied loading. Curve B represents a pavement with inadequate structural and material design specifications in which plastic flow dominates; this is typical of either overloading or an unstable asphalt mix. Curve C represents either the influence of water ingress on a sound granular or cemented pavement which has cracked and, as a result of inadequate maintenance, allowed the ingress of water and the

10 Modelling Rutting in Flexible Pavements subsequent weakening of the pavement, or a sound asphalt pavement for which an increase in material temperature resulted in the creep of the layer under traffic. In general, the permanent deformation of pavement materials as a function of time or the number of load repetitions, may be divided into the following three phases: Initial consolidation: This phase, sometimes also referred to as bedding in or post construction compaction, describes the relatively rapid initial increase in rutting on a newly constructed pavement once it is opened to traffic. The phase is characterised by a decreasing deformation (strain) rate, and the amount of initial consolidation is mainly influenced by the compaction achieved during construction, material type, pavement strength and traffic load. B Deformation Increased deformation C A Constant deformation Initial consolidation Traffic or Time Figure 2.6: General trends of rut depth including effects of cracking, water ingress and maintenance (Freeme, 1983; Paterson, 1987) Constant deformation: During this phase the rate of deformation (strain) tends to stabilise, resulting in a constant increase of deformation over time or traffic load. The rate of deformation is mainly influenced by traffic loading, pavement strength, material type and environmental influences. Increased deformation: This is the third and final phase in the development of deformation, and it is characterised by an increased rate of deformation (strain). For granular and cemented bases this phase normally begins after load-associated cracking allows the ingress of moisture and causes a subsequent decrease in pavement strength. For asphalt materials, this phase is normally the result of an increase in pavement

Modelling Rutting in Flexible Pavements 11 temperatures to levels close to or exceeding the softening point of the binder. In the absence of immediate intervention, this phase will lead to pavement failure should the moisture ingress or high temperatures continue. Various models are available to quantify these observed trends and phases. These models are either mechanistic, based on laboratory material characterisation and theoretical structural analysis of the stresses and strains induced in each layer under traffic loading, or empirical, based on correlations between field data of rut depth trends and explanatory parameters representing the pavement and loading. The original HDM-III models, which were based on the performance results of in service pavements, are classified as empirical. 2.4.2 Existing HDM-III Rut Depth Models Background The empirical models included in HDM-III were developed from the Brazil study and were the first empirical models to incorporate both the mechanisms of traffic-associated permanent deformation discussed previously. Two separate models, one for the mean rut depth and one for the rut depth standard deviation were developed by Paterson (1987). A strong relationship between the mean rut depth and the rut depth standard deviation was found: both are non-linear functions of pavement age, cumulative equivalent standard axle loadings, modified structural number, average relative compaction, cracking and rainfall. The generalised expressions developed by Paterson (1987) were modified in HDM-III to allow cracking to be a progressive variable with age and to allow HDM-III to compute the first-year rut depth for a new pavement using the same variables. The models in HDM-III are (Watanatada et al., 1987): Mean rut depth at the end of the first year (rut depth at start of first year is zero): RDM = Krp [39800 SNC -0.502 COMP -2.30 YE4 ERM ] ERM = 0.09 + 0.0384DEF - 0.0009RH + 0.00158 MMP CRX a Subsequent annual incremental increase in mean rut depth as a result of road deterioration: RDM d = Krp [((0.256-0.0009RH + 0.0384DEF + 0.00158MMP CRX a )/AGE3 + 0.0219 MMP CRX d ln (MAX(1, AGE3 YE4)))RDM a ] Mean rut depth at the end of the analysis year, with a limit of 50 mm, is given by: RDMb = min (50; RDM a + RDM d ) Where: RDM = Initial mean rut depth value in both wheelpaths at end of first year, in mm; only used if mean rut depth (RDM a ) at start of first year is zero.

