Академический Документы
Профессиональный Документы
Культура Документы
Abstract: The permeability of asphalt concrete has been the subject of much study by pavement engineers over the last decade. The work
undertaken has tended to focus on high air voids as the primary indicator of permeable asphalt concrete. This paper presents a simple
approach for understanding the parameters that affect permeability. Principles explained by Taylor in 1956 in channel theory work for soils
are used to derive a new parameter—representative pore size. Representative pore size is related to the air voids in the compacted mix and
the D75 of the asphalt mix grading curve. Collected Superpave permeability data from published literature and data collected by the writers
at the Queensland Department of Transport and Main Roads is shown to be better correlated with representative pore size than air voids,
reducing the scatter considerably. Using the database of collected field and laboratory permeability values an equation is proposed that
pavement engineers can use to estimate the permeability of in-place pavements.
DOI: 10.1061/(ASCE)MT.1943-5533.0000151
CE Database subject headings: Asphalts; Concrete; Permeability; Voids; Regression analysis; Binders, material; Porosity;
Pavements.
Author keywords: Asphalt concrete; Permeability; Grading; Air voids; Regression analysis; Binder .
Introduction ent (the head loss per unit length); m = experimental flow index;
and A = total cross-sectional area.
Permeability of asphalt concrete has received a lot of attention by The m value will vary from approx 0.5 for turbulent flow to
pavement engineers over the last decade. Excessive permeability 1.0 for laminar flow. The values of k and m vary depending upon
leads to infiltration into road surfacings which leads to damage the flow rate. This generalized flow equation was first proposed
caused by an increase in pore pressures that build up under traffic by Prony in 1804 (refer to Parkin 1971 and Jaeger 1956, p. 397).
loads. Predicting the permeability of asphalt is important so that The secret of getting appropriate k and m values is to test at the
appropriate compaction criteria can be set for road construction hydraulic gradients of interest (Tan et al. 2000).
projects. Darcy’s original formula which holds for laminar flow
[Eq. (1)] defines the coefficient of permeability
Literature Review
Q = kiA (1) The literature is extensive on the factors that affect asphalt con-
where Q = rate of fluid (water) flow moving through the specimen; crete permeability with the following being commonly cited cf.
k = coefficient of permeability; i = hydraulic gradient (the head loss Abdullah et al. (1998): the percentage of bitumen binder in the
per unit length); and A = total cross-sectional area. asphalt mix; compaction effort; type of aggregate; nominal maxi-
The coefficient of permeability, k, varies depending on the mum aggregate size (NMAS); lift thickness; aggregate mix gra-
material being studied. It is this parameter that is usually quoted dation; and air voids in the mixture. Cooley et al. (2002)
when the permeability of an asphalt specimen is requested. Tan et presented data from 23 hot mix asphalt (HMA) construction
al. (2000) preferred the generalized flow equation projects and showed the effect of density (air voids) on perme-
ability. They also noted that 9.5- and 12.5-mm NMAS mixtures
Q = k'imA (2) had similar permeability characteristics, though larger stone sizes
tended to have higher permeability readings. On the issue of lift
where Q = rate of fluid (water) flow moving through the specimen; thickness it was shown that increasing lift thickness decreased
k'= experimental coefficient of permeability; i = hydraulic gradi- permeability in the field.
Tarefder et al. (2005) developed a neural network model using
1
Graduate Researcher, Dept. of Engineering, Cambridge Univ., 100 field cores from 50 asphalt mixtures. They found that five
Trumpington St., Cambridge, CB2 1PZ U.K.; formerly, Graduate Engi- main factors affect the permeability of asphalt concrete: air voids;
neer, Dept. of Transport and Main Roads, Queensland, Australia (corre- D10 (grain size through which 10% of the materials pass); D30
sponding author). E-mail: pjv27@cam.ac.uk (grain size through which 30% of the materials pass); saturation
2
Senior Physicist, Dept. of Transport and Main Roads, 35 Butterfield characterized by the CoreLok infiltration coefficient and the ef-
St., Herston, Brisbane, Queensland 4006, Australia. fective asphalt to dust ratio (Pbe P0.075). A higher coefficient of
Note. This manuscript was submitted on August 13, 2009; approved
on June 25, 2010; published online on July 10, 2010. Discussion period
determination was found using the neural network model as op-
open until July 1, 2011; separate discussions must be submitted for indi- posed to standard multiple linear regression methods. Masad et al.
