 Research
 Open access
 Published:
Model development and prediction of antiicing longevity of asphalt pavement with saltstorage additive
Journal of Infrastructure Preservation and Resilience volume 3, Article number: 1 (2022)
Abstract
This study established a systematic simulation framework to predict the antiicing longevity of a thin overlay of asphalt pavement with saltstorage additive (APSSA). The water and chloride transport in the overlay when subjected to varying precipitation, temperature, thermal cracking, and fatigue cracking over time were modeled using a Finite Element Method based software. The simulation included two parts: water transport followed by chloride transport. Water transport that obeys the law of conservation of mass was modeled using the phase transport in porous media (phtr) interface of COMSOL, while chloride transport based on Fick’s second law was modeled with the transport of diluted species (tds) interface. The simulation results show that the antiicing function of a 16mm thick overlay was fully effective in 2 years and 5 years for the minimum pavement temperature above 3.4 °C and 2.4 °C, respectively. These two pavement temperatures are equivalent to 97.4percentile and 96.3percentile of historical hourly pavement temperature near Pullman, Washington.
Graphical abstract
Introduction
A few studies have reported that the antiicing life of asphalt pavement with saltstorage additive (APSSA) can be calculated with predictive equations incorporating laboratory experiments [11, 24, 25, 55,56,57,58]. However, these estimations were overly simplified, rough, and inaccurate because of lack of a mechanistic model. Another significant deficiency of those predictions is neglect of a comprehensive consideration of the factors influencing the antiicing life of APSSA, such as environmental conditions (e.g., precipitation and temperature) and pavement conditions (e.g., thermal cracking and fatigue cracking). Although many studies have worked on modeling ions (e.g., chloride) transport from the outside to the inside of cement concrete [2, 9, 18, 22, 31, 36, 49], no one has simulated outward ions (e.g., chloride) transport in asphalt pavement or cement concrete. No model has been developed to predict the antiicing life of APSSA. Therefore, it is necessary to establish a systematic model for a more mechanistic and reliable prediction of the antiicing life of APSSA.
The COMSOL Multiphysics is a crossplatform finite element method (FEM) based multiphysics simulation software. This software supports the simulation of designs involving single or coupled physics such as electromagnetics, structural mechanics, acoustics, fluid flow, heat transfer, chemical engineering, etc. [7, 17, 20, 26, 32, 36, 47]. Among them, the phase transport in porous media (phtr) interface in the subsurface flow module and the transport of diluted species (tds) interface in the chemical reaction engineering module have been widely applied for simulating multiphase flow (e.g., water–air flow) in porous media [28, 46, 48, 51] and simulating ionic ingress into cement concrete due to concentration gradient [2, 9, 22, 36, 49], respectively. For APSSA, ionic transport from the inside to the surface must be facilitated by water, and the process include two activities: water transport in asphalt mastic and saltstorage additive (SSA) followed by ion (e.g., chloride) transport due to concentration gradient. The two activities can be well modeled and linked by the cooperation of the two interfaces of the FEM software.
This work pioneered a numerical model with the COMSOL Multiphysics to predict the antiicing life of a thin overlay of APSSA when subjected to varying precipitation, temperature, thermal cracking, and fatigue cracking over time. In this study, the APSSA sample used for the experiments to obtain the parameters of the model was asphalt mixture with 5.1 wt% laboratoryprepared functional additive (CaCl_{2}zeolite/pepoxy) #16. The additive was fabricated through a surface treatment approach, in which zeolite particles containing calcium chloride were coated by a microporous epoxy layer. The #16 denotes the gradation of the additive, which passed the sieve #16 and stayed on the sieve #30 (1.2 mm – 2.4 mm). More details of the preparation of CaCl_{2}zeolite/pepoxy #16 and the APSSA sample can be found in the laboratory study [53]. The simulation was separated into two steps, which are water transport into the overlay first then followed by chloride transport from the inside to the surface. The water transport process follows the law of conservation of mass, while the chloride transport from highconcentration areas to lowconcentration areas obeys Fick’s second law.
Model description
To activate the antiicing function of the functional additives in the overlay, there are two critical processes: 1) wetting of solid calcium chloride (CaCl_{2}) in the additives and 2) outward transport of the chloride ions. The wetting process is the one by which water intrudes into the asphalt mastic from the surface and finally enters the functional additives first through various pathways (e.g., pores and defects) in the overlay and then through the micropores on the epoxy coating layer. Subsequently, CaCl_{2} turns from its combined state (i.e., solid CaCl_{2} inside the additives) to ionic state (Ca cations and Cl anions dissolved in water) in the additives. Thereafter, the chloride ions in water migrate to the surface of the overlay through the same pathways of water ingress, mainly by diffusion (due to the chloride concentration gradient inside the overlay). The schematic illustration of the two processes is shown in Fig. 1.
In reality, the overlay counters wetdry cycles due to weather variations over time. The wetdry cycles lead to water–air transport in the overlay, which is a problem of a twophase flow in porous media that obeys the law of conservation of mass. On the other hand, Fick’s second law is used to describe the transporting process of the chloride ions in the overlay, which is a nonsteady state diffusion due to the chloride concentration gradient between the inside and the surface of the overlay.
Wetdry cycles of the overlay
Sufficient water in the overlay is an essential precondition that makes the chloride ions being transported outward from inside of the additives to the surface of the overlay. The sources of water are rainfall in summer and snowfall in winter. To solve water–air phase transport in the overlay, the macroscopic mass conservation equation was used [3, 43, 44, 46]:
where ε_{p} is the porosity of porous media, ρ_{i} is the density of phase i, s_{i} is the volume fraction of phase i, and u_{i} is the volumetric flux (velocity vector) of phase i.
u_{i} is determined with the extended Darcy’s law as shown in Eq. 2 [5, 44]:
where k_{ri} is the relative permeability of phase i, u_{i} is the dynamic viscosity of phase i, k is the permeability of porous media, p_{i} is the pressure field of phase i, and g is the gravitational acceleration vector.
The pressure difference across the interface between the nonwetting phase (e.g., air) and the wetting phase (e.g., water) is called capillary pressure (p_{c}), and the pressure of the nonwetting phase is greater than the pressure of the wetting phase because of the curvature and interfacial tension of the interface (Brooks and Corey, 1964; [43, 44, 46]:
where \({p}_{{s}_{n}}\) is the pressure in the nonwetting phase, and \({p}_{{s}_{w}}\) is the pressure in the wetting phase.
In the Brooks and Corey model, p_{c} depends on phase saturation as shown in Eq. 4 (Brooks and Corey, 1964, 1966; [46]:
where \(\overline{{s }_{w}}\) is the effective volume fraction of the wetting phase, which is expressed as:
where \({s}_{w}\) is the volume fraction of the wetting phase, \({s}_{rw}\) is the residual volume fraction of the wetting phase, and \({s}_{rn}\) is the residual volume fraction of the nonwetting phase.
In the Brooks and Corey model, the relative permeability of the wetting phase (\({k}_{r{s}_{w}}\)) and the nonwetting phase (\({k}_{r{s}_{n}}\)) is expressed as Eqs. 6 and 7 (Brooks and Corey, 1964, 1966), respectively:
where \(\overline{{s }_{n}}\) is the effective volume fraction of the nonwetting phase, which is expressed as:
where \({s}_{n}\) is the volume fraction of the nonwetting phase.
In the “phtr” interface, there are two phases in the model: the water phase and the air phase. The “phtr” interface solves for the averaged volume fraction of water and air. The dependent variables are the volume fraction of water (s_{1}) and the volume fraction of air (s_{2}). The sum of the two volume fractions follows the volume constraint equation:
Equations 1 through 9 were implemented into the “phtr” interface of the COMSOL Multiphysics to simulate the water–air phase transport in overlay when exposed to weather variations.
Chloride transport
The transport of the chloride ions from highconcentration areas to lowconcentration ones in the overlay obeys Fick’s second law, which is expressed as Eq. 10 [9, 19, 23]:
where C is chloride concentration, t is time, D is chloride diffusion coefficient, and x is a position variable.
In the “tds” interface of the COMSOL Multiphysics, Eq. 10 is written as Eqs. 11 and 12:
where C_{i} is the concentration of species i, t is time, J_{i} is diffusive flux vector, R_{i} is a reaction rate expression for species i (R_{i} = 0 in this study), and D_{i} is the diffusion coefficient of species i.
Model assumptions
To simplify the modeling process, the following assumptions were proposed for the twodimensional axisymmetric model in this study:

1.
Asphalt mastic, and the additives are homogeneous, isotropic, and incompressible material.

2.
The overlay simulated in the model has been constructed and started to be used on May 1, 2012.

3.
The subbase layer under the overlay is an impermeable asphalt layer. Thus, no water or chloride ions flows into the subbase layer.

4.
The effect of gravity is neglected for water transport.

5.
CaCl_{2} is not deposited on the surface and pores of the overlay.
Model settings in COMSOL multiphysics
Model geometry
Figure 2 depicts the geometries of the overlay in the model. Figure 2a is the crosssectional image of an asphalt mixture specimen with 5.1 wt% CaCl_{2}zeolite/pepoxy #16. The dimension of the crosssection is 150 mm × 32 mm. The geometry of the model was pictured proportionally according to the original crosssectional image with the aid of the computeraided design and drafting software AutoCAD, as shown in Fig. 2b. The geometry in Fig. 2b was used to determine the thermal crackingrelated coefficient and fatigue crackingrelated coefficient of the chloride diffusivity of asphalt mastic. To simplify the geometry, the area in the red box in Fig. 2a and b was adopted for the simulations of water transport and chloride transport, as shown in Fig. 2c. In the Fig. 2b and c, the gray circles are CaCl_{2}zeolite/pepoxy particles, the gray irregular polygons are coarse aggregates, and the purple area is the asphalt mastic. Two diameters of the additives were used: 1 mm for the small additives and 2 mm for the big additives. Three asphalt mixture samples were pictured, and the one with the median and thus representative levels of areas of mastic, aggregates, and additives was selected for the modeling study. The software ImageJ was used to measure the areas of mastic, aggregates, and additives in each of the samples, and Table S1 shows the estimated values.
The coarse aggregates in the overlay were assumed to be highly impermeable and thus had little influence on the simulation results of water transport and chloride transport. Therefore, the domain of the coarse aggregates in the geometry was excluded from the simulations.
Phase transport in porous media (phtr) interface
Phase and porous media transport properties of asphalt mastic and additives
In the Brooks and Corey capillary pressure model, the entry capillary pressure (P_{ec}) can be estimated using Eq. 13 [16, 51]:
where σ is interfacial tension of a fluid, θ is contact angle between the fluid and the capillary tube, and r_{c} is radius of the capillary tube. The σ of water is related to pavement temperatures with a unit of Kelvin (K) [10]. The temperature range for the data in the literature is from 5 °C to 45 °C. The plot of the relationship between the σ of water and pavement temperatures is shown in Figure S1. The contact angles of water to the epoxy layer and the asphalt mastic were measured using water contact angle test. For the pores in the epoxy layer on the surface of the additive, r_{c} is the average value of the pore’s radii measured from scanning electron microscope (SEM) images. For the pores in asphalt mastic, the average r_{c} was calculated with Eq. 14 [41]:
where AV% is air void percentage or porosity in asphalt mastic, and D_{75} is the particlesize diameter corresponding to 75% on the cumulative particlesize distribution curve of aggregates (Figure. S2). In the literature, the equation was determined based on the channel theory [38], which is suitable for asphalt pavement with an air void content of higher than 4%. According to Eq. 3, the positive value of the calculated P_{ec} of water in the pores of asphalt mastic was used in the model.
In the Brooks and Corey capillary pressure model, the pore size distribution index of the epoxy layer (λ_{pepoxy}) on the surface of the additive was calculated using Eq. 15:
where W_{i} is the weight fraction of pores and W_{i} = N_{i}*A_{i}/∑(N_{i}*A_{i}), N_{i} is the number of pores, and A_{i} is the surface area of pores. N_{i} and A_{i} were obtained from SEM images (e.g., Fig. 3) of the surface of the additive with assisting with AutoCAD. An example of calculating the λ_{pepoxy} for Fig. 3 is shown in Table S2. Three SEM images were analyzed, and the results are shown in Table S3.
The pore size distribution index of asphalt mastic (λ_{pmastic}) was calculated using Eq. 16 [33]:
where k_{mastic} is the water permeability of asphalt mastic. Equation 16 was converted from Eq. 2, and the parameters in the equation were determined based on the results of the experiments conducted by the authors of the literature [33].
For simplicity, the density (ρ) of water was approximated to be 1000 kg/m^{3} for all pavement temperatures. The dynamic viscosity (μ) of water and the ρ and μ of air entailed three equations related to pavement temperatures with a unit of K [27, 42]. In the literature, the temperature range for the μ of water is from 0 °C to 60 °C at atmospheric pressure, and the equations for the ρ and μ of air are valid for a temperature range from 20 °C to 40 °C and a pressure of 1000 bm. The plots of the relationship between the properties and pavement temperatures are shown in Figures S3, S4 and S5. The determination of pavement temperature will be described in the Sect. 3.3.1. The residual saturation (s_{r}) of a water–air system was assumed to be 0 in this study.
The (average) porosity of the epoxy layer on the additive was 2.1%, which can be found in our previous study [53]. The water permeability of the epoxy layer was assumed to be a very small positive value, which was expressed as 'eps [m^{2}]' in COMSOL Multiphysics.
The values or equations of the terms mentioned above are shown in Table 1.
Initial value and boundary condition
The initial value of the water volume fractions (s_{0,s1}) in the model was assumed to be 0.
The wetdry cycles of the overlay are mainly affected by precipitations, which are the source of water driving chloride transport. Thus, the hourly data of precipitations including rainfalls and snowfalls from May 1, 2012 to November 27, 2020 was introduced into the model. The weather data at Pullman, Whitman County, Washington was obtained from “AgWeatherNet” of Washington State University (https://weather.wsu.edu/). In the model, the boundary condition on the top surface of the geometry was that either the water volume fraction was 1 for precipitation or the water volume fraction was 0 for no precipitation, which can be implemented using an “Interpolation” function of COMSOL Multiphysics.
Transport of diluted species (tds) interface
Transport properties of asphalt mastic
For a onedimensional finite diffusion problem, assuming the chloride concentration at surface is constant, the analytical solution of the partial differential Eq. 10 can be expressed as Eq. 17 [9, 19, 22, 23]:
where C(x,t) is the chloride concentration at a location x at a time t, C_{s} is the chloride concentration at surface, C_{0} is the initial chloride concentration at location x, erf(z) represents the error function:
Thus, a steadystate chloride diffusion coefficient (D_{ss}) in asphalt mastic with a certain C_{s} at a given temperature can be estimated with Eqs. 17 and 18 incorporating a mastic diffusion test, which will be described in the Sect. 4.2.
In reality, the D of asphalt pavement is nonsteadystate due to the variations of pavement temperature, chloride concentration, the depth, width, and number of thermal cracks and fatigue cracks, which have been considered in the model for more accurate results.
A previous study [37] reported that the D of cement concrete nonlinearly decreases with increasing the square root of C. The plot of the relationship between D and (C + 0.01 mol/m^{3}) of cement concrete with a watercement ratio of 0.6 is shown in Fig. 4, which was used to represent the relationship between D and C of the studied asphalt overlay, considering their comparable level of chloride diffusivity. A negligibly low background concentration of chloride, 0.01 mol/m^{3}, was introduced to represent chlorides potentially sourced from aggregates and binder, aimed to avoid the nonconverged calculation in COMSOL Multiphysics. The watercement ratio of 0.6 was selected because the D of this cement concrete was the closest to the D of asphalt mixture with 5.1 wt% CaCl_{2}zeolite/pepoxy #16 [6, 29, 40, 52]. Crelated coefficient to D (A_{c}) is expressed as:
where C_{s} is the surface chloride concentration used in the mastic diffusion test.
Pavement temperature also has significant influence on D. The effect of pavement temperature on D can be well expressed by the Arrhenius equation [50]:
where D_{0} is maximal diffusion coefficient, E_{a} is activation energy for diffusion, R is the universal gas constant, and T is the absolute temperature of asphalt pavement. According to [30], the D and T of cement concrete with a watercement ratio of 0.6 had the relationship as shown in Fig. 5, which can be used for the relationship between D and T of asphalt pavement. Therefore, Trelated coefficient to D (A_{T}) is expressed as:
where T_{t} is the testing temperature of the mastic diffusion test.
The hourly data of pavement temperature T from May 1, 2012 to November 27, 2020 was calculated with Eq. 22 adapted from [15], and it was implemented using the “Interpolation” function of COMSOL Multiphysics.
where T_{a} is air temperature, R_{s} is daily solar radiation, and d is depth from pavement surface. Arguably the functional pavement might retain more moisture than conventional asphalt pavement and thus feature a slightly different relationship than Eq. (22); but this difference is not considered at this stage of the modeling study.
Moreover, the initiation and propagation of the cracks in cement concrete are known to increase D [8, 12, 45]. In this study, considering the location (Pullman, WA), one can assume that thermal cracking and fatigue cracking were the dominant mechanisms for the development of cracking in the overlay over time. Other risks, such as rutting, raveling, stripping, alligator cracking, were not considered. Similar to cracked cement concrete, the thermal cracks and fatigue cracks in asphalt pavement increase D as well. The thermal cracking mainly results from low temperature environments, while the fatigue cracking is caused by repeated traffic loadings at moderate ambient temperatures.
There is no study reporting how D evolves in asphalt pavement over time as a function of thermal cracking or the relationship between D and the level of thermal cracking. To determine the Dtime relationship under the effect of thermal cracking in asphalt pavement, a physical simulation was conducted using COMSOL Multiphysics incorporating a set of thermal crackingtime data adapted from Fig. 13 [14]. Four data points where the thermal cracking percentage of the asphalt pavement was 0% (0 year), 27% (4 years), 38% (9 years), and 59% (11 years) were selected from [14] for this study. According to the definition of the thermal cracking percentage in the reference, the number of the cracks per 150 m length of the survey section was 0, 13.5, 19, and 29.5 corresponding to 0%, 27%, 38%, and 59%, respectively. For a 150 mm length overlay, the number of cracks at 0%, 27%, 38%, and 59% was 0, 0.0135, 0.019, and 0.0295 respectively. In other words, the corresponding number of the 150mm section per crack was 0, 74, 53, and 34, respectively. To estimate D at different thermal cracking levels, the variation of the average C (C_{ave}) at the bottom of the simulation geometries over time was first determined with Eq. 23:
where N_{n} is the number of the noncracking 150mm section, C_{n} is the chloride concentration at the bottom of the geometry of a noncracking 150mm section, N_{c} is the number of the 1crack 150mm section, and C_{c} is the chloride concentration at the bottom of the geometry of a 1crack 150mm section. C_{n} and C_{c} can be obtained from the simulation described below.
Figure 6 shows the geometries used to determine C_{n} and C_{c} in COMSOL Multiphysics. The geometry in Fig. 6a is the same as the geometry used in the Sect. 3.1. The geometries in Fig. 6bd include inverted triangle cracks with different depths and widths. The circles in the geometry were used as traditional aggregates instead of the additive. For simplicity, the model had the following assumptions:

1.
the max crack depth of the overlay was 30 mm at 59%;

2.
the max crack width was half of the max crack depth;

3.
the ratio of two max crack depths was proportional to the ratio of the two thermal cracking percentages.
In the “tds” interface, the aggregates were excluded from computing. The D_{ss} in asphalt mastic had been determined with the mastic diffusion test. It was assumed that the initial value of the chloride concentration in the asphalt mastic was 0. The C_{s} on the top and crack surfaces of the geometry was 560 mol/m^{3}. The output was the relationship between the chloride concentration (e.g., C_{n} and C_{c}) at the bottom of the geometry and time. Eleven points were selected to calculate C_{n} and C_{c} over time. The simulated C_{ave} over time at each cracking level was determined with Eq. 23. Meanwhile, the chloride transport follows the Fick’s second law. Thus, D at each cracking level was estimated where the Ctime curve calculated according to Eqs. 17 and 18 matched best with the C_{ave}time curve simulated with COMSOL Multiphysics. Figures S6, S7, S8 and S9 show the simulated and calculated Ctime curves and the goodness of fit at different cracking levels. The Dtime relationship related to thermal cracking is shown in Fig. 7. Therefore, thermal crackingrelated coefficient to D (A_{tc}) is expressed as:
The fatigue crackingrelated coefficient to D (A_{fc}) was determined with the same method of A_{tc}. The datapoints of equivalent single axle loads (ESALs) and crack properties (e.g., crack rate and crack width) used for this study were adapted from [21]. Lane #9 was a 90mm asphalt layer constructed on a 153mm inplace stabilized soil cement (10%) subbase under a 102mm layer of crushed stone. The relationship between ESALs and crack rate is shown in Figure S10, which was adapted according to the curve of Lane #9 in Fig. 7.2 in the main text of a reference [21]. The ESALscrack width relationship as shown in Figure S11 was obtained based on Figures S12 and S13, which were adapted according to the figure of the sample No.25 with a load of 7.25 KN in Appendix D of the reference [21] and the single load equivalency factor (LEF) data of a design guide (Transportation Officials, 1993) respectively. The crack widthcrack depth relationship as shown in Figure S14 was adapted from Fig. 5.22 of a reference [21]. The timeESALs datapoints at State Route 195 near Pullman in Washington State from 2010 to 2019 as shown in Table S4 were obtained from the longterm pavement performance (LTPP) program of Federal Highway Administration, United States Department of Transportation.
It was assumed that the lane width of the 150 mmsection model was 3.66 m and all fatigue cracks were transverse fatigue cracks. According to the equation in Figure S10, Table S4, and Eq. 25, the length of the transverse fatigue crack (L) in a 150 mmsection model was determined. L then was converted to the number of 150mm sections at different years when one transverse fatigue crack existed and crossed the whole lane of the model, as shown in Table S5. The width of the transverse fatigue crack was determined according to the equation in Figure S11, and the total depth of the transverse fatigue crack was determined according to the equation in Figure S14. It was assumed that the transverse fatigue crack started cracking from the bottom of a 50mm binder course under the overlay. Thus, the depth of the transverse fatigue crack in the model was calculated using the total depth minus 50 mm. The depth and the width of the transverse fatigue crack are shown in Table S6. In COMSOL Multiphysics, the geometries used to determine C_{n} and C_{c} related to the fatigue cracking of asphalt pavement are shown in Fig. 8. The settings in the “tds” interface for fatigue cracking and the calculating procedures were similar with that for thermal cracking.
Figures S15, S16, S17 and S18 show the simulated and calculated Ctime curves and the goodness of fit at different years for fatigue cracking. The Dtime relationship related to fatigue cracking is shown in Fig. 9. Therefore, A_{fc} is expressed as:
Combining D_{ss}, Eq. 19, Eq. 21, Eq. 24, and Eq. 26, the final equation for the D of asphalt mastic (D_{mastic}) in the tds interface was expressed as:
In addition, the threshold of moisture volume fraction (s_{1}) in concrete for chloride diffusion was 20% [35]. As such, D_{mastic} was expressed as Eq. 27 when s_{1} was equal or greater than 0.2, while D_{mastic} was 0 when s_{1} was less than 0.2, in which s_{1} was the result of the “phtr” interface.
The parameters related to D_{mastic} mentioned above are provided in Table 2.
Transport properties of additives
The D in the additives in the presence of water (D_{additive}) was estimated with the cooperation of the curve of chloride concentration vs. time in water from an additive immersion test and the curve of chloride concentration vs. time in water simulated with the tds interface of COMSOL Multiphysics. D_{additive} was determined when the two curves had the best fit. The additive immersion test is described in the Sect. 4.2.
In the “tds” interface, the geometry used to simulate the Ctime curve and the original image of the layer of the compacted asphaltcoated additives in the beaker for the test are shown in Fig. 10a and b, respectively. The total height and the width of the geometry were 132.5 mm and 120 mm respectively, which were the height and the width of the water in the beaker. In the geometry, the gray part is water and the rest represent the compacted asphaltcoated additives with a height of 17.5 mm, which was pictured using AutoCAD according to Fig. 10c. Figure 10c was the processed image according to Fig. 10b using ImageJ.
In the settings of transport properties, D_{additive} was a value that needed to be solved to achieve the best match between the experimental vs. simulated chloride concentration curves, and the D in water was 1.2E9 m^{2}/s derived from the reference [13]. The initial value of the C in water was 0, and the initial value of the C in the additives (C_{initial}) was another value to be solved for a match between two chloride concentration curves.
The simulation output was the simulated Ctime curve. Simulation results showed the two curves (Figure S19) had the best match when D_{additive} was 2E11 m^{2}/s and C_{initial} was 1500 mol/m^{3}. The data points before 71 h in Figure S19b were removed due to the unstable diffusion at the very early stage of the test. Same as the expression of D_{mastic}, D_{additive} was affected by s_{1} from the result of the “phtr” interface. D_{additive} was equal to 2E11 m^{2}/s when s_{1} was equal or greater than 0.2, while D_{additive} was 0 when s_{1} was less than 0.2.
Initial values and boundary conditions
In reality, CaCl_{2} exists in a liquid state when water is sufficient, otherwise it is in a solid state. The C in the additives was supposed to be affected by s_{1} from the water transport model as well, but this effect was difficult to be simulated in COMSOL Multiphysics at this stage. As such, this study assumed that in a very short time the C in the additives reached to the maximum value (7325 mol/m^{3}), which was calculated based on the average volume of the additives. For simplicity, the amount of available water within the sphere of influence of the additives was assumed to be equivalent to the volume of the additives. Subsequently, the C in the additives decreased over time until the system reached a balanced C. To this end, the initial value of the C in the additives in the model was set to be 7325 mol/m^{3}, while the initial value of the C in the asphalt mastic was assumed to be 0 mol/m^{3}.
The C in asphalt mastic and on the top surface of the overlay may gradually increase over time until it reaches to the maximum C. In the service environment, part of the salt on overlay surface is washed away by rainfalls and snowfalls. There is no study reporting how much of the salt is removed by rainfalls or snowfalls. Based on field observations, this study assumes that approximate 5% and 50% of the C on the top surface of the overlay remained after the hour of rainfall and snowfall, respectively. The boundary condition related to C on the top surface of the geometry was implemented by the functions of “Flux” and “Interpolation” in COMSOL Multiphysics. The snowfall season was defined as from November 13 to February 25 every year in Pullman, Washington.
Figure 11 shows the flowchart of the water transport model, the chloride transport model, and how the output of the water transport model affects the chloride transport model.
Mesh and study
The asphalt mastic, additives, and top surface boundary were meshed according to their sizes. The free triangular mesh was selected as the mesh of the model. This study selected the normal size, fine size, and finer size of mesh for the asphalt mastic, additives, and top surface boundary, respectively.
Both the water transport timedependent model and the chloride transport timedependent model were studied for 44,280 h (approximate 5.05 years) with an interval of 1 h. This is because in practice such an overlay would be replaced after 5 years in service and some of the model assumptions would no longer hold true if the asphalt overlay serves beyond 5 years in the field where moisture damage, UV aging, etc. start to notably compromise the integrity of the overlay.
Experimental
Materials
In this study, the same materials including asphalt binder and aggregates less than 4.75 mm, the same mixing design, and the same mixing and compaction procedure as in our previous study [53] were used to prepare asphalt mastic. The aggregates greater than 4.75 mm were not included in asphalt mastic. The air void of the asphalt mastic samples was 7% ± 0.5%.
The same fabrication procedure and materials including CaCl_{2}, zeolite, epoxy, halloysite nanoclay, and acetone as in our previous study [53] were used to prepare CaCl_{2}zeolite/pepoxy #16. The calculated density of CaCl_{2}zeolite/pepoxy #16 is 1.98 g/cm^{3}.
Calcium nitrate tetrahydrate (Ca(NO_{3})_{2}·6H_{2}O) was purchased from Thermo Fisher Scientific.
Testing methods
To evaluate the diffusion behavior of chloride ions in asphalt mastic, a mastic diffusion test was adopted with a customdesigned apparatus as shown in Fig. 12. A mastic disk was sandwiched between two plastic cells, which were used to store the electrolyte solutions of CaCl_{2} and Ca(NO_{3})_{2}, without any leakage. Mastic disk samples with a thickness of 12 mm ± 0.5 mm were cut from the prepared asphalt mastic samples and polished with sandpapers. The two cells were filled with DI water to saturate the mastic disk samples for 24 h before the mastic diffusion test, to minimize the effect of wick action (water transport) during ionic penetration [22]. Then, DI water was replaced by 0.28 mol/L (3 wt%) CaCl_{2} electrolyte solution in the left cell and 0.28 mol/L Ca(NO_{3})_{2} electrolyte solution in the right cell. To avoid the evaporation of the water, the cells were covered by rubber caps. The variation of the chloride concentration over time in the right cell was periodically measured using a laboratoryprepared Ag/AgCl chloride sensor, which has been calibrated using standard NaCl solutions before each of measurements. The testing temperatures were 15 °C. The results of the test are shown in Figure S20. It is assumed that the D has been stable at the last point of the curve in Figure S20b, therefore D_{ss} was equal to 1.14E12 m^{2}/s.
The insidetooutside diffusion of chloride ions in CaCl_{2}zeolite/pepoxy #16 was investigated by an additive immersion test, in which the manually compacted mixture of 133.6 g CaCl_{2}zeolite/pepoxy #16 and 7.2 g asphalt (5.1 wt% binder content) was soaked in 1.5 L DI water at 15 °C as shown in Fig. 13. The mixing temperature was 145 °C, and the porosity of the mixture was 40%. The chloride concentration was monitored periodically by the same method used in the mastic diffusion test.
A falling head test [1], which is suitable for dense asphalt mixture samples with a low permeability, was used to evaluate k_{mastic}. Figure 14 shows the schematic diagram of the setup of the falling head test. The diameter and thickness of mastic disk samples for the falling head test were 58 mm ± 0.5 mm and 12 mm ± 0.5 mm. The asphalt mastic samples were saturated for 24 h before the test. The upper and lower levels of water in the standpipe, the volume of water passing through a sample, and the time required for the water passing through the sample were measured. The equations used to calculate k_{mastic} are expressed as Eqs. 28 and 29. The parameters and results of the test is shown in Table S7.
where K_{mastic} is the hydraulic conductivity of asphalt mastic, H_{1} is the initial height of water in standpipe, H_{2} is the final height of water in standpipe, A_{1} is the crosssectional area of sample, A_{2} is the crosssectional area of standpipe, L_{s} is the length of sample, and t is the elapsed time of test.
where μ_{w} is the dynamic viscosity of water at 25 °C, ρ_{w} is the density of water, and g is the gravitational acceleration.
Water contact angle test was used to measure the values of θ of water to the epoxy layer and the asphalt mastic. An example of water contact angle tests on asphalt mastic is shown in Fig. 15. The results of the water contact angle test are shown in Table S8.
SEM (JEOL JXA8500F Field Emission Electron Microprobe) was adopted to measure the radii of the micropores in the epoxy coating layer of CaCl_{2}zeolite/pepoxy. More details can be found in our previous study [53].
Results and discussion
Water transport
Figures 16, 17, and 18 depict the simulation results of the water transport model. Figure 16a16f display the representative images of the evolution of the s_{1} in the overlay when exposed to varying precipitation and temperature. At the very beginning (Fig. 16a, t = 0 h), the s_{1} regardless of the location was in the model was equal to 0 because there was no rainfall or snowfall on the overlay. At 72 h (Fig. 16b), the s_{1} increased to different levels, depending on the location in the model. Two rainfalls between 0 and 72 h accounted for the increase in the s_{1}, which could be observed in Fig. 17. Figure 17a illustrates the temporal evolution of the s_{1} on the top surface of the overlay up to 44,280 h. To see the lines clearly, a local enlarged figure (Fig. 17b) up to 100 h was plotted. Figure 17b reveals that the first rainfall appeared between 56 and 60 h and the second one rained between 63 to 65 h. In contrast, the drying process between 60 and 63 h and after 65 h was responsible for the different values of the s_{1} at the different locations. Water could only evaporate through the top surface, which explains why the s_{1} near the top surface had lower values than that around the middle and the bottom. At 96 h (Fig. 16c), the s_{1} changed slightly due to the corporation of the third 1 hour rainfall (as shown in Fig. 17b) and the drying process after 90 h. Between 96 and 480 h, there was no rainfall. Therefore, the s_{1} gradually decreased to low values over time, as shown in Fig. 16d. Figure 16e reveals that all the s_{1} in the overlay reached to 1 at 840 h, resulting from sufficient amount of rains falling on the surface before 840 h. The last status of the overlay about s_{1} for this study was shown in Fig. 16f (44,280 h, about 5.05 years). Figure 18 illustrates the temporal evolution of the s_{1} at the center of the geometry (in asphalt mastic), which was consistent with the results shown in Figs. 16 and 17.
It should be noted that pavement temperature is a variable in several equations of the model’s inputs, including the entry capillary pressure of water in pores of asphalt mastic and functional additive, the dynamic viscosity of water and air, and the density of air. Pavement temperature affects the water fraction through influencing these parameters.
Chloride Transport
Figure 19 displays the representative images that illustrate the temporal evolution of the C in the overlay when exposed to wetdry cycles, varying pavement temperature, pavement thermal cracking, and pavement fatigue cracking. The wetdry cycles in the model were expressed by s_{1} (> = 0.2 for a wet condition and < 0.2 for a dry condition), which resulted from the prediction of the water transport model. The C in asphalt mastic was 0 mol/m^{3} from 0 to 48 h as shown in Fig. 19ac, because the overlay was in a dry condition (s_{1} < 0.2) as shown in Fig. 17b. After the first rainfall, the s_{1} in asphalt mastic was greater than 0.2, thus the C gradually increased over time (as shown in Fig. 19df) until the s_{1} decreased to be lower than 0.2 due to water evaporation (when there was no rainfall on the pavement). Figure 19fh indicate that the C in asphalt mastic remained unchanged from 144 to 240 h. The reason was that the D_{mastic} was equal to 0 when the s_{1} was lower than 0.2 during that period as shown in Fig. 18b. At 4272 h (Fig. 19i), the C in asphalt mastic greatly increased while the C in functional additives significantly decreased, relative to those before 240 h. Chloride ions in the overlay moved gradually from the locations with high concentrations to the locations with low concentrations. The last status of the C in asphalt mastic and functional additive for this study was shown in Fig. 19j (44,280 h, about 5.05 years). In this figure, the C in asphalt mastic near the top surface (the blue area) was visibly lower than that at the middle and bottom, which is due to dilution and removal of chloride ions from the top surface by precipitation of rain or snow. The C at deeper areas has higher values, mainly because the chloride ions in the additives at deeper areas need to travel longer distances to reach the top surface and their transport takes more time than the time needed by the dilution.
In addition to moisture availability (due to precipitations), pavement temperature, pavement thermal cracking, and pavement fatigue cracking were three coefficients in the equation of the diffusion coefficient of asphalt mastic, all of which influenced the change of the C in asphalt mastic and functional additive accordingly.
Although Fig. 19 qualitatively illustrates the overall variation of the C in the overlay, it is desirable to quantitatively monitor the continuous evolution of the C in functional additive and asphalt mastic and on the top surface, respectively. Therefore, Figs. 20, 21, and 22 were plotted accordingly.
Figure 20 contrasts the temporal evolution of the C in three functional additives at three different locations: top, middle, and bottom. They all shared a same trend and gradually decreased over time. In some periods the curve of the C exhibited a stable plateau phase due to that the Ds of asphalt mastic and functional additive were 0 m^{2}/s because the s_{1} was equal to 0. The slight differences among the three Cs resulted from the surrounding environment of the three additives, such as the quantity of functional additives and aggregates nearby, the shape of adjacent aggregates, and the cumulative time when s_{1} was equal or greater than 0.2.
Figure 21 illustrates the temporal evolution of the C on the top surface of the overlay. Five points on the top surface were selected, and the averaged C was plotted. It can be seen that the C on the top surface was reduced following the pattern of precipitations shown in Fig. 17. The amount of the reduction in C depended on the type of precipitation; the reduction in C was not exact 95% and 50% for rainfalls and snowfalls respectively. This is because that the setting in COMSOL Multiphysics for the chloride removal at top surface was the boundary condition ‘Flux’, instead of directly removing the available C. Arguably, the plot shown in Fig. 21 better represents the reality.
The effective antiicing concentration of Cl^{−} in water for different temperatures is provided in Table 3, which was converted based on the phase diagram for the waterCaCl_{2} system as shown in Figure S21. One can assume that there is a linear relationship between temperature and the CaCl_{2} concentration (in wt%) when the temperature is higher than 10 °C. The equation is expressed as:
where C_{CaCl2} is the effective mass percentage of CaCl_{2} for antiicing at T and T is temperature. The temperature of asphalt pavement near surface is usually 1 to 2 °C higher than air temperature during winter [34], and snow events generally occur when air temperature is lower than 0 °C in winter. As such, this study assumes that snow events occurred when the temperature of the overlay dropped below 2 °C, and the historical hourly pavement temperature data (near Pullman, WA) below 2 °C were thus used to calculate the statistical parameters (e.g., percentiles mentioned later).
Considering the compressed or vibrated pavement surface due to the effects of loading and friction from moving vehicles [54] and the average size of functional additive being 2 mm, this study assumes that the C in the overlay at a depth of 4 mm approximately determines whether the antiicing function is effective or not. The averaged C of five points in asphalt mastic at the depth of 4 mm was plotted as shown in Fig. 22. At the beginning, the C increased rapidly and reached to 1,118 mol/m^{3} within two months. The fast diffusion resulted from the big difference of the C between asphalt mastic and functional additives. Subsequently, the C gradually decreased over time, due to chloride removal on the top surface. According to the plot in Fig. 22 and the data in Table 3, the antiicing function of the overlay was fully effective in 17,520 h (2 years) and 43,800 h (5 years) for the minimum pavement temperature above 3.4 °C and 2.4 °C, corresponding to its lowest C of 808 mol/m^{3} and 569 mol/m^{3}, respectively. These two pavement temperatures translate to 97.4percentile and 96.3percentile of historical hourly pavement temperature near Pullman, Washington, respectively. In other words, this specific 16mm thick functional overlay could prevent the surface from icing, for 97.4% of the snow events within 2 years and for 96.3% of the snow events within 5 years. The simulation results agree well with the estimated antiicing life using Eq. 3.2 in our previous study [53]. It should be noted that these effective antiicing service lives are conservatively calculated, because beyond them there are still residual friction benefits of this functional overlay, because part of the overlay surface would still be able to keep the moisture in a liquid instead of frozen state.
Sensitivity analysis
Sensitivity analysis was conducted to investigate how the C of the overlay was affected by variations in the overlay thickness (h) and D_{mastic}, because the simulation suggested that these two were dominant factors in determining the C of the overlay. With other parameters unchanged, the simulation model was rerun when the thickness of the geometry in Fig. 2c or D_{mastic} was increased or decreased by 20%.
The results of the water transport model under the new settings were very similar to or the same as those under the original setting, while the changes in h and D_{mastic} had more pronounced effects on the results of the chloride transport model. Under the new settings, the values of the C on top surface or in functional additives at a same point of time slightly changed from those under the original setting. Figure 23a depicts the variation of the C over time at 4 mm below the top surface as a function of the h and D_{mastic} settings, over the simulated 43,800 h (5 years). In the first 1,500 h, as shown in Fig. 23b, the C decreased with reducing h or D_{mastic}, and a combined action of reducing both h and D_{mastic} further decreased the C. In contrast, the C increased with increasing h or D_{mastic}, and the C further increased when h and D_{mastic} increased simultaneously. For a longterm evaluation (as shown in Fig. 23a), interestingly, the reduction of D_{mastic} or both D_{mastic} and h resulted in slight increase in the C on the top surface of the overlay, likely due to a smaller amount of chloride ions being removed by precipitations. The other six changes of the parameters slightly decreased the C, indicative of multiple mechanisms at work.
Table 4 shows the predicted antiicing life of the functional overlay for the different settings, assuming the goal is to prevent the surface from icing for 96.3% of the snow events. Under the original setting (h = 16 mm, D_{mastic} = Eq. 27), the antiicing function of the overlay is predicted to be effective for the overlay temperature above 2.4 °C, and the corresponding Cl^{−} concentration was 557 mol/m^{3}, as shown in Table 3. Compared with the original setting, the antiicing life of the overlay increased by 2% when D_{mastic} decreased by 20% and increased by 1% when both D_{mastic} and h decreased by 20%. On the other hand, it decreased by 6%, 9%, 9%, 10%, 1%, and 11% when D_{mastic} increased by 20%, h decreased by 20%, h increased by 20%, both D_{mastic} and h increased by 20%, D_{mastic} decreased by 20% and h increased by 20%, and D_{mastic} increased by 20% and h decreased by 20% respectively.
Indirect validation of the predictive model
This study is limited to exploratory investigation. The finite element methodbased model was built by following first principles of water and chloride transport, incorporated experimental results from laboratory, and relying on case study data and assumptions where necessary. For the prediction of antiicing life of the overlay with the functional additive, direct validation of this model remains challenging due the lack of directly usable data in the published domain and the lack of resources to conduct longterm laboratory or field investigation for longevity validation. Nonetheless, this section describes encouraging results that indirectly validate the predictive model established thus far.
Figure 24 displays the temporal evolution of water fraction on the top surface vs. at center point of the overlay, and temporal evolution of chloride concentration at center point of the overlay vs. in an additive near center point, up to 1500 h. The variations of water fraction on the top surface of the overlay closely matches the pattern of how precipitation occurred on an hourly basis. It can be clearly seen that the water fraction at the center point rapidly increases during precipitations, whereas it gradually decreases due to natural water evaporations when there is no precipitation. Moreover, the chloride concentration at the center point increases over time when the water fraction is equal to or greater than 0.2, whereas the chloride concentration remains unchanged when the water fraction is less than 0.2. In contrast, the chloride concentration in the additive decreases over time when the water fraction is equal to or greater than 0.2, whereas it remains unchanged when the water fraction is less than 0.2. These predicted trends for both the water fraction and the chloride concentration at different points in the overlay are reasonable and consistent with the reality.
In our previous study [53], the results of the fogfreezing tests indicated the friction coefficient of the asphalt mixture with 5.1 wt% additive #16 decreased from 0.5 at 3.9 °C to 0.4 at 9.4 °C. This can be explained by the temporal evolution of the chloride concentration at a depth of 4 mm below the surface of the overlay. Figure 22 illustrates the predicted profile of the chloride concentration in the functional overlay at 1248 h. The model results indicate that the lowest temperature for an effective antiicing of this overlay was about 4.6 °C (corresponding to its highest chloride concentration of 1,118 mol/m^{3} on the plot, see Fig. 23a). At this temperature or above, all the water remained a liquid state and would not turn into ice. Thus, the antiicing function was fully effective at 3.9 °C, which explains the high friction coefficients from the fogfreezing experiment at that temperature. It is apparent that 1,118 mol/m^{3} would not be sufficient to keep all the water (at the overlay surface) at a liquid state at 9.4 °C; instead, a minimum C of 2,400 mol/m^{3} is needed, by the theoretical evaluation based on the eutectic curve (see Figure S21). The simulation result at 1248 h revealed the chloride concentration at some locations at the 4mm depth was higher than 2,400 mol/m^{3}, as shown in the areas marked with purple ellipses in Fig. 25. These localized high chlorideconcentration locations help explain the experimental result at 9.4 °C. Because not all the water on the sample surface was prevented from icing, the friction coefficient reduced to 0.4 at 9.4 °C from 0.5 at 3.9 °C.
Conclusions
This study employed a Finite Element Method based software to predict the antiicing life of a thin overlay of asphalt pavement with saltstorage additive, by simulating chloride transport in the overlay when exposed to the changes in weather conditions in terms of precipitation and temperature and pavement conditions in terms of thermal cracking and fatigue cracking. The FEM based simulation included water transport modeled with the “phtr” interface and chloride transport modeled with the “tds” interface of the software. The output of the water transport model was water fraction, which was used as one input parameter of the chloride transport model. The water fraction plays a crucial role in controlling the chloride diffusion coefficients of asphalt mastic and functional additive.
The results of the water transport model show the temporal and spatial evolutions of the water fraction in the overlay as a function of (historical) precipitation and pavement temperature. The water fraction increased rapidly during precipitations, whereas it gradually decreased due to water evaporation through the surface of the overlay when there was no precipitation. It was easier to lose water at the locations near the surface of the overlay than that around the middle and bottom. Pavement temperature was a variable affecting many parameters of the model, thus affecting the water fraction. These parameters included entry capillary pressure of water in pores of asphalt mastic and functional additive, dynamic viscosity of water and air, and density of air.
For the chloride transport model, the chloride concentration in asphalt mastic highly depended on the value of the water fraction, i.e., moisture availability limited the Cl^{−} diffusion coefficients of asphalt mastic and functional additive. The two Cl^{−} diffusion coefficients were equal to 0 when the water fraction was lower than 0.2. As the water fraction was equal to or higher than 0.2, the Cl^{−} diffusion coefficient of asphalt mastic was defined as a function of the surrounding chloride concentration, pavement temperature, pavement thermal cracking, and pavement fatigue cracking, and the Cl^{−} diffusion coefficient of functional additive remained constant. Chloride ions migrated from highconcentration locations to lowconcentration ones when the water fraction was equal to or higher than 0.2, until a balanced concentration profile was obtained in the system. The chloride concentration in asphalt mastic increased over time and the chloride concentration in functional additive decreased over time when the water fraction was equal to or higher than 0.2, otherwise it remained unchanged. Due to chloride removal on the top surface, the chloride concentration near the top surface was lower than that around the middle and bottom of the overlay.
Although the model considered the effect of pavement thermal cracking and fatigue cracking on increasing the Cl− diffusion coefficient of asphalt mastic, other pavement distresses (moisture damage, rutting, longitudinal cracking, UV aging, etc.) and the interactions among these distresses were not included. Additional research is needed to further improve this model to allow better representation of the field behavior of the overlay over the long run, under various service environments.
This study revealed that the antiicing function of the overlay was fully effective in 2 years and 5 years for the minimum pavement temperature above 3.4 °C and 2.4 °C, corresponding to its lowest chloride concentration of 808 mol/m^{3} and 569 mol/m^{3} at 4 mm depth of the 16mm overlay, respectively. These two pavement temperatures translate to 97.4percentile and 96.3percentile of historical hourly pavement temperature near Pullman, WA, respectively. This was consistent with the estimated effective antiicing life (5.0 years for the sample with 5.1 wt% additive #16) based on Eq. 3.2 in our previous study [53]. The current setting of the overlay thickness (16 mm) and the Cl^{−} diffusion coefficient of asphalt mastic (Eq. 27) also predicts a 96.3percentile effective antiicing life of 5.00 years, whereas the sensitivity analysis revealed that 20% changes in the overlay thickness or the Cl^{−} diffusion coefficient of asphalt mastic would change this service life to 4.55 to 5.09 years.
Availability of data and materials
The datasets generated and/or analyzed during the current study are available upon request.
Abbreviations
 APSSA:

Asphalt pavement with saltstorage additive
 phtr:

Phase transport in porous media
 tds:

Transport of diluted species
 SSA:

Saltstorage additive
 CaCl_{2}zeolite/pepoxy:

Laboratoryprepared functional additive
 CaCl_{2} :

Calcium chloride
 ε_{p} :

Porosity of porous media
 ρ_{i} :

Density of phase i
 s_{i} :

Volume fraction of phase i
 u_{i} :

Volumetric flux (velocity vector) of phase i
 k_{ri} :

Relative permeability of phase i
 u_{i} :

Dynamic viscosity of phase i
 k:

Permeability of porous media
 p_{i} :

Pressure field of phase i
 g:

Gravitational acceleration vector
 p_{c} :

Capillary pressure
 \({p}_{{s}_{n}}\) :

Pressure in the nonwetting phase
 \({p}_{{s}_{w}}\) :

Pressure in the wetting phase
 \(\overline{{s }_{w}}\) :

Effective volume fraction of the wetting phase
 \({s}_{w}\) :

Volume fraction of the wetting phase
 \({s}_{rw}\) :

Residual volume fraction of the wetting phase
 \({s}_{rn}\) :

Residual volume fraction of the nonwetting phase
 \({k}_{r{s}_{w}}\) :

Relative permeability of the wetting phase
 \({k}_{r{s}_{n}}\) :

Relative permeability of the nonwetting phase
 \(\overline{{s }_{n}}\) :

Effective volume fraction of the nonwetting phase
 \({s}_{n}\) :

Volume fraction of the nonwetting phase
 s_{1} :

Volume fraction of water
 s_{2} :

Volume fraction of air
 C:

Chloride concentration
 t:

Time
 D:

Chloride diffusion coefficient
 x:

A position variable
 C_{i} :

Concentration of species i
 J_{i} :

Diffusive flux vector
 R_{i} :

A reaction rate expression for species i
 D_{i} :

Diffusion coefficient of species i
 AutoCAD:

Computeraided design and drafting software
 P_{ec} :

Entry capillary pressure
 σ:

Interfacial tension of a fluid
 θ:

Contact angle between the fluid and the capillary tube
 r_{c} :

Radius of the capillary tube
 K:

Kelvin
 SEM:

Scanning electron microscope
 r_{c} :

Average value of the pore’s radii
 AV%:

Air void percentage or porosity in asphalt mastic
 D_{75} :

Particlesize diameter corresponding to 75% on the cumulative particlesize distribution curve of aggregates
 λ_{p} _{epoxy} :

Pore size distribution index
 W_{i} :

Weight fraction of pores
 N_{i} :

Number of pores
 A_{i} :

Surface area of pores
 λ_{p} _{mastic} :

Pore size distribution index of asphalt mastic
 k_{mastic} :

Water permeability of asphalt mastic
 ρ:

Density
 μ:

Dynamic viscosity
 s_{r} :

Residual saturation
 s_{0,s1} :

Initial value of the water volume fractions
 C(x,t):

Chloride concentration at a location x at a time t
 C_{s} :

Chloride concentration at surface
 C_{0} :

Initial chloride concentration at location x
 erf(z):

Error function
 D_{ss} :

Steadystate chloride diffusion coefficient
 A_{c} :

Crelated coefficient to D
 D_{0} :

Maximal chloride diffusion coefficient
 E_{a} :

Activation energy for diffusion
 R:

Universal gas constant
 T:

Absolute temperature of asphalt pavement
 A_{T} :

Trelated coefficient to D
 T_{t} :

Testing temperature of the mastic diffusion test
 T_{a} :

Air temperature
 R_{s} :

Daily solar radiation
 d:

Depth from pavement surface
 C_{ave} :

Average chloride concentration
 N_{n} :

Number of the noncracking 150 mm sections
 C_{n} :

Chloride concentration at the bottom of the geometry of a noncracking 150 mm section
 N_{c} :

Number of the 1crack 150 mm sections
 C_{c} :

Chloride concentration at the bottom of the geometry of a 1crack 150 mm section
 A_{tc} :

Thermal crackingrelated coefficient to D
 ESALs:

Equivalent single axle loads
 LEF:

Load equivalency factor
 LTPP:

Longterm pavement performance
 L:

Length of the transverse fatigue crack
 A_{fc} :

Fatigue crackingrelated coefficient to D
 D_{mastic} :

Chloride diffusion coefficient of asphalt mastic
 D_{additive} :

Chloride diffusion coefficient in the additives
 C_{initial} :

Initial value of the chloride concentration in the additives
 Ca(NO_{3})_{2}·6H_{2}O:

Calcium nitrate tetrahydrate
 H_{1} :

Initial height of water in standpipe
 H_{2} :

Final height of water in standpipe
 A_{1} :

Crosssectional area of sample
 A_{2} :

Crosssectional area of standpipe
 L_{s} :

Length of sample
 μ_{w} :

Dynamic viscosity of water at 25 °C
 ρ_{w} :

Density of water
 C_{CaCl2} :

Effective mass percentage of CaCl_{2} for antiicing at a temperature
 h:

Overlay thickness
 FEM:

Finite element method
References
ASTM D5084 2020 Test Methods for Measurement of Hydraulic Conductivity of Saturated Porous Materials Using a Flexible Wall Permeameter ASTM International https://doi.org/10.1520/D508403
Bentz DP, Garboczi EJ, Lu Y, Martys N, Sakulich AR, Weiss WJ (2013) Modeling of the influence of transverse cracking on chloride penetration into concrete. Cem Concr Compos 38:65–74. https://doi.org/10.1016/j.cemconcomp.2013.03.003
Bogdanov, I.I., El Ganaoui, K., Kamp, A.M., 2007. COMSOL 2D Simulation of Heavy Oil Recovery by Steam Assisted Gravity Drainage. Proc. Eur. COMSOL Conf.
Brooks B, R.H., Corey, A.T., (1964) Hydraulic properties of porous media. Diss Colo. State Univ, Libr
Brooks RH, Corey AT (1966) Properties of Porous Media Affecting Fluid Flow. J Irrig Drain Div 92:61–88
de Vera G, Climent MA, Viqueira E, Antón C, Andrade C (2007) A test method for measuring chloride diffusion coefficients through partially saturated concrete. Part II: The instantaneous plane source diffusion case with chloride binding consideration. Cem Concr Res 37:714–724
Dickinson EJF, Ekström H, Fontes E (2014) COMSOL Multiphysics®: Finite element software for electrochemical analysis. A minireview Electrochem Commun 40:71–74
Djerbi A, Bonnet S, Khelidj A, Baroghelbouny V (2008) Influence of traversing crack on chloride diffusion into concrete. Cem Concr Res 38:877–883
Garboczi EJ, Bentz DP, Davis JM, Lu Y (2012) Modeling Chloride transport in Cracked Concrete: A 3D imagebased microstructure simulation 1–15
Gittens GJ (1969) Variation of surface tension of water with temperature. J Colloid Interface Sci 30:406–412
Giuliani F, Merusi F, Polacco G, Filippi S, Paci M (2012) Effectiveness of sodium chloridebased antiicing filler in asphalt mixtures. Constr Build Mater 30:174–179
Gu C, Ye G, Sun W (2015) A review of the chloride transport properties of cracked concrete: experiments and simulations. J. Zhejiang Univ.Sci. A 16:81–92
Hall JR, Wishaw BF, Stokes RH (1953) The diffusion coefficients of calcium chloride and ammonium chloride in concentrated aqueous solutions at 25. J Am Chem Soc 75:1556–1560
Hoplin C (2016) Cracking Performance Evaluation of Minnesota Asphalt Pavements
Islam, M., 2015. THERMAL FATIGUE DAMAGE OF ASPHALT PAVEMENT. Civ. Eng. ETDs.
JoekarNiasar V, Hassanizadeh SM, Leijnse A (2008) Insights into the Relationships Among Capillary Pressure, Saturation, Interfacial Area and Relative Permeability Using PoreNetwork Modeling. Transp Porous Media 74:201–219
Kanakaraj U, Lhaden T, Karthik Raj V (2015) Analysis of structural mechanics of solid microneedle using COMSOL software, in: 2015 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS). pp 1–5 (Presented at the 2015 International Conference on Innovations in Information, Embedded and Communication Systems (ICIIECS))
Khan MU, Ahmad S, AlGahtani HJ (2017) ChlorideInduced Corrosion of Steel in Concrete: An Overview on Chloride Diffusion and Prediction of Corrosion Initiation Time. Int. J. Corros 2017:e5819202
Kwon SJ, Na UJ, Park SS, Jung SH (2009) Service life prediction of concrete wharves with earlyaged crack: Probabilistic approach for chloride diffusion. Struct Saf 31:75–83. https://doi.org/10.1016/j.strusafe.2008.03.004
Li Q, Ito K, Wu Z, Lowry CS, Ii SPL (2009) COMSOL Multiphysics: A Novel Approach to Ground Water Modeling. Groundwater 47:480–487
Li Y (1999) Asphalt Pavement Fatigue Cracking Modeling
Liu Y, Shi X (2012) Ionic transport in cementitious materials under an externally applied electric field: Finite element modeling. Constr Build Mater 27:450–460
Liu Y, Shi X (2012) Stochastic Modeling of Service Life of Concrete Structures in ChlorideLaden Environments. J Mater Civ Eng 24:381–390
Liu Z, Xing M, Chen S, He R, Cong P (2014) Influence of the chloridebased antifreeze filler on the properties of asphalt mixtures. Constr Build Mater 51:133–140
Ma T, Geng L, Ding X, Zhang D, Huang X (2016) Experimental study of deicing asphalt mixture with antiicing additives. Constr Build Mater 127:653–662
Ma, X., Cheng, B., Mao, J., Liu, W., Zi, D., 2009. Finite element modelling of coupled heat and moisture transfer in typical earthsheltered building envelope. Proc. Elev. Int. IBPSA Conf. Glasg. Scotl. Build. Simul. 1850–1856.
Montgomery RB (1947) VISCOSITY AND THERMAL CONDUCTIVITY OF AIR AND DIFFUSIVITY OF WATER VAPOR IN AIR. J Atmospheric Sci 4:193–196
Nardi A, Idiart A, Trinchero P, de Vries LM, Molinero J (2014) Interface COMSOLPHREEQC (iCP), an efficient numerical framework for the solution of coupled multiphysics and geochemistry. Comput Geosci 69:10–21
Nielsen EP, Geiker MR (2003) Chloride diffusion in partially saturated cementitious material. Cem Concr Res 33:133–138
Page CL, Short NR, El Tarras A (1981) Diffusion of chloride ions in hardened cement pastes. Cem Concr Res 11:395–406
Paul SK, Chaudhuri S, Barai SV (2014) Chloride diffusion study in different types of concrete using finite element method (FEM). Adv Concr Constr 2:39–56
Rachedi BA, Babouri A, Berrouk F (2014) A study of electromagnetic field generated by high voltage lines using COMSOL MULTIPHYSICS. 2014 International Conference on Electrical Sciences and Technologies in Maghreb (CISTEM). pp 1–5 (Presented at the 2014 International Conference on Electrical Sciences and Technologies in Maghreb (CISTEM))
Saadoon T, Garcia A, GómezMeijide B (2017) Dynamics of water evaporation in cold asphalt mixtures. Mater Des 134:196–206
Solaimanian, M., Thomas, W.K., 1993. Predicting maximum pavement surface temperature using maximum air temperature and hourly solar radiation. Transp. Res. Rec. 1–1.
Spragg RP, Castro J, Li W, PourGhaz M, Huang PT, Weiss J (2011) Wetting and drying of concrete using aqueous solutions containing deicing salts. Cem Concr Compos 33:535–542
Tan, Y., Wang, S., Xu, F., Liu, W., Chen, X., Liang, Y., 2017. Application of COMSOL Multiphysics in Research of Concrete DurabilityA Short Review. J. Chin. Ceram. Soc. 05.
Tang L (1999) Concentration dependence of diffusion and migration of chloride ions: Part 2. Experimental evaluations Cem Concr Res 29:1469–1474
Taylor DW (1956) Fundamentals of Soil Mechanics 100
Transportation Officials, 1993. AASHTO Guide for Design of Pavement Structures. Aashto Vol. 1.
Truc O, Ollivier JP, Carcassès M (2000) A new way for determining the chloride diffusion coefficient in concrete from steady state migration test. Cem Concr Res 30:217–226
Vardanega PJ, Waters TJ (2011) Analysis of Asphalt Concrete Permeability Data Using Representative Pore Size. J Mater Civ Eng 23:169–176
Vennard JK, Street RL (1975) Elementary Fluid Mechanics, 5th edn. Wiley, NY
Wang, C.Y., 1998. Modeling multiphase flow and transport in porous media. Transp. Phenom. Porous Media 383–410.
Wang CY, Cheng P (1996) A multiphase mixture model for multiphase, multicomponent transport in capillary porous media—I. Model development. Int J Heat Mass Transf 39:3607–3618
Wang XY, Zhang LN (2016) Simulation of Chloride Diffusion in Cracked Concrete with Different Crack Patterns. Adv. Mater. Sci. Eng 2016:e1075452
Wegner, J., Ganzer, L., 2012. Numerical Simulation of Oil Recovery by Polymer Injection using COMSOL. Excerpt Proc. 2012 COMSOL Conf. Milan.
Wei Z, Weavers LK (2016) Combining COMSOL modeling with acoustic pressure maps to design sonoreactors. Ultrason Sonochem 31:490–498
Wu LZ, Zhang LM, Zhou Y, Li BE (2017) Analysis of multiphase coupled seepage and stability in anisotropic slopes under rainfall condition. Environ Earth Sci 76:469
Yang P, Dhandapani Y, Santhanam M, Neithalath N (2020) Simulation of chloride diffusion in fly ash and limestonecalcined clay cement (LC3) concretes and the influence of damage on servicelife. Cem. Concr. Res 130:106010
Yuan, Q., Shi, C., De Schutter, G., Audenaert, K., 2008. Effect of temperature on transport of chloride ions in concrete. Concr. Repair Rehabil. Retrofit. II CRC Press. 159–160.
Zamel N, Li X (2008) A parametric study of multiphase and multispecies transport in the cathode of PEM fuel cells. Int J Energy Res 32:698–721
Zhang P, Li D, Qiao Y, Zhang S, Sun C, Zhao T (2018) Effect of Air Entrainment on the Mechanical Properties, Chloride Migration, and Microstructure of Ordinary Concrete and Fly Ash Concrete. J Mater Civ Eng 30:04018265
Zhang Y, Shi X (2021) Laboratory evaluation of a sustainable additive for antiicing asphalt. Cold Reg. Sci. Technol 189:103338
Zhao XH, Zhang XD (2011) The Research on the AntiIcing and Pavement Performance of the ChlorideStored Asphalt Mixture. Appl Mech Mater 97–98:321–326
Zheng M, Wang C, Han L, Sun Y, Li Y, Ma Z (2016) Laboratory evaluation of longterm antiicing performance and moisture susceptibility of chloridebased asphalt mixture. Int J Pavement Res Technol 9:140–148
Zheng M, Zhou J, Wu S, Yuan H, Meng J (2015) Evaluation of longterm performance of antiicing asphalt pavement. Constr Build Mater 84:277–283
Zhong K, Sun M, Chang R (2018) Performance evaluation of highelastic/saltstorage asphalt mixture modified with Mafilon and rubber particles. Constr Build Mater 193:153–161
Zhou J, Li J, Liu G, Yang T, Zhao Y (2019) LongTerm Performance and Deicing Effect of SustainedRelease Snow Melting Asphalt Mixture. Adv Civ Eng 2019:1940692. https://doi.org/10.1155/2019/1940692
Acknowledgements
The authors also would like to acknowledge professional engineers Alex Guo and Yue Huang who work for COMSOL Inc. for their tremendous help in using the software.
Funding
The authors thankfully acknowledge financial support by the Tier 1 University Transportation Center CESTiCC (Center for Environmentally Sustainable Transportation in Cold Climates).
Author information
Authors and Affiliations
Contributions
Y. Zhang did final interpretation of the results of all the tests, established part of the model with COMSOL Multiphysics, and wrote part of the draft manuscript. Y. Deng established part of the model with COMSOL Multiphysics and wrote part of the draft manuscript. X. Shi designed the scope of this work and advised the team throughout the process (from conceptualization to manuscript preparation). All authors read and approved the final manuscript.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interest.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Additional file 1: Figure S1.
Relationship between σ of water and pavement temperatures. Figure S2. Particlesize distribution curve of aggregates used for asphalt mixture. Figure S3. Relationship between μ of water and pavement temperatures. Figure S4. Relationship between μ of air and pavement temperatures. Figure S5. Relationship between ρ of air and pavement temperatures. Figure S6. a) Simulated and calculated Ctime curves and b) goodness of fit at 0% cracking level for thermal cracking. Figure S7. a) Simulated and calculated Ctime curves and b) goodness of fit at 27% cracking level for thermal cracking. Figure S8. a) Simulated and calculated Ctime curves and b) goodness of fit at 38% cracking level for thermal cracking. Figure S9. a) Simulated and calculated Ctime curves and b) goodness of fit at 59% cracking level for thermal cracking. Figure S10. ESALsfatigue crack rate relationship of asphalt pavement. Figure S11. EASLscrack width relationship of asphalt pavement. Figure S12. Loading numbercrack width relationship of asphalt pavement. Figure S13. Loadsingle LEF relationship of asphalt pavement. Figure S14. Crack widthcrack depth relationship of asphalt pavement. Figure S15. a) Simulated and calculated Ctime curves and b) goodness of fit at 0 year for fatigue cracking. Figure S16. a) Simulated and calculated Ctime curves and b) goodness of fit at 8 years for fatigue cracking. Figure S17. a) Simulated and calculated Ctime curves and b) goodness of fit at 9 years for fatigue cracking. Figure S18. a) Simulated and calculated Ctime curves and b) goodness of fit at 10 years for fatigue cracking. Figure S19. a) Ctime curves from simulation and experiment and b) goodness of fit to determine D of additives (datapoints before 71 h were removed in Figure S19b due to unstable diffusion). Figure S20. a) Cl concentration–time relationship and b) Cl diffusion coefficienttime relationship from mastic diffusion test. Figure S21. Phase diagram for waterCaCl2 system (adopted from Phase diagrams for binary salt solutions  Phasediagram). Table S1. a) Area fractions of asphalt mastic, aggregates, and additives in each of samples. Table S2. An example of calculating λ_{pepoxy} for Figure 2. Table S3. Estimated λ_{pepoxy} based on three SEM images of the surface of CaCl_{2}zeolite/pepoxy #16. Table S4. TimeESALs datapoints. Table S5. # of 150mm sections over time related to fatigue cracking of asphalt pavement. Table S6. Total depth, width, and partial depth (in the model) of the transverse fatigue crack at different years. Table S7. Parameters and results of falling head test. Table S8. Results of water contact angle test.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Zhang, Y., Deng, Y. & Shi, X. Model development and prediction of antiicing longevity of asphalt pavement with saltstorage additive. J Infrastruct Preserv Resil 3, 1 (2022). https://doi.org/10.1186/s4306502100047w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1186/s4306502100047w