12 Modelling Rutting in Flexible Pavements RDM d = The predicted change in mean rut depth during the analysis year due to road deterioration, in mm. RDM a = Mean rut depth at start of year, in mm; equals the mean rut depth at end of previous year (RDM b ). For first year, RDM a = 0. At start of second year, RDM a = RDM. RDM b = Mean rut depth at end of analysis year, in mm. Krp = User specified factor for rut depth progression (default = 1). SNC = Modified structural number of pavement. COMP = The average relative compaction weighted by layer thickness in the base, subbase and selected subgrade layers, as percentage. YE4 = Number of equivalent standard axle loads in millions/lane for the analysis year, based on an axle load equivalency exponent of 4. DEF = Mean Benkleman Beam rebound deflection, in mm, of the surfacing in both wheelpaths under an 80 kn standard axle load, 520 kpa tire pressure, and 30 C average asphalt temperature. RH = Rehabilitation indicator, where RH = 1 for surface types asphalt concrete overlay (OVSA) or open-graded cold-mix (OCMS) overlays; RH = 0 otherwise. MMP = CRX a = ACRA a = ACRW a = AGE3 = CRX d = ACRA b = ACRW b = Mean monthly precipitation, in m/month. Total area of indexed cracking at beginning of the analysis year, as percentage of total carriageway area: = 0.62 ACRA a + 0.39 ACRW a Area of all cracking at beginning of the analysis year, as percentage of total carriageway area. Area of wide cracking at beginning of the analysis year, as percentage of total carriageway area. Construction age, defined as the time since latest overlay, reconstruction or new construction activity, in years. Predicted change in the area of indexed cracking due to road deterioration in the analysis year, in percentage of carriageway area. = 0.62 (ACRA b - ACRA a ) + 0.39 (ACRW b - ACRW a ) Area of all cracking at the end of the analysis year, as percentage of carriageway area. Area of wide cracking at end of the analysis year, as percentage of carriageway area. Rut depth standard deviation at end of first year (if mean rut depth at start of year is zero): RDS = Krp [4390 RDM 0.532 d SNC -0.422 COMP -1.66 YE4 ERS ] ERS = -0.009RH + 0.00115MMP CRX a Subsequent annual incremental increase in rut depth standard deviation as a result of road deterioration: RDS d = Krp [(0.532 (RDMb - RDMa) RDMa + (-0.0009RH + 0.00115 MMP CRX a )/AGE3 + 0.0159 MMP CRX d ln (MAX(1, AGE3 YE4)))RDS a ] Standard deviation of rut depth at the end of the analysis year, with an upper limit equal to the mean rut depth, is given by:

Modelling Rutting in Flexible Pavements 13 RDS b = min(rdm b, RDS a + RDS d ) where: RDS = Rut depth standard deviation across both wheelpaths at end of first year, in mm; only used if mean rut depth at start of first year is zero. The prediction is halved to suppress the sharp initial increase. RDS d = The predicted change in mean rut depth during the analysis year due to road deterioration, in mm. RDS a = Rut depth standard deviation at start of year, in mm; equals the rut depth standard deviation at end of previous year. For first year, RDS a = 0. At start of second year, RDS a = 0.5 (RDM). RDS b = Rut depth standard deviation at end of analysis year, in mm. Evaluation of HDM-III models The mean rut depth model is not used directly in the roughness model of HDM-III, but is instead used as a means to estimate the rut depth standard deviation which contributes directly to roughness. The prediction of first-year mean rut depth by the HDM-III model in terms of parameters included in the model is illustrated in Figure 2.7, and first-year rut depth standard deviation is shown in Figure 2.8. From Figure 2.7 it is evident that the expression for first-year mean rut depth development is the most sensitive to pavement strength (at lower strengths), and to a lesser degree to compaction, with virtually no sensitivity to axle loading. The sensitivity to the other terms within the expression, namely rainfall and cracking, is assumed to be negligible since the pavement is new and assumed to be uncracked. As seen in Figure 2.8, basically the same is applicable to first-year rut depth standard deviation, except that the magnitude of the predicted values are far lower than those of mean rut depth. The prediction of mean rut depth development and subsequent progression over the life of a pavement are illustrated in Figure 2.9 for various structural strengths, and Figure 2.10 for various pavement base types (SNC = 2 for all) and rainfall. Only routine maintenance are allowed over the life of the pavement, allowing the evaluation of the HDM-III rutmodels under extreme conditions. In Figure 2.9, the influence of the modified structural number (lines SNC2, SNC4 and SNC6) on the initiation and progression of rut depth on a granular base course with cracking and a rainfall of 3600 mm/year under an annual directional traffic load of 1 million ESA (MESAL) is evident. It is also interesting to note the influence that cracking has on the development of rutting by observing the difference between the lines SNC4 and SNC4-No crack, where both have an annual traffic load of 1 million ESA (per direction) and a rainfall of 3600 mm/year. This difference illustrates that rut depth progression within the existing HDM-III equation is based on the cracking of the surface layer and the subsequent ingress of water, resulting in a decrease of pavement strength and shear failure of layers when over-stressed. If the traffic load is not sufficient to induce critical stresses within the pavement layers, but only stresses within the elastic range of the material, the cracking and subsequent weakening of the pavement through water ingress would not have a severe influence on rut depth progression. This is illustrated in Figure 2.9 by line SNC4-0,1MESA for an annual traffic loading of 0.1 million ESA (per direction), where the rut depth tends towards an ultimate value determined largely by the amount of densification and the pavement strength. As

