Feng, S. et al. Géotechnique [https://doi.org/10.1680/jgeot.19.D.

005]

DISCUSSION

Permeability assessment of some granular mixtures

SHUYIN FENG
, PAUL J. VARDANEGA†, ERDIN IBRAIM‡, ISWANDARU WIDYATMOKO§,
CHIBUZOR OJUM∥, BRENDAN C. O’KELLY¶ and MARIA NOGAL**

Contribution by Brendan C. O’Kelly and Maria Nogal the various granular mixtures presented in Table 2.
As part of their paper (Feng et al., 2019), the authors Compared to the authors’ model (equation (27): R 2 = 0·90,
presented the novel application of the grading entropy n = 30, p , 0·0001), the analysis for the proposed expanded
framework for hydraulic conductivity assessments, along model with deduced fitting coefficients C1 = 662·75 mm/s,
with a model (equation (27)), requiring only the normalised C2 = 5·55, C3 = 1·32 and C4 = 4·58 resulted in a slightly
grading entropy coordinates A and B as inputs, for reliably better fit (R 2 = 0·96, n = 30, p , 0·0001). In terms of the
predicting the permeability coefficient (k20°C ) values of adjusted R-squared, R2a , which penalises the number of pre-
30 compacted, crushed basalt–gritstone gravel mixtures dictors employed in the model, the two-variable model
investigated. The discussers note that the gradation charac- exhibits a value of 0·88, compared to the proposed three-
teristics of these gravel mixtures were such that their B values variable model R2a value of 0·95, again indicative of the latter
negatively correlate with A (R 2 = 0·50), meaning that the model’s better fit.
k20°C values could also be reasonably predicted based entirely In order to further test the advantage of the proposed
on their A values, or less reliably based solely on their expanded model for other soil types (classifications), the
B values (equations (25) and (26), respectively). Further, discussers employed the same investigative approach in con-
considering all 30 gravel mixtures investigated, their reported sidering the data set comprising A, B, e and k20°C values
values of void ratio (e) linearly correlate positively and reported for 20 silty sand and sand materials in the paper by
negatively with A (R 2 = 0·64) and B (R 2 = 0·53), respectively. Arshad et al. (2019). Compared to the gravel mixtures inves-
Importantly, the values of coefficients deduced for these tigated by the authors (D10 = 0·72–7·02 mm, e = 0·51–0·85
various correlations are dependent on compaction level. and k20°C = 4·19–561·20 mm/s), these sandy soils had
As such, equations (25)–(27) would generally underestimate particle shape classes varying from angular to sub-angular,
the actual k20°C values for the same gravel mixtures placed D10 = 0·01–0·50 mm, e = 0·32–0·60 and substantially lower
at lower densification levels (higher e values). In these values of k20°C in the range 0·0007–3·50 mm/s. Further, in
instances, the inclusion of the e parameter in the model terms of linear correlation, the B values of these sandy soils
(equation (27)) would seem appropriate, thereby extending only weakly correlated with their A values (R 2 = 0·27), such
its scope and reliability for other field applications that when analysed independently, k20°C only weakly corre-
(e.g. assessing loosely placed materials as potential lated with A and B. Overall, the same general trends are
drainage/filter media). Therefore, the following regression evident for both Arshad et al. (2019) and Feng et al. (2019)
model is proposed data sets, namely: log k20°C correlates positively with log A
and negatively with log B, whereas log D10 and e both
k20°C ¼ C1 AC2 BC3 eC4 ð28Þ correlate positively with A and negatively with B. In terms
where C1, C2 , C3 and C4 are the fitting coefficients. Note that of the k20°C –e relationship: for the Arshad et al. (2019) data
C1 is expressed in the same units as k20°C, whereas the other set, power fitting produced superior R 2, whereas for the Feng
coefficients are dimensionless. This avoids the mathematical et al. (2019) data set, power and exponential fitting were
and physical inconsistencies discussed in the papers by found to produce comparable results (R 2  0·86).
Castillo et al. (2014a, 2014b). The discussers found that equation (27) cannot reliably
Towards demonstrating this point, the discussers per- represent the described sandy soil data set, with 90% of the
formed multiple linear regression analysis for the proposed Arshad et al. (2019) points falling outside the prediction
model (equation (28)) utilising the listed A, B and e values for intervals (α = 5%). Further, with fitting coefficient values
deduced for the basalt–gritstone gravel mixtures as inputs,
the proposed expanded model also cannot reliably represent
the sandy soil data set, with 95% of the Arshad et al. (2019)

Department of Civil Engineering, University of Bristol, Bristol, points falling outside the prediction intervals (α = 5%). In
other words, it would appear that the deduced fitting
UK (Orcid:0000-0002-3837-6762).
† Department of Civil Engineering, University of Bristol, Bristol, coefficient values are specific to the particular test material
UK (Orcid:0000-0001-7177-7851). under investigation. This is not unexpected, but consistent
‡ Department of Civil Engineering, University of Bristol, Bristol, with the fact that the two samples (data sets) do not represent
UK. the same statistical population.
§ Pavement Engineering, Centre of Excellence for Asset The discussers also performed multiple linear regression
Consultancy, AECOM, Nottingham, UK analysis of the data set (n = 20) for the silty sand and sand
(Orcid:0000-0003-1252-3979). materials to investigate the goodness of fit achieved for
∥ Pavement Engineering, Centre of Excellence for Asset both prediction models. Compared to the model without the
Consultancy, AECOM, Nottingham, UK. e parameter included (R 2 = 0·45, R2a = 0·30 and p = 0·006
¶ Department of Civil, Structural and Environmental Engineering,
Trinity College Dublin, Dublin, Ireland
for C4 = 0), the proposed model (equation (28)) produced
(Orcid:0000-0002-1343-4428). values of R 2 = 0·68, R2a = 0·56 and p = 0·0003. That is, con-
** Department of Civil, Structural and Environmental Engineering, sideration of the e parameter significantly improves the
Trinity College Dublin, Dublin, Ireland fitting, with less than 50% of the variability of the data
(Orcid:0000-0001-5405-0626). predicted by the two-variable model, whereas equation (28)

Downloaded by [] on [08/09/19]. Copyright © ICE Publishing, all rights reserved.

 2 DISCUSSION
accounts for almost 70% of the variability of the model. 1000·00
Further, the k20°C values for these sandy soils were mostly Out of ±50% range
In ±50% range
captured by B and e, whereas A adjusts the final k20°C value
(small C2 value). Compared to the basalt–gritstone gravel
mixtures, the substantially different values of the deduced Line of equality
fitting coefficients and significantly lower R2 values for the
sandy soils may be explained by greater variation in their 100·00

Measured k20°: mm/s

gradation characteristics (with B substantially independent
of A), differences in shape factor for the solid particles and
their significantly greater specific surface area (Ss), as well as
possibly the relatively small sample size of this data set. As
with the evolution of conventional permeability models, the
inclusion of a particle shape factor and the Ss parameter in 10·00
the formulation of the discussed models may further enhance

%
their performance.

50
us
Pl

%
50
us
Authors’ reply

in
M
The authors are delighted by the discussers’ interest in the 1·00
1 10 100 1000
research presented in the paper by Feng et al. (2019). The
Predicted k20°: mm/s
discussers have proposed an improvement to a regression
model, equation (27) (Feng et al., 2019), for the hydraulic
Fig. 13. Predicted k20°C plotted against measured k20°C for equation
conductivity (or coefficient of permeability) which considers
(28a), calibrated with the data set from the paper by Feng et al. (2019)
the compaction level of soil mixtures (equation (28)). As an (n = 30)
indicator for compaction level, the void ratio (e) is now
incorporated along with the normalised grading entropy
coordinates A and B as predictors of permeability. That said, the authors wish to reiterate that the aim of the
Equations (27) and (28) were then examined by the discussers research presented in the paper by Feng et al. (2019) was to
using the data set of gravel mixtures from the paper by Feng determine whether the normalised grading entropy coordi-
et al. (2019) (n = 30) and the silty-sand and sand materials data nates alone can be used to give an estimate of the coefficient
set (n = 20) from the paper by Arshad et al. (2019). The of permeability of gravel mixtures. Pavement engineers often
authors agree that the inclusion of the void ratio (e) does specify gradation curve envelopes for mixtures when design-
enhance the prediction accuracy of equation (27) to some ing such mixtures and therefore a method to estimate
extent, as illustrated by the discussers in their contribution. coefficient of permeability straight from the gradation
This is expected as the void ratio (e) has been long recognised curve itself is useful. To this end, equation (27) did not
as a primary predictor for coefficient of permeability of soils include void ratio: the effect of void ratio having been tested
and soil mixtures and is included in many empirical and by way of the ‘Kozeny–Carman’ (e.g. Kozeny, 1927; Carman,
semi-empirical models (e.g. Kozeny, 1927; Carman, 1937, 1937, 1939) and Chapuis (2012) formulations (see Figs 7 and
1939; Carrier, 2003; Chapuis & Aubertin, 2003; Chapuis, 8) in the original technical note. However, the authors agree
2004, 2012; Vardanega et al., 2017). that forms of equation (28) should be considered for future
The authors have further refined the fitting coefficients research work in this area.
C1 to C4 in equation (28) given by the discussers using The discussers also note that equation (27) cannot
the original data files from the paper by Feng et al. (2019), to successfully predict the data set presented in the paper by
generate equation (28a) Arshad et al. (2019) (n = 20). As equation (27) is calibrated
based on the data of one soil type only, this is not surprising.
k20°C ðmm=sÞ ¼ 67183A559 B130 e458 The authors never suggested that equation (27) was a
ð28aÞ
R2 ¼ 096; n ¼ 30; p , 00001 universal equation for all soils or soil mixtures. As stated in
the paper by Feng et al. (2019: p. 653), ‘the coefficients and
The slight discrepancy between equations (28) and (28a) correlations calculated in equations (18)–(27) only hold for
(i.e. the values of C1 to C4) is because the data presented in the specific soil tested in this research’. With a larger database
the technical note were rounded to two decimal places to save of more diverse soil types and gradations, the calibrated
space. To further illustrate the improvement in prediction of coefficients of equation (27) may be made more representa-
the coefficient of permeability for the data set from the paper tive. It is good to read that the discussers report similar
by Feng et al. (2019) (n = 30) using equation (28a) instead of overall trends for the Arshad et al. (2019) data set as for the
equation (27), the predicted against measured value plot data set from the paper by Feng et al. (2019). This further
(Fig. 13) was drawn. Compared to the predicted against confirms that some degree of correlation exists between the
measured plot generated using equation (27) (Fig. 12, Feng coefficient of permeability and the normalised grading
et al., 2019), it can be seen that a few more points migrate entropy coordinates.
into the ±50% bounds. The authors therefore agree with the The authors agree with the discussers that inclusion of
discussers that equation (28a) has better predictive power more information such as specific surface and particle shape
than equation (27) for this data set. It is also worth pointing factor should improve the correlations. However, as already
out that Fig. 13 has the more points within the ±50% bounds stated, the authors were attempting to see how well a smaller
than shown on Fig. 8 (i.e. for the ‘Kozeny–Carman’ model number of parameters could predict the coefficient of
(Kozeny, 1927; Carman, 1937, 1939). Therefore, for the data permeability of the gravel mixtures studied.
set from the paper by Feng et al. (2019) (n = 30) equation The authors would like to restate their appreciation to the
(28a) is the most preferred choice (of the models examined) discussers for their effort in producing equation (28) and
on the basis of both the coefficient of determination values re-examining the two data sets presented in the papers by
and the number of points falling within the ±50% bounds on Feng et al. (2019) and Arshad et al. (2019). Drawing
the predicted against measured plots. predicted against measured plots similar to those shown in

Downloaded by [] on [08/09/19]. Copyright © ICE Publishing, all rights reserved.

 DISCUSSION 3
Fig. 13 for the data set in the paper by Arshad et al. (2019) Chapuis, R. P. (2004). Predicting the saturated hydraulic conduc-
would be helpful to demonstrate graphically the predictive tivity of sand and gravel using effective diameter and void ratio.
power of the fittings presented by the discussers. Can. Geotech. J. 41, No. 5, 787–795, https://doi.org/10.
1139/t04-022.
Chapuis, R. P. (2012). Predicting the saturated hydraulic conduc-
REFERENCES tivity of soils: a review. Bull. Engng Geol. Environ. 71, No. 3,
Arshad, M., Nazir, M. S. & O’Kelly, B. C. (2019). Evolution of 401–434, https://doi.org/10.1007/s10064-012-0418-7.
hydraulic conductivity models for sandy soils. Proc. Instn Civ. Chapuis, R. P. & Aubertin, M. (2003). On the use of the Kozeny
Engineers – Geotech. Engng, https://doi.org/10.1680/jgeen.18. Carman equation to predict the hydraulic conductivity of soils.
00062. Can. Geotech. J. 40, No. 3, 616–628, https://doi.org/10.1139/
Carman, P. C. (1937). Fluid flow through granular beds. Trans. Instn t03-013.
Chem. Engrs 15, 150–166. Feng, S., Vardanega, P. J., Ibraim, E., Widyatmoko, I. & Ojum, C.
Carman, P. C. (1939). Permeability of saturated sands, soils and (2019). Permeability assessment of some granular mixtures.
clays. J. Agric. Sci. 29, No. 2, 262–273, https://doi.org/10.1017/ Géotechnique 69, No. 7, 646–654, https://doi.org/10.1680/jgeot.
S0021859600051789. 17.T.039.
Carrier, W. D. III (2003). Goodbye, Hazen; Hello, Kozeny–Carman. Kozeny, J. (1927). Über kapillare Leitung des Wassers im Boden
J. Geotech. Geoenviron. Engng 129, No. 11, 1054–1056, (Aufstieg, Versickerung und Anwendung auf die Bewässerung).
https://doi.org/10.1061/(ASCE)1090-0241(2003)129:11(1054). Sitzungsber. Akad. Wiss. Wien. 136, No. 2a, 271–306 (in
Castillo, E., Calviño, A., Nogal, M. & Lo, H. K. (2014a). On the German).
probabilistic and physical consistency of traffic random variables Vardanega, P. J., Feng, S. & Shepheard, C. J. (2017). Some recent
and models. Comput.-Aided Civ. Infrastructure Engng 29, No. 7, research on the hydraulic conductivity of road materials.
496–517, https://doi.org/10.1111/mice.12061. In Bearing capacity of roads, railways and airfields, proceedings
Castillo, E., O’Connor, A. J., Nogal, M. & Calviño, A. (2014b). On of the 10th international conference on the bearing capacity of
the physical and probabilistic consistency of some engineering roads, railways and airfields (BCRRA 2017), Athens, Greece
random models. Struct. Saf. 51, 1–12, https://doi.org/10.1016/ (eds A. Loizos, I. Al-Qadi and T. Scarpas), pp. 135–142.
j.strusafe.2014.05.003. London, UK: Taylor & Francis.

Downloaded by [] on [08/09/19]. Copyright © ICE Publishing, all rights reserved.