PERFORMANCE PREDICTION OF PAVEMENTS WITH COLD IN-PLACE RECYCLING ASPHALT MIXTURES

The pavement rehabilitation and reconstruction method with Cold In-Place Recycling (CIR) is an alternative that can effectively conserve materials and energy, preserve the environment and reduce the cost. An attempt was made to predict the performance, particularly low-temperature cracking resistance characteristics of CIR mixtures prepared with the mix design procedure developed at the University of Rhode Island (URI) for the Federal Highway Administration (FHWA). The mix design procedure was developed to reduce wide variations in CIR mixture production and to develop a nation-wide standard. This standard was applied to a Rhode Island (RI) reclaimed asphalt pavement (RAP) to produce CIR mixtures with CSS-1h asphalt emulsion as the additive. By adjusting the number of gyrations of the Superpave Gyratory Compactor (SGC) for compaction, the field density of 130 pcf was achieved in the laboratory. To secure a base line, hot mix asphalt (HMA) samples were produced first according to the Superpave volumetric mix design procedure with an air void content of 4.0%. These were tested and analyzed parallel to the CIR specimens to compare the performances. The specimens were tested using the Indirect Tensile (IDT) tester at temperatures of -20, -10 and 0°C (-4, 14, and 32°F, respectively) in accordance with the AASHTO T 322 procedure. The creep compliance and tensile strength values were used as input data for the Mechanistic Empirical Pavement Design Guide (MEPDG) analysis. This software predicts the performance of roadways with different pavement structures, traffic, environmental conditions, and material properties using several mathematical models for different types of distresses. The analysis results indicated that no thermal or low-temperature cracking is expected over the entire analysis period of 20 years for both HMA and CIR mixtures. It confirms with the field performance in Arizona. Thus, it appears that CIR is a sustainable rehabilitation technique, and it justifies further research on and investigation of load-related distresses such as rutting and fatigue cracking.


INTRODUCTION
Roadways are exposed to various loadings and stresses that reduce their serviceability like other infrastructures. Traffic as well as environmental stresses wear out pavement structures. Despite careful consideration in the design and construction process, distresses cannot be avoided in the pavement surfaces. The most efficient means to deal with these distresses and wearing appears to be rehabilitating the pavement at a point where its condition can be improved with a reasonably affordable amount of resources. This could be an adequate maintenance practice in order to avoid expensive reconstruction, and it needs to be planned over the expected pavement service period.
Whenever rehabilitation or reconstruction is required, it is necessary to rebuild portions of the pavement structure. Certain layers of the roadway are typically milled up to a determined depth, which can include the surface and even the base course. If the milled materials are replaced with virgin materials, it requires purchasing and transporting new material which consumes time, energy, and money. Furthermore, the old material becomes waste, harming the environment and incurring further costs tied to disposal.
A method of reducing these issues is in-place recycling. It allows the user to reuse materials that are already in the pavement. This process includes milling, screening and crushing of the broken pavement materials. Additives such as emulsified asphalt or fly ash are then incorporated. This mixture is put back in place and compacted. Finally, a protective overlay is placed above the recycled layer of asphalt concrete, which is typically Hot Mix Asphalt (HMA).
In general, major advantages of this recycling procedure are as follows:  Less trucking  Conservation of materials, energy, and time  Preservation of the environment  Cost reduction These advantages pose major incentives to promote and support in-place recycling and allow the roadway rebuilding procedure to be conducted in a sustainable way.
In-place recycling can be performed at different temperatures. Cold recycling typically uses materials at ambient temperatures, i.e., at around 25°C (77°F). The absence of the necessity to heat up the material provides major advantages.
Furthermore, pollution in the form of smoke, heat and noise is reduced. Less time is needed for cooling off, therefore allowing sooner openings for traffic. Thus, in-place recycling is an approach for green highways and streets.

COLD IN-PLACE RECYCLING (CIR)
While various pavement construction practices have been used for centuries and a considerable amount of empirical experiences has been accumulated, in-place recycling is a relatively new technique, and performance records are limited. The special feature of this technique is the use of reclaimed asphalt pavement (RAP), and its properties and behavior are not uniform. Assumptions used for virgin material may not necessarily be applicable for RAP.
Standardized regulations for procedures, testing, and quality control are still needed for a wide-spread application of this technique. Based on the URI Mix Design procedure, this Master thesis deals with the prediction of performance of pavements with CIR, particularly, low-temperature cracking resistance characteristics. In order to fairly evaluate the predictions, the pavement structures with CIR materials were compared to the ones with virgin materials, or HMA, with the same boundary conditions. Even with varying thicknesses of different layers in the pavement structure (which are explained in section 5.1), exposure to the same environmental conditions ensure a fair comparison.

RESEARCH PLAN
To evaluate the performance of pavement for CIR materials, the first step was to get accustomed to this construction method and to investigate the current status of knowledge.
A parallel HMA base line had to be established to which the results of the CIR materials could be compared. The results, obtained similarly to those with CIR material, represent the performance of materials which are being used presently for a variety of road construction or rehabilitation projects. The results for the CIR mixture give expectations in a situation where CIR materials can be used instead of the HMA using raw materials.
Cylindrical specimens with almost equal dimensions were prepared for HMA and CIR materials. These are then tested with the Indirect Tensile (IDT) testing device, which allows the user to prepare input data for performance prediction. After a data analysis, the creep compliance, a measure to determine resistance against deformation with respect to time, and the tensile strength, were obtained. These are the input data for a software which can predict the performance of both materials for a given analysis period. Comparing the outputs for both materials give the user the expectable performance. Interpretations of these were used for a technical recommendation for not only future road construction projects, but also further research.

CURRENT STATUS OF KNOWLEDGE
The literature review comprised many fields of pavement engineering, starting from the specimen preparation, i.e., basic material handling, all the way to softwarebased predictions on the basis of analyzed test data. This chapter also explains why certain decisions were made and why certain steps were taken in order to accomplish the objective. It includes steps of the CIR Mix Design, the indirect tensile test, and application of gained data in the computer software.

EARLIER WORKS BY URI RESEARCH TEAM
A URI research team developed a performance-based mix design for bituminous materials with CIR under the contract of the Federal Highway Administration (FHWA) (Lee, Brayton and Milton 2002). Firstly, they analyzed the difficulties that existed for the standardized application of this technique. Overall it was found that there were wide variations concerning procedures, testing and quality control, e.g., Oregon and California have state-wide methods which differ considerably from one another. Hence, the main objective was to develop a standard procedure to use throughout the US.
The research project started with forming an expert task group (ETG) whose members comprised representatives from CIR contractors, universities and state agencies. They assisted setting up a work plan consisting of five phases which are shown in Table 2-1. Unlike the hammer blows with the Marshall compactor, the Superpave gyratory compactor "imparts a constant vertical pressure of 600 kPa to the sample while rotating (or gyrating) the sample with an eccentricity of 1.25° from the vertical axis" (Coree and VanDerHorst 1998). After every gyration, the height of the sample is determined, which allows any desired density, once the initial mass was measured. With field densities known, representative specimens can be prepared.
Since curing plays an important role, the appropriate duration and temperature needed to be determined. In the study, two levels for both curing time as well as temperature were compared. Results showed that the temperature of 60°C (140°F) was more effective than 25°C (77°F), and a curing time of 24 hours was more practical than 6 hours. It was decided to add an asphalt emulsion as the additive in the study. Once the RAP with the asphalt emulsion is put back in place, compaction causes the emulsion to break and the asphalt eventually coats the aggregate particles. What needed to be determined was the adequate amount of water as well as emulsion.
To optimize both parameters, a criterion was chosen. In this case, the bulk specific gravity was supposed to become maximum. Optimizing both parameters with only one criterion is a very challenging task, thus the parameters were optimized one at a time.
Initially, the water content is to be kept constant (3.0% suggested) while the amount of emulsion varies. Four different amounts of emulsion are to be used with 2 specimens each for improved reliability. For both contents, values may be based on experience, but the procedure suggests starting values that are expected to surround the optimum contents. After specimen production and determination of unit weight (or bulk specific gravity) according to the procedure of AASHTO T 166 (AASHTO 2011) and theoretical maximum specific gravity according to AASHTO T 209 (AASHTO 2011), both values are plotted over the percentage of emulsion in the first, or water in the second step.
For both the emulsion and water content, the optimum levels are determined by selecting the "content at which the unit weight is at its maximum value" or, "if a maximum unit weight is not achieved, the content at which the unit weight is similar to those found in the field" should be the optimum value (Lee, Brayton and Milton 2002).
The Superpave design method was recommended to prepare specimens for the prediction of future performance of the pavement with CIR. Software, such as VESYS® or the Mechanistic-Empirical Pavement Design Guide (MEPDG®) which includes more comprehensive prediction models, can be used to predict performance of the roadway in the future using the input data of the materials and environmental conditions.
A visco-elastic model, VESYS, requires two parameters, K 1q and K 2q , which describe the material's fatigue characteristics, to predict the area that is expected to crack, with certain stochastic deviation. These parameters can be determined from material testing; the flexure fatigue test on beams (Steen 2001). Those parameters need to be determined at a set of different temperatures because the behavior of asphalt concrete, as a viscoelastic material, is highly dependent on the temperature. The test results are utilized to determine the parameters with which the models predict the performance.
For field verification, a test site was constructed in Arizona. The procedure described above was applied, and the optimum emulsion content (OEC) and optimum water content (OWC) used were 2.5 % and 2.0 %, respectively. A 2-inch milling produced a 2-inch CIR layer, but instead of the planned chip seal, a 1.5-inch HMA overlay was placed on top of CIR base. It has been reported that no significant distresses were found, which proves a very good performance.
For this thesis work, the developed performance-based Mix Design for CIR mixtures was used. The entire procedure can be found in APPENDIX A. However, the concise steps are described in the following chapter.

ASPHALT EMULSIONS
RAP materials still contain some asphalt binder; however it has aged and does not coat the particles well after milling. Without heating, it cannot be distributed evenly throughout the pavement material to glue it together.
Generally, there are several options for the type of additives for CIR projects, but an asphalt emulsion was used in the study. An emulsion's versatility and adjustability to many different boundary conditions was one reason to become the most common additive for CIR projects (California Department of Transportation 2008).
In general, "emulsified asphalt is simply a suspension of small asphalt cement globules in water, which is assisted by an emulsifying agent". As liquid asphalt is based on oil and does not dissolve in water, a chemical agent is needed to disperse the liquid asphalt in water. It forms droplets with a diameter ranging from 1 to 10 μm (Walker 2012). however. The following label yields information about the "setting", breaking rate.
"RS" stands for rapid set, "SS" for slow set, and "QS" means quick set. Setting or breaking of the emulsion is the process in which the tiny droplets begin to coalesce and form a film to cover the aggregate. Careful handling according to the applicable safety and handling instructions prevent premature breaking.
After breaking, sufficient time must be allowed for curing. This process comprises the attainment of stiffness when the water separates and disperses. Also, the ambient weather can offer good conditions for this while high humidity and low temperatures, on the other hand, can deter proper curing (The Asphalt Institute 1979).
For laboratory specimen preparation, a sufficient time for curing must be ensured (24 hours) in an oven.
CSS-1h emulsions are usually a good choice for CIR projects. The slow setting rate is suitable for this type of road construction. The number refers to the emulsion's viscosity. In this case, the emulsion has a lower viscosity than a "2". The "h" designates the use of harder base asphalt, which is usually the case.

INDIRECT TENSILE TESTING
When conducting the indirect tension (IDT) test, a cylindrical specimen with a diameter of 150 mm (6 in) and a thickness of 38 to 50 mm (1½ to 2 in) is exposed to a single load imposed perpendicularly to the longitudinal axis with a static support on the opposite side. Figure 2-2 1 offers a front view of the specimen during the test.
It was found that the IDT test is able to represent the most critical location of a pavement under a wheel load. Roque and Buttlar (1992)     where the future pavement is to be constructed, the pavement structure can be entered into the program. Then, the program output is a prediction of the serviceability of the roadway that can be expected under the given boundary conditions. MEPDG combines a variety of sub-programs that each treat different distresses such as rutting, fatigue cracking and thermal cracking.
TC MODEL is the program which analyzes the stresses due to low temperatures and is able to quantify the thermal distresses. It should be noted at this point, that "TC MODEL does not consider traffic effects" (Marasteanu, et al. 2007). This program is very user-friendly, as it offers a convenient way to enter the data input. It simplifies the problem by using linear elastic fracture mechanics in a one-dimensional stress evaluation model. Newer software aims to treat this problem with nonlinear finite element analysis engines. This allows for better reliability of the predictions due to a more accurate mathematical representation of the problem. This software is still under development for the full-scale deployment (Leon, Dave and Park 2011).
TCMODEL offers a mechanistic approach to treat thermal cracking in flexible pavements by means of mathematical modeling. Unlike a variety of other mechanistic programs, it offers the possibility to quantify the amount of low-temperature cracking which is a major advantage. However, this is only the final step after first calculating the thermal stress and, second, calculating the crack propagation in the pavement structure over time.
Based on the creep compliance with respect to real time t determined for different temperatures, master curves are obtained. From that, a 4-parameter Prony series is retrieved for the creep compliance D with respect to reduced time ξ. Using a known strain history, the thermal stress over time can be obtained using a onedimensional hereditary integral. In order to integrate with respect to real time t instead of reduced time ξ, a finite difference solution was developed. This requires for the solution to be calculated for time intervals Δt in which the strain changes by Δε.
The next major step is the calculation of the crack propagation. A crack is modeled as a single vertical crack in the bituminous layer. To express the growth of the crack length, Paris' law is applied and simplified as TC MODEL computes the results on a daily basis. With both the thermal stress and the crack propagation treated mathematically, now the particular step of TC MODEL, the calculation of the crack amount, can be started. The amount of cracking is obtained from a calibrated probability function with respect to the crack length being at least as long as the thickness of the bituminous layer, as shown in Eq 2-1.
observed amount of thermal cracking β regression coefficient, determined through field calibration P( ) probability function D thickness of surface layer C crack length Field calibration resulted in β to be determined as 353.5. Also, not only the character of the probability function, but the program itself inherits certain restrictions concerning the outputs for crack amount. A crack is only considered a crack if it reaches the bottom of the asphalt layer. Also, the program predicts no more than 50% of the total possible amount of thermal cracking (Marasteanu, et al. 2007).
The results from the performed calculations are shown in the output files. It shows the results after every month for the desired analysis period, which is usually 20 years. So for every month the amount of transverse cracking will be shown in feet per mile. Since MEPDG includes a variety of sub-programs to predict multiple types of distresses, the regular result includes their outputs as well. For the analyses conducted in this thesis, however, the results are limited to the distress of thermal cracking. Then, they are interpreted in order to give an evaluation of the thermal cracking resistance of the recycled material to contribute to a recommendation for the use of CIR in future projects.

ASPHALT PAVEMENT MATERIALS
The primary objective of this study was the evaluation of the performance of a CIR material for rehabilitation or reconstruction projects. A base line was established with HMA first. For HMA specimen production, aggregates were acquired from a quarry.
The new type of material was CIR, which makes techniques using this material being more sustainable. The present, old pavement material is milled off the road, screened, and emulsion is added. Then the material is put back in place and compacted.
After a sufficient time for curing, the surface needs to be protected by a sealing course or an overlay.
To measure performance in the laboratory, specimens are necessary that represent field conditions as accurately as possible. This chapter explains how this requirement was dealt with, how the material was obtained, how the recipes for both types were developed, and how the final specimens for testing were produced.

HOT-MIX ASPHALT (HMA) SPECIMENS
HMA represents the current practice of road construction, and consists of mineral aggregates and asphalt binder. In this study, the aggregates were acquired from the PJ Keating Company in Cranston, Rhode Island. These were sieved and weighed in accordance with the gradation of RI Class I-1 (see APPENDIX C).
Specimens with 4 different binder contents from 5.5 to 7.0% with increments of 0.5% were tested, and the optimum binder content (OBC) was determined at the air void content of 4.0%. The details can be found in APPENDIX C.
A loose (uncompacted) sample with a mass of approximately 1,000 g (2.2 lbs) was used to determine the theoretic maximum specific gravity according to the procedure of AASHTO T 209 (AASHTO 2011). The aggregate particles were separated manually in order to avoid small cavities that would contain air and therefore distort the volume determination. After that, the sample was weighed into a container of known volume. Then, enough water was filled into the container to completely cover the sample. The sealed container was agitated for 15 minutes with a mechanical device while a vacuum pump creates negative pressure to remove all the air that is being expelled from the sample. Finally, the container was filled with water entirely and its mass was measured. The maximum specific gravity could be determined according to It should be mentioned at this point that the calculated test result is very sensitive to slight deviations concerning the determination of the masses. Not only is it important to avoid any air bubbles when sealing the container with the lid, but also water can be in the tap that is required for the connection of the vacuum pump. That is the reason why this test was repeated several times to get reliable and repeatable results.
This test was performed only for one binder content. Using the correlation between the maximum theoretical specific gravity (G mm ) and the effective specific gravity (G se ) of the mixture, G mm for the other binder contents can be calculated according to Eq 3-2 (Mamlouk and Zaniewski 2006).
These results are the only ones obtained from the uncompacted specimens. The results are shown in The compacted specimens were then tested for their bulk specific gravity (G mb ) and water absorption according to the procedure of AASHTO T 166 (AASHTO 2011).
After determining the dry mass of all specimens, they were submersed in water for 4 minutes and the weight under water was determined. Then, they were taken out of the water and the surface water was removed with a damp towel in order to achieve a saturated surface-dry state. Finally, the mass was determined again. The G mb was calculated according to Eq 3-4.

C B
mass of saturated surface-dry sample in air [g] C mass of specimen in water [g] For this test, duplicate specimen results were averaged, the results are shown in  The bulk SG after every gyration can be estimated because the specimen's mass is known and the gyratory compactor yields the height of the specimen after every gyration in the compaction mold of known dimensions. By dividing the mass by the calculated volume and the density of water at 4.0°C (0.999972 g/cm 3 ), the estimated bulk SG (G sb,est ) can be computed. A correction factor C is introduced to obtain the corrected bulk SG (G sb,corr ). It is determined and applied to every gyration according to Eq 3-5 and Eq 3-6.
Eq 3-6 By dividing the corrected bulk SG by the theoretical maximum SG, the compaction level as a percentage of the maximum theoretical SG is computed.  As one can see, the compaction level after 100 gyrations is 96.2% which corresponds to an air void content of 3.8%. This calculation was done for duplicate specimens at 4 binder contents. As an example, the spreadsheet for the first sample with a binder content of 5.5% can be found in APPENDIX D.
Finally, the binder content is plotted against the averaged air void contents. The results are shown in Figure 3 Air Voids over Binder Content

COLD IN-PLACE RECYCLED (CIR) MIXTURES
RAP was acquired from a construction site of Rhode Island Route 3.
Unfortunately, the material was stored uncovered for an unknown amount of time, and the influence of aging, oxidation and freezing, especially on the binder, may have been significant. However, since another source was not available, the RAP was used despite concerns over different properties from fresh one.
After drying, the moisture content and the gradation were determined in accordance with the procedure of AASHTO T 255 and T 27, respectively (AASHTO 2011). The results indicated that neither the moisture content of 4.1% nor the gradation of Figure 3-3 exhibited any significant deviation from expected ones. The details can be found in APPENDIX E.

Figure 3-3 Gradation of RAP Material
As reported earlier, the CSS-1h emulsion was used in this study. According to the CIR Mix Design procedure which was developed by a URI research team (Lee, Brayton and Milton 2002), the optimum emulsion and water content needed to be determined. The process was carried out by first keeping the water content constant and Weight passing varying the emulsion content. For the determination of the optimum content, the unit weight is the parameter to be compared to field conditions. Either the maximum value or, if a maximum cannot be determined, the best representation of field condition should be chosen. Before any specimen can be produced, the appropriate number of gyrations for compaction needs to be determined. Since the standard procedure for HMA (AASHTO 2011) is not applicable here, another method needed to be used. The URI procedure states that, "The load shall be applied for the number of gyrations that will result in achieving densities similar to those found in the field." Therefore, a method was used that is somewhat similar to the determination of the amount of air voids for HMA materials (Lee, Brayton and Milton 2002). It is based on representing field density, and the value of previous URI study, i.e., 130 pcf was used (Steen 2001).
Following steps were used in this study: The test specimen for this procedure was made with a water content of 3.0% and an emulsion content of 1.0%. After compaction, the bulk SG was determined and all the required calculations were completed as seen in APPENDIX F. The compacting behavior of the specimen can be seen in Figure 3  As can be seen from this graph, the degree of compaction can very closely be approximated using a logarithmic curve. Based on the values shown in Table F-1 (APPENDIX F), 116 gyrations were used for all following specimens.
For any set of emulsion and water contents, 9,000 g of RAP were used. This was because duplicate specimens for bulk SG determination with about 4,000 g each were needed along with one sample for theoretical maximum SG determination, for which a mass of about 1,000 g was sufficient. To ensure sufficient mass, 9,000 g for the RAP was chosen because emulsion and water still had to be added, which lead to a mass of more than 9,200 g before curing.
For the determination of the OEC, emulsion contents varied from 0.5% (of total mix mass) to 2.0% with increments of 0.5%, while the water content stayed constant at 3.0%. A summary of the masses of ingredients for all specimens that were produced for OEC determination can be found in Table 3  After production, specimens were put in an oven at a temperature of 60°C (140°F) for a period of 24 hours for curing. This time was needed for the water to leave the specimen and for the binder to coat RAP and attain stiffness.
While the bulk SG specimens were being cured, the theoretical maximum SG of  In one aspect, the regression fit the data points very well which is indicated by the R 2 value above 0.99 in both cases. However, the behavior of the left curve was somewhat different than expected since it indicated that a higher unit weight was achieved using less emulsion. A similar behavior was observed (Lee, Brayton and Milton 2002) and a solution corresponding to the applied Mix Design procedure was applied: "Due to the highly variable nature of RAP materials and their mixture with emulsion and water, the relationship between unit weight and emulsion content, as described earlier, occasionally does not hold true for CIR mixtures. Such a case occurred with the Kansas mixture. The highest unit weight was achieved at the lowest emulsion content of 0.5%. However, 0.5% emulsion does not supply enough asphalt to properly coat the RAP particles. Under such conditions, the OEC should be selected at the emulsion content that produces the same unit weight as found in the field." In this study, the same option was chosen. Thanks to the close fit of the regression curve, it could be used to numerically determine at which emulsion content a unit weight of 130 pcf was achieved. Thus, the optimum emulsion content was determined to be 0.7%.
With the emulsion content optimized, the next step was to determine the optimum water content (OWC). This was very similar to the previous step with the exception that the emulsion content was kept constant at the optimum level while the gyrations. However, this time the ingredient weights shown in Table 3-3 were used. After these specimens were produced and cured, testing for the maximum theoretical and bulk SG was performed. The obtained results are shown in APPENDIX G. Figure 3-6 represents the unit weight and the air void content versus the water content. Again, the R 2 error refers to a parabolic regression curve calculated by the spreadsheet program.  Here, a clear maximum can be seen which identifies the optimum water content to be 3.0%.
In summation, the optimum contents for the emulsion and water were determined to be 0.7 and 3.0% of total mix weight, respectively. Another observation was made after the first set of CIR specimens was completed. The specimens, especially in direct comparison between HMA and CIR, revealed that some of the fine particles were not as thoroughly integrated into the material as in the HMA specimens. Even with only slight rubbing motions with a glove or palm over the surface of the CIR specimens, small amounts of fine materials were separated from the specimen. Of course, these amounts were negligible by far, but it was a behavior that suggests a weakness of the material and was expected to be the reason for the absolute necessity of a sealing layer of HMA or seal coat when CIR is applied in a reconstruction project.
Another observation was that even with a limited number of specimens to be produced for both materials, the amount of laboratory work was extensive. The effort for acquiring, drying, and sieving the aggregate as well as specimen production and testing must not be underestimated.

INDIRECT TENSILE (IDT) TESTING
This chapter describes the different steps from the specimen preparation to the test results.

SPECIMEN PREPARATIONS
A testing machine, Instron® 5582 was available in the Rhode Island Transportation Research Center (RITRC) laboratory at URI. An attempt was made to calibrate the machine. A malfunction was found in data acquisition system. Several tests showed that the machine could work on a sample, but the deflections and applied force could not be logged during or stored after testing. Efforts were initiated towards the purchase of a new data acquisition system. The existing machine was purchased before the turn of the millennium and purchasing an entirely new system would exceed the available funding. Therefore, suitable new parts that would work with the machine needed to be identified. In cooperation with the manufacturer with extensive communication via email, the proper upgrades could be found and a quotation was filed (see APPENDIX B). The order was placed, but unfortunately the delivery and complete installation of the new software was too late for this project to take effect. Therefore, another way of testing needed to be found.
Fortunately, the University of Connecticut (UConn) provided the required testing system for this project. With experienced guidance, a total of five appointments were agreed upon first to inspect the testing equipment at the Connecticut Transportation Institute (CTI). Thereafter, four more appointments were made for two different sets of specimens to be sawed and tested. Since sawing the cylindrical specimens to suitable heights of 38 to 50 mm (1½ to 2 in.) from their original heights was accomplished with wet sawing, an appropriate time frame of at least 24 hours was necessary to allow the samples to dry before testing. At this point, gratitude should be expressed to UConn for their generous support towards this project.
In cooperation with the University of Connecticut (UConn), five appointments were made. The first meeting was to evaluate the sawing and testing equipment, coordinate future testing, and review requirements and restrictive boundary conditions for performing the test. The following four appointments were sawing and testing both materials.
At least one day was required after sawing as the specimens had to be dried. The cooling water of the saw pervades the specimen. Since testing is conducted at temperatures far below the freezing point of water, the specimens' performance is highly susceptible to any water content.
The cylindrical samples have a height of about 110 mm (4.3 in), thus two specimens with the required height could be produced. Due to the quality of the saw, a high level of accuracy could be maintained. Figure 4-1 shows one HMA and one CIR specimen in the saw fixture and during sawing. It was observed that the different behavior of the materials could be seen even during specimen preparation. The fine materials of the recycled samples were less strongly integrated into the material and therefore chipping was increased during sawing. In order to still obtain usable specimens, care had to be taken to saw the specimens fast enough to minimize wobbling of the blade and at the same time slow enough not to rip out particles instead of cutting through them.
After labeling, the average heights were determined and listed. Eight specimens were used from both mixtures for a total of 3 different temperatures. The labels are shown in APPENDIX J.
Next, metal mounting buttons were glued onto the specimens. The loading ram was lowered onto the specimen as slowly as possible to prevent an impact when the ram came into contact with the specimen. The load was kept constant at such a magnitude that a horizontal displacement, measured by the horizontal strain gauge or linear variable differential transformer (LVDT), was between 1.25 and 19.0 μm, as stated in the procedure of AASHTO T 322 (AASHTO 2011). This ensured that the deflections were within the viscoelastic range.
To achieve this by adjusting the magnitude of the load, it should have been controlled by input from the horizontal deformation measurement device, but this was difficult to accomplish due to the limitations of the closed-loop control system. Therefore, the load was estimated on the basis of experience and previous testing, and the resulting deformation was verified. In practice, the given range was hard to comply to, and experiences showed that slight exceedances still led to reliable results.
This load was constant for the complete creep test time of 1,000 seconds. During testing, the testing apparatus has to constantly move the loading ram in order to establish a constant load. This is possible electronically through a proportional-integralderivative (PID) controller. Its task is to minimize the error which is the difference between the desired and the detected load. The loading piston position is corrected constantly; three parameters determine the mode of readjustment. They are chosen mainly based on experience and can be readjusted during testing. The better these values are, the less fluctuation is detected in the course of load over time. Table 4-1 shows the values for the conducted tests. After establishing this constant load, the deflections were measured for both faces of the specimen. This is intended to reduce influences on the obtained measurements due to material inhomogeneities by allowing the user to average them.  After this testing phase, the load was removed. Although the specimen was not destroyed, it exhibited permanent deformation. However, it can still be used for the tensile strength test, as can be seen in the standard test method (AASHTO 2011). In this test, the loading ram had to maintain a constant movement of 12.5 mm/min (0.5 in/min), while only the imposed load needed to be measured. The deflections did not matter here. Since this test destroys the specimen, the strain gauges were removed to prevent damage that might have occurred due to the specimen's collapse. The ram movement must be maintained until the load sustained by the specimen starts to decrease. This is regarded to as failure, and the maximum load is used to calculate the failure load. When testing for the material's strength at different temperatures, it was observed that the temperature had a significant influence on the behavior. The lower the temperature, the more brittle failing occurred. While the material allowed some rather ductile deflection before completely falling apart rather slowly at freezing point. At -20°C (-4°F) there was mainly one sudden, loud crack, and the specimen collapsed.
APPENDIX I shows the testing schedule. The first testing took more time than the schedule. But, the experience helped tremendously to shorten the testing time at the following meeting. The data were retrieved and analyzed to obtain the creep compliance as well as the tensile strength for the planned simulations.

DATA ANALYSIS
It should be noted that the testing software includes analysis algorithms.
However, since academic testing is mostly used for educational purposes, in this study only raw data were used. Also, in order to understand the entire sequence of events and to check for any mistakes that may have happened during testing, manual calculations were more useful than automated processing.

Table 4-2 Creep Test Steps
Step 1  Balance strain gauges and load cell Step 2  Lower loading ram  Load ≤ 0.5 kN ('noise') Step 3  Load > 0.5 kN  Increase load to desired level Step 4  Keep load constant for test duration of 1,000 s Step 5  Remove load  End test The first three steps had to precede the actual testing. The strain gauges and the load cell were zeroed and the loading piston was moved towards the specimen. The important data for the analysis were produced in step 4. For supervisory reasons; however, data from all steps were checked first and only step 4 data was used for the creep analysis.
First, the behavior of the load was checked for every specimen to see its deviation around the desired load. As explained earlier, the PID controller algorithm constantly attempts to minimize the error which in this case is the difference between the desired and the detected load. However, this can practically never be perfect and it gets worse with PID settings that are not optimized.   This is a good result as it proves consistency within the specimen and the probability of  getting close to the "true" value is increased by calculating the average. The vertical displacements deviate a little more. This can be due to multiple reasons, starting from inhomogeneous material, influence of large or interlocked pieces of aggregate, or aggregate gaps on the path from the loading piston to the center of the specimen that are filled with binder or air voids. Also, sometimes the fault can lie within the strain gauges, although this is rather rare. Additionally, Figure 4-5 shows that the offset for all deflections is almost zero. One might argue that this must be the case automatically as a consequence of zeroing in step 1, but in one case initial measurements that varied from 0 significantly were observed. In such a case the error needs to be corrected by shifting all values of that curve by the error's magnitude towards zero to compensate the inaccuracy.
Subsequently, the horizontal and vertical deformations of all specimens at the analyzed temperature were averaged and normalized in order to compare them. This was accomplished using Eq 4-1 and Eq 4-2 (AASHTO 2011).    curves or arrays, the accuracy would decrease due to these two curves that deviate from Eq 4-6 while Eq 4-7 must be true.
It may be noted that Poisson's ratio ν should always be between 0.05 and 0.50 (AASHTO 2011). The calculations were carried out in a spreadsheet program (Microsoft Excel ©). They were performed for all temperatures for both mixtures. Table   4-3 shows the results for HMA at -10°C, while a summary of all calculations is provided in APPENDIX L.  Table 4-4, the results were averaged over the surrounding 5 time points in increments of 0.5 seconds. Also, the used version of the program only allows the input of the results in US customary units, while the test method consistently uses SI units. Therefore, the creep compliance is firstly calculated in [1/GPa] (SI unit) since it results from the calculations above. Technically, the obtained unit was [1/TPa], but by dividing by 10 3 , the unit [1/GPa] was obtained. Then the conversion factor of (145000) -1 was applied to obtain the customary unit of [1/psi]. In summary, Table 4-5 and Table 4-6 were prepared as input material parameters for MEPDG analysis. It may be noted that the CIR mixture has a higher compliance than HMA does, as expected.  The specimen is aligned in the same way as for the compliance test, but this time the loading piston is to move at a constant speed of 12.5 mm/min. During testing only the sustained load is measured until a decrease is detected. This may or may not come along with a brittle collapse of the sample, but the maximum load is interpreted as the failure load and is used to determine the tensile strength.
When the failure load is known, Eq 4-10 allows computation of the tensile strength (AASHTO 2011).  As it can be seen, the temperature increase from -20 to -10°C does not affect the averaged tensile strength. It is rather consistent at a magnitude of 6.0 MPa. Even the increase in tension until failure seems to be linear when increasing, and both temperatures have an even steeper decline after failure.
Both samples at 0°C, however, behave differently. The increase in tension is slower, and both of them peak at about 4.4 MPa. This is due to the more ductile behavior of the binder as it allows more deformation with increasing temperatures, but it also leads to a lower strength.  Along with the results from the previous section which are shown in APPENDIX L, the tensile strength at a temperature of -10°C (14°F) serves as input data for the prediction software. They are used to predict the low-temperature cracking for a fictional roadway project which is the topic of the following chapter.

PERFORMANCE PREDICTION OF BOTH MIXTURES
This chapter's topic is the final step of performance prediction using the obtained data for a fictional project to evaluate the performance of the CIR mixture comparatively. It includes the input to the model and program for the selected boundary conditions of the site. Consequently, the outputs of the program would include resistance to low temperatures cracking and would be interpreted to formulate recommendations.

INPUT FOR TCMODEL AND MEPDG
The software MEPDG offers a user-friendly input framework, characterized by a checklist layout. Each bullet point allows the adjustment of parameters for general information, traffic, climate, and pavement structure. This window is shown in Figure   5-1. MEPDG also includes several prediction models including TCMODEL for thermal cracking.
For this project a "minor arterial rural highway" was selected: Rhode Island Route 2 leads from South Kingstown to North Kingstown (Lee, Marcus, et al. 2003). A section of the road in the southern part of the State was chosen since from that area a report offers a variety of data for traffic and subgrade soil information. The study was conducted by a URI research team for the Rhode Island Department of Transportation The next inputs comprise the data for traffic amounts and distributions. It may be noted that thermal cracking is not load-related and does not depend on the amount of vehicles for this project. However, the programs requires a completed set of information for any project, so the traffic amount of 1,346 annual average daily truck traffic (AADTT) was entered (Lee, Marcus, et al. 2003). A summary of the traffic data is provided in APPENDIX N.

Figure 5-1 MEPDG Input Checklist
The climate plays a very important role for this distress. In MEPDG, the climate files are created based on the history that is known for weather stations in the vicinity of the project site. There are three stations in the State of Rhode Island: Westerly, Newport and Providence. Newport, RI, is located on an island, and is also rather far away from the planned location. In addition, climate data from stations in neighboring States were available and were used for improved accuracy. In total, the climate file was created through interpolation among 5 stations, as shown in Figure 5-2.
Next to the weather data, the location for which the data is to be interpolated needs to be entered. As shown, a position was chosen in the southern part of the State; its coordinates are N 41.52, E 71.55. The elevation is approximately 220 ft; both information were found using "Google Earth". The depth of water table was entered as 10 ft (Esri 2011). A summary of the weather data is shown in APPENDIX N.

Figure 5-2 Climate File Generation via Interpolation
The next input is the pavement structure. The bottom layer is subgrade soil classified by AASHTO standards as A-1-b, i.e., that a maximum of 50 and 25% of the aggregate would pass the No. 40 and No. 200 sieves, respectively (Atkins 2002). The resilient modulus was 14,300 psi (Lee, Marcus, et al. 2003). It is semi-infinite while the above layers are assigned finite thicknesses. The layer above the subgrade is granular subbase with a thickness of 12 in., as it is common in Rhode Island for a fill or embankment section. In the case of cut or excavation section, 18 in. would be appropriate.
The base and surface courses need to be provided in the prediction software.
This is the point where MEPDG, unfortunately, limits the possibilities to enter a pavement that would accurately reflect the way CIR is supposed to be used in reality.
After application of CIR and curing, protection is required for "the surface of the CIRtreated material by either a surface wearing course, such as a seal or HMA overlay" (FHWA 2005). Therefore, CIR would most likely be applied in combination with HMA. However, the program allows entering of only one set of creep compliance and strength test data for all bituminous layers of the entire setup, which prevents pavement structures that comprise layers of both HMA and CIR. So for comparison reasons, basically the two bituminous layers are entered with the creep compliance and tensile strength results from HMA in one case and from CIR in the other case. As it is practice in Rhode Island for deep strength pavements, the base and surface courses have thicknesses of 5 and 2 in., respectively. In addition, a third setup was simulated with a base course thickness of only 2 in. This, of course, must never be used in reality.
However, it is intended to reveal how a very thin course of CIR mixture would perform in terms of thermal cracking. Figure 5-3 shows the three setups used in this study.

Figure 5-3 MEPDG Pavement Setups
The parameters to characterize HMA and CIR, the densities and percent of air voids must be entered into the program. The values are shown in Table 5-1.

PREDICTION RESULTS AND INTERPRETATION
After simulating all three different cases, summaries of distresses revealed that none of the cases are expected to exhibit any distresses over the entire analysis period of 20 years. Not even the third case, where the pavement is by far too thin, showed any distresses. As an example, the output for the distresses of the pavement including CIR mixtures with a base course thickness of 5 in. ("CIR (1)") is shown in Table 5-2. This is a rather unexpected result. Especially because of the observations about the texture which appeared weaker than the HMA, apparently this property does not affect the distress of thermal cracking. It can be expected to affect the performance for a variety of load-related distresses such as longitudinal or alligator cracking; however this was not within the scope of this project but is rather recommended to be analyzed in future research.
The question of the performance cannot be answered entirely with the result that neither material will crack at the simulated weather conditions. Still, the result that CIR material performs very well in a climate found in the South of Rhode Island is more than desirable and supports this approach towards a more sustainable reconstruction practice immensely.
How can a CIR mixture with a tensile strength of less than 12% of HMA's strength perform just as well? Apparently, not only the tensile strength, but the material's behavior before failure plays a significant role for cracking. This distress is not load-, but temperature-related. The stresses do not arise from imposed loads that need to be sustained, but rather from (blocked) shrinkage. As mentioned earlier, the creep compliance is a measure of deflection over stress. The graphs of creep compliance over time reveal for both materials that CIR exhibits a more ductile behavior, i.e., allows more deflection. As both mixtures are exposed to the same climatic situations, they both do not show distresses, but behave differently. CIR mixtures reduce stresses by allowing higher deflections. HMA mixtures behave more brittle and deflect less, but do not fail because the tensile strength is higher than the actual stresses.
The results from MEPDG are shown in APPENDIX O. The output contains the magnitudes of the distress after every month, yet for clarity reasons only every full year is shown.
In summary, these results represent a more than desirable result for the analyzed problem. Thermal cracking of this type of recycled pavement material is of such a low extent that it can be recommended, although more works in terms of performance regarding load-related distresses are necessary.

CONCLUSIONS AND RECOMMENDATIONS
The conclusions and recommendations based on the findings and observations of this study have been summarized below.

CONCLUSIONS
(1) As an alternative to conventional roadway reconstruction practices, the Cold In-Place Recycling (CIR) possesses properties which meet conventional requirements.
(2) Based on the Mix Design for CIR mixtures developed at URI, specimens were produced and tested. In comparison to conventional HMA, the texture of the CIR specimens appeared not to glue the particles to one another as thoroughly.
(3) IDT testing revealed that the tensile strength is, in fact, reduced by up to 90%. However, also the creep deflection is increased, allowing the material to increase strain at a given load. This is expected to be a major reason for its very good performance.
(4) The simulated results showed that no thermal cracking is expected to occur over the entire analysis period of 20 years.
The results of this study, however, support CIR as a viable option for roadway reconstruction. Through the reduced stresses to the environment and the people, CIR can provide a greener and more sustainable approach.

RECOMMENDATIONS
(1) This positive result strongly supports the CIR technique as a rehabilitation strategy.
(2) While the performed simulations were limited to exposure to Rhode Island climate, further investigations should be conducted for the severe weather conditions in other US states.
(3) Further questions offer plenty of research possibilities regarding other types of distresses, variations for the additives and more.
(4) Acquisition of newer RAP is recommended. CIR projects are often realized with recycling trains that perform the construction projects automatically.
Retrieving RAP from such a train can reflect field conditions more accurately.
(5) Concerning the sample production, two major improvements are recommended. One of them is the production of more specimens to obtain an increased statistical reliability. In this project, the scope was limited; however for future research projects more data is recommended. Also, a wet saw with a bigger blade is necessary. It should be able to cut specimens with a diameter of 150 mm in one motion, i.e., without having to turn the specimen during cutting.
(6) For the prediction, the used software simplifies the problem as it applies linear elastic fracture mechanics. Newer software is recommended for future use, programs such as "LTC Model" contain non-linear approaches that contain a higher level of accuracy in representing the problem.
(7) In addition, it is suggested to continue research in terms of different types of distresses, especially those depending on the amount and distribution of load applications, e.g., longitudinal and/or alligator cracking.

APPENDIX A MODIFIED SUPERPAVE MIX DESIGN FOR CIR MIXTURES
This section contains the modified mix design procedure for CIR mixtures. It was developed by a research team under leadership of the URI to establish a standard and reduce the variations in the practice of CIR applications.
1. Scope 1.1. This method covers the design of mixtures for cold in-place recycling (CIR) using the Superpave Gyratory Compactor. The procedures presented are applicable only for mixtures containing asphalt emulsion and reclaimed asphalt pavement (RAP). This method consists of two parts. The first is the determination of the optimum emulsion content and the second is the determination of the optimum mixing water content. 2. Apparatus 2.1.
See (AASHTO 2011) 3. Test Specimens 3.1. Preparation of RAP 3.1.1. RAP samples shall be obtained from the roadway that will be recycled by taking cores to the specified depth. These cores will then be crushed in order to have representative samples. 3.1.2. Dry a portion of the RAP to a constant mass at 110° C (230° F) to determine the moisture content. Dry the remainder of the RAP to a constant mass at 60° C (140°F) to remove the existing water. 3.1.3. Separate the RAP into particle sizes according to Table A-1, by screening through a series of sieves. Eliminate the material retained on the 31.75 mm (1¼ in) sieve either by removing or crushing the material such that excess fines are not produced. Preparation of Mixtures 3.3.1. The first part of the mix design involves the determination of the optimum emulsion content, while keeping the mixing water content constant. A minimum of two specimens shall be prepared for a minimum of four emulsion contents in 0.5 % increments. All specimens will be prepared with 3.0% mixing water (different water contents can be used based on experience). In addition, one loose sample shall be prepared for each additive content for determination of maximum theoretical specific gravity.

Weigh into individual pans a sufficient amount of RAP (~ 4000 grams)
based on the gradation determined in section 3.1.3 to fabricate specimens 150mm (6 in) in diameter and 115 mm (4.5 in) in height. 3.3.3. Heat RAP samples at 25°C  2°C (77°F  4°F) for a minimum of one hour. In addition, heat emulsion at the specified temperature (Section 3.2.1) for one hour. 3.3.4. Add mixing water to each sample and mix thoroughly for one minute.
Mixing may be performed either by hand or through the use of a mechanical mixer. 3.3.5. Add emulsion to each sample according to section 3.3.1 and mix thoroughly until the emulsion is uniformly dispersed but for no longer than two minutes. If the sample is not uniformly mixed after two minutes, additional mixing water may be required to improve emulsion dispersion. Otherwise, another emulsion type may be required. 3.3.6. Spread the mixture in a pan and allow the sample to cure until it 'breaks' (when sample changes from a brown to a black color).

3.4.
Compaction of Specimens 3.4.1. Preheat the molds at 60°C (140°F) for a minimum of one hour. 3.4.2. Apply load using the Superpave Gyratory Compactor (SGC). The loading pressure shall be 600 kPa at an angle of gyration of 1.25°. The load shall be applied for the number of gyrations that will result in achieving densities similar to those found in the field. 3.4.3. Remove specimens from their molds immediately after compaction. 3.4.4. Oven-cure the specimens at 60°C (140°F) for 24 hours. 3.4.5. Remove the specimens from the oven and allow to cool to room temperature. 3.5.
Bulk Specific Gravity, G mb 3.5.1. Perform testing according to T 166(AASHTO 2011). 3.5.2. This test method should be used when the samples absorb less than 2 % of water by volume as determined by section 6.2 of T 166(AASHTO 2011). Otherwise use(AASHTO 2011). 3.5.3. Determine maximum theoretical specific gravity for each emulsion content using (AASHTO 2011).
3.6. Determine Optimum Emulsion Content (OEC) 3.6.1. Plot unit weight versus percent emulsion content for each emulsion content. 3.6.2. Plot percent air voids versus percent emulsion content for each emulsion content 3.6.3. OEC is the emulsion content at which the unit weight is at its maximum value. 3.6.4. If a maximum unit weight is not achieved, the OEC should be the emulsion content at which the unit weight is the same as that found in the field. 3.7.
Determine Optimum Mixing Water Content (OWC) 3.7.1. OWC is determined by following steps 3.1 through 3.5, with the following exceptions. 3.7.2. A minimum of two specimens will be prepared at the Optimum Emulsion Content (OEC) with each of four varying water contents, 0.5% and 1.0% above and below the mixing water content used in step 3.3.1. 3.7.3. Plot unit weight versus percent water content for each water content. 3.7.4. Plot percent air voids versus percent water content for each water content. 3.7.5. OWC is the water content at which the unit weight is at its maximum value. 3.7.6. If a maximum unit weight is not achieved, the OWC should be the water content at which the unit weight is the same as that found in the field. 3.8.
Moisture Sensitivity 3.8.1. Prepare six specimens at OEC and OWC, three for dry testing and three for conditioned testing and determine moisture sensitivity of the specimens in accordance with T 283(AASHTO 2011 Basically, what we need is an apparatus that can perform the required ISDCT test. So we would need a setup with 4 channels. I figure, that then a variety of other tests would be possible as well, which would come in great. Considering your options, I would like to know how well Bluehill 2 performs in terms of reliability, user-friendliness and data recording.
Furthermore, for the decision I would like to know if you could give a quotation or an estimate of what expenses we are looking at. I assume that the least expensive option will be the very first one you mentioned will be BH 2 with the VersaChannel option in combination with 4 Channels. 4 channels would allow 4 displacements to be monitored, correct? So a quotation for that would be extremely helpful.
In addition, how much would retrofitting the 5582 model to 5900 and the Expansion Channel Module be?
Thank you for your support! We will have a meeting this afternoon, so the more information I have, the better I can present it to my supervisor. When the desired test temperature is entered according to the manual, the chamber will keep on heating up, although the desired temperature is exceeded. Furthermore, the data acquisition system seems to malfunction in a way that although LVDTs are connected as described in the manual and can be read in the software, no real-time data are monitored during testing and no recordings are being stored. I would be happy to receive an evaluation about likely malfunction and ideas to get the machine back to working properly. I am going to work on my Master thesis next semester and I rely on this machine to provide accurate test results.

APPENDIX G DETERMINATION OF OPTIMUM CONTENTS OF CIR MATERIAL
The following Table G-1 through Table G-4 show the values measured and calculated in order to identify the optimum emulsion and water contents for CIR material.

APPENDIX I PLANNED TESTING SCHEDULE
The following Figure I-1 shows the planned testing schedule for one material in one day. Of course, testing rarely works as planned, however rough planning should at least be attempted to get an idea of the approximate time frames.

APPENDIX K NORMALIZED HORIZONTAL DEFORMATIONS
The following shows a summary of the normalized horizontal deformations of all specimens made from both materials.

APPENDIX L CREEP COMPLIANCE SUMMARY
In this section, the dimensions and loads for all tested specimens are shown in    will lead to an 18-kip-ESAL amount of 10,549,906 after the analysis period of 20 years (Lee, Marcus, et al. 2003).  This Appendix contains the simulation results for all three pavement structures. The results were stripped as the original shows the distress magnitudes after every month which is not suitable here. Instead, the results after every year are shown for the HMA pavement ("HMA"), CIR with 5 in. of base material ("CIR (1)"), and CIR with 2 in. of base material ("CIR (2)").