14 Modelling Rutting in Flexible Pavements seen from Figure 2.10, there is no difference between the initiation and progression of rut depth for an asphalt mix on granular (AMGB-360) or asphalt base course layers (AMAB- 360), or a surface treatment on a granular base course (STGB-360) under an annual rainfall of 360 mm per year and traffic loading of 1 million ESA (per direction). The same is applicable for the first 10 years for asphalt mix under an annual rainfall of 3600 mm per year, after which the mean rut depth tends to progress to a slightly higher level for the asphalt base (AMAB-3600). The rut depth on the surface treatment (STGB-3600) tends to develop faster, but not to levels as high as that observed for the asphalt mix. There is no model for a surface treatment on an asphalt base. 12 Traffic in MESAL FIRST YEAR MEAN RUT DEPTH (mm) 10 8 6 4 2 85% 85% 85% Relative Compaction 90% 95% 0.1 0.5 1.0 100 % 0 2 3 4 5 6 MODIFIED STRUCTURAL NUMBER (SNC) Figure 2.7: First-year mean rut depth (RDM) prediction

Modelling Rutting in Flexible Pavements 15 12 Traffic in MESAL FIRST YEAR RDS DEVIATION (mm) 10 8 6 4 2 85% 85% 85% Relative Compaction 90% 95% 100 % 0.1 0.5 1.0 0 2 3 4 5 6 MODIFIED STRUCTURAL NUMBER (SNC) Figure 2.8: First year rut depth standard (RDS) deviation prediction 35 30 25 20 15 SNC2 SNC4 SNC6 SNC4-No Crack SNC4-0,1MESA 10 5 PREDICTED MEAN RUT DEPTH (mm) 0 0 2 4 6 8 10 12 14 16 18 20 PAVEMENT AGE (YEARS) Figure 2.9: Mean rut depth progression for various pavement strengths

16 Modelling Rutting in Flexible Pavements 40 PREDICTED MEAN RUT DEPTH IN (mm) 35 30 25 20 15 10 5 SNC = 2 for all pavements AMGB-3600 AMSB-3600 AMAB-3600 STGB-3600 AMGB-360 AMSB-360 AMAB-360 STGB-360 0 0 2 4 6 8 10 12 14 16 18 20 PAVEMENT AGE (YEARS) Figure 2.10: Mean rut depth progression for various base types and rainfall The above differences between the various pavement types, as well as the anomaly observed for the models under high rainfall (3600mm/year), where the rut depth decreases at a certain age and then progresses again at a constant rate, are the result of the influence of the area of cracking in the HDM-III rut model. The reduction in rut depth occurs when the area of cracking is decreased to allow for the initiation and progression of other surface defects. Since the maximum area of damage for a pavement surface can only be 100 percent, the area cracked has to be decreased to allow for potholes and ravelling. The maximum area allowed for potholing is 30 percent, resulting in cracking being held at a constant 70 percent (assuming no ravelling). This is why the rate of rut depth increase is constant after the apparent drop. The anomaly observed in the predicted rut depth value might be justified by assuming that potholes generally appear in the most over-stressed pavement areas, which are normally the areas with extensive cracking. Based on the definition of rut depth progression these will also be the areas of severe rutting. Thus where potholes exist these areas can no longer be included in the calculation of rut depth, since their contribution to roughness is already included through the pothole function; the exclusion of these areas of severe rutting results in a decrease in both the mean rut depth and the rut depth standard deviation value. In practise this is not what is normally found, and this anomaly is to be addressed in the new rut model. The performance of pavements with cement-treated bases is different from those with bituminous and granular base course layers, with lower rut depth values predicted. The difference is the result of different relationships being used for granular and cemented

Modelling Rutting in Flexible Pavements 17 base pavements to endogenously compute the deflection from the modified structural number. The relationships used are: Granular Base: Def = 6.5 SNC -1.6 Cemented Base: Def = 3.5 SNC -1.6 It can be seen that the difference in predicted rut depth agrees more or less with the ratio between the coefficients of the two relationships, which is about 0.5 for a cemented base to a granular base. As mentioned previously, it is only the incremental change in rut depth standard deviation that is included in the roughness model of HDM-III. Comparing Figure 2.11 to Figure 2.9 shows that rut depth standard deviation is a strong function of the mean rut depth, generally following the same general pattern for each pavement, although having much lower values than the mean rut depth. This is illustrated in Figure 2.12 by the more or less constant ratio observed between rut depth standard deviation (RDS) and mean rut depth (RDM) over a pavements life. As illustrated by Figure 2.7 through to Figure 2.11, the value of rut depth standard deviation depends on a number of variables. The effect of changes within these parameters can be quantified by evaluating the influence that changes in rut depth standard deviation has on roughness. This is illustrated in Figure 2.13, from which it can be seen that the relationship between roughness and incremental change in rut depth standard deviation (RDS) is a linear one. Recalling the worst scenario in Figure 2.11, the largest change in rut depth standard deviation between 2 years is about 2 mm. Substituting this into Figure 2.13 results in an increase of roughness of 0,2 IRI, which is not a significant change. Despite the small contribution of rut depth to roughness in the HDM-III model, the prediction of it is still important since it is one of the parameters affecting road safety, especially in wet whether. Thus, although the effect of rut depth on vehicle operating costs may be small, its effect on user costs through accidents could be substantial. For these reasons it is important that the existing rut models within HDM-III be improved, not only to address the various limitations identified during the use of the models over the past years, but also to incorporate the latest international research findings. HDM-III Rut Model Limitations Figure 2.9 and Figure 2.10 show that the HDM-III rut depth models are capable of predicting the generally observed phases of rutting illustrated in Figure 2.6 to a certain extent, allowing for material, traffic and environmental influences. But despite the acceptable representation of the development and progress of rutting in flexible pavements similar to the range of materials and thickness in Brazil, the applicability of HDM-III rut models are limited by the following: The use of a single set of coefficients in the model to quantify the various phases of rut depth development results in the initial consolidation and stable phases overshadowing the increased deformation phase. Thus instead of a sharp transition between uncracked and cracked phases as observed for pavements, the current coefficients and model averages the cracking effect over the pavement life. This could contribute to the under-

18 Modelling Rutting in Flexible Pavements prediction of rut depth observed for some pavements. Also, the anomaly observed when area cracking decreases due to initiation of potholes needs to be reconsidered. The plastic flow incorporated in the rut models is only based on the plastic flow resulting from the shear failure of pavement layers when overstressed, and does not represent the plastic flow (shoving) of asphalt layers (soft asphalt at high road temperatures) or long-term plastic deformation (creep) of thick asphalt (> 150 mm) pavements, as illustrated in Figure 2.10. This limitation was identified by Paterson (1987) during the initial validation of the rut models on rut data obtained on thick asphalt pavements, and results in the under prediction of rutting for these pavement types. Another limitation identified by Paterson (1987) during validation studies is the influence of seasonal effects in wet-nonfreezing and wet-freezing climates, also resulting in under-prediction of rutting for pavements under these environmental conditions. The models do not address mechanical wear, which occurs in countries where winter snow and ice on roads necessitates the use of snow chains or studded tires on vehicles. 40 RUT DEPTH STANDARD DEVIATION (mm) 35 30 25 20 15 10 5 SNC2 SNC4 SNC6 SNC4-No Crack SNC4-0,1MESA 0 0 2 4 6 8 10 12 14 16 18 20 PAVEMENT AGE (YEARS) Figure 2.11: RDS progression for various pavement strengths

Modelling Rutting in Flexible Pavements 19 1 0.9 RATIO BETWEEN RDS and RDM 0.8 0.7 0.6 0.5 0.4 0.3 0.2 SNC2 SNC4 SNC6 SNC4-No Crack SNC4-0,1MESA 0.1 0 0 2 4 6 8 10 12 14 16 18 20 PAVEMENT AGE (YEARS) Figure 2.12: Relation between RDS and RDM 1.2 CHANGE IN ROUGHNESS (IRI) 1 0.8 0.6 0.4 0.2 0 0 2 4 6 8 10 INCREMENTAL CHANGE IN RDS (mm) Figure 2.13: Influence of RDS changes on roughness

20 Modelling Rutting in Flexible Pavements In addition to the above-mentioned limitations, the following general factors are also believed to contribute to the differences observed between predicted and observed rut depths: The method employed by the algorithm used for calculating the rut depth from the measured transverse profile obtained from some high-speed profilometers. The straight-edge length used during manual measurements, or during calculations of rut depth from the transverse profile. This length is important, since HDM-III models were developed from rut measurements under a 1.2 m straight edge, a factor not always considered by users. The sampling interval used during measurement of rutting and the subsequent aggregation of the data into different interval lengths. This is especially important with data measured with high-speed profilometers. The aim of the study is thus to first focus on the general factors influencing the rut depth data available, and then address the identified limitations in the existing equation. It is believed that the incorporation of these modifications will result in improved predictions of rut depth by the HDM-4 models. 2.5 Additional models available from literature 2.5.1 Background Based on the identified limitations of the existing HDM-III models an international literature study was conducted to identify the availability of models with which the existing HDM-III models could be replaced (or enhanced) to improve rutting predictions. Despite the existence of numerous studies, the number of models available to the study team for evaluation were severely limited by the following: Mechanistic principles: A large number of the models or transfer functions developed within recent years rely on the prediction of stresses and strains within the pavement layers under the applied traffic loading using linear elastic theory. Since HDM-4 does not incorporate the ability to predict stresses and strains, these models or transfer functions are not appropriate and were not further evaluated. Specialised equipment: Some of the models or transfer functions developed are based on output from tests conducted with specialised and very expensive equipment. Since the aim is to develop a universally applicable model, these models or transfer functions are not appropriate and nor were they further evaluated.

Modelling Rutting in Flexible Pavements 21 2.5.2 Models identified Background The models that were examined range from simplistic linear equations based on observed trends to sophisticated multi-linear models incorporating most of the materials parameters that could be measured. The additional models that were identified are discussed for the various basic flexible pavement base types included within HDM-4, and then for wear observed in cold climates. Granular base pavements No additional models could be identified within the literature that would improve on the logic of the current HDM-III models. The current HDM-III models will, however, be reevaluated with additional data to confirm the coefficients included and to evaluate whether allowing a before cracking and after cracking phase will improve the prediction capabilities. Cement-treated base pavements Once again no models could be identified that would improve on the logic of the current HDM-III models. The existing models will, however, be re-evaluated with additional data to confirm the coefficients included. Asphalt-treated base pavements From the available literature most of the models developed are for predicting the rutting observed in asphalt base pavements, the material type for which no acceptable predictions are made within the existing HDM-III models. The most basic model studied was based on unpublished research in England by TRRL for use in the assessment of future costs associated with different construction and maintenance options. The following rut development model is employed to predict the expected rut depth: R = a + bt Where: R = Predicted rut depth, in mm. a, b = Parameters whose values depends on the roadbase type. T = Cumulative traffic carried, in million standard axles, which is equal to YE4 used in HDM-III, assuming a load equivalency factor of 4. The coefficients for the various roadbase types are summarised in Table 2.1. These coefficients have been determined from the analysis of the performance of TRL full-scale experimental pavements. According to the author, the a parameter represents the rut depth assumed for new surfaces at the time of construction and b is the expected linear increase in rut depth with traffic until the next resurfacing (excluding surface dressing) or strengthening treatment, as illustrated in Figure 2.14.

22 Modelling Rutting in Flexible Pavements Table 2.1: Parameters for the rut model (based on research by TRL) Roadbase Type Parameters a b Dense Bitumen Macadam 1,95 0,46 Hot Rolled Asphalt 2,21 0,38 Cemented 2,19 0,31 Granular 2,66 0,84 As seen from Figure 2.14, rutting is assumed to change linearly with the traffic carried, and the same equation is applied each time maintenance or rehabilitation is conducted. This linear behaviour is in line with the stable phase discussed for the various pavement materials. The use of a constant linear increase is justified by the author, by assuming that the weaker pavement areas experiencing rapid increase in rut depth as the pavement approaches failure would be repaired through structural patching, thus allowing the use of the model beyond the stable phase. Also, the relative behaviour observed for the various materials, with cement base pavements having lower rutting than the others, is in line with what was shown previously. It is important to note, however, that these initial rut depths and the subsequent rates of increase are based on pavement materials, designs and construction standards practised in England; their applicability to other conditions would thus have to be evaluated. To avoid the prediction of unacceptable increases in rut depth at high traffic loads, resulting in an unacceptable number of resurfacings needed, it is recommended that the deterioration rate within the model be amended to allow for improvements in mix design and materials used. At this stage no additional models for possible incorporation into HDM-4 are identified; the current literature available from the SHRP study only contains models for the design of asphalt mixes, and no prediction models have of yet been developed from observed data. It is believed, however, that the factors included within these models could give insight into an attempt to quantify rutting within asphalt layers in the future. However, although the prediction capabilities, and thus the general acceptability, of the HDM-4 rut depth models might be improved by incorporating such detailed models, the detail required to utilise these models would generally make them impractical for network level applications.

Modelling Rutting in Flexible Pavements 23 45 40 35 Predicted Rut Depth (mm) 30 25 20 15 10 DBM HRA Cem Gra 5 0 0 5 10 15 20 25 30 35 40 45 50 Cumulative Traffic Loading (MESA) Figure 2.14: Prediction of rut depth (based on research by TRL) Wear of surfaces (Whether this is part of cold climates model or general model is still uncertain) From the literature available to the study team only one model could be identified for predicting rutting as a result of studded tyre wear. The model quantified surface wear (SW) as follow (Ullidtz, 1987): SW = Ef x { CS A } x Nb Where: SW = Surface wear, in mm. Ef = Environmental factor which depends on the season. CS = Contact stress, which is the tyre pressure, in kpa. A, b = Coefficients depending on the type of aggregate. There is no indication of any calibration coefficients available by the author. The Malaysian team is currently awaiting feedback from the cold climates team regarding models for studded tyre wear. 2.6 Conclusions

24 Modelling Rutting in Flexible Pavements Despite the many attempts by researchers and highway agencies during the last two decades to develop models that could predict the deterioration of a pavement over time, including models for the prediction of rut depth propagation of a pavement over time, no additional models could be identified within the literature that would improve on the prediction capabilities and the logic of the current HDM-III models. As discussed above, this is primarily a result of the emphasis that is placed in new models on mechanistic equations and specialised equipment. Some models are identified that address certain limitations of the current HDM-III models. It is believed that the incorporation of these models in HDM-4 will result in improved rutting prediction capabilities. Thus the current HDM-III models will be re-evaluated with additional data to confirm the coefficients included and to evaluate whether allowing before-crack and after-crack phases will improve the prediction capabilities. Also, the models identified will be evaluated for inclusion into the revised models for HDM-4. 3. PROCESSING AND EVALUATION OF RUT DEPTH DATA 3.1 Introduction The standard measure of roughness is the International Roughness Index (IRI). The IRI is a standard, time-stable and transportable algorithm that is used for calculating and comparing roughness results from around the world. There is no such standard for rut depth calculations, and as a result different algorithms and straight-edge lengths are being used around the world to calculate rut depths from measured transverse profiles. One result is that rut depth measurements can not automatically be compared from one agency to another. Furthermore, with 100 percent sampling, an immense amount of data is generated. To overcome this, outputs from laser-based equipment are reported over interval lengths that are typically several orders of magnitude longer than the sampling length interval. This Chapter presents the results of a study conducted into the evaluation of: The influence that the algorithm used to calculated the rut depth from a transverse profile has on the calculated rut depth. The interval spacing at which transverse profiles are uncorrelated. This will ensure that no bias is introduced within transformations developed. Also the influence of the aggregation of rut depth data on the descriptive statistics and, more importantly, the subsequent assessment of the condition. The straight-edge length used in HDM-4, and the development of a transfer function to guide users during the transformation of measured rut depth data to the straight-edge length used by HDM-4.

Modelling Rutting in Flexible Pavements 25 3.2 Algorithms used for rut depth calculations Since each manufacturer of high speed data collection vehicles develops their own software for the calculation of rut depth from the measured transverse profile, a range of algorithms are in use internationally. The width of the transverse profile measured ranges from 1.8 m to 3.7 m between the various vehicles, depending upon the number of sensors used and their spacing. The various algorithms are, however, based on one of the following mathematical models: Wire model: This method simulates a string or wire being stretched transversely across the road profile following convex curves and overlaying hollows as a straight line, as illustrated in Figure 3.1. The rut depth is then calculated as the greatest perpendicular distance from the simulated wire, shown as W1 and W2. The advantage of this method is that it is mathematically easy to use, and it gives very good results for convex road profiles because it follows the general shape of the transverse profile and still takes hollows at the top into consideration (Larsen et al., 1994). This model has the disadvantage in that it can give very large rut depths where the mid-profile is one large basin, as illustrated in Figure 3.2. Although resulting in larger rut depths for the above case, it might be argued that this method gives a more accurate indication of the depth of free water for the large basin. Secondly, only the largest rut depth value will normally be calculated for each cross-profile; thus for the large basin in Figure 3.2, only the rut depth value W1 would be calculated. Based on these anomalies the rut depth calculated by the wire model could differ substantially from the rut depth measured under a straight-edge. A discrepancy between manual rut measurements and measurements calculated with the wire model are also noted in New Zealand (Cenek 1994). Straight-edge model: This method simulates a straight line of a given length similar to a straight-edge being attached to the transverse profile. Various methods exist of attaching this straight line; some attach it to the outer-most laser at each end, after ensuring that the reading at this point is not that of a curb, and then calculate the largest vertical difference between the transverse profile and the reference line (Larsen et al., 1994). This results in two rut depth values, of which only S2 is illustrated in Figure 3.2. Also illustrated is the difference in rut depth calculated by the two models. The other method simulates rapidly moving a straight line of a given length over the profile, performing up to over 200 rut calculations to identify the greatest rut depth in each wheel path. Both methods are able to simulate different straight-edge lengths, and as such allow the evaluation of the influence of straight-edge length on the calculated rut depth. Although both models have their advantages, the nature of the data required in this study favours the use of rut depth values calculated by the straight-edge model. These rut depth values are used to evaluate the interval at which the rut profiles are uncorrelated, as well as the influence of aggregation of data on the assessment of condition. These values are also used in the development of a transfer function for correlating rut depths of various straight-edge lengths to the length used in HDM-4.