vidual papers. This paper is part of the Journal of Materials in Civil (2004, 2006) developed an empirical equation based on the
Engineering, Vol. 23, No. 2, February 1, 2011. ©ASCE, ISSN 0899- Kozeny-Carman equation to describe permeability of asphalt con-
1561/2011/2-169–176/$25.00. crete. This equation expresses permeability as a function of air
y = 1.674x - 12.442
20 2
R = 0.997
75
15
y = 0.611x - 2.097
50 2
R = 0.841
10
25
5
0 0
0.01 0.1 1 10 100 0 5 10 15 20 25
Fig. 1. Gradings for Mix 1 and Mix 2 Fig. 2. Relationship between accessible and total voids for Mix 1 and
Mix 2
Connected Voids
Permeability Database
The concept of accessible voids in asphalt mixes was first intro-
duced by Smith and Gotolski (1969) and refers to those voids in In the present paper, the use of both air voids alone and RPS as
the mix which are available for the passage of water. They found indicators of permeability are investigated. The power relation-
that “one can force all air from a sample by cupping one’s hand ships have been determined for both. In order to determine the
over the sample and by pumping the hand as a plunger.” They relationships a large database of both field and laboratory perme-
found that a relationship existed between total and accessible ability data, similar to that used by Masad et al. (2004, 2006) is
voids. Smith and Gotolski (1969) also related these voids to (air) used. Table 2 shows the source of the data and the references in
permeability. the original publications. In this database of collected data, 47
Follow up studies have been carried out by Queensland Main asphalt mixtures were used with NMAS of 9.5, 10, 12.5, 14, 19,
Roads on two asphalt mixes, referred to as Mixes 1 and 2. The and 25 mm included. A total of 467 individual lab and field per-
first mix (Mix 1) was investigated as part of a study on “Voids in meability measurements are in the data set. This large database
Asphalt” and reported by Waters (1986). In this earlier report, an allows a study of the parameters that affect permeability with a
outline of the test method and a procedure for determining the variety of stone sizes, air void measurement techniques, binder
accessible voids is given. The relationship between total and ac- contents, lift thicknesses, and grading curves present in the data-
cessible voids is also given. base. To be sure, the aforementioned factors will influence the
The second mix was investigated as part of the ongoing field data set and will increase scatter. The purpose is to test the valid-
and laboratory permeability studies at Queensland Main Roads. ity of the RPS approach. Fig. 4 shows the grading data for the
The tests were carried out on asphalt cores and involved deter- studied mixtures. Fig. 5 shows the plot of measured air voids
mining the total and accessible voids as well as the permeability. against permeability for the database. The coefficient of determi-
In this study, relationships are given between accessible voids and nation (R2) is 0.41 for the 467 data points; which is reasonable
total voids, total voids and permeability, and accessible voids and considering the large sample size. In other words, 41% of the
permeability. In addition, implications for asphalt design are dis- variation can be explained by the air void level using the power
cussed. The grading curves of the two mixes tested are displayed function shown in Fig. 5.
graphically in Fig. 1.
For both Mixes 1 and 2 the total and accessible voids were
determined. In the case of Mix 1, the tests were carried out on
laboratory prepared pats while for Mix 2 the tests were carried out
on cores though tested in the laboratory. The relationship between 10
4.3%. It could be said, in this case, the degree of connectivity is 2.3866 k = 3E-08(Total Voids)
5.9184
k = 0.0006(Accessible Voids)
43%. Fig. 3 shows the relationship between voids (total and ac- 2
R = 0.7591
2
R = 0.7764
0.1
cessible) and permeability for Mix 2. It is clear that at low per-
meability values the two fitted lines diverge the most as there is
more chance of isolated voids at low permeabilities and less so at 0.01
high permeabilities. Also to note is the power change: in the case
of accessible voids the power is 2.4 and for total voids it is 5.9.
The above work has shown the effect of air void connectivity 0.001
1 10 100
on the permeability relationship. Given the power change from Voids (%)
total to accessible voids (5.9 to 2.4) it is reasonable to suggest that Total Voids Accessible Voids Power (Total Voids) Power (Accessible Voids)
the degree of connectivity will vary the power when mixtures are
prepared, especially if total voids are used. In the following ex- Fig. 3. Relationship between voids (accessible and total voids) and
amples, air voids refers to total air voids. permeability
Parametric Study characterizes the database the best. In passing, it is noted that the
Consideration of grading and hence effective particle size is im- power on the effective particle size and air voids was found not
plicit in the RPS ideas. It is proposed to use Eq. (9) to estimate to be too dissimilar for the data subsets (i.e., no significant im-
the permeability of asphalt concrete, using the database. What is provement in fit) and hence a single power was adopted, allowing
also required is a determination of the effective particle size that Eq. (9) to be written as follows:
Coarse 01-1086
01-1089 01-1091
60 01-1093 01-1095
01-1097 01-1101
50 01-1103 01-1105
01-1106 Mix 1, 12.5mm
Mix 2, SM-1 Mix 3, 9.5mm
40
Mix 4, 12.5mm Mix 5, 12.5mm
9.5mm (fine) 9.5mm (coarse)
30 12.5mm (coarse) 19mm (coarse)
25mm (coarse) Project 1 (12.5mm)
Project 2 (9.5mm) Project 3 (19mm)
20
Project 4 (12.5mm) Project 5 (25mm)
Project 6 (9.5mm) Project 7 (25mm)
10 Project 8 (25mm) Project 9 (12.5mm)
Project 10 (12.5mm) Project 11 (19mm)
0 01-1098
0.01 0.1 1 10
Sieve Size (cm)
k = A(Rp)b (9) as the key percentage passing for tests done at Queensland Trans-
port and Main Roads. A wider ranging examination of the 47
where Rp = representative pore size in mm; Rp =2 3
grading curves in the database (Fig. 4) was undertaken for this
×(AV(%) 100)× Deff; k = coefficient of permeability in mm/s;
and A and b = regression parameters study. The following particle sizes were determined by linear in-
Vardanega et al. (2008) identified the D40, D50, D60, and D70 terpolation from the supplied grading data: D10, D25, D40, D50,
Hewitt (1991)
0.1
Permeability (k) (mm/s)
Maupin (2009)
0.01
Maupin (2001)
0.001
0.0000001
0.8
Co-efficient of Determination (R2)
D60, D70, D75, D80, and D90 and these were substituted as Deff in
Eq. (8) and the coefficient of determination (R2) calculated for
each. Fig. 6 shows the graphical results of this process. It is seen
that for the seven subsets of data, there is generally an increase of the fitted line relative to the scatter about the mean y line. Using
fit at larger effective particle sizes. This is not unexpected as just air voids as a predictor of permeability (Fig. 5), the RD is
small voids are filled with binder at the low end of the grading 76.8%. Using representative pore size, the RD is 52.0%. There-
curve. In the case of Cooley et al. (2001), Mallick et al. (2003), fore, the scatter about the fitted line is reduced by 32%, as a result
Kanitpong et al. (2001), and Maupin (2001), there is an upswing of using Rp instead of AV (%).
of R2 around D70 – D90. There is a reduction of R2 for the data of
Vardanega et al. (2008) and Maupin (W. Maupin, personal com-
munication, 2009) at D80 and D90. It was decided that D75 was a Summary and Conclusions
good approximation for Deff as it would yield a good R2 for each
of the data subsets. The changes in coefficient of determination The use of a large database to determine the effective particle size
for the entire database are shown as Fig. 7. Examination of Fig. 7 of asphalt is not without its drawbacks. Permeability can vary
shows that for the whole database it is clear that no significant with method of testing, in particular the variation of the hydraulic
improvement in fit occurs once Deff is taken as greater than D 50 .A gradient. For the field data, assumptions are made about the ef-
very good R2 is present for D75, making it a reasonable choice to fective thicknesses of the asphalt being considered and hence the
characterize the database. hydraulic gradient and the assumption of one-dimensional flow.
Fig. 8 shows the regression of representative pore size, calcu- Remember that Darcy’s Law only holds only for laminar flow.
lated using the D75 versus permeability. The R2 = 0.74 which is
The regression relationship developed only used D75 and air
very good considering the large sample size (n = 467) and other
voids to compute the permeability, via the representative pore
influencing parameters. The equation adopted for permeability is
size. The fact that only two parameters were needed and an R2 of
shown as Eq. (10)
0.74 for 467 data points obtained is extremely pleasing, given the
k = 0.46(Rp)3.70 R2 = 0.74 (10) many minor factors that will affect the results, such as measure-
ment of voids and permeability, will differ slightly between
where sources. The following conclusions can be drawn:
1. Grading of the asphalt mix is a major influence on perme-
Rp = 23 × (AV(%)100) × D75 ability;
k = coefficient of permeability in mm/s. Rewritten in expanded 2. Air voids in the asphalt mix have a major influence on per-
form the regression model adopted for estimation of asphalt per- meability;
meability is 3. The power on the air voids versus permeability relationship
is affected by the degree of connectivity of the pore struc-
k(mms) =
46
100 3
[ 2
×
AV(%)
100 × D 75(mm) ] 3.70
(11) 4.
ture;
The representative pore size, based on channel theory, is
shown to be a significantly better predictor of asphalt perme-
A useful term for comparing relationships is the relative deviation ability than air voids alone;
(RD). The RD is given by (Waters and Vardanega 2009) 5. The Deff for asphalt was computed to be greater than the D50,
with D75 being selected as a good representation for the da-
RD = 100(1− R2)0.5 (12)
tabase being studied; and
This coined statistical parameter is useful for quantifying the 6. Eq. (11) can generally predict the permeability to within an
reduction of scatter between two dependent variables, both pre- order of magnitude for the seven data subsets examined (467
dicting the same quantity. The RD is essentially the scatter about data points in total).
0.01 0.1 1 10
10
3.695
k = 0.460Rp
2
Vardanega et al. (2008)
R = 0.740
1
Hewitt (1991)
0.1
Permeability (k) (mm/s)
Maupin (2009)
0.01
Maupin (2001)
0.001
RH = eDS6 (18)
Appendix: Derivation of RPS
The void ratio is related to the porosity (n) by
In channel theory, the term hydraulic radius is used as a measure e = n (1− n) (19)
of the size of the flow channels (Taylor 1956). The hydraulic
radius is defined as the ratio of the volume to surface area of flow It follows that the hydraulic radius is given by
channel. That is, the hydraulic radius is given by
RH = nD S 6(1− n) (20)
RH = VCAC (13) For small porosity values, Eq. (15) is approximately given by
RH = nDS6 (21)
where RH = hydraulic radius; VC = volume of flow channel; and
AC = surface area of channel. Taking the simple case of a pipe, the By analogy to the case of water flow through a pipe, it is
hydraulic radius is given by proposed that the RPS is equal to four times the hydraulic radius.
It follows that, for small porosity values, the RPS is given by:
RH = ALPL (14)
RP = 2nDS3 (22)
where A = cross-sectional area of pipe; P = perimeter of pipe; and
L = length of pipe. Since the cross-sectional area of the pipe is
equal to uD2 4 and the perimeter of the pipe is uD, it follows References
that the hydraulic radius of a pipe is equal to one quarter of the
diameter of the pipe (D 4). For a porous medium, the RPS will Abdullah, W., Obaidat, M., and Nazem, M. (1998). “Influence of aggre-
be defined in a similar way to the diameter of the channel. In the gate type and gradation on voids of asphalt concrete pavements.”
case of a porous medium made up of single size spherical par- J. Mater. Civ. Eng., 10(2), 76–85.
ticles, the volume of the flow channel is given by Arambula, E., Masad, E., and Martin, A. E. (2007). “Influence of air void