Professional Documents
Culture Documents
net/publication/242465851
Assessment of Direct Cone Penetration Test Methods for Predicting the Ultimate
Capacity of Friction Driven Piles
CITATIONS READS
104 1,537
2 authors:
Some of the authors of this publication are also working on these related projects:
WHRP/WisDOT Project ID 0092-15-05: Evaluation of WisDOT Quality Management Program (QMP) Activities and Impacts on Pavement Performance View project
Calibration of Resistance Factor for Drilled Shaft according to New FHWA Design Method View project
All content following this page was uploaded by Murad Abu-Farsakh on 06 March 2014.
Abstract: This paper evaluates the applicability of eight direct cone penetration test 共CPT兲 methods to predict the ultimate load capacity
of square precast prestressed concrete 共PPC兲 driven friction piles. Analyses and evaluation were conducted on 35 driven friction piles of
different sizes and lengths that were failed during pile load testing. The CPT methods, as well as the static ␣ and  methods, were used
to estimate the load carrying capacities of the investigated piles (Q P ). The Butler–Hoy method was used to determine the measured load
carrying capacities from pile load tests (Q m ). The pile capacities determined using the different methods were compared with the
measured pile capacities obtained from the pile load tests. Four criteria were selected as bases of evaluation: the best fit line for Q p versus
Q m , the arithmetic mean and standard deviation for the ratio Q p /Q m , the cumulative probability for Q p /Q m , and the histogram and log
normal distribution for Q p /Q m . Results of the analyses showed that the best performing CPT methods are the LCPC method by
Bustamante and Gianeselli as well as the De Ruiter and Beringen method. These methods were ranked number one according to the
mentioned criteria.
DOI: 10.1061/共ASCE兲1090-0241共2004兲130:9共935兲
CE Database subject headings: Driven piles; Pile friction; Statistical analysis; Cone penetration tests; Bearing capacity.
兺
capacity was determined with the corresponding pile head settle-
Q u ⫽Q t ⫹Q s ⫽q b A t ⫹ f i A si (1) ment of less than 19 mm 共0.75 in.兲.
i⫽1
Different methods can be used to determine the ultimate pile
where q b ⫽unit end bearing capacity; A t ⫽cross-section area of capacity from the load-settlement curve obtained during the static
the pile tip; f i ⫽average unit skin friction of the soil layer i; A si pile load test 共e.g., Davisson 1972; Butler and Hoy 1977兲. The
⫽pile shaft area interfacing with layer i; and n⫽number of soil Butler–Hoy method 共1977兲 was selected because it is the primary
layers along the pile shaft. The allowable load capacity (Q all) of load test interpretation method used by LA DOTD. A study con-
the pile is usually calculated by dividing the pile capacity (Q u ) by ducted by Titi et al. 共1999兲 on 34 friction test piles driven into
an appropriate factor of safety 共FS兲. Louisiana soils showed that the pile capacity obtained using the
Several methods are available for calculating the axial pile Davisson method 共1972兲 was almost equal to the capacity ob-
capacity from CPT data. These methods can be classified into two tained by the Butler–Hoy method 共1977兲 with a coefficient of
approaches: determination of 0.99. Therefore the Butler–Hoy method 共1977兲
1. Direct approach: The unit end bearing capacity of the pile can be considered analogous to the Davisson method 共1972兲 for
(q b ) is evaluated from the cone tip resistance (q c ), and the friction piles driven into Louisiana soils. Fig. 1 describes how the
unit skin friction of the pile 共f 兲 is evaluated from either the Butler–Hoy method is used to determine the ultimate load capac-
sleeve friction ( f s ) profile or q c profile. ity of 356 mm square PPC pile driven at the site of Houma Intra-
2. Indirect approach: The CPT data (q c and f s ) are first used to coastal Waterway Bridge.
evaluate the soil strength parameters such as the undrained
shear strength (S u ) and the angle of internal friction 共兲.
These parameters are then used to evaluate the unit end bear- Soil Identification
ing capacity of the pile (q t ) and the unit skin friction of the In order to use the CPT methods to determine the ultimate load
pile 共f 兲 using formulas derived based on semiempirical/ capacity of piles, it is important first to classify the soils using the
theoretical methods. CPT data. Moreover, correlation factors must be applied to the
A summary of the CPT methods used in this study is presented cone tip resistance and sleeve friction to determine unit end bear-
in Table 1. These methods were described in detail in a research ing (q b ) and unit skin friction 共f 兲 in the pile capacity prediction
report by Titi and Abu-Farsakh 共1999兲 and in the references given methodology. These correlation factors depend mainly on the soil
in Table 1. type. Therefore the variation of soil type with depth needs to be
identified so that the appropriate correlation factors relevant to
Research Methodology each method can be selected.
Several methods have been used to classify soil utilizing CPT
data 共e.g., Douglas and Olsen 1981; Robertson et al. 1986; Rob-
Characteristics of Investigated Piles
ertson 1990; Olsen and Mitchell 1995; Zhang and Tumay 1999兲.
Results from static load tests on 35 square PPC piles that failed After a thorough review, the authors selected the probabilistic
under load tests were collected from the LA DOTD files. The region estimation method 共Zhang and Tumay 1999兲 to be used in
piles embedment lengths range from 9 to 38 m and the cross this study. This method is similar to the classic soil classification
sections range from 356 to 762 mm. They are driven into different methods since it is based on soil composition. In this method, a
soil types at different locations. A summary of the characteristics conformal transformation is used to determine the soil classifica-
of the investigated friction piles is presented in Table 2. These tion index 共U兲 from the CPT sounding parameters, the cone tip
piles are categorized based on their size and the type of dominant resistance, and friction ratio. A statistical correlation was estab-
soil they were driven into. Among the 35 piles, 26 piles were lished between the U index and the compositional soil type given
driven in clay soils while the other nine piles were driven in by the Unified Soil Classification System 共USCS兲. A normal dis-
layered soils. tribution of U was established for each reference USCS soil type
d⫽0
d
8D
f sA s⫹ 兺
L
d⫽8D
f sA s 册
k value depends on d/D ratio
De Ruiter and In clay: In clay:
Beringen 共1979兲
q b ⫽N c S u ⭐15 MPa f ⫽␣S u ⭐120 kPa
S u ⫽q ca /N k , N c ⫽9, N k ⫽15 to 20 ␣⫽1 for NC clay and 0.5 for OC clay
再
In sand: In sand
similar to Schmertmann 共1978兲 fs
qca/300 共 compression兲
f⫽min
q ca /400 共 tension兲
120 kPa
LCPC 共Bustamante q b ⫽k b1 q eq(tip) f ⫽q eq(side)/k S1 ⭐ max
and Gianeselli 1982兲
k b1 ⫽0.15– 0.60 depending on soil type k S1 ⫽30– 150 depending on soil type, pile type,
and installation procedure and installation procedure
q eq : equivalent average of q c values of zone ranging
from 1.5D below pile tip to 1.5D above pile tip.
Tumay and Similar to Schmertmann 共1978兲 f ⫽m f ca ⭐72 kPa
Fakhroo 共1982兲
m⫽0.5⫹9.5e ⫺0.09f ca
f ca : average friction 共kPa兲
Aoki and q b ⫽q ca (tip)/F b ⭐15 MPa f ⫽q ca (side)␣ 1 /F S2 ⭐120 kPa
De Alencar 共1975兲
F b : depends on pile type⫽1.75 for PPC ␣ 1 ⫽1.4– 6 depends on soil type F S2 :
driven piles depends on pile type⫽3.5 for PPC driven piles
Prince and q b ⫽k b2 q ca (tip) f ⫽␣ 2 f s
Wardle 共1982兲
k b2 : depends on pile type⫽0.35 for driven piles ␣ 2 : depends on pile type⫽0.53 for driven piles
Philipponnat 共1980兲 q b ⫽k b3 q ca (tip) f ⫽q ca (side)␣ 3 /F S2
k b3 : depends on soil type⫽0.4 for sand, 0.45 ␣ 3 : depends on pile type⫽1.25 for PPC
for silt, and 0.5 for clay driven piles F S2 ⫽50– 200 depending on soil type
Penpile
共Clisby et al. 1978兲 qb⫽ 再 0.25q ca 共 tip 兲 for pile tip in clay
0.125q ca 共 tip 兲 for pile tip in sand
f⫽
fs
1.5⫹14.47f s
f and f s 共MPa兲
Note: q ca ⫽average q c values over a specified zone that depends on the method.
共GP, SP, SM, SC, ML, CL, and CH兲. Each U value corresponds to
several soil types with different probabilities. Soil types were fur-
ther rearranged into three groups: sandy and gravelly soils 共GP,
SP, and SM兲, silty soils 共SC and ML兲, and clayey soils 共CL and
Table 2. Characteristics of Investigated Piles CH兲.
Pile size Clay Clay and Sand The Zhang and Tumay method 共1999兲 provides a profile of the
mm 共in.兲 共number of piles兲 共number of piles兲 probability or the chance of having each soil type with depth;
which shows the chance of misclassifying the soils, similar to
356 共14兲 16 2
other CPT soil classification methods. Fig. 2 shows a comparison
406 共16兲 4 —
of soil classifications obtained using the Robertson et al. 共1986兲
457 共18兲 1 1
method, the probabilistic region estimation method 共Zhang and
610 共24兲 2 —
Tumay 1999兲, and the actual soil profile obtained from conven-
762 共30兲 3 6
tional soil boring at the site of Houma Intracoastal Waterway
Total 26 9
Bridge.
Fig. 2. Cone penetration test 共CPT兲 data and boring log at the site of Houma Intracoastal Waterway Bridge: 共a兲 Profile of cone tip resistance; 共b兲
profile of friction ratio; 共c兲 soil stratification from soil boring; 共d兲 CPT soil classification using Robertson et al. 共1986兲; and 共e兲 CPT soil
classification using Zhang and Tumay 共1999兲.
f 共 x 兲⫽
1
冑2 ln x
冋 冉
exp ⫺
1 ln共 x 兲 ⫺ ln
2 ln 冊册
2
(2)
⫽0.95) tends to underestimate the measured capacity by 3% and
therefore ranks number 2 (R1⫽2). According to this criterion,
the Schmertmann 共1978兲, Philipponnat 共1980兲,  method, and
where x⫽Q p /Q m ; ln⫽mean of ln(Qp /Qm); and ln⫽standard Tumay and Fakhroo 共1982兲 methods tend to overestimate the
deviation of ln(Qp /Qm). measured pile capacity, while the Aoki and De Alencar 共1975兲,
In statistical analyses, the mean 共兲 and standard deviation 共兲 Price and Wardle 共1982兲, Penpile, and ␣ methods tend to under-
of Q p /Q m are important indicators of the accuracy and precision estimate the measured pile capacity. The Penpile method showed
of the prediction method. An accurate and precise method gives the least performance with Q fit⫽0.57Q m (R 2 ⫽0.95) and there-
(Q p /Q m )⫽1 and (Q p /Q m )⫽0, respectively, which means fore was given the rank R1⫽10.
that for each pile, the calculated pile capacity equals the measured It should be noted that the factors used in the CPT methods to
one. This case is ideal, however, in reality the method is better evaluate q b and f are empirical correlations developed by com-
when (Q p /Q m ) is closer to 1 and (Q p /Q m ) is closer to 0. paring pile load test and CPT results. Therefore they can vary
Long and Wysockey 共1999兲 suggested that the cumulative according to the procedure followed in performing and interpret-
probability could also provide an additional evaluation criterion ing the pile capacity from the pile load test. For example, Tumay
of the prediction methods when the ratio Q p /Q m is plotted versus and Fakhroo 共1982兲 used the maximum load on the pile head to
the cumulative probability. The concept is to arrange Q p /Q m val- estimate the ultimate pile capacity from pile load test, which is in
ues for each method in an ascending order (1,2,3,...,i,...,n), and all cases higher than the capacity obtained by the Butler and Hoy
then estimate the cumulative probability 共P兲 from the following method 共1977兲.
equation 共Long and Wysockey 1999兲: The arithmetic mean 共兲 and standard deviation 共兲 of the
ratio Q p /Q m values for each method were calculated and used as
i a second evaluation criterion. Based on and , the CPT meth-
P⫽ (3) ods were given ranks as summarized in Table 3 共column 7兲. Ac-
共 n⫹1 兲
cording to this criterion, the De Ruiter and Beringen 共1979兲
where i⫽order number given for the considered ratio; and n method with ⫽0.982 and ⫽0.25 ranks number one with R2
⫽number of piles. The 50 and 90% cumulative probabilities ( P 50 ⫽1 followed by the LCPC method (R2⫽2). The Schmertmann
and P 90) are calculated and used to quantify the ability of the 共1978兲, Philipponnat 共1980兲, -method, Tumay and Fakhroo
methods to estimate the measured pile capacities. For example, 共1982兲, and ␣ methods all have ⬎1, which means that these
the P 50 is used as a measure of the tendency of the method to methods on average are overpredicting the measured pile capac-
overpredict or underpredict the measured pile capacity. The closer ity. On the other hand, Aoki and De Alencar 共1975兲, Price and
the ratio Q p /Q m is to one the better the method is. Wardle 共1982兲, and the Penpile methods all have ⬍1, which
means that these methods on average are underpredicting the were determined and used to identify the log normal distribution
measured pile capacity. of the density function for each method. The histogram and log
The third evaluation criterion is based on the 50 and 90% normal distribution of Q p /Q m for the investigated methods are
cumulative probabilities, P 50 and P 90 of Q p /Q m . The cumulative presented in Fig. 6. Fig. 7 depicts the comparison of log normal
probabilities versus the ratio Q p /Q m for the investigated methods distributions for the different methods considered in this study.
are presented in Fig. 5. P 50 and P 90 values were determined and The area underneath each curve in Fig. 7 is equal to one.
presented in Table 3 共columns 8 and 9兲. The pile capacity predic- The histogram and log normal probability distribution were
tion method with a P 50 value closer to one and with a lower used to calculate the probability of predicting the ultimate load
P 50⫺ P 90 range is considered the best method. Based on this cri- capacity within 20% accuracy. At a specified accuracy level, the
terion, the LCPC method with P 50⫽1.03 and P 90⫽1.42 ranks probability of predicting ultimate pile capacity is determined by
number one (R3⫽1) followed by De Ruiter and Beringen 共1979兲 calculating the total area underneath the curve within the accuracy
with R3⫽2. The Penpile method has very low P 50 and P 90 values limits. The prediction accuracy obtained from the log normal dis-
and therefore ranks number ten (R3⫽10) according to this crite- tribution for the different methods is plotted in Fig. 8. The higher
rion. the probability the better is the performance of the method. The
The fourth criterion used to evaluate the methods is based on probability corresponding to 20% accuracy is the likelihood that
the histogram and the log normal distribution of Q p /Q m . First, the estimated pile capacity will be within 0.8Q m ⭐Q p ⭐1.2Q m .
the ratio Q p /Q m and the natural logarithm of the ratio ln(Qp /Qm) The probability of predicting the ultimate load capacity within
for each pile were calculated. Then, the mean ( ln), standard 20% accuracy was estimated and presented in Table 3 共columns
deviation ( ln), and coefficient of variation 共COV兲 of ln(Qp /Qm) 11 and 12 for log normal distribution and histogram, respec-
tively兲. Based on the 20% accuracy level, the Bustamante and piles. They concluded that the LCPC method is the best method
Gianeselli 共1982兲 method showed the highest probability values for predicting the ultimate load capacity of piles.
共59.5 and 60%兲 and therefore ranks number one (R4⫽1). De
Ruiter and Beringen 共1979兲 ranks number two (R4⫽2). The Implementation of Cone Penetration Test
Penpile method has the lowest probability value at this accuracy Technology for Pile Design
level and therefore ranks number ten (R4⫽10).
The overall performance of the CPT methods was then evalu- The past practice of LA DOTD in the design and analysis of piles
ated using the four different criteria via RI. For example, RI for was based solely on the static ␣ method. The soil parameters were
the De Ruiter and Beringen 共1979兲 method is determined as RI obtained from laboratory tests on samples from conventional sub-
⫽R1⫹R2⫹R3⫹R4⫽1⫹1⫹2⫹2⫽6. The RI values for all the surface exploration procedures. This procedure is expensive, time
methods are presented in Table 3 共column 14兲. The final overall consuming, and often resulting in a conservative pile design. Pile
rank for each CPT method obtained based on RI values is pre- load tests are also required to verify the design load based on the
sented in column 15 of Table 3. actual field performance of the pile. The number of the required
Based on the results of this analysis, the Bustamante and Gi- pile load tests depends on the project size and importance. The
aneselli 共LCPC兲 共1982兲 and De Ruiter and Beringen 共1979兲 meth- estimated average cost of conventional exploration is $50/ft,
ods have rank number one. These two methods showed the best which takes 7 weeks to complete. This is compared to $20/ft and
performance according to the evaluation criteria. The static ␣ 1 week in the case of the CPT. While the cost of driving and
method ranks number three while the  method ranks number six. loading a test pile in Louisiana usually ranges from $15,000 to
The Penpile method showed the lowest performance among all $25,000, the use of CPT technology will reduce the number of
methods and therefore it ranks number ten. The results of this test piles and soil borings needed for a project resulting in cost
investigation are consistent with the findings of a study performed savings.
by Briaud and Tucker 共1988兲, in which they evaluated the perfor- The use of CPT for pile applications in Louisiana was limited
mance of 13 methods to predict the ultimate load capacity of 98 to identifying the depth of the sand layers to tip the end bearing
Acknowledgments