Composite Pavement Systems, Volume 2: PCC/PCC Composite Pavements (2013)

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.

47 Introduction The analysis of laboratory and field data along with the devel- opment of models to better understand PCC/PCC design and behavior are detailed in this chapter. Although composite pavements generally are known to perform quite well, there has been little formal research into the performance of new PCC/PCC structures. The effort conducted through this SHRP 2 R21 project involved the review and modification of existing models in the Mechanistic-Empirical Pavement Design Guide (MEPDG) for bonded-PCC-over-JPCP projects and the use of lattice models for fracture to better understand the risk of debonding at the PCC/PCC interface. Table 3.1 introduces anticipated modes of failure for PCC/PCC, prediction models associated with these distresses (including MEPDG), changes made to the associated models if applicable, and overall com- ments on the model modifications and development in the project work. Overall, the structural models in place for the MEPDG structural analysis of bonded PCC overlays were found to be sufficient for PCC/PCC composite pavements. The main dif- ficulty in MEPDG modeling for PCC/PCC was in the EICM, used by versions of MEPDG before MEPDG v. 1.3000:R21, to determine thermal gradients through the pavement system and assign k-value to the subgrade. For this reason, the bulk of changes to MEPDG models were in the EICM, as detailed in the following sections and in Appendix R. Analysis of Test Section Laboratory Data As detailed in Chapter 2, three concrete mixes were used at MnROAD. These include concrete with RCA aggregate and 40% fly ash replacement (denoted as RCA concrete), concrete with Class A aggregate with 60% fly ash replacement (denoted as low-cost concrete or LC), and concrete with high-quality granite aggregate with specified gradation for the exposed aggregate texture, and 15% fly ash replacement (denoted as EAC concrete). The notations RCA, LC, and EAC are used in the following sections to represent these three concrete mixes. Hardened PCC Properties Compressive Strength Compressive strength tests were performed on 4-in. and 6-in. EAC, RCA, and LC cylinders between 1 and 28 days. The concrete for these cylinders was taken from the trucks delivering the concrete to the construction site. Samples were cured on-site for 24 hours and then transported to laborato- ries. Figures 3.1, 3.2, and 3.3 show the compressive strength values of EAC, RCA, and LC samples, respectively. The figures show there was significant variability in the compressive strengths reported for the EAC, RCA, and LC. The variability in compressive strengths of the EAC samples ranged between 1,000 and 2,000 psi. This strength variance was the largest of the three concrete types, although this concrete had the most consistent plastic properties during placement. The compressive strengths of the RCA and LC ranged within a few hundred psi until approximately 7 days and increased to approximately 2,000 psi by 28 days. Although the Minnesota DOT data points appear more variable than the FHWA data points, it should be noted that all of the FHWA data points are average compressive strengths of two samples. All samples exceeded the 4,000 psi requirement by 28 days. A comparison of the average daily compressive strengths of all EAC, RCA, and LC samples is shown in Figure 3.4. The EAC samples had the highest average compressive strengths, whereas the RCA and LC compressive strengths were similar. The high compressive strengths for the EAC samples were thought to be attributable to the high quantity of cementi- tious material in the mixture compared with that in the RCA and LC mixes. The average 28-day compressive strengths for EAC, RCA, and LC field samples are within a few hundred C h A p T e r 3 PCC/PCC Analysis and Performance Modeling

48 Table 3.1. Modes of Failure, MEPDG Models Available, R21 Project Modifications for PCC/PCC Design Failure Mechanism MEPDG Model Available SHRP 2 R21 Model Modification Comment Bottom-up transverse cracking (fatigue) Available, Limited calibration Limited Existing JPCP model in MEPDG found sufficient for PCC/PCC with limited modification (thickness and slab/base friction). Additional work included modification to Enhanced Integrated Climatic Model (EICM). Top-down transverse cracking (fatigue) Available, Limited calibration Limited Existing JPCP model in MEPDG found sufficient for PCC/PCC with limited modification (thickness and slab/base friction). Additional work included modification to EICM. Longitudinal cracking (fatigue) Not available Not available Models for longitudinal cracking in PCC/PCC are not needed because they are not typically observed in the field. Joint faulting Available, Limited calibration None Existing JPCP model in MEPDG found sufficient for PCC/PCC composite pavement. Debonding between PCC layers Not available Not available Used lattice models for fracture to investigate debonding and determined that debonding in newly constructed PCC/PCC is not a concern. Figure 3.1. EAC compressive strength values for test sections constructed at MnROAD. Figure 3.2. RCA compressive strength values for test sections constructed at MnROAD. Figure 3.3. LC compressive strength values for test sections constructed at MnROAD. Figure 3.4. Average EAC, RCA, and LC compressive strengths for all cylinder samples for test sections constructed at MnROAD.

49 Figure 3.5. EAC flexural strength values for test sections constructed at MnROAD. Figure 3.6. RCA flexural strength values for test sections constructed at MnROAD. Figure 3.7. LC flexural strength values for test sections constructed at MnROAD. Figure 3.8. Average EAC, RCA, and LC flexural strengths for all beam samples for test sections constructed at MnROAD. psi of those determined for the laboratory mixed versions of these concretes by an independent, third-party testing agency. Flexural Strength Flexural strength tests were performed on 6- × 6- × 24-in. con- crete beams at the ages of 5, 7, 14, and 28 days. The concrete for these beams was taken from the trucks delivering the con- crete to the construction site. Samples were cured on-site for 24 hours and then transported to laboratories. Figures 3.5, 3.6, and 3.7 show the measured flexural strength values of EAC, RCA, and LC samples, respectively. The figures indicate that a 75 to 200 psi variation in flexural strength was common for each concrete mixture between 5 and 28 days. It should be noted that all of the FHWA data points are the average of two beam specimens. Figure 3.8 shows the average daily flexural strengths of all EAC, RCA, and LC beam samples. The EAC samples had the highest average flexural strengths, whereas the RCA had a slightly higher (50 to 100 psi) flexural strength than did the LC. The average 28-day flexural strength for the field EAC sample was approximately 200 psi less than that of the laboratory-mixed sample, as determined by an independent testing agency. The RCA field sample flex- ural strength was almost 300 psi less than that of the RCA labo- ratory flexural strength. The difference in the flexural strengths of the LC field and laboratory samples was around 50 psi. Poisson’s Ratio, CTE, and Split Tensile Strength In addition to compressive and flexural strengths, the FHWA Mobile Concrete Laboratory measured the modulus of elas- ticity, Poisson’s ratio, split tensile strength, and coefficient of thermal expansion (CTE) properties of the EAC, RCA, and LC samples, as summarized in Table 2.5 in Chapter 2. The concrete for these samples was taken from the trucks delivering the concrete to the construction site. Samples were cured on-site for 24 hours and then transported to the FHWA laboratory.

50 Table 2.5 showed that LC had the highest modulus of elas- ticity of the three concretes, which confirms that it is not a low-quality concrete. The aggregates could have been used for conventional concrete, but it is considered as low cost because fly ash was substituted for 60% of its cement. Gener- ally, because the lower layer of composite pavements does not receive tire wear, lower quality aggregates that are less polish- resistant could be used in the layer. For the construction at MnROAD, the contractor did not have access to lower quality aggregates that were not polish resistant and instead had to use normal or higher quality aggregates. Slant Shear Bond Strength Minnesota DOT’s Department of Materials and Road Research made concrete samples for ASTM C882: Standard Test Method for Bond Strength of Epoxy-Resin Systems Used with Concrete by Slant Shear. The specimens were made at the construction site by bonding two layers of concrete—EAC over LC or EAC over RCA—at an angled plane in a cylinder. An example of a slant shear specimen is shown in Figure 3.9. Figure 3.10 shows the measured bond strengths of either EAC over LC or EAC over RCA. Notice that LCC = LC and RCC = RCA (Akkari and Izevbekhai 2011). Freeze–Thaw Durability of MnROAD PCC/PCC Concretes Despite a relaxation in constituent standards, it is still important for the lower layer pavement concrete to be freeze–thaw durable because, although the surface concrete layer will act as an insu- lator in reducing temperature extremes, it will not prevent the lower layer from experiencing regular freeze–thaw cycles. To test the freeze–thaw durability of the proposed concrete mixtures, the International Union of Testing and Research Laboratories for Materials and Structures (Paris) (RILEM) CIF concrete freeze–thaw standard (Setzer 1997; Setzer 2009) was imple- mented, instead of the AASHTO T161 and AASHTO T277 standards. RILEM is an international union of laboratories Source: Akkari and Izevbekhai 2011. Figure 3.9. Example of a slant shear specimen. Source: Akkari and Izevbekhai 2011. Figure 3.10. Measured bond strengths of EAC/LC or EAC/RCA concretes.

51 and experts in construction materials, systems, and structures. The CIF test evaluates the capillary suction, surface scaling resistance, and internal damage of concrete samples exposed to 3% by volume sodium chloride solution, whereas AASHTO T161 evaluates the internal damage of concrete submerged in water from rapid freeze–thaw cycles and AASHTO T277 evalu- ates the scaling resistance of concrete exposed to 3% sodium chloride solution and freeze–thaw cycles. Capillary SuCtion Figure 3.11 shows the mass of water or solution absorbed per unit sample area during the capillary suction period (cycle 0) and during freeze–thaw cycles. The test surfaces of the field RCA and EAC samples were submerged in pure water, and the test surfaces of the laboratory RCA, LC, EAC, and control samples were submerged in 3% sodium chloride solution. The dotted line on Figure 3.11 indicates the moisture uptake of the concrete samples at 56 freeze–thaw cycles. Tables 3.2 and 3.3 show the mean and standard devia- tions (SD) of mass increase after the capillary suction phase and before freeze–thaw cycles began (cycle 0), at the criti- cal freeze–thaw cycle (56 cycles), and at the end of the test, which was 106 cycles for the field samples and 98 cycles for the laboratory samples. During the 7-day preconditioning period, the field RCA and EAC samples averaged a 0.49% mass gain because of capillary suction of pure water. Comparatively, the laboratory RCA and EAC samples averaged mass gains of 1.58% and 1.03%, respectively, attributable to capillary suction of 3% sodium chloride solution. After the preconditioning period and during the freeze–thaw cycles, the rate of mass increase was approxi- mately equal for all field and laboratory samples. At approxi- mately 30 freeze–thaw cycles, the rate of mass increase for all field and laboratory samples decreased to almost zero. At first glance, it seems that the disparity in the initial uptake of test liquid between the field and laboratory samples is a function of the test liquid: pure water for the field samples and 3% sodium chloride solution for the lab samples. How- ever, consideration of the compressive strengths (Table 3.4) and delineation of the moisture uptakes between the field and the laboratory samples (Figure 3.11) suggest that the dis- parity in moisture uptake has more to do with the samples’ capillary porosity than with the test liquid. In other words, despite being made with identical mix designs and concrete constituents, the laboratory samples had a greater volume Figure 3.11. Field and laboratory samples’ average moisture uptake as a percentage of sample mass during freeze–thaw cycles. Table 3.2. Field Sample Relative Increase in Mass Mean and Standard Deviation Cycle Field RCA Mean Mass Increase (%) Field RCA SD Mass Increase (%) Field EAC Mean Mass Increase (%) Field EAC SD Mass Increase (%) 0 0.49 0.0012 0.50 0.0010 66 1.48 0.0016 1.47 0.0016 106 1.59 0.0024 1.60 0.0022

52 and average capillary pore size than did the field samples, as evidenced by the lower 7- and 28-day strengths. The average RCA laboratory sample’s 28-day compressive strength was 50% that of the average RCA field sample. Similarly, the aver- age EAC laboratory sample’s 28-day compressive strength was 75% that of the average EAC field sample. The reason for the strength, and thus capillary suction, disparity between the field and laboratory samples remains unknown. Two production disparities between the field and laboratory samples may have contributed to the disparity in moisture uptake. One production disparity was that the con- crete used to make the field samples was batched at a ready-mix concrete plant, and the concrete used to make the laboratory samples was batched in a 1-yd3 mixer. A second production disparity was that the field samples were consolidated on a vibrating table and the laboratory samples were consolidated by rodding. Note that the laboratory samples were prepared twice because the first samples failed to gain the expected strength. The compressive strength of the second batch of laboratory samples also did not gain the expected strength. In a RILEM CIF test study of high-performance concrete mixtures with water-to-cement ratios of approximately 0.30 and average cement contents of 742 lb/yd3, the moisture uptake was between 1.5% and 3.0% of the samples’ mass (Setzer 1997). In another study that used the CIF test on concrete samples with 0.5 w/c ratios and with between 50% and 100% coarse RCA substitution, the moisture uptake was between 3% and 4.5% (Setzer 1997). For the MnROAD samples, as the moisture uptake ranged from 0.5% to 3%, it is reasonable to conclude that the water uptake by all of the field and laboratory samples was within an acceptable range. DeiCing Salt SCaling Figure 3.12 shows the scaled test surfaces of an EAC sample after 56 freeze–thaw cycle. Table 3.3. Laboratory Sample Relative Increase in Mass Mean and Standard Deviation Cycle Laboratory RCA Mean Mass Increase (%) RCA SD Mass Increase (%) Laboratory Mean LC Mass Increase (%) LC SD Mass Increase (%) Laboratory Mean EAC Mass Increase (%) EAC SD Mass Increase (%) Laboratory Mean Control Mass Increase (%) Control SD Mass Increase (%) 0 1.58 0.17 0.94 0.07 1.03 0.16 1.34 NA 64 3.06 0.20 1.65 0.11 2.55 0.06 2.67 NA 98 3.31 0.18 1.78 0.06 2.63 0.07 2.70 NA Note: NA = not available. Table 3.4. Plastic and Hardened Properties of Field and Laboratory Samples Field RCA Field EAC Laboratory RCA Laboratory LC Laboratory EAC Laboratory Control Plastic Properties Slump (in.) 1.75 2.25 4.5 2 1 4.75 Air (%) 6.5 4.5 7.5 6.2 4.7 8 Unit weight (lb/ft3) 145 145 140 140 151 141 Hardened Properties 7-day compressive strength (psi) 3,599 5,314 1,738 2,950 3,836 1,769 28-day compressive strength (psi) 4,305 5,855 2,365 4,185 4,404 2,294 Figure 3.12. Scaled test surfaces of an EAC sample after 56 freeze–thaw cycles.

53 Figure 3.13 shows the cumulative mass of scaled material per unit area for both field and laboratory concrete samples as the number of freeze–thaw cycles increases. It is important to reiterate here that the test surfaces of the field RCA and EAC samples were submerged in pure water and the test sur- faces of the laboratory RCA, LC, EAC, and Control samples were submerged in 3% sodium chloride solution. The hori- zontal dashed line on Figure 3.13 indicates the maximum recommended scaling mass after 56 freeze–thaw cycles. The vertical dotted line indicates mass of scaled material at the 56th freeze–thaw cycle. Tables 3.5 and 3.6 show the mean value and the standard deviation of scaled material at the 56th freeze–thaw cycle and after termination of the CIF test for the field and laboratory concrete samples, respectively. As shown by Figure 3.13, all field and laboratory samples, except the laboratory LC sample, averaged below 1,500 g/m2 of scaled material at 56 freeze–thaw cycles. The laboratory LC sample average was 1,913 g/m2, which is within 2 SD of the limit and thus can be cautiously accepted as adequately able to resist surface scaling. The field RCA and EAC samples scaled significantly less than the laboratory RCA and EAC samples. This disparity was expected and is primarily attributable to the difference in test liquid (pure water for field samples and 3% sodium chloride solution for laboratory samples). The disparity was not caused by capillary suction (Setzer 2009), nor was it likely caused by a lack of entrained air in the surface concrete. The entrained air measurements of the plastic concretes were suf- ficient (between 4.5% and 8%) for all field and laboratory Figure 3.13. Cumulative mass of scaled material per unit area subjected to freeze–thaw cycles. Table 3.5. Mean and Standard Deviation of Field Sample Scaled Material per Surface Area Cycle Field RCA Mean Scaled Material (g/m2) Field RCA SD (g/m2) Field EAC Mean Scaled Material (g/m2) Field EAC SD (g/m2) 66 126 19 95 19 106 204 30 143 25 Table 3.6. Mean and Standard Deviation of Laboratory Sample Scaled Material per Surface Area Cycle Laboratory RCA Mean Scaled Material (g/m2) RCA SD (g/m2) Laboratory Mean LC Scaled Material (g/m2) LC SD (g/m2) Laboratory Mean EAC Scaled Material (g/m2) EAC SD (g/m2) Laboratory Mean Control Scaled Material (g/m2) Control SD (g/m2) 64 1,438 71 1,913 225 1,651 342 1,139 NA 98 2,296 251 2,816 270 2,520 348 1,642 NA Note: NA = not available.

54 samples. For the PCC/PCC composite pavements constructed at MnROAD, the EAC was the only concrete that would have been subjected to deicing salt, and it proved to be an adequate concrete mixture for resisting deicing salt scaling. internal Damage Figure 3.14 shows the relative modulus of elasticity of ultra- sonic transit time (relative modulus) for field and laboratory samples. The vertical dotted line indicates the relative modulus at 56 freeze–thaw cycles. The horizontal dashed line indicates the modulus below which the sample is considered distressed. Tables 3.7 and 3.8 show the mean and standard deviation rela- tive modulus values at 54 or 56, and either 106 or 98 freeze– thaw cycles for the field and laboratory CIF test samples, respectively. After 56 freeze–thaw cycles, the relative moduli of the field RCA and EAC samples were approximately 94% and 98%, respectively, indicating very little internal damage. As con- firmed by the low uptake of moisture, the capillary porosity of the field samples was low and the capillaries likely were disconnected, which are indicators of a concrete paste that is resistant to freeze–thaw damage. In addition, these samples were adequately air entrained. CIF experiments by others have shown that internal damage is not a function of test liquid (sodium chloride solution), surface texture, or carbonation but rather depends mostly on the sample’s water-to-cement Table 3.7. Average Relative Modulus of Elasticity of Ultrasonic Transit Time for Field RCA and EAC Samples Cycle Field RCA Average Ru,n (%) Field RCA SD (%) Field EAC Average Ru,n (%) Field EAC SD (%) 0 100 0 100 0 54 94 4 98 2 106 93 3 97 2 Note: The values shown represent an average from five samples. Ru,n = relative modulus of elasticity of ultrasonic transit time. Table 3.8. Average Relative Modulus of Elasticity of Ultrasonic Transit Time for Laboratory RCA, LC, EAC, and Control Samples Cycle Laboratory Average RCA Ru,n RCA SD Laboratory Average LC Ru,n LC SD Laboratory Average EAC Ru,n EAC SD Laboratory Average Control Ru,n Control SD 0 100 0 100 0 100 0 100 NA 56 80 17 93 2 88 3 96 NA 98 84 3 98 12 78 26 106 NA Note: Ru,n = relative modulus of elasticity of ultrasonic transit time; NA = not available. Figure 3.14. Modulus of elasticity of ultrasonic pulse velocity.

55 ratio and the quantity and spacing of entrained air bubbles (Setzer 1997). As indicated by the erratic peaks and valleys of the labora- tory sample modulus measurements, the data did not follow a neat downward line, and sometimes the modulus values exceeded 100%. Despite the fluctuations, the data trended toward a decreasing modulus as the number of freeze–thaw cycles increased. The relative moduli of the samples also remained at or above the 80% after 56 cycles, which indi- cates that the samples adequately resisted internal damage caused by frost action. As stated, it is the strength of the concrete matrix, rather than the test liquid, that influences the resistance of a sample to internal damage. Compared with the decrease in relative modulus of other concrete samples studied with the RILEM CIF procedure, the decreases in relative moduli of all of the field and labora- tory samples were relatively small (Setzer 1997; Setzer 2009). The difference between the concretes used for the MnROAD construction and those in the referenced studies is that all of the MnROAD construction samples were air entrained. One study evaluated a 0.5 water-to-cement ratio concrete with- out entrained air, and the relative modulus values decreased below 80% after between 10 and 30 freeze–thaw cycles (Setzer 1997). Another study showed that even in a CIF experiment that evaluated 0.3 water-to-cement ratio and high cement content samples, only those concrete mixtures that contained entrained air were able to adequately resist internal cracking according to the RILEM CIF standard. For the SHRP 2 R21 composite pavement project, the lack of internal damage in both the RCA and LC mixtures after 56 freeze–thaw cycles indicated that these mixtures are suitable for use in long-life concrete pavements, despite containing RCA or having a 60% cement replacement with fly ash, respec- tively. It was expected that the EAC samples would experience minimal internal damage caused by frost action because of its superior granite aggregates and high cement content paste. Analysis of Field Data at MnrOAD The two PCC/PCC test sections constructed at the MnROAD facilities were labeled Cell 71 and Cell 72. In addition, as detailed in Volume 1, an HMA/PCC section was also constructed and labeled Cell 70. Cell 71 had the same lower lift PCC mix as Cell 70 with RCA in the PCC mix, whereas Cell 72 did not have RCA in PCC mix but had higher fly ash content (60% replacement versus 40% replacement). Results from Cell 70 are included in some of the figures and discussions in this section. PCC Slab Temperature Profiles The simplest way to characterize the temperature distribu- tion in the slab is by assuming a linear distribution for the temperature throughout the depth of the slab. The linear tem- perature gradient (LTG) is calculated as the temperature dif- ference between the top and bottom of the slab taken over the distance between the two. However, several field studies have shown that the distribution of temperature throughout the slab depth is primarily nonlinear (Armaghani et al. 1987; Yu et al. 1998). To account for the nonlinearity of the temperature distribution in the slab, the equivalent temperature gradient concept was developed (Thomlinson 1940; Choubane and Tia 1992; Mohamed and Hansen 1997). The equivalent linear tem- perature gradient (ELTG) is a linear gradient that would pro- duce the same curvature in the slab as the original nonlinear temperature gradient. The ELTG concept was later generalized for nonuniform, multilayered slabs (Khazanovich 1994; Ioan- nides and Khazanovich 1998). The latter method, in which the ELTG is established for an effective slab, with a thickness and stiffness equivalent to that of a composite multilayer section, is used in this analysis. Effect of Location on Slab (Midslab versus Edge versus Corner) An assessment was made for the LTGs between locations within each slab. It is possible that the magnitude of temperature gra- dients in a slab can differ between locations because of the dif- ferent boundary conditions at midslab, edge, and corner. To assess the variation in the temperature gradients that develop in the slab with location, the LTG in the composite of the upper and lower layers for Cells 71 and 72 was calculated over a typi- cal day. Temperature data collected on July 19, 2010, was used for this analysis. The ambient temperature and solar radiation for this day is presented in Figure 3.15. The maximum ambient temperature of 80°F observed on July 19, 2010, is not represen- tative of the extreme conditions observed at MnROAD during the summer. However, because the objective of this analysis is only to determine the variation in temperature gradients with Figure 3.15. Ambient temperature and solar radiation for July 19, 2010, measured at the project site.

56 thermocouple depths are summarized in Table 3.9 for both cells. The depths available for the sensors, shown in Table 3.9, indicate that the sensors were installed at similar depths for each of the locations. Based on the comparative analysis of the variation in the LTG for each location within the slab, it can be concluded that the largest positive and negative LTGs occur at the midslab location. Therefore, the LTG corresponding to this location is used for subsequent analyses. The variation between LTGs for different locations is not significant and is not expected to affect the analysis. Variation in LTG between Cell 71 and Cell 72 Figures 3.18 and 3.19 show histograms of the LTG distribution in the slabs for Cells 71 and 72, respectively. The figures show that the variation in LTG between the slabs within the same cell is negligible. To compare the LTGs between Cells 71 and 72, LTGs only for Slab 2 are compared in Table 3.10. The tempera- ture data for Slab 1 in Cell 72 was not reliable for a considerable period, so the data for Slab 2 in both the cells is considered. Based on the data in Table 3.10, the variation between the two cells is apparent. Cell 71 experiences larger temperature -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 0 2 4 6 8 10 12 14 16 18 20 22 24 LT G , ° F/ in Time of day (July 19th 2010) Panel 1: Mid slab Panel 1: Edge Panel 2: Corner Panel 2: Mid slab Panel 2: Edge Figure 3.16. LTGs for various slab locations in Cell 71 on July 19, 2010. -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 0 2 4 6 8 10 12 14 16 18 20 22 24 LT G , ° F/ in Time of day (July 19th 2010) Panel 1: Corner Panel 1: Mid slab Panel 1: Edge Panel 2: Corner Panel 2: Mid slab Panel 2: Edge Figure 3.17. LTG at various locations in the slabs in Cell 72 on July 19, 2010. Table 3.9. Approximate Depths of the Top and Bottom Thermocouples for Cell 71 and Cell 72 Thermocouple Depths, in. Cells 71 and 72: Slab 1 Cells 71 and 72: Slab 2 Location Corner Midslab Edge Corner Midslab Edge Top 0.75 0.75 0.75 0.75 0.75 0.75 Bottom 9.25 9.25 9.25 9.25 9.25 9.25 location over a typical day, the LTGs computed for this date from the temperature data at all the available thermocouple locations were used. The LTGs calculated at different locations within Cells 71 and 72 are shown in Figures 3.16 and 3.17. The midslab loca- tions in both Slabs 1 and 2 for both cells experience higher magnitude of LTG compared than do other locations. The Figure 3.18. LTG distribution for Slabs 1 and 2 in Cell 71.

57 not sufficient for characterizing the actual variation of the temperature throughout the depth. The nonlinearity of the temperature distribution should also be considered, which is done using ELTG. The ELTGs were estimated for Cells 70 and 71, using Equations 3.1 to 3.3. ∆T h T z T zdz E x h x eff eff top bot b = × ( )−( ){ + − −∫122 0 α ot top top top top bot α E T z T zdz h x h h x ( )−( )  − + −∫ 0  ( . )3 1 where DTeff = difference between temperatures at the top and bottom surfaces of the effective slab, T(z) = temperature distributions through the PCC concrete, T0 = zero-stress temperature, z = vertical coordinate measured downward from the neutral axis of the composite pavement, htop = thickness of the upper layer, hbot = thickness of the lower layer, Etop = elastic modulus of the upper layer, Table 3.10. Statistics for LTG Over the Analysis Period, Comparing Cells 71 and 72 LTG, F/in. Cell 71 Cell 72 Average 0.03 0.01 Maximum 4.14 2.93 Minimum -2.22 -1.91 SD 1.26 0.93 Median -0.37 -0.25 Number of observations per slab = 20,814 Figure 3.19. LTG distribution for Slabs 1 and 2 in Cell 72. Figure 3.20. Temperature distribution in the PCC for Cell 70. Figure 3.21. Temperature distribution in the composite section for Cell 71. gradients more frequently than does Cell 72. As discussed in Appendix H, the weighted average temperature (WAT) for Cell 71 was higher than that for Cell 72. This finding could be the result of the limited data collected or the difference in thermal properties (such as heat capacity and thermal con- ductivity) between the two cells. This analysis should be per- formed again once a larger database of data is available. Comparison of Exposed Concrete (Cell 71) to HMA-Covered Concrete (Cell 70) Figures 3.20 and 3.21 illustrate the temperature distribu- tions in the slabs throughout a single summer day for Cells 70 and 71, respectively. The figures show that the devia- tion of the temperature profiles from a linear gradient is more pronounced in Cell 71, which is not covered by the HMA, as is the case for Cell 70. The difference in nonlinear- ity between Cell 70 and Cell 71 suggests that LTG alone is

58 Ebot = elastic modulus of the lower layer, atop = coefficient of thermal expansion of the upper layer, abot = coefficient of thermal expansion of the lower layer, and heff = effective thickness of the pavement, which can be determined from Equation 3.2: h h E E h h x h eff top bot top bot top top = + + −     3 3 12 2  + + −        2 2 2 E E h h x hbot top top bot bot  3 3 2( . ) x = distance between the neutral plane and the top sur- face of the upper layer, which can be determined from Equation 3.3: x h E E h h h h E = + +   + top bot top bot top bot top 2 2 2 bot top bot E h ( . )3 3 Figures 3.22 and 3.23 show the variation in the LTGs and ELTGs between Cells 70 and 71. The summary statistics for the ELTGs that developed in both cells is presented in Tables 3.11 and 3.12. Cell 70 shows a higher frequency of occurrence of ELTGs close to zero than does Cell 71. As seen in Figures 3.22 and 3.23, the shapes of the frequency distribution curves of LTGs and ELTGs for Cell 70 are quite similar, whereas a con- siderable difference in shape is observed in Cell 71. Unlike the LTGs, the ELTGs for Cell 71 are evenly distributed over a broader range. To investigate the significance of variation in LTGs and ELTGs within a cell and the variation of each of these gradients between the cells, a paired t-test for two sample means was performed. Results from the t-test are given in Tables 3.11 and 3.12. The variation between the LTGs and ELTGs in Cell 70 is not significant, whereas it is quite significant for Cell 71. This indicates that the gradients that develop in the HMA/PCC pavements tend to be much more linear than in the PCC/PCC pavements, as is supported by the temperature profiles provided in Figure 3.21. The results of the paired t-test for the ELTGs between Cells 70 and 71 also conclude the difference in the ELTGs is significant at a 95% confidence level. The ELTGs in Cell 71 (PCC/PCC) are much higher over a larger period of time than the ELTGs for Cell 70 (HMA/PCC). It can clearly be seen in Figure 3.23 that the magnitude of the temperature gradients, as well as the frequency at which these higher gradients develop, is signifi- cantly greater for a PCC/PCC pavement than for an HMA/PCC pavement. PCC Slab Moisture Profiles In addition to temperature variation within the slab, the moisture variation through the depth is also important. This is because the moisture variation through the depth of the slab Figure 3.22. Comparison of relative frequencies for LTGs in Cells 70 and 71. Figure 3.23. Comparison of relative frequencies for ELTGs in Cells 70 and 71. Table 3.11. Comparing the LTG and ELTG Within Each Cell for Cells 70 and 71 Cell 70 Cell 71 LTG ELTG LTG ELTG Average, °F/in. -0.07 -0.07 -0.08 -0.20 Variance, °F/in. 0.79 0.829 1.29 3.85 Observations 28,649 Hypothesized mean difference 0 Degrees of freedom 28,648 t-Statistics 1.43 14.64 p-value 0.15 0 t Critical two-tail 1.96 1.96

59 produces an upward warping of the slab (because the bot- tom of the slab is almost always saturated and the top typi- cally goes through wet-dry cycles). To capture the variation in the moisture content through the depth of the concrete layers, relative humidity was measured at different depths and locations. A total of 72 humidity sensors were installed in the two PCC/ PCC cells. Ambient relative humidity, temperature, and solar radiation also were measured with the weather station on-site. The variation in the daily average ambient relative humidity at the project location, during the analysis period (May 2010 to March 2011), is shown in Figure 3.24. The range of the average daily ambient relative humidity is between 50% and 100%. The variation in the relative humidity in the concrete at very shallow depths might show a similar variation to the ambient relative humidity with time. However, it is more likely that the variation in the relative humidity in the concrete follows seasonal trends because of a slow diffusion of water through concrete. To assess the seasonal trends in the ambient relative humidity, the average ambient relative humidity for each month of the analysis period was calculated as shown in Figure 3.25. The figure shows that the ambient relative humidity increases in the winter, with the highest values in November and December and the lowest value in September. As detailed in Appendix H, the relative humidity data col- lected by some sensors were not of an acceptable quality. Therefore, an initial quality check was performed on the data to select sensor locations with a suitable data set. On this basis, the relative humidity data for the midslab and corner of Slab 2 for Cell 71 and edge of Slabs 1 and 2 for Cell 72 were selected for analysis. Figures 3.26 through 3.29 present the variation in relative humidity with depth for the locations in Cells 71 and 72. The common observation for all four figures is that for the first 2 to 3 weeks after paving, there is a significant drop in rela- tive humidity that is uniform for all sensors throughout the depth of the concrete. This may be attributed to hydration of the concrete. It can be seen that in late November, the varia- tion in the relative humidity increases suddenly, and contin- ues throughout the winter and early spring. The increase in relative humidity during the winter months is the result of a decrease in the temperature and not a change in the moisture content. Unfortunately, the moisture content in the concrete cannot be measured directly and must be estimated based on the measured relative humidity. Therefore, when interpreting these data, it is important to remember that the relative humid- ity will increase when the temperature decreases, even when Table 3.12. Comparing the ELTG Statistics for Cells 70 and 71 ELTG, F/in. Cell 70 Cell 71 Average °F/in. -0.070 -0.195 Maximum 3.540 5.951 Minimum -1.864 -4.699 SD 0.911 1.961 Median -0.292 -0.336 Paired t-test results Variance, °F/in. 0.829 3.845 Observations 28,649 Hypothesized mean difference 0 Degrees of freedom 28,648 t-Statistics 15.11 p-value 0 t Critical Two-tail 1.96 Figure 3.24. Daily average ambient relative humidity at MnROAD.

60 Figure 3.25. Monthly average ambient relative humidity at MnROAD. 70 75 80 85 90 95 100 105 110 5/5 52/5 41/6 4/7 42/7 31/8 2/9 22/9 21/01 1/11 12/11 11/21 13/21 02/1 9/2 1/3 12/3 % ,ytidi m u H e vitale R 0.75-in 7.25-in 3.25-in Figure 3.26. Relative humidity in the concrete at midslab for Slab 2 in Cell 71. 70 75 80 85 90 95 100 105 110 5/5 52/5 41/6 4/7 42/7 31/8 2/9 22/9 21/01 1/11 12/11 11/21 13/21 02/1 9/2 1/3 12/3 % ,ytidi m u H e vitale R 2.75-in 6.25-in 7.25-in Figure 3.27. Relative humidity in the concrete for the corner of Slab 2 in Cell 71. the moisture content remains constant. For this reason, it is important to make comparisons between the concrete relative humidity measurements made at the same time of year over a period of 5 or 6 years. It typically takes about 5 to 7 years before all of the irrecoverable drying shrinkage develops at the surface of the slab. Unfortunately, the complete interpretation of the moisture data is not possible because less than a year’s data were available at the time the analysis for this report was performed. The most reasonable measurements of the relative humid- ity in the concrete were obtained from Cell 72. These mea- surements are shown in Figures 3.28 and 3.29. The relative humidity measured at the top of Slabs 1 and 2 for this cell show the largest daily variations. This is expected because the sensors close to the surface are most heavily influenced by the ambient conditions. However, the two top sensors in Cell 71 do not show the same behavior over the analysis period. The variation in the relative humidity between the lower sensors is small and remains constant over the year after initial drying of the concrete. It is possible that this variability is the result of variations in the sensor depth because the exact as-built depths are unknown. A better interpretation will be possible as more data become available over time. Establishing Built-In Gradients The built-in gradient includes the temperature and moisture gradient that “lock” into the slab at the zero-stress time (TZ). TZ occurs after the final set and is the point in time when the slab has grown sufficient strength (essentially changing from a semisolid to solid state), to respond to temperature changes. Although moisture gradients at TZ have been shown to be close to zero (Wells et al. 2006), temperature gradients at this point in time can have influential values. Built-in tempera- ture gradients are important because as a result of this gradi- ent the slab does not remain flat during its service life, even when temperature and moisture gradients are zero. Before TZ, the slab is flat regardless of the temperature gradient in

61 the slab. The temperature gradient that is present in the slab at TZ is “locked” into the slab. The TZ, WAT, and built-in temperature gradient were estab- lished for each instrumented cell at MnROAD. To establish TZ, two methodologies were used, one based on the variation seen in the measured strain with respect to temperature changes in the slab (Method 1) and the other based on the initiation of curling in the slabs with respect to LTG (Method 2), as detailed in Appendix H. For Cell 71, TZ was established as between 15 and 17 hours using Method 1 and between 14 and 15 hours using Method 2. The early-age data were unfortunately missing for Cell 72. TZ was determined based on the maturity concept for the upper layer in Cell 72 as between 16 and 20 hours, whereas TZ for the lower layer was estimated as sometime between 16 and 24 hours. Overcast conditions at the time of paving resulted in relatively constant temperature conditions in the slab over the first 24 hours after paving. Therefore, the climatic conditions 70 75 80 85 90 95 100 01/5/5 01/52/5 01/41/6 01/4/7 01/42/7 01/31/8 01/2/9 01/22/9 01/21/01 01/1/11 01/12/11 01/11/21 01/13/21 11/02/1 11/9/2 11/1/3 11/12/3 0.75-in 1.75-in 7.25-in 8.75-in Figure 3.28. Relative humidity in the concrete at the edge of Slab 1 in Cell 72. 70 75 80 85 90 95 100 01/5/5 01/52/5 01/41/6 01/4/7 01/42/7 01/31/8 01/2/9 01/22/9 01/21/01 01/1/11 01/12/11 01/11/21 01/13/21 11/02/1 11/9/2 11/1/3 11/12/3 0.75-in 7.25-in 8.75-in Figure 3.29. Relative humidity in the concrete at the edge of Slab 2 in Cell 72.

62 in the slab at the TZ could be established relatively well, even with the limited data available. The slab WAT at TZ is another parameter that needs to be established. This parameter is significant because it defines the amount of uniform thermal expansion and contraction in the slab. The WAT at hour 15, selected as the TZ, is approxi- mately 79°F for the lower PCC in Cell 71. For the upper PCC layer in Cell 71, the WAT at TZ (hour 15) is approximately 75°F. For Cell 72, the WAT at the selected TZ (hour 20) is approximately 58°F for the upper PCC layer and approxi- mately 62°F for the lower PCC layer. The ELTG at TZ is the built-in temperature gradient that locks into the slab and influences its future shape. For Cell 71, the average TZ (based on both methodologies), for top and lower layers is between 15 and 16.25 hours. The ELTGs esti- mated using thermocouple data from Slab 2 between 14 and 17 hours after paving are shown in Figure 3.30. The thermo- couple data from Slab 1 on this cell was not usable for the first 5 days. Figure 3.30 shows that the ELTG is approximately -0.7°F/in., which corresponds to a built-in temperature dif- ference of approximately -6°F for a 9-in. PCC slab. For Cell 72, the ELTG calculated based on the thermocou- ple readings over the time period of 16 to 24 hours after pav- ing (the estimate of TZ), is shown in Figure 3.31. Because the ELTG in the entire time span shows a significant variation, hour 20, at which the ELTG stabilizes and remains constant thereafter, is selected as TZ for both layers in this cell. The ELTG at TZ is approximately -0.8°F/in., which corresponds to a built-in temperature difference of approximately -7°F for a 9-in. PCC slab. PCC/PCC Interface Tensile Bond Strength Test During the survey of agencies regarding PCC/PCC construc- tion, a major concern expressed by many agencies had to do with the interface bond between the two PCC layers. A good bond between both the PCC layers is essential for the long- term performance of PCC/PCC composite pavements. The bond between the two PCC layers was tested by the FHWA Mobile Concrete Laboratory in August 2011, more than 1 year after construction. ASTM C1583-04 (Standard Test Method for Tensile Strength of Concrete Surfaces and the Bond Strength or Tensile Strength of Concrete Repair and Overlay Materials by Direct Tension [Pull-Off Method]) was used to evaluate the bond. The pull-off test involves applying a direct tensile load to a partial core (one that is advanced completely through the upper PCC layer but only partially through the lower PCC layer) until failure occurs. The tensile load is applied to the partial core through the use of a metal disk with a pull pin, bonded to the surface of the upper layer with an epoxy. A loading device with a reaction frame applies the load to the pull pin. The load is applied at a constant rate, and the ultimate load is recorded. Figure 3.32 illustrates the principle of the pull-off test. The pull-off strength (SPO) is defined as the tensile (pull- off) force (FT) divided by the area of the fracture surface (Af), using Equation 3.4: SPO F AT f= ( . )3 4 The Mobile Concrete Laboratory staff used Proceq’s DYNA Z15 to perform all the pull-off tests. Pull-off testing was per- formed on the passing lanes of Cell 71 and Cell 72. The weather conditions were windy, and testing on both the cells was per- formed on the same day. The results of the pull-off tests are shown in Tables 3.13 and 3.14 for Cell 71 and Cell 72, respec- tively. Figure 3.33 shows the fractured cores of the pull-off tests from both the cells. The results from the tensile pull-off tests show the following: • The tensile bond strength results of all the tests were higher than 1 MPa, which is generally considered good bond strength. Except for Test B4, results of all the tests in the two Figure 3.30. ELTG over the range of TZs for Slab 2 in Cell 71, using thermocouple data. Figure 3.31. ELTG estimated using thermocouple data in Slab 2 for the range of the TZs for Cell 72.

63 Source: FHWA 2000. Figure 3.32. Schematic of pull-off test. Table 3.13. Pull-off Test Results for Cell 71 Core ID Tensile Strength, (MPa) Location of Fracture, Comment Average Core Length (in.) B1 na Bond failure at the surface, epoxy na B2 na Bond failure at the surface, epoxy na B3 2.0 Below interface in substrate concrete 3.4 B4 1.3 Below interface in substrate concrete 3.5 B5 1.8 Below interface in substrate concrete, pulled-off aggregate 3.8 B6 1.9 At the interface 3.3 Note: Core lengths are the average of three readings; na = not applicable. Table 3.14. Pull-off Results for Cell 72 Core ID Tensile Strength (MPa) Location of Fracture, Comment Average Core Length (in.) A1 2.2 Below interface in substrate concrete 3.4 A2 2.4 Predominantly along the bond surface and partially in the substrate 3.3 A3 3.0 In the top layer 2.0 A4 3.5 In the top layer 1.0 A5 2.3 Partially along the bond surface and partially in the substrate 3.2 A6 1.8 Below interface in substrate concrete 3.6 Note: Core lengths are the average of three readings. Figure 3.33. Fractured cores: Cell 72 (left) and Cell 71 (right).

64 cells were equal to or higher than 1.8 MPa. Thus, the com- posite pavement layers in Cell 71 and Cell 72 are well bonded to each other. • Except for A3 and A4, failures occurred either in the lower PCC layer or partly at the interface and partly in the lower PCC layer. This shows that the upper PCC layer is stron- ger than the lower PCC layer, which is consistent with the results from the compressive and flexural strength testing. • In the case of A3 and A4, fracture occurred in the upper PCC layer and the tensile strengths of both the tests were significantly higher (3 and 3.5 MPa) than were those of other tests (2.4 MPa or lower). It appears that the partial cores at A3 and A4 did not advance beyond the bond interface. This could be attributable to the upper PCC layer having greater thickness at the A3 and A4 locations. Therefore, at A3 and A4, tensile stress was applied only to the upper PCC layer, so fracture occurred at the weakest point in the upper PCC layer. The high tensile strength at A3 and A4 is because of the high strength of the upper PCC layer. • The standard deviations of the results of four tests in Cell 71 and the four tests in Cell 72 (excluding the A3 and A4 tests) were low, which shows good repeatability of the test. The average tensile strength of the four tests of Cell 71 was 1.8 MPa and the average tensile strength of the four tests in Cell 72 was 2.2 MPa. This shows that the LC mix was slightly stronger than the RCA mix. The higher amount of fly ash in the LC mix compared with the RCA mix (60% versus 40%) could be the reason for the higher strength. • Tests B1 and B2 had epoxy failure. The diamond grinding texture in this cell provided smaller surface area for the epoxy and may have contributed to the failure. MEPDG JpCp Transverse Cracking Models for pCC/pCC The MEPDG considers two mechanisms of transverse crack- ing: bottom-up and top-down cracking. When the truck axles are near the longitudinal edge of the slab, midway between the transverse joints, a critical tensile bending stress occurs at the bottom of the slab, as shown in Figure 3.34. This stress increases greatly when there is a high positive temperature gradient through the slab (on a hot sunny day, the top of the slab is warmer than the bottom of the slab). The critical loading condition for top-down cracking involves a combination of axles that loads the opposite ends of a slab simultaneously. In the presence of a high negative tempera- ture gradient, such load combinations cause a high tensile stress at the top of the slab near the middle of the critical edge, as shown in Figure 3.35. During the analysis of the MEPDG for the SHRP 2 R21 proj- ect, numerous strange predictions for both bottom-up and top-down cracking resulting from changes of the thickness of layers in a PCC/PCC pavement were encountered. For exam- ple, equivalent PCC/PCC systems with a composite thickness of 9 in. and identical material properties were shown to pro- duce vastly different performances in bottom-up damage, as shown in Table 3.15. Note that all pavements are equivalent (i.e., both layers have identical properties); their only difference is the top layer thickness. The MEPDG should not be sensitive to this if the pavements are otherwise equivalent. However, a 2.8 in.-over-6.2 in. PCC/PCC pavement is predicted to have ~200% more damage in bottom-up cracking than is a 3 in.- over-6 in. PCC/PCC pavement. To address this damage issue, two points were considered: 1. The assumption of zero friction between PCC slab and the base layer is a reasonable one for bonded-PCC-over- JPCP, but for new construction of PCC/PCC slabs, such an assumption, is not reasonable. As a result, the model for transverse cracking, as applied to bonded-PCC-over- JPCP, was modified to account for the presence of full fric- tion at the base over the initial period of the pavement life, as it is for new JPCP projects. 2. The EICM temperature analysis was analyzed for the com- posite slab. In previous versions of MEPDG, the raw EICM thermal data (at variably spaced nodes) was used to com- pute the temperature at equally spaced nodes through the slab. In subsequent versions of the MEPDG, the EICM raw outputs (temperature at variably spaced nodes) are used directly for the stress analysis. This creates a discrepancy between bonded-PCC-over-JPCP and structurally equiva- lent single-layer results. Because the design process for the single-layer JPCP involved extensive validation and cali- bration as part of NCHRP 1-37A and NCHRP 1-40B, it is Figure 3.34. Curling of PCC slab caused by night-time negative temperature difference, plus critical traffic loading position. This resulted in high tensile stress at slab bottom. Figure 3.35. Curling of PCC slab caused by nighttime negative temperature difference plus critical traffic loading position. This resulted in high tensile stress at slab top.

65 reasonable to validate two-layer predictions against those of a structurally equivalent single-layer JPCP as predicted by MEPDG. To make this possible, modification to the EICM analysis in the two-layer PCC case were made, as detailed in the following sections, so that it would be more compatible with the single-layer design. Aside from changes to the MEPDG to allow for thinner upper lifts and full-friction between the slab and base, no modifications of substance were made to the MEPDG JPCP transverse cracking models. Provided that the appropri- ate revisions were made to the EICM, these models were deemed sufficient for predicting bottom-up and top-down cracking in PCC/PCC pavements using the Bonded-PCC- over-JPCP project. Longitudinal Cracking Models for pCC/pCC Longitudinal cracking caused by repeated loadings and large slab upward curling in JPCP is a phenomenon that has received little attention in pavement engineering but has occurred on several projects, especially with widened slabs on Specific Pavement Studies-2 (SPS-2) sites. The SHRP 2 R21 tour of PCC/PCC pavements identified no longitudinal cracking in the Netherlands, Austria, or Germany. The MEPDG does not consider longitudinal cracking; however, this is a potential topic for future research and development. MEPDG JpCp Faulting Model for pCC/pCC Repeated heavy axle loads crossing transverse joints cre- ate the potential for joint faulting. Faulting can become severe and cause loss of ride quality and require premature rehabilitation if any of the following conditions occurs: repeated heavy axle loads; poor joint load transfer effi- ciency (LTE); presence of an erodible base, subbase, or subgrade beneath the joint; and presence of free moisture under the joint. The evaluation of the MEPDG JPCP faulting model deter- mined that the model was applicable to PCC/PCC composite pavements. It was found that the program implemented in the MEPDG framework for the bonded-PCC-over-JPCP case assumes the modulus of the PCC overlay for the composite slab. In cases in which the elastic moduli of the two layers do not vary greatly, the error this creates in the faulting output is hardly detectable because the differences in faulting in the two cases are less than the precision of reported results. However, when the two layers have substantially different elastic moduli, as illustrated in the sensitivity analyses in Chapter 4, the error in implementation is higher. In addition to the elastic modulus, other material prop- erties of one layer in the system are being assumed for the composite slab in the MEPDG, rather than computing an effective property for an equivalent single-layered slab. In the case of the elastic modulus, in MEPDG v. 1.3000:R21 the program was modified so that the system uses an equiv- alent modulus generated from the moduli of the two layers, rather than simply forcing the upper-lift elastic modulus as the modulus of the overall system. An approach that accounted for the inequality of the CTEs by modifying the equivalent temperature gradients was developed. This approach is described in the MEPDG EICM modifications section below. Aside from the EICM modifications and their impact on faulting predictions, no other modifica- tions of substance were made to the MEPDG JPCP Faulting Model, and this model was deemed sufficient for predicting faulting in PCC/PCC pavements using the Bonded-PCC- over-JPCP project. Table 3.15. Comparison of Predicted Performance in a 9-in. JPCP and Structurally Equivalent Bonded PCC Over JPCP Systems Using MEPDG Version 1.003 Top-Layer Thickness (in.) Bottom- Layer Thickness (in.) IRI (in./mi) Percent Slabs Cracked Faulting (in.) Bottom-up Top-down Damage Percent Differencea Damage Percent Differencea 3.1 5.9 68.8 0.8 0.006 0.0252 13.0 0.0808 -4.5 3.0 6.0 68.9 0.8 0.006 0.0223 na 0.0846 na 2.9 6.1 69.3 0.8 0.006 0.0434 94.6 0.0750 -11.3 2.8 6.2 69.4 0.9 0.006 0.0681 205.4 0.0624 -26.2 2.5 6.5 69.4 0.8 0.006 0.0488 118.8 0.0715 -15.5 2.0 7.0 69.4 0.7 0.006 0.0404 81.2 0.0722 -14.7 1.5 7.5 69.1 0.6 0.005 0.0136 -39.0 0.0740 -12.5 a Compared with 3-in. over 6-in. PCC/PCC pavement.

66 MEPDG enhanced Integrated Climatic Model (eICM) for pCC/pCC In the original EICM thermal analysis (version 1.003 and ver- sions before 1.014:9030A), 10 nodes were distributed through the PCC slab with an additional node at the bottom of the base layer, resulting in a total of 11 nodes used to represent the tem- perature through the PCC slab and base with respect to a refer- ence temperature (NCHRP 2004; Larson and Dempsey 1997; Lytton et al. 1989). This distribution of nodes was then used to calculate the nonlinear stresses at the top and bottom of the slabs for damage calculations. During subsequent developments, to provide consistent results between PCC/PCC composite pave- ments and an equivalent JPCP, the EICM thermal analysis was revised in versions 1.014:9030A and 1.206:R21 for SHRP 2 R21. Rather than the 10 nodes being applied to the entire compos- ite slab (approximated by the bonded PCC overlay project), each PCC layer was assigned 10 nodes, which resulted in the use of a minimum of 20 temperature nodes for the entire slab and base. These additional nodes presented two key challenges. The first was that their inclusion dramatically increased the run time for the damage calculation in PCC/PCC composite pavement. The second, and more important, was that the sys- tem with additional nodes threatened the self-consistency of the MEPDG. As noted, for a single-layer PCC pavement, EICM uses only 10 nodes. For a new PCC/PCC pavement, the thermal gradient was approximated by EICM using 20 or more nodes through the composite slab. This modeling difference rippled through the project runs and provided results for structurally equivalent systems that were significantly different, although they should have been nearly identical. To address this inconsistency, the thermal gradient for a bonded PCC overlay was modified to use PCC layer thicknesses and the base layer thickness to develop 11 equally spaced nodes through the composite slab and a twelfth node at the bottom of the base layer, thereby creating 10 intervals in the composite slab and one for the base layer. The thermal node arrangement used in MEPDG versions 1.014:9030A and 1.206:R21 is described in Figure 3.36a. The modified thermal node arrangement (used in (a) (b) (c) Figure 3.36. Modified thermal nodes through slab thickness in MEPDG for (a) MEPDG version 1.014:9030A, (b) MEPDG v. 1.3000:R21, and (c) both approximations relative to nonlinear thermal gradient.

67 MEPDG v. 1.3000:R21) is described in Figure 3.36b. The recom- mended modification implemented in MEPDG v. 1.3000:R21 ensured that PCC/PCC projects and their structurally equiva- lent single-layer JPCP projects have the same number of inter- vals through the PCC layers, whether the project is single-lift or two-lift. eICM Calculation of Subgrade response for Single-Layer and Composite Two-Layer rigid pavement Systems In adapting the MEPDG and EICM to model newly constructed PCC/PCC composite pavements as bonded PCC overlays of existing pavements, a major consideration was the effect of this modeling choice on the calculation of the subgrade response, or k-value. For a typical single-layer PCC pavement, the effective dynamic k-value is obtained by first determining the deflection profile of the PCC surface using an elastic layer pro- gram, modeling all layers specified for the design (Figure 3.37). The subgrade resilient modulus is adjusted to reflect the lower deviator stresses that typically exist under a concrete slab and base course compared with the deviator stress used in labora- tory resilient modulus testing. Next, the computed deflection profile is used to back calculate the effective dynamic k-value. Thus, the effective dynamic k-value is a computed value, not a direct input to the MEPDG design procedure (except in rehabilitation). The effective k-value used in the MEPDG is a dynamic k-value, as opposed to traditional static k-values used in pre- vious design procedures. The effective dynamic k-value of the subgrade is calculated for each month of the year and used directly to compute critical stresses and deflections in the incremental damage accumulation over the design life of the pavement. Factors such as water table depth, depth to bedrock, and frost penetration depth (frozen material) can significantly affect effective dynamic k-value. All of these fac- tors are considered in the EICM. However, this procedure is different for bonded PCC overlay projects. For a bonded PCC overlay, only the existing PCC layer is used to determine the deflection profile of the PCC using an elastic layer program. Thus, the stiffness con- tribution of the overlay is discounted. Figure 3.38 shows the monthly difference in k-values between structurally identi- cal bonded PCC overlay of JPCP and new JPCP using a pre- vious version of MEPDG. The bonded PCC overlay of JPCP (the proxy for PCC/PCC composite pavement) is 3 in. over 6 in, whereas the JPCP is 9 in., with all material properties for all PCC layers being identical. As described above, this difference arises because the k-value is backcalculated using the elastic layer deflection profile and 9 in. PCC for the JPCP but only 6 in. PCC for bonded PCC overlay of JPCP and consequently for PCC/JPCP. For MEPDG (v. 1.3000:R21) the bonded PCC overlay of PCC pavement was modified to include the overlay, or in terms of composite PCC/PCC, the subgrade response calculation includes both lifts of the two-layer PCC slab. Figure 3.39 illustrates that modifications to calculation of the subgrade k-value in the bonded PCC overlay project have reduced the extreme differences in the k-value for the structur- ally equivalent systems. However, it may be valuable for future research to note that small differences (approximately 3%) still remain in the monthly calculation of the subgrade reaction, suggesting that additional modifications to the EICM calcula- tion will be necessary to make the two designs identical. Comparison of PCC/PCC and Structurally Equivalent Single-Layer JPCP An important trial in the evaluation of MEPDG (v. 1.3000: R21) was the verification of the performance predictions for PCC/PCC (or bonded PCC over JPCP in the MEPDG) as Concrete Slab (JPCP, CRCP) Bedrock Base Course (Unbound, Asphalt, Cement) Subbase Course (Unbound, Stabilized) Compacted Subgrade Natural Subgrade Ebase E c Concrete Slab (JPCP, CRCP) Base Course (Unbound, Asphalt, Cement) Effective k-value obtained through backcalculation Figure 3.37. Structural model for rigid pavement structural response computations.

68 Note: OL = overlay. Figure 3.38. Subgrade k-value calculation for a PCC/PCC pavement and its JPCP single-layer structural analog (versions 1.003 and 1.014:9030A of MEPDG). Note: OL = overlay. Figure 3.39. Subgrade k-value calculation for a PCC/PCC pavement and its JPCP single-layer structural analog (MEPDG v. 1.3000:R21).

69 compared with that of a structurally equivalent single layer JPCP. Table 3.16 shows the results of comparing MEPDG performance prediction of two structurally equivalent 9-in. PCC systems (the baseline PCC/JPC and a single-layer JPCP). Note that all pavement layer properties for each case are exactly identical. It was expected that the performance of the systems would be very similar, if not identical. This trial yields similar performance across all perfor- mance measures. It is worth noting that it took three revi- sions of the MEPDG as part of the SHRP 2 R21 project to ensure that structurally similar pavements perform similarly. Influence of PCC Layer Thickness in the Performance of Structurally Equivalent PCC/PCC Tests were also run to determine if the predicted cracking damage calculated using MEPDG (v. 1.3000:R21) in structur- ally equivalent PCC/PCC pavements are similar. These test runs involved structurally equivalent PCC/PCC pavements that differed only with respect to the relative thickness of their PCC layers. The combined thickness for both layers was always 9 in., and all properties for both layers were identical. The results are shown in Table 3.17. The results show that similar performance is achieved for all selected PCC layer thicknesses, as expected, and point to the fact that the MEPDG (v. 1.3000: R21) is functioning properly with respect to this issue. Note that the MEPDG will not allow a top-lift thickness of less than 1 in.; selecting 1 in. or less for the upper PCC will result in instabilities in the program. Additional layer thicknesses were evaluated for a different base PCC/PCC case. Again, these test runs involved structur- ally equivalent PCC/PCC pavements that differed only with respect to the relative thickness of their PCC layers. The com- bined thickness for both layers was always 9 in., and all prop- erties for both layers were identical. The results are shown in Table 3.18, which illustrates a pavement that fails in faulting (according to specified performance measures) consistently for all cases. One case is a JPCP project, and the rest are its PCC/ PCC structural equivalents. Pass or fail, similar performance is achieved for all selected PCC layer thicknesses and project types for structurally equivalent PCC pavements. Again, the Table 3.16. Comparison of 9-in. Total Thickness PCC/PCC and Structural Equivalent 9-in. JPCP Using MEPDG and Identical Properties for All PCC Project International Roughness Index (IRI), in./mi (Limit: 172) Percentage Slabs Cracked (Limit: 15) Mean Joint Faulting, in. (Limit: 0.12) Bottom-up Cracking Damage Top-down Cracking Damage PCC/PCC 68.6 0.4 0.006 0.0427 0.0439 JPCP 67.1 0.4 0.006 0.0444 0.0430 Note: MEPDG (v. 1.3000:R21). Table 3.17. Comparison of Structurally Equivalent 9-in. PCC/PCC Pavements Using MEPDG (v. 1.3000:R21) h1 (in.) h2 (in.) IRI, in./mi (Limit: 172) Percentage Slabs Cracked (Limit: 15) Mean Joint Faulting, in. (Limit: 0.12) Bottom-up Cracking Damage Top-down Cracking Damage 3.1 5.9 68.5 0.4 0.006 0.0427 0.0439 3.0 6.0 68.6 0.4 0.006 0.0427 0.0439 2.9 6.1 68.6 0.4 0.006 0.0427 0.0439 2.8 6.2 68.7 0.4 0.006 0.0427 0.0438 2.5 6.5 68.8 0.4 0.006 0.0426 0.0438 2.0 7.0 69.1 0.4 0.006 0.0427 0.0438 1.5 7.5 69.4 0.4 0.006 0.0426 0.0438 Note: h1 = thickness of the upper layer; h2 = thickness of the lower layer.

70 Table 3.18. Comparison of Additional Structurally Equivalent 9-in. PCC/PCC Pavements Using MEPDG h1 (in.) h2 (in.) IRI, in./mi (Limit: 172) Percentage Slabs Cracked (Limit: 15) Mean Joint Faulting, in. (Limit: 0.12) Bottom-up Cracking Damage Top-down Cracking Damage NA 9.0 147.5 11.1 0.136 0.3494 0 1.5 7.5 150.4 11.0 0.130 0.3475 0 2.0 7.0 150.0 11.0 0.131 0.3475 0 2.5 6.5 149.4 11.0 0.131 0.3475 0 3.0 6.0 149.0 11.0 0.131 0.3477 0 3.5 5.5 148.7 11.0 0.132 0.3474 0 4.0 5.0 148.2 11.0 0.132 0.3474 0 4.5 4.5 148.0 11.0 0.132 0.3477 0 5.0 4.0 147.6 11.0 0.132 0.3477 0 5.5 3.5 147.4 11.0 0.133 0.3478 0 6.0 3.0 147.1 11.0 0.133 0.3478 0 6.5 2.5 146.9 11.0 0.133 0.3474 0 Note: MEPDG (v. 1.3000:R21); NA = not available; h1 = thickness of the upper layer; h2 = thickness of the lower layer. results point to the fact that the modifications to the MEPDG in v. 1.3000:R21 are robust. Comparing these results with those obtained with previ- ous versions of the MEPDG (an example of which is dis- cussed earlier in this chapter and shown in Table 3.15), clearly indicates that the modifications to the MEPDG as part of SHRP 2 R21 project were necessary and valid and have resulted in a version of the MEPDG that is more appropri- ate for designing PCC/PCC composite pavements. Overall MEPDG performance Modeling Table 3.19 presents MEPDG performance predictions for the R21 PCC/PCC database sections. These results were generated using the final version of the MEPDG modified as part of the SHRP 2 R21 project (MEPDG v. 1.3000:R21). An additional measure of model capabilities is to examine the damage in top-down cracking and bottom-up cracking for spe- cific sections relative to their observed field measured cracking. Figures 3.40 and 3.41 illustrate the ability of the revised MEPDG to account for the damage of a pavement relative to its percent- age of cracked slabs. All PCC/PCC sections were used to make this important comparison. The s-shaped curves represent the national calibration curves obtained for JPCP one-layer slabs relating accumulated damage to transverse fatigue cracking of JPCP. The composite PCC/PCC pavements show good corre- spondence to these curves. Thus, the national models used in JPCP design also can be used for PCC/JPCP design of compos- ite pavement. The same results were obtained for HMA/PCC composite pavements as described in Volume 1. Lattice Modeling of pCC/pCC Interface Behavior (Debonding) Debonding between the PCC layers may lead to premature failure of the pavement. However, debonding is not currently modeled by the MEPDG. Granju (2001) illustrates the process of debonding: crack initiation as a result of volume changes, thermal loads, and traffic loads, debonding occurring in the region of the crack, peeling of the top-lift (overlay) from the lower lift over time. Lattice models for composite beam and composite slab simulations was determined by the SHRP 2 R21 research team to be the most effective manner of determining if debonding was a legitimate concern for PCC/PCC. A lattice model consists of a triangular grid of points con- nected by one-dimensional spring elements. This network of springs represents the discretized medium. The lattice can be deformed by internal strains resulting from diffusive, thermal, or hygral processes or by external displacements or forces. Lat- tice models can differ in the representations of the constitutive relations used for individual springs (or, instead, the minimi- zation of the stored elastic energy) (Schlangen and van Mier 1992). The varying properties of springs allow the lattice to sim- ulate the behavior of heterogeneous media, such as concrete.

71 Table 3.19. Final Performance Predictions for R21 PCC/PCC Database Sections Using R21 Revised MEPDG Section, Location Climate Traffic PCC/PCC, Year Constructed, Joint Space, Dowels IRI (in./mi) Transverse Cracking (% Cracked Slabs) Faulting (in.) Comment FL45 (3A), Fort Myers, Florida Wet, Non- freeze 5 million trucks, 30 years 3 in. PCC, 9 in. Econocrete, 1978, 20 ft, 1 in. Predicted: 96.9 Measured: 104 Predicted: 0.0 Measured: 0 Predicted: 0.081 Measured: 0.04 Econocrete had fc = 2,000 psi FL45 (3B), Fort Myers, Florida Wet, Non- freeze 5 million trucks, 30 years 3 in. PCC, 9 in. Econocrete, 1978, 20 ft, 1 in. Predicted: 101.0 Measured: 112 Predicted: 6.8 Measured: 3 Predicted: 0.077 Measured: 0.07 Econocrete had fc = 1,250 psi FL45 (2A), Fort Myers, Florida Wet, Non- freeze 5 million trucks, 30 years 3 in. PCC, 9 in. Econo- crete, 1978, 15 ft, none Predicted: 71.1 Measured: 119 Predicted: 0 Measured: 0 Predicted: 0.011 Measured: 0.13 Econocrete had fc = 2,000 psi FL45 (2B), Fort Myers, Florida Wet, Non- freeze 5 million trucks, 30 years 3 in. PCC, 9 in. Econo- crete, 1978, 15 ft, none Predicted: 74.1 Measured: 113 Predicted: 0 Measured: 0 Predicted: 0.017 Measured: 0.09 Econocrete had fc = 1,250 psi FL45 (2C), Fort Myers, Florida Wet, Non- freeze 5 million trucks, 30 years 3 in. PCC, 9 in. Econocrete, 1978, 15 ft, none Predicted: 71.5 Measured: 166 Predicted: 0.6 Measured: 5 Predicted: 0.011 Measured: 0.18 Econocrete had fc = 750 psi K96, Haven, Kansas Wet, freeze 2.1 million trucks, 14 years 3 in. PCC, 7 in. JPCP, 1997, 15 ft, 1 in. Predicted: 92 Measured: NA Predicted: 0 Measured: 0 Predicted: 0.043 Measured: 0.02 Predicted estimates measured well K96, Haven, Kansas Wet, freeze 2.1 million trucks, 14 years 3 in. PCC, 7 in. JPCP, 1997, 15 ft, 1 in. Predicted: 90.4 Measured: NA Predicted: 1 Measured: 0 Predicted: 0.041 Measured: 0.02 Predicted estimates measured well I-10, Kansas Wet, freeze 3.0 million trucks, 4 years 1.5 in. PCC, 11.8 in. JPCP, 2007, 15 ft, 1.25 in. Predicted: 73.9 Measured: NA Predicted: 0 Measured: 0 Predicted: 0.010 Measured: 0.03 Predicted estimates measured well I-75, Detroit, Michigan Wet, freeze 71.7 million trucks, 18 years 2.5 in. PCC, 7.5 in. JPCP, 1993, 15 ft, 1.25 in. Predicted: 146.5 Measured: NA Predicted: 0 Measured: 0 Predicted: 0.129 Measured 0.059 Predicted estimates measured well I-94, MnROAD Wet, freeze 0.7 million trucks, 1 year 3 in. PCC, 6 in. JPCP, 2010, 15 ft, 1.25 in. Predicted: 65.3 Measured: NA Predicted: 0 Measured: 0 Predicted: 0.001 Measured: NA Predicted estimates measured well A1, Austria Wet, freeze 47.2 million trucks, 14 years 2 in. PCC, 7.9 in. JPCP, 1994, 18 ft, 1 in. Predicted: 172.6 Measured: NA Predicted: 4.4 Measured: 0 Predicted: 0.226 Measured: 0.10 Overpredicted faulting resulting from MEPDG model not accounting for EU base/subgrade prep to reduce faulting A1, Austria Wet, freeze 26.2 million trucks, 15 years 1.6 in. PCC, 8.3 in. JPCP, 1993, 18 ft, 1 in. Predicted: 144.5 Measured: NA Predicted: 0.2 Measured: 0 Predicted: 0.167 Measured: 0.10 Overpredicted faulting resulting from MEPDG model not accounting for EU base/subgrade prep to reduce faulting A93, Germany Wet, freeze 52.6 million trucks, 13 years 2.8 in. PCC, 7.5 in. JPCP, 1995, 16 ft, 1 in. Predicted: 126.4 Measured: NA Predicted: 0 Measured: 0 Predicted: 0.131 Measured: 0.10 Overpredicted faulting resulting from MEPDG model not accounting for EU base/subgrade prep to reduce faulting N279, The Netherlands Wet, freeze 11.9 million trucks, 8 years 3.5 in. PCC, 7 in. JPCP, 2000, 18 ft, 1 in. Predicted: 119.8 Measured: NA Predicted: 0.1 Measured: 0 Predicted: 0.122 Measured: 0.08 Overpredicted faulting resulting from MEPDG model not accounting for EU base/subgrade prep to reduce faulting Note: MEPDG (v. 1.3000:R21); NA = not available; fc = 28-day compressive strength.

72 Figure 3.40. PCC/PCC measured cracking versus MEPDG (v. 1.3000:R21) predicted bottom-up damage. Figure 3.41. Measured cracking versus MEPDG (v. 1.3000:R21) predicted top-down damage. The characterization of fracture, the essence of the debond- ing problem, often is simulated in pavement engineering using finite element methods (FEM) that begin with contin- uum equations. Although FEM can be successfully applied to a fracture problem, the success of the application depends largely on the homogeneity of the medium and the lack of disorder in crack propagation. Modeling of crack propagation in FEM requires the use of special elements in the crack path that must be specified a priori. Thus, if the medium is relatively homoge- neous and the crack path can be anticipated, the FEM is more than adequate for simulating cracking. In cases exhibiting heterogeneity or nontrivial crack paths, the placement of the

73 specialized elements for fracture becomes a nontrivial prob- lem. This problem is commonly solved using trial-and-error methods, with computationally expensive and cumbersome remeshing during the fracture process. In light of these challenges, lattice models are a viable alter- native to or a candidate to be coupled with FEM. In the lattice model, the simulation of cohesive cracking involves a reduc- tion of stiffness and strength and the removal of individual springs. The lattice model does not require a priori knowledge of the crack path. The application of a specific lattice model (that is, a random network of spring elements) to the PCC/ PCC debonding problem is prefaced with a summary of the model formulation and its simulation of fracture. The formu- lation of a lattice network of spring elements and the fracture rules assigned to the body are detailed below. Basic Model Formulation The lattice model applied to PCC/PCC composite pavements is a rigid-body-spring network (Bolander and Saito, 1998). Begin- ning with a region containing randomly distributed points, a Delaunay tessellation is used to define a network through the connection of these random points. The model then uses Voronoi diagrams to establish facets surrounding random points (to create nodes) and define nodal/facet volumes for later stress calculations. For each pair of neighboring nodes, we define an element ij connecting these nodes (Figure 3.42). Element ij can be more easily characterized by the shared facet between the two nodes. Each facet is associated with a total of 6 spring constants corresponding to displacements in the x, y, z directions and rotations about each of these axes (see Equations 3.5 and 3.6). k k k E A h x y z ij ij = = = ( . )3 5 k E J h k E I h k E I h x p ij y ij z I ij ϕ ϕ ϕ= = =, , ( . ) 22 1 3 6 The basis of the elastic equations for the lattice model is then, for representative element ij, F = Kd where F is a load vector, K is a 12 × 12 stiffness matrix, and d is a vector of displacements. Taken for all elements in the lattice, this model represents a linear elastic system of equations. (Other factors not included in this formulation are those accounting for thermal and hygral diffusive processes, which can be and have been incorporated into lattice models.) More detail on the formulation and solu- tion of these problems can be found in Schlangen and Garbocki (1996). This formulation can be applied to any arrangement of elements in the lattice. The model described here uses a random geometry network to define the desired domain. The geometry is based on a Voronoi diagram for a given number of randomly generated points within the region. A major advantage to the use of random geometry net- works is that these networks, as implemented in this model, are such that the model preserves elastic uniformity under loading. That is, the strain for each element in the ran- dom geometry network agrees with the global strain for the body under consideration (Schlangen and Garbocki 1996; Bolander and Sukumar 2005). Figure 3.43 illustrates the issue of elastic uniformity with histograms for elemen- tal strains in a random geometry network and elemental strains in a regular lattice network for a cube under uni- form tensile loading. The quantities Aij and hij describe the area and length of element ij, respectively. In the instance of a regular or ordinary lattice, elemental area Aij is described as uniform area A _ for all elements. Furthermore, it should be noted that although regular, symmetric networks also ensure elastic uniformity, regular networks create bias in crack propagation. Random geometry networks eliminate this bias. Source: Bolander 2008. Figure 3.42. Element ij within network (left); facet and springs associated with ij (right). Source: Bolander 2008. Figure 3.43. Random geometry networks are implemented in such a way that the model describes uniform elastic behavior under loading.

74 Mixed-Mode Fracture Criteria A key feature of this model, as alluded to previously, is that it does not adopt a continuum approach for fracture. Rather, its discrete representation allows for the development of param- eters governing fracture in the body, as suggested by Jagota and Bennison (1994). These parameters are rules based on element response to critical stresses that are applied to the body after the constitutive equation has been solved for a given load-step. In this model, crack initiation and propagation are gov- erned by the tensile and shearing stresses at the facet-defining element ij. These stresses are considered in terms of a Mohr- Coulomb fracture criterion, in which both the tensile and resultant shear stresses acting on element ij are held against critical strength values for the element to determine if break- ing occurs. For this model, after the solution of a given load- step, for each element the criterion R = r/rf is calculated, where r ij ij= +σ τ2 2 , sij is the normal stress for the facet, tij is the resultant shear stress for the facet, and rf is the distance to the fracture criterion curve in Figure 3.44. Other important parameters in the figure are the tensile strength ft, pure shear failure criterion tc = rft, approximate shear strength under critical tensile stress t* = r2ft, and the angle g = tan-1(r-r2) to further specify the slope of the fracture criterion curve, where r and r2 are specified material parameters through tc and t*. Where R > 1, an element is defined as having undergone a fracture event. For instance, where more than one element has a value of R in excess of 1, only the element with the largest value of R is considered. More detail on the Mohr-Coulomb criterion and rules governing fracture for this model can be found in Bolander and Saito (1998). Later figures will refer back to the mixed-mode fracture curve of Figure 3.44 to define fracture events in simulations. Furthermore, the frac- ture rules assigned by this model include the degradation of spring stiffnesses to simulate softening. The model described here uses a bilinear softening relationship. For every successive fracture event according to the rules above, both the modulus of elasticity E and the tensile strength ft are reduced according to a bilinear softening relation such as that of Figure 3.44. Thus, there is a recursive effect between the softening rela- tion and the mixed-mode fracture criterion itself. As fracture events are characterized for an element ij according to the Mohr-Coulomb criterion, both E and ft for the element are consequently reduced. The softening in ft then causes the s and t intercepts of the Mohr-Coulomb criterion, as repre- sented in the curve in Figure 3.44 by ft and tc, respectively, to become reduced as well. When extensive fracturing occurs, the mixed-mode fracture criterion will “collapse” gradually toward the origin as a result of the bilinear softening. This effect is summarized in Figure 3.45. Later figures will depict simulations that involve mixed- mode fracture and characterizing these events, and the Figure 3.44. (Left) Bilinear softening to describe degrada tion of stiffnesses of elements under breaking. (Right) Mohr- Coulomb fracture criterion defines breaking in the described lattice model. Figure 3.45. This figure illustrates the so-called collapse of mixed-mode fracture criterion curve toward origin in the event of extensive fracture for a given element ij.

75 beam. Furthermore, each simulated beam was developed to accommodate nodes expressly to act as sites of forcing to accommodate three-point loading. The next cases allow a closer examination of mixed-mode failure at a given interface. That is accomplished by loading a beam at midspan and gradually moving the interface away from the load toward a support. It was hypothesized that, as the interface was moved away from midspan, the nature of fracture would shift from one that is tensile to one that is mixed. In the interface strength problems, a controlled displacement of 0.01 mm is applied until 0.1 mm is reached, and a controlled displacement of 0.001 mm is applied thereafter. The controlled displacement is applied at midspan (L/2 = 500 mm) until simulated failure occurs. The problems use six distinct random geometry lattices for a domain of 80 × 1,100 × 250 mm, where supports are placed 50 mm from the end to yield an effective span of L = 1,000 mm. The six lattices differ in the location of the interface, which is near the load at l1 = 490 mm, near the support at l6 = 25 mm, and at four locations in between the load at midspan and the support (l2 = 330 mm, l3 = 190 mm, l4 = 110 mm, l5 = 50 mm). Beam properties are indicated in Table 3.20. Layers PCC1 and PCC2 were chosen to be equivalent to isolate the effects of the interface. The value of ft = 0.2 MPa was selected to provide a suf- ficiently weak interface. For the six cases described here, specimen failures in the two extremes of this problem are illustrated in Figure 3.47. One would expect that the location of the weak interface has an influence on the load capacity and failure of the beam. Figure 3.48 illustrates the reduction in the ultimate load as the interface is moved from the support toward midspan. Note again that span length L = 1,000 mm. The hypothesis being tested by the six cases is the expec- tation that as the interface was moved within the span, the nature of the initial fracture would shift from being largely tensile under midspan (as is commonly observed in three- point bending tests) to a mixed-mode. Thus, the analysis visibility of reduced criterion curves, such as is seen in Figure 3.45, will be visible in the simulations. Once the strength has been reduced to zero, the spring is removed from the lattice, and its contributions to the global stiffness matrix are conse- quently negated. Beam Simulations to Understand Fracture and Interface Failure To better develop models for debonding in composite pave- ments, laboratory tests for bond interface integrity were reviewed. Of the many available tests, the test developed at the Road Laboratory of Barcelona (Laboratorio de Caminos de Barcelona) was chosen because of its simple and direct exper- iment to induce the loss of bond integrity in a layered asphalt field core or laboratory sample. In the LCB procedure, cylin- drical asphalt cores are clamped into the testing apparatus and subjected to loading under modified three-point bend- ing. The intention is to generate shear stresses at the interface and avoid a bending moment by placing the interface very near the support. This procedure measures resistance to tan- gential stress at the interface and the displacement of the lay- ers with respect to one another (Recasens et al. 2006). Although the procedure is designed for asphalt composite cores or samples only, the LCB test also could be used for con- crete composites. The specimens would be cast using wet-on- wet techniques to mirror the techniques used in the field. The LCB test was adopted for this research as the physical ana- logue for the debonding simulations. Figure 3.46 illustrates the general composition of the beams and an example of a two-layered beam with random lattice geometry and paired- node interface for simulation. Multiple random geometry networks were generated to locate the interface at different locations measured from the roller support, as illustrated by li and corresponding hashed lines in Figure 3.46. The random point generators used to develop all specimens were oriented such that more points were distributed in the region surrounding the Figure 3.46. General composition of three-point beam (left). Composite beam with paired-node interface near support (right).

76 Beginning with the case l2 = 330 mm, we observe that the nature of the fracture is mostly tensile; however, some events feature a shear component as high as ~0.1 MPa. As we continue clockwise, the shear component becomes more pronounced and reaches values as high as ~0.3 MPa at li = 950 mm. The final case is the placement of the interface at l6 = 25 mm. The first 1,000 fracture events for the final case are shown in Figure 3.51. Figure 3.51 illustrates the observation that near the support, the shear component of fracture events is larger than the tensile events; shear stresses are nearly 0.4 MPa in this case. It is not until later events (only a few of which occur within the first 1,000) that normal (tensile) stresses exceed those of shear. Fur- thermore, the normal stresses for the first of the fracture events are compressive in nature. This phenomenon is most likely attributable to the proximity of the support, which acts to both compress and shear elements in the vicinity to instigate fracture. The six cases support the concept of mixed-mode fracture. Although the idea is not novel given that the model determines the criterion, the simulations satisfy expectations of fracture behavior in particular situations: in Mode I (opening) fracture situations, the tensile strength contributes more to the determi- nation of failure, whereas in predominately Mode II (in-plane Table 3.20. Beam Properties for Multiple Interface Locations E (MPa) ft (MPa) tc (MPa) t* (MPa) NFE COBP (mm) TFCO (mm) PCC1 32000 4.0 6.0 2.80 100 0.002 0.008 PCC2 32000 4.0 6.0 2.80 100 0.002 0.008 Interface 32000 0.2 0.3 0.14 100 0.002 0.008 Note: t* = approximate shear strength under critical tensile stress; NFE = maximum allowable number of fracture events; COBP = crack opening at the break point; TFCO = traction-freecrack opening. begins with l1 = 490 mm, when the interface is almost directly beneath the controlled displacement at midspan. In this instance, we would expect the failure to be classical Mode I fracture. The first 1,000 fracture events are plotted against the fracture criterion curve from above (Figure 3.49). Fracture events in Figure 3.49 are represented by a single black dot. The aggregate of these dots into one large mass illustrates that the initial fracture events are purely tensile in nature and in the neighborhood of 0.25 MPa, which exceeds the interface critical strength of ft = 0.2 MPa. In this regard, the model simulation confirms the initial hypoth- esis. Only the initial (for these purposes, the first 1,000) events are plotted to make plots more legible and to further distinguish the separate cases, whose most striking differ- ences in fracture are in initiation. The interface is moved gradually toward the roller support, the beam loaded, and the nature of fracture recorded. The first 1,000 fracture events in each of the following four cases (where l2 = 330 mm, l3 = 190 mm, l4 = 110 mm, and l5 = 50 mm) are presented together in Figure 3.50. Figure 3.48. Reduction of ultimate load as weak interface is moved toward midspan. Figure 3.47. Failure in composite beams with weakened interface near midspan and near support under loading.

77 shearing) fracture situations, the shear strength weighs more heavily in predicting failure. These capabilities in the combined lattice–FEM model for fracture in composite slabs give the ability to better simulate fracture behavior at the interface. Slab Simulations to Characterize Debonding under Thermal Gradients If the two lifts of a PCC/PCC pavement are constructed within a reasonable time frame (less than 2 hours), there will be no debonding at the PCC/PCC interface. This conclusion is sup- ported by field observations of PCC/PCC composite pavements, from discussions with R21 project consultants in Europe, by ultrasound measurements of PCC/PCC at the MnROAD dem- onstration slab and mainline sections, and the pull-off tests conducted by the FHWA Mobile Concrete Laboratory. However, we briefly consider the possibility that there is a possibility of unexpected delays in the placement of the upper PCC layer that is significantly greater than 2 hours. Figure 3.49. Failure in tension predominates at a weakened interface near midspan (l1 5 490 mm). Figure 3.50. Evolution of mixed-mode fracture as interface is moved from midspan toward support (beginning top left with l2 5 330 mm, moving clockwise, and finishing bottom left at l5 5 50 mm).

78 constant temperature equivalent to the temperature at the bot- tom of the upper PCC layer. For this example, a half-slab with dimension 90 × 72 in. with finite element size 6 in. × 6 in. was used. The mesh for the plate is depicted in Figure 3.52. ISLAB2005 input parameters are indicated below. Layer properties are adapted from those of the MnROAD PCC/ PCC test section. The coefficient of thermal expansion (CTE) of the upper layer is exaggerated to develop a “worst case sce- nario,” wherein the thermal properties of the two layers are vastly different. Note that this was not the case at MnROAD, as described in Chapter 2, where the measured CTE results for the two layers were nearly identical. Lattice properties for the two concretes in the composite slab are shown in Table 3.21, properties of the assumed interface are listed in Table 3.22, and the undeformed two-lift lattice is shown in Figure 3.53. a. ISLAB2005, Geometry • Custom mesh. • x-direction. 1. Length: 72 in. 2. Number of nodes: 13 (equally spaced). • y-direction. 1. Length: 90 in. 2. Number of nodes: 16 (equally spaced). b. ISLAB2005, Layers • Layer 1. 1. Element type: Plate. 2. Thickness: 3 in. 3. Poisson’s ratio: 0.15. 4. Elastic modulus: 4,266,000 psi. 5. Coefficient of thermal expansion: 7 × 10-6/°F. If significant delays occur, full bond at the interface may be compromised. This may cause debonding resulting from dif- ferential (thermal and hygral) shrinkage strains in the upper and lower PCC layers. A coupled Lattice-ISLAB2005 model was used to simulate this situation. The following simulation considers a lattice embedded in a composite slab undergoing differential shrinkage strains in the top PCC layer. This uses a bilinear temperature gradient, for which the upper layer of the composite slab experiences a nega- tive temperature gradient while the lower layer experiences a Figure 3.51. Shear events characterize mixed-mode failure at a weakened interface near support (l6 5 25 mm). Figure 3.52. Finite element mesh for composite slab with location of lattice region shaded.

79 Table 3.21. Slab Properties for Two Lifts of PCC in Differential Strain Example E (psi) ft (psi) tc (psi) t* (psi) NFE CTE PCC1 (upper) 4,266,000 725 1087.5 507.5 100 7E-06 PCC2 (lower) 3,827,000 627 940.5 438.9 100 5E-06 Note: t* = approximate shear strength under critical tensile stress; NFE = maximum allowable number of fracture events; CTE = coefficient of thermal expansion. Table 3.22. Slab Properties for the Weakened Interface of Two Lifts of PCC in Differential Strain Example Assumed Interface E (psi) ft (psi) tc (psi) t* (psi) NFE CTE 100% of PCC2 3,827,000 627.00 940.500 438.900 100 5.00E-06 60% of PCC2 2,296,200 376.20 564.300 263.340 60 5.00E-06 40% of PCC2 1,530,800 250.80 376.200 175.560 40 5.00E-06 20% of PCC2 765,400 125.40 188.100 87.780 20 5.00E-06 15% of PCC2 574,050 94.05 141.075 65.835 15 5.00E-06 10% of PCC2 382,700 62.70 94.050 43.890 10 5.00E-06 5% of PCC2 191,350 31.35 47.025 21.945 10 5.00E-06 Note: t* = approximate shear strength under critical tensile stress; NFE = maximum allowable number of fracture events; CTE = coefficient of thermal expansion. • Layer 2. 1. Element type: Plate. 2. Thickness: 6 in. 3. Poisson’s ratio: 0.15. 4. Elastic modulus: 3,827,000 psi. 5. Coefficient of thermal expansion: 5 × 10-6/°F. c. ISLAB2005, Subgrade • Subgrade k-value: 200. d. ISLAB2005, Temperature (Case 1, 220F in upper layer) • Layer 1. 1. Type: Nonlinear. 2. Reference: 65°F. 3. Node 1 (0 in.): 45°F. 4. Node 2 (1.5 in.): 55°F. 5. Node 3 (3 in.): 65°F. • Layer 2. 1. Type: Nonlinear. 2. Reference: 65°F. 3. Node 1 (0 in.): 65°F. 4. Node 2 (3 in.): 65°F. 5. Node 3 (6 in.): 65°F. e. ISLAB2005, Temperature (Case 2, 250F in upper layer) • Layer 1. 1. Type: Nonlinear. 2. Reference: 65°F. 3. Node 1 (0 in.): 15°F. 4. Node 2 (1.5 in.): 40°F. 5. Node 3 (3 in.): 65°F. • Layer 2. 1. Type: Nonlinear. 2. Reference: 65°F. 3. Node 1 (0 in.): 65°F. 4. Node 2 (3 in.): 65°F. 5. Node 3 (6 in.): 65°F. The simulations of the composite slab used seven differ- ent assumed interfaces for the two temperature differences through the upper PCC lift, as shown in Table 3.22. These simulations considered the two cases of differential thermal Figure 3.53. Undeformed lattice of the composite slab, located in the corner of the plate mesh.

80 strain described in ISLAB2005 inputs: a -50°F temperature difference through the top layer and a -20°F temperature dif- ference though the top layer. The -50°F is an extreme situ- ation and is clearly unrealistic for an upper layer only 3 in. thick and even -20°F is unrealistic for only the top layer. How- ever, these conditions were assumed to exaggerate the thermal differences between the two layers: to determine extremes for simulation and determine what interface properties are required for failure. For the assumption of a -20°F temperature difference through the upper PCC layer, debonding at the interface did not occur in the simulations until the interface was degraded to 10% of the lower layer PCC (Table 3.22). This corresponds to the interface having flexural strength 62.7 psi and shear strength of 44 psi. The fracture and debond- ing that occurred are depicted in Figures 3.54 and 3.55. Displacements are scaled higher to depict debonding (frac- ture) behavior in Figures 3.54 and 3.55, which complicates depicting the deformation of the surrounding slab. The curl of the entire slab is more easily viewed using ISLAB2005, depicted in Figure 3.56. Given that (1) measured shear bond strengths from MnROAD PCC/PCC laboratory specimens were well in excess of these levels, (2) CTE properties in the upper layer are over- estimated to exaggerate thermal strains, and (3) thermal gra- dients through the slab are not as demanding as the bilinear gradient assumed, debonding in PCC/PCC appears to be an unlikely event, even in extreme cases. For the assumption of a -50°F temperature difference through the upper lift, debonding at the interface did not occur in the simulated cases until the interface was degraded to 15% of the strength properties of the lower layer PCC (see Table 3.22). For the composite structure to bear this kind of extreme, unrealistic thermal load and differential strains in the PCC layers in these simulations once again suggests that debonding would require very special and unconventional circumstances that far exceed degraded interface strength properties, material differences in PCC layers, and extreme conditions. Conclusions from Lattice Modeling Overall, the lattice models support the conclusion reached in the field many decades ago by PCC/PCC practitio- ners and researchers in Europe: if the two lifts of a PCC/ 9” 90” 72” Figure 3.54. Debonded composite lattice in global slab view. Figure 3.55. Local view of debonding at a degraded interface (10% strength properties of lower layer PCC) under 220F temperature difference in upper PCC layer.

81 PCC pavement are constructed within a reasonable time frame there will be no debonding. Furthermore, the R21 2008 Survey of European Composite Pavements team was unable to locate field observations of PCC/PCC debond- ing with the assistance of project consultants in Europe. No PCC/PCC composite pavements surveyed in the United States exhibited any debonding either, even after 30 years for the Florida PCC/PCC sections. Finally, the pull-off tests conducted by FHWA Mobile Concrete Laboratory for the MnROAD sections confirmed that the bond strength is just as strong, if not stronger, than the strength of the lower PCC layer. Thus, debonding was determined to be a concern only in the case of PCC overlays of existing old PCC pavements, which is out of the scope of the SHRP 2 R21 project. Figure 3.56. Slab curl in the simulated MnROAD PCC/PCC composite slab under 220F temperature difference in upper PCC layer.