Moisture Content-Based Longitudinal Cracking Prediction and Evaluation Model for Low-Volume Roads over Expansive Soils

Pavement-related information was obtained by falling weight deflectometer (FWD) and dynamic cone penetration (DCP) tests performed by the research team. Subgrade-related inputs were obtained from laboratory testing of the retrieved soil samples from the sites.

Six representative clayey sites in Texas, including five high-PI ($\mathrm{PI}>25$) sites (Houston, Fort Worth, San Antonio, Paris, and Bryan) and one low-PI site (El Paso), were selected to develop the MC-based constitutive material models. Soil index property tests conducted include hydrometer, Atterberg limits, and moisture-density tests to obtain MC, PI, liquid limit (LL), optimum MC (OMC), and maximum dry density (MDD). Strength tests performed include unconfined compressive strength (UCS) tests, indirect tensile strength (IDT) tests, and four-point flexural bending tests (Sabnis et al. 2008). Stiffness tests conducted include the free-free resonant column tests (FFRC) (Nazarian et al. 2006), resilient modulus (RM) tests, and permanent deformation (PD) tests. Volumetric shrinkage tests were conducted to measure the decrease in the total volume of soil specimens due to loss of MC from predetermined initial MC to a completely dry state. A test method developed by Puppala et al. (2004) was used. Table 1 presents a summary of various index properties of the soils used in the development of the MC-based material models. These soils are classified in accordance with the ASSHTO soil classification system and the Unified Soil Classification System (USCS).

To better understand the drying and wetting effects on soil properties, three moisture-conditioning regimes were used for the aforementioned tests: (1) samples were dried to constant weight from optimum [dried from optimum (DFO)]; (2) samples were wetted to reach complete saturation from optimum [saturated from optimum (SFO)]; and (3) samples were first wetted to saturation and then dried to constant weight [dried from saturation (DFS)]. As described in detail by Sabnis et al. (2008), at least three identical samples were prepared at their corresponding OMC for each test. The IDT, flexural, and RM/PD tests on saturated specimens could not be carried out because the specimens were too soft to withstand the loads.

To be specific, the following three MC-based relationships were developed: (1) ${\epsilon }_{\mathrm{ss}}$ as a function of MC; (2) $E$ as a function of MC; and (3) ${\sigma }_t$ as a function of MC. The material models presented here are based on DFO. Although relationships for specimens of DFS were also developed, they are not included in this paper for brevity’s sake. The work by Sabnis et al. (2008) contains detailed information about those models.

According to Sabnis et al. (2008), the shrinkage strain (${\epsilon }_{\mathrm{ss}}$) of an individual expansive soil can be estimated from the following relationship:where $A^{*}$ is a curve-fitting parameter and NMC is the normalized moisture content (NMC) defined as the actual MC divided by the OMC for a given soil. Similarly, subgrade stiffness increases when moisture is lost, and a relationship between normalized subgrade modulus ($E_n$) and NMC was also obtained as follows:where $B^{*}$ and $C^{*}$ are also curve-fitting parameters for a particular soil.

For each site, the aforementioned relationships in Eqs. (2) and (3) described the testing data well with $R^2$ values greater than 0.97. However, expansive soil mechanical properties are highly site specific, and thus soil index properties were introduced in the correlational study. Material from Houston was not used in the development of the constitutive material models but rather used as validation. The parameter $A^{*}$, used to estimate lateral shrinkage strain (${\epsilon }_{\mathrm{ss}}$), was correlated to different soil index properties for the remaining five clayey sites (Fig. 2). Based on $R^2$ values, parameter $A^{*}$ is well correlated to PI, OMC, and LL and marginally correlated to MDD. Therefore, parameter $A^{*}$ can be estimated fromwhere $A_i$ = estimated linear correlation parameter from index property $i$ by $A_i$ = $a{x}_i+b$; $x_i$ = index property value of the particular soil; $a$ and $b$ = slope and intercept of correlation line, respectively, as shown in graph; and $W_i$ = weighting factor for index parameter $i$. The subscript $i$ refers to PI, LL, OMC, and MDD, respectively. For $R^2$ values greater than 0.8, between 0.6 and 0.8, and less than 0.6, $W_i$ values of 4, 2, and 1 are recommended, respectively. Similarly, $B^{*}$ and $C^{*}$ can be obtained as functions of the weighted average of linear trend lines $B_i$ and $C_i$. Fig. 3 compares the predicted and measured shrinkage strains versus NMC for three Houston specimens. The results compared favorably, with approximately 90, 70, and 75% of the values within 20% margin of error for vertical, lateral, and volumetric strains, respectively.

The IDT tests were performed at six different MCs, from optimum to dry conditions for each site. The variations in the average IDT strengths and NMC for all soils are shown in Fig. 4. All soils, except Bryan, followed a unique trend. To build in conservatism in the model, the results from Bryan were excluded from the curve-fitting process. Also, data analysis showed poor correlations between peak strengths and index property parameters; thus, soil tensile strength (${\sigma }_t$) is estimated as a function of NMC

As depicted in Fig. 1, MC-based constitutive material models are used in a two-dimensional plane–strain linear elastic [Eq. (1)] model (henceforth referred to as the elastic model) to provide shrinkage-induced strain (${\epsilon }_{\mathrm{ss}}$), modulus ($E$), and strength (${\sigma }_t$) for a specific MC level via Eqs. (2), (4), and (5), respectively. The elastic model simulates and compares maximum shrinkage stresses (${\sigma }_{\mathrm{ss}}$) to predicted strength (${\sigma }_t$); if ${\sigma }_{\mathrm{ss}}<{\sigma }_t$, then MC is further reduced to simulate the subgrade drying process. The process is repeated until the threshold is reached. The locations of the top 50 most critical stress points and critical MC thresholds for crack initiation (${\mathrm{MC}}_{\mathrm{I}}$) are identified. The elastic model was developed in MATLAB and linked to the ExSPRS program. Mesh generation and optimization were automated so that pavement engineers would not have to spend time on FEA modeling details but rather focus on analyzing the results. Once a crack starts in a subgrade (${\sigma }_{\mathrm{ss}}\ge {\sigma }_t$), a more sophisticated nonlinear plastic–elastic model (referred to as a fracture model) developed in commercial FEA software is used to further study crack propagation within pavement layers. [Wanyan et al. (2008a) provides FEA developmental details.] The fracture toughness ($K_{\mathrm{IC}}$) and the stress intensity factor ($K_{\mathrm{I}}$) for Mode I fracture (Griffith 1921; Irwin 1957) are used as crack propagation criteria to further examine whether the initial shrinkage crack is stable or whether it will propagate through the pavement structure bywhere $f(a/W)$ = dimensionless parameter also referred to as a geometric factor. As its name implies, $f(a/W)$ depends on the geometries of both the specimen and the crack. Parameter $2a$ is the through-thickness crack length; $\sigma$ is the (remotely, not on crack tip) applied stress. Parameters $K_{\mathrm{I}}$ and $K_{\mathrm{IC}}$ can be compared using the same Eq. (6) but different applied linear tensile stresses (${\sigma }_{\mathrm{ss}}$ versus ${\sigma }_t$) and different geometry factors (semi-infinite subgrade vs. four point bending test dimensions) respectively. When $K_{\mathrm{I}}<K_{\mathrm{IC}}$, the crack is stable and will not grow. The controlled parameter MC is then further reduced. On the other hand, however, when $K_{\mathrm{I}}\ge K_{\mathrm{IC}}$, the crack will start to propagate upward. The progression of the initial shrinkage crack is critical to the development of the surface longitudinal crack. The fracture model identifies the other threshold (${\mathrm{MC}}_{\mathrm{p}}$) at which the cracks will grow through pavement layers and appear at the surface. The final outputs of the LSC model include the two moisture thresholds ${\mathrm{MC}}_{\mathrm{I}}$ and ${\mathrm{MC}}_{\mathrm{P}}$ and the coordinates of the critical locations for longitudinal cracking.

The two FEA models were set up using the same geometry to present typical low-volume pavement sections consisting of a HMA layer over a flexible base, an optional subbase, and a subgrade. Because of the symmetry, a half-width 3.66-m (12-ft) wide pavement with a 1.22-m (4-ft) wide shoulder was studied to reduce calculation efforts. The pavement shoulder was modeled as a uniform block fully bounded at the pavement interface. As few as three layers and as many as five layers can be introduced into the FEA models. Pavement layers were assumed to be homogeneous and isotropic.

The fracture model yielded similar results for typical three- and four-layer low-volume pavement sections as observed from the elastic model: the overall location of the largest tensile stress points shifted toward the centerline when the combined pavement thickness above the subgrade increased (with either an increased single-layer thickness or an increased number of layers), and the magnitude of the maximum shrinkage stress in the subgrade was not very sensitive (less than 5% difference) to the pavement structure variation [the HMA layer thickness varied from 0.01 to 0.11 m (0.5 to 4.5 in.) and modulus varied from 2.1 to 4.8 GPa (300 to 700 ksi); the flexible base thickness varied from 0.15 to 0.46 m (6 to 18 in.), and modulus varied from 137.9 to 551.6 MPa (20 to 80 ksi)]. However, with the more sophisticated fracture model, the benefit of using a better subgrade stands out. Fig. 6 compares the typical stress contours of the best (El Paso) and worst (Paris) case scenarios for three- and four-layer pavements that underwent the same moisture change (dried from optimum to 0.8 NMC). The typical pavement layer thicknesses are 0.06-m (2.5-in.) HMA, 0.30-m (12-in.) flexible base, 0.30-m (12-in.) lime stabilized subgrade (for four-layer case), and a semi-infinite natural subgrade. For the same subgrade, damage caused by subgrade shrinking is less severe for four-layer pavements, but the maximum ${\sigma }_{\mathrm{ss}}$ is close in magnitude. On the other hand, for the same pavement, a better subgrade shows noticeable improvement in terms of withstanding drying damage. These trends also agree with the trends from the elastic model.

As mentioned earlier, different drying/wetting paths were studied. When soil specimens are dried through different paths, the correlations between MC and soil mechanical property parameters (such as ${\epsilon }_{\mathrm{ss}}$, $E$, ${\sigma }_t$) change. Fig. 7 depicts typical trends in subgrade lateral shrinkage strain (top) and modulus (bottom) with different moisture variation paths, respectively. In the drying case, for example, the same expansive soil has bigger $E$ but smaller ${\epsilon }_{\mathrm{ss}}$ values at the same MC when comparing DFO to DFS, which is more optimistic. The great flexibility of the MC-based approach lies in the fact that different drying/wetting cycle effects can be incorporated into the material models, which can be switched very easily in FEA models to study different drying/wetting situations. In addition, the parameters used are standard soil index property tests, which are easy to measure and less susceptible to testing variations compared to suction measurements.

To evaluate and validate the LSC model, five representative FM roads built over high-PI clayey subgrade were selected for case study (Fort Worth, Houston, San Antonio, Paris and Atlanta in Texas) with the following preferred attributes: (1) they were reasonably newly constructed, (2) design records were available, (3) construction records were reasonably completed, (4) they contained some areas with typical distresses encountered due to high-PI clay, and (5) the clay subgrade was reasonably uniform. To ascertain the soil properties, soils from study sites were sampled for the aforementioned laboratory tests. As described by Manosuthikij (2008), two types of field sensor (Gropoint moisture sensors with data logger and Fredlund thermal conductivity matric suction sensors) were embedded at each test site and data collection was carried out every 1 to 2 months during each site visit. The pavement section information and subgrade properties from all sites are summarized in Table 2. The moduli of the pavement layers were obtained from FWD tests at the center of the lane (1.83 m or 6 ft from the shoulder) and checked with core sample test results from the district offices. The subgrade moduli listed in the table correspond to their laboratory values at the OMC. The actual subgrade moduli used in the LSC model was calculated based on Eq. (5).

The Paris site was used as a worst-case scenario in parametric studies, and field observations showed the worst pavement conditions at the initial site visit. Hence, details of this site are described here as an example. This road was rehabilitated in April 2007 and again in July 2007; minor cracks still reappeared on the pavement surface shortly after rehabilitation. Fig. 8 summarizes the Paris site field measurement results. The Paris site suffered from a poorly maintained drainage ditch and large trees near the pavement section. The LSC model predicted that the MC threshold for crack initiation and appearance was ${\mathrm{MC}}_{\mathrm{I}}=21\%$ and ${\mathrm{MC}}_{\mathrm{P}}=16\%$, respectively. This accords very well with field observations. As shown in Fig. 8, new cracks were observed in late September 2007. The lowest average MC was 15–17%. Elevation survey data also reached the lowest value near that time, confirming significant shrinkage behavior in the underlying and adjacent soils. Fig. 9 shows the predicted (a) and actual (b) location of the newly developed longitudinal crack. The top figure presents the LSC model plot of the simulated top 50 largest tensile stress points within the subgrade. The predicted most likely location of the crack is near 1 m (40 in.) from the edge. The actual longitudinal cracking was developed near outer-wheel lane at about $1/3$ lane width from edge, around 4–5 mm (1.5–2 in.) wide and in some areas over 0.46 m (18 in.) deep. Cracks were also visible along the shoulders at some locations.

Similar analyses were performed to compare results from other sites. Table 3 summarizes the field measurements with site attributes and LSC model results. All results from the LSC model show good correlations with field measurements. The measured and estimated MCs when the longitudinal cracks first appeared are close for the Fort Worth, San Antonio, and Paris sites. The model suggested that the most likely location of the crack at the Fort Worth site was about 0.91 m (3 ft) from the edge. The actual distressed area, which was observed within the outer wheel path, corresponded well to the model estimation. The San Antonio site was newly constructed, and no cracks were observed at the initial field visit. Severe longitudinal cracks, some of which were approximately 1 in. wide, were observed 2 years later near the pavement edge, which matches the predicted 0 distance from pavement–shoulder interface. The Houston site should have experienced longitudinal cracking damage when the MC dropped below 14%. However, no longitudinal cracks were observed at this site because even during the dry season, the average MC was above 21%, which was above the predicted crack initiation threshold. The high MC can be attributed to a creek in the vicinity of the pavement. The Atlanta site showed longitudinal cracks close to the shoulder during the initial visit. Due to equipment vandalism, field data for this site were incomplete. However, based on historical performance, this site is highly susceptible to moisture variation and, thus, is expected to experience substantial longitudinal cracking damage.