Vous êtes sur la page 1sur 8

A new approach to the analysis of high-strain dynamic $ile test data

-%

YVESROBERT
Quiforrnat Ltd., 591 Le Breton, Longueuil, QC J4G IR9, Canada
Received November 10, 1992
Accepted November 22, 1993

High-strain dynamic pile tests using the pile driving analyzer (PDA) and the Case method have been available in
Canada for over 15 years. During that period of time, the hardware has evolved considerably but the way the data
is interpreted has basically remained the same. The evaluation of the bearing capacity of the tested pile still uses the
Case method, with the Case J factor being calculated either from the results of a static load test or, more often,
from the results of one or more CAPWAP (Case pile wave analysis program) analyses. A computer program has been
developed to estimate the bearing capacity of piles using the dynamic test results produced by the PDA. This pro-
gram uses direct correlations between PDA data and CAPWAP results. It also includes an artificial intelligence module
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20

trained to predict CAPWAP pile capacity. It is not designed to replace the CAPWAP program, but rather to indicate
when an additional CAPWAP analysis might be required.
Key words: dynamic load test, computer program, artificial intelligence, CAPWAP, Case method.

Les essais de chargement dynamiques de pieux h grande deformation utilisant I'analyseur de battage de pieux
(PDA) et la mCthode Case sont disponibles au Canada depuis environ 15 ans. Pendant cette pCriode, I'Cquipement
a CvoluC considCrablement, mais la mCthode d'interprktation est demeurCe sensiblement la m2me. L'evaluation de la
capacitC portante d'un pieu d'essai se fait toujours h l'aide de la mCthode Case, le facteur Case J Ctant calculk h partir
des rCsultats d'un essai de chargement statique ou, plus souvent, h partir des rCsultats d'une ou plusieurs analyses
CAPWAP. Un programme d'ordinateur a CtC dCveloppC pour estimer la capacitC portante des pieux en utilisant les
rCsultats des essais dynamiques produits par l'analyