Interpreting DPSH TECHNICAL PAPER

penetration values Journal of the South African

Institution of Civil Engineering

in sand soils ISSN 1021-2019

Vol 59 No 3, September 2017, Pages 11–15, Paper 1551

C J MacRobert DR CHARLES JOHN MACROBERT, who is an

Associate Member of the South African
Institution of Civil Engineering, received his BSc
(Eng) degree from the University of Cape Town,
Site investigations to classify the underlying soil for geotechnical purposes often rely on in-situ and both his MSc (Eng) and PhD from the
penetrometer tests. Two common tests used in southern Africa are the Standard Penetration University of the Witwatersrand. He is currently
Test (SPT) and Dynamic Probe Super Heavy (DPSH) test. Although the specific work per blow is a lecturer of geotechnical engineering in the
School of Civil and Environmental Engineering
essentially the same in both tests, the resulting penetration values are not equivalent. The DPSH
at the University of the Witwatersrand. His research interests lie
tends to be more variable than the SPT and has higher blow counts. A comparison of SPT and predominantly in the interaction of seepage with geo-materials, both from
DPSH penetration values at a series of strata below sites has been undertaken. From this, new the grain-scale to the macro-scale level.
relative density descriptor boundaries, based on DPSH penetration values, are suggested for
Contact details:
sand soils. School of Civil and Environmental Engineering
University of the Witwatersrand
Private Bag 3
Introduction tests, such as the SPT and DPSH tests, have PO Wits
2050
The Standard Penetration Test (SPT) and been criticised for their poor repeatability, due
South Africa
Dynamic Probe Super Heavy (DPSH) test in particular to hammer energy inefficien- T: +27 11 717 7156
are two common in-situ penetrometer tests cies and rod friction in the case of the DPSH E: charles.macrobert@wits.ac.za
employed in geotechnical site investigations in (Broms & Flodin 1988). In southern African
southern Africa. Although both tests have the practice, the SPT blow count is counted
same specific work per blow (Table 1), the SPT over 300 mm and referred to as an N value;
is carried out in an open hole and the DPSH is likewise, the DPSH blow count is counted over
driven continuously into the soil. Despite this 300 mm and is referred to as an N30SB value.
difference, the two tests are often assumed Reliance on this single qualitative parameter
equivalent (Byrne & Berry 2008). MacRobert to determine requisite engineering design
et al (2011) found this not to be the case and parameters has also been questioned (Mayne et
proposed an empirical correlation between al 2009). Consequently, static tests such as the
the two tests. Since the publication of that cone penetration test (CPT) are increasingly
paper, questions regarding the observed being advocated due to higher accuracy and
variability within the reported data and dif- repeatability (Shukla 2015). Traditional CPT
ferences from other data sets have been raised equipment measures both tip resistance and
(Harrison & A’Bear 2011; Shahien & Farouk sleeve friction, with modern equipment meas-
2013). Furthermore, additional data sets have uring pore pressure and shear wave velocity
become available to the current author. The (Robertson 2009). Engineering parameters can
aim of this paper is to shed light on this vari- therefore be determined from a greater pool
ability and propose new descriptor boundaries of measurements.
to classify the relative density of sand soils Despite the serious deficiencies of dynamic
using DPSH penetration values. tests, they are still popular. This is particularly
because they are cheap and have a long history
of use (Broms & Flodin 1988). Virtually every
Interpretation of geotechnical engineering design para­meter
Penetrometer Results has been correlated with SPT penetration
In-situ penetrometer tests are either “dynamic” values, although many of these correlations
or nominally “static”, that is the probe is either do not give any indication of statistical scat-
hammered or pushed into the soil. Dynamic ter (Mayne et al 2009). Robertson and Cabal

Table 1 Dynamic probe classifications

Test designation DPSH SPT

Hammer mass (kg) 63.5 63.5

Hammer fall (m) 0.76 0.76

Probe 51 mm diameter 90° cone 51 mm diameter split spoon sampler

Specific work per blow (kJ/m 2) 240 240 Keywords: site investigations, in-situ testing, penetrometers, soil classification,
statistical analysis

11
MacRobert CJ. Interpreting DPSH penetration values in sand soils.
J. S. Afr. Inst. Civ. Eng. 2017:59(3), Art. #1551, 5 pages. http://dx.doi.org/10.17159/2309-8775/2017/v59n3a2
Table 2 R
 elative density N descriptor Table 3 Descriptive geology of each site
boundaries for sand soils Average relative
Site Description Water table
(after Terzaghi & Peck 1948) density

SPT resistance value Bellville Transported fine- to medium-grained locally No water

Relative density Medium-dense
(N) South Africa calcareous sand. strikes
Chloorkop Clayey silty sand with fine gravel becoming more No water
0–4 Very loose Medium-dense
South Africa abundant with depth. Reworked residual granite. strikes
4 – 10 Loose Dunkeld Clayey silty coarse sand with traces of sub-angular No water
Very loose
South Africa quartz gravel. Reworked residual granite. strikes
10 – 30 Medium dense
Glenhazel Silty sand with fine gravel at depth. Fill, hillwash No water
Loose
30 – 50 Dense South Africa and reworked residual granite. strikes
Matutuíne No water
Over 50 Very dense Medium and fine sand. Medium-dense
Mozambique† strikes
Namakwa No water
Non-plastic screen-separated sand. Dense
South Africa† strikes
(2012) conclude that SPT penetration values
are suited to determining relative density of Mt
On average
Edgecombe Slightly clayey transported sand. Medium-dense
predominantly sand profiles, but only moder- below 21.5 m‡
South Africa†
ately so. This is typically done by comparing N
Parktown Profile of mixed origin, predominantly silt and No water
values to the descriptor boundaries proposed Loose
South Africa sand. strikes
in Table 2 (Terzaghi & Peck 1948).
Milnerton Transported loose to medium slightly silty-fine On average
Variability in penetration values arises Medium-dense
South Africa sand. below 0.6 m
from lateral variation in the soil profile and
Chicalla Fine to medium-grained sand with abundant shell On average
different testing procedures. It is difficult Loose
Angola fragments. below 14.7 m
to separate these two sources of variability. Umdloti Slightly moist to moist, fine to medium through On average
Medium-dense
Serota and Lowther (1973) determined that South Africa coarse-grained sand. below 7.3 m
the coefficient of variation (COV), defined Gope Transported sand cover with a thin variable layer No water
Medium-dense
by Equation 1, for N values in a calibration Botswana of poorly developed calcified pedogenic material. strikes
chamber is between 12 and 10% for automatic Matola On average
Silty sand dune deposit. Dense
trip hammers. Mozambique below 21.0 m
† Due to the extensive nature of these sites, a single site profile was not developed. Instead, tests in close
s proximity were compared.
COV = (1) ‡ One probe over a depth of 1 m was in saturated soil.
x̄

where x̄ is the sample mean and s is the sam- that the average COV in sand was 54% and in the former study predominantly reflects
ple standard deviation. ranged between 19% and 62%. The larger variation within the testing procedure. No
variability in the later study reflects lateral studies on variability in the DPSH are appar-
Phoon and Kulhawy (1999), considering variation in site soil profiles and various ent in literature; however, similar variability
published site investigation data, suggested hammer mechanisms, whereas the variability to that reported for the SPT is likely.

Table 4 Summary of penetration testing data from each site

Average Maximum SPT (N) Summary statistics DPSH (N30SB) Summary statistics
Initial
refusal refusal
Site depth
depth depth Number of Average COV Number of Average COV
(m)
(m) (m) profiles† (COV range) profiles (COV range)

Bellville,South Africa 1.2 4.2 5.1 1 - 2 25% (50 – 6%)

Chloorkop, South Africa 0.9 5.6 8.1 3 33% (55 – 15%) 2 19% (54 – 6%)

Dunkeld, South Africa 1.2 4.5 5.1 1 - 2 52% (100 – 25%)

Glenhazel, South Africa 0.9 2.0 4.2 1 - 3 41% (91 – 15%)

Namakwa, South Africa 2.1 11.7 26.7 2 (2) - 8 (2) 33% (127 – 5%)

Matutuíne, Mozambique 0.9 3.1 6.6 4 (2) 29% (76 – 6%) 19 (2) 68% (96 – 4%)

Mt Edgecombe, South Africa 1.5 9.2 15.3 15 (6) 17% (64 – 1%) 18 (6) 19% (90 – 2%)

Parktown, South Africa 0.9 4.1 6.3 8 105% (164 – 78%) 19 139% (222 – 88%)

Milnerton, South Africa 0.9 3.0 3.9 2 14% (35 – 2%) 2 26% (56 – 1%)

Chicalla, Angola 1.2 7.2 10.8 3 31% (47 – 12%) 8 27% (56 – 4%)

Umdloti, South Africa 1.2 3.7 6.3 3 42% (88 – 8%) 10 37% (96 – 6%)

Gope, Botswana 1.2 6.5 9.3 5 23% (46 – 3%) 11 43% (84 – 28%)

Matola Mozambique 1.5 9.6 13.8 17 26% (33 – 19%) 17 22% (41 – 7%)

† Values in parentheses indicate number of subsites considered (see note in Table 3)

12 Journal of the South African Institution of Civil Engineering  •  Volume 59  Number 3  September 2017
Correlation between for all strata at each site and gives the average.  N30SB N 
N and N30SB The average values are generally towards the 200 150 100 50 0 50 100
lower range of the limits (19% to 62%) reported 0
Strata:
Test sites by Phoon and Kulhawy (1999). This suggests
1 1.5
To investigate the correlation between N and that there was limited lateral variation in the
N30SB, data from 13 site investigations were soil profiles. One site that exhibited signifi-
2 3
analysed. Data was collected from various cantly greater variability was the Parktown
engineering and contracting companies, site. This site was characterised by material
3 4.5
with probing carried out according to best of mixed origin that included coal, ash and
practice in southern Africa (MacRobert et al refuse which contributed to the large variation
4 6
2010). Consequently, relationships developed observed. Disregarding this site as anomalous,

Depth (m)
may not be applicable for different hammer the average COV for N30SB was 32% and for
5 7.5
efficiencies and where probing practices N was 25%. Although the average COV values
differ. Table 3 shows that all profiles probed for the N30SB and N are similar, it is clear from 6 9
consisted of sand soils, with Table 4 giving the ranges of COV values for the two tests that
details of the probing undertaken at each N30SB showed greater variability. 7 10.5
site. Most sites were small and borehole logs
indicated similar soil profiles, and so all N Statistical methodology 8 12
and N30SB profiles for such sites could be In light of the variability in penetration
compared. For sites where probing was over values, individual values were not compared. 9 13.5
a large area, N and N30SB profiles were sepa- Rather, the range of N values were compared
rated into subsites with similar soil profiles to the corresponding range of N30SB values 15
based on borehole logs. within a stratum across a site. Consequently
Soil profiles below sites (or subsites) were energy corrections, such as proposed by Figure 1 Matola strata
divided into 1 m thick strata centred at the Skempton (1986), were not applied, as these
depths where N values were determined. An are more appropriate when considering assigned a relative density based on which
example of these strata for the Matola site is individual N values. Assuming N values to relative density resulted in the highest prob-
given in Figure 1. The average and standard be normally distributed, the probabilities of ability. A comparison was then made between
deviation of all N values within a stratum each stratum being classified into each of the relative density assigned by N values and
were then determined. For sites with only the five relative density ranges (Table 2) were N30SB values. The N30SB boundaries were sub-
one N profile, the standard deviation was calculated. Each stratum was then assigned sequently optimised, using the entire data set,
calculated assuming a COV of 25%. The cor- a relative density based on which relative to maximise the number of strata assigned the
responding range of N30SB values was deter- density resulted in the highest probability. same relative density by both tests.
mined as the average and standard deviation Equivalent N30SB relative density
of all N30SB values within each 1 m thick boundaries (Table 6) were calculated using the
stratum. This resulted in a series of strata for empirical correlation proposed by MacRobert Results
which average N and N30SB values and asso- et al (2011) from N boundaries in Table 2. Prior to considering the entire data set,
ciated standard deviations were known. Assuming N30SB values to be normally distrib- results for the Matola site are discussed.
For each of these strata the COV values uted, the probabilities of each stratum being For each of the strata (Figure 1) average
of the N and N30SB values were calculated. classified into each of these relative density and standard deviation of N and N30SB
Table 4 summarises the range of COV values ranges were calculated. Each stratum was then values are given in Table 5. The calculated

Table 5 Matola site statistical output

MacRobert et al (2011)
Terzaghi and Peck (1948) boundaries† Optimised boundaries‡
boundaries‡
N Probability associated N30SB Probability associated
Probability associated
with relative density with relative density
Strata with relative density

COV Very Med- Very COV Very Med- Very Med-

x̄ s n Loose Dense x̄ s n Loose Dense Loose Dense
(%) loose dense dense (%) loose dense loose dense

1 25 8 32 15 0.004 0.026 0.704 0.265 0.001 41 17 41 39 0.013 0.021 0.834 0.132 0.023 0.033 0.933 0.011

2 27 11 41 17 0.018 0.043 0.546 0.374 0.018 64 24 38 39 0.006 0.007 0.422 0.566 0.009 0.010 0.729 0.252

3 35 8 23 15 0.000 0.001 0.265 0.704 0.030 86 28 33 39 0.002 0.002 0.173 0.823 0.002 0.003 0.410 0.585

4 39 10 26 17 0.000 0.002 0.182 0.680 0.136 107 23 21 39 0.000 0.000 0.020 0.979 0.000 0.000 0.120 0.880

5 36 8 22 15 0.000 0.001 0.226 0.733 0.040 124 17 14 31 0.000 0.000 0.000 1.000 0.000 0.000 0.005 0.995

6 39 12 31 17 0.002 0.006 0.219 0.594 0.180 138 13 9 24 0.000 0.000 0.000 1.000 0.000 0.000 0.000 1.000

7 37 8 22 16 0.000 0.000 0.190 0.757 0.052 140 13 9 12 0.000 0.000 0.000 1.000 0.000 0.000 0.000 1.000

8 50 13 26 15 0.000 0.001 0.061 0.438 0.500 143 5 3 6 0.000 0.000 0.000 1.000 0.000 0.000 0.000 1.000

9 57 28 49 6 0.029 0.017 0.121 0.234 0.599 149 1 1 3 0.000 0.000 0.000 1.000 0.000 0.000 0.000 1.000

†see Table 2  ‡see Table 6

Journal of the South African Institution of Civil Engineering  •  Volume 59  Number 3  September 2017 13
probability for each relative density, for Table 6 Relative density N30SB descriptor boundaries for sand soils
each stratum, based on N values is given.
N30SB boundaries by
Each stratum’s assigned relative density and Relative density Optimised N30SB boundaries
MacRobert et al (2011)
associated probability are indicated by bold
Very loose 0–3 0–7
type. Results from calculations performed
on N30SB values are presented in a similar Loose 3 – 10 7 – 14
fashion. Based on the MacRobert et al (2011) Medium-dense 10 – 60 14 – 80
boundaries, 6 stratum are assigned the same
Dense > 60 > 80
relative density by both tests, 1 is assigned a
higher relative density by N30SB values and 2
are assigned a lower relative density by N30SB define a single equation to obtain equivalent strata assigned a lower relative density by
values. With the optimised boundaries, N from N30SB. However, it is possible to N30SB values than by N values. Ordinates are
7 stratum are assigned the same relative assign a relative density to a stratum with the probabilities that the N values would give
density by both tests and 2 are assigned some confidence. the same lower relative density. In this case,
a lower relative density by N30SB values. With the MacRobert et al (2011) descrip- the average probability that N values would
Considering the two strata assigned lower tor boundaries, 49% of the strata were give the same lower relative density is 0.15.
relative densities by N30SB values, the prob- assigned the same relative density by both Whilst this probability is small, it is greater
abilities that these strata would be assigned tests, 29% were assigned a lower relative than 0.05, suggesting there is nevertheless a
the same lower relative density by N values density by N30SB values (i.e. a conservative significant chance that the strata are correct-
are significant (> 0.05). This suggests that estimate), and 22% were assigned a higher ly defined by N30SB values. Figure 3(c) shows
the optimised boundaries are adequate for relative density by N30SB values. With the strata assigned a higher relative density by
categorising strata. optimised descriptor boundaries, 57% of the N30SB values than by N values. Ordinates
Figure 2 illustrates the average N and strata were assigned the same relative density are the probabilities that the N value would
N30SB values for each stratum and for each by both tests, 31% were assigned a lower give the same higher relative density. In this
site. In general, N30SB values are greater than relative density by N30SB values (i.e. a con- case, the average probability that N values
respective N values. This is due to N30SB servative estimate), and 12% were assigned a would give a similar higher relative density
values increasing with depth at a greater rate higher relative density by N30SB values. is 0.07, so there is a much lower chance that
than N values. Harrison and A’Bear (2011) Calculated probabilities associated with the strata are correctly defined by N30SB
attributed this to rods bowing during prob- defining a stratum’s relative density by N values. Referring to Figure 2 it is evident
ing, causing jamming and sidewall collapse. values and N30SB values (optimised descrip- that most of the strata assigned a higher
This suggests that a correlation varying with tor boundaries) are considered in Figure 3. relative density based on N30SB are from the
depth may be appropriate. However, no such Figure 3(a) considers strata assigned the Chicalla site. As pointed out by MacRobert
relationship was apparent when analysing same relative density by both tests. As et al (2011), the ground profile at this site
the data. It is evident that the equation expected, the confidence with which these contained numerous shell fragments which
proposed by MacRobert et al (2011) is not strata are assigned a relative density is high may have resulted in the higher N30SB values.
sufficiently accurate to obtain equivalent N in both tests. However, individual prob- This highlights the need for a local know­
values from N30SB values. From the scatter abilities are not comparable, as points do not ledge of geology when interpreting N30SB
in the graph, it is evidently impossible to lie along a line of equality. Figure 3(b) shows values. Points marked with a red asterisk in

70
Bellville

Terzaghi and Peck (1948) Chloorkop

60 boundaries
Line of equality Dunkeld

Glenhazel
50
Matutuine
MacRobert et al (2011)
40 Namakwa

Mt Edgecombe
N

30 Parktown
Optimised boundaries
Milnerton
20
Chicalla

Umdloti
10
Gope

Matola
0
0 20 40 60 80 100 120 140 160
N30SB

Figure 2 Correlation between N and N30SB values

14 Journal of the South African Institution of Civil Engineering  •  Volume 59  Number 3  September 2017
 Probability based on N 1.0 1.0 1.0

Probability based on N

Probability based on N
0.5 0.5 0.5

0 0 0
0 0.5 1.0 0 0.5 1.0 0 0.5 1.0
(a) Probability based on N30SB (b) Probability based on N30SB (c) Probability based on N30SB

Figure 3 P
 robability associated with assigning the same relative density to a stratum by N and N30SB for: (a) similar cases, (b) conservative cases and
(c) unconservative cases

Figure 2 indicate strata where the probability kindly provided valuable comments on testing. Proceedings, 17th International Conference
N would give the same relative density, as the manuscript. on Soil Mechanics and Geotechnical Engineering,
N30SB is less than 0.05. These points make up Alexandria, Egypt. Amsterdam: IOS Press,
12% of the data set. 2777–2872.
References Phoon, K-K & Kulhawy, F H 1999. Characterization
Broms, B B & Flodin, N 1988. History of soil penetration of geotechnical variability. Canadian Geotechnical
Conclusions testing. Proceedings, 1st International Symposium on Journal, 36(4): 612–624.
A statistical analysis of 13 site investigations, Penetration Testing, ISOPT-1, 20–24 March, Orlando, Robertson, P 2009. Interpretation of cone penetration
in which 65 SPT and 121 DPSH profiles FL. Rotterdam: AA Balkema, 157–220. tests – a unified approach. Canadian Geotechnical
were determined, was undertaken. This Byrne, G & Berry, A D 2008. A guide to practical Journal, 46(11): 1337–1355.
was used to propose new N30SB relative geotechnical engineering in southern Africa. Sandton: Robertson, P & Cabal, K 2012. Guide to cone
density descriptor boundaries for sand Frankipile South Africa. penetration testing for geotechnical engineering.
soils. Considering the inherent variability of Harrison, B & A’Bear, T 2011. The dynamic probe Signal Hill, CA: Gregg Drilling & Testing Inc.
penetration values obtained from penetro­ super heavy penetrometer and its correlation with Serota, S & Lowther, G 1973. Discussion: Accuracy of
meters, it is clear that defining a single the Standard Penetration Test. In: Quadros, C & relative density measurements. Géotechnique, 23(2):
equation to determine equivalent N values Jacobsz S W (Eds.), Proceedings, 15th African 301–303.
from N30SB values is futile. The practice Regional Conference on Soil Mechanics and Shahien, M M & Farouk, A 2013. Estimation of
of using N30SB values to obtain anything Geotechnical Engineering, Maputo, Mocambique. deformation modulus of gravelly soils using dynamic
more than an estimate of relative density is Amsterdam: IOS Press, 571–579. cone penetration tests. Ain Shams Engineering
therefore unwarranted. MacRobert, C, Kalumba, D & Beales, P 2010. Journal, 4(4): 633–640.
Penetration testing: Test procedures and design use Shukla, S K 2015. Core concepts of Geotechnical
in South Africa. Civil Engineering, 18(3): 29–38. Engineering. London, ICE Publishers.
AcknowledgEments MacRobert, C, Kalumba, D & Beales, P 2011. Skempton, A 1986. Standard penetration test
The author acknowledges the various geo- Correlating standard penetration test and dynamic procedures and the effects in sands of overburden
technical consultants and contractors for probe super heavy penetration resistance values in pressure, relative density, particle size, ageing and
kindly providing the analysed data. Stuart sandy soils. Journal of the South African Institution overconsolidation. Géotechnique, 36(3): 425–447.
Hoepper’s assistance in collecting the data of Civil Engineering, 53(1), 46–54. Terzaghi, K & Peck, R B 1948. Soil Mechanics In
is acknowledged; his untimely passing is a Mayne, P W, Coop, M R, Springman, S M, Huang, A-B Engineering Practice. New York: Wiley.
reminder of life’s fragility. Dr Irvin Luker & Zornberg, J G 2009. Geomaterial behavior and

Journal of the South African Institution of Civil Engineering  •  Volume 59  Number 3  September 2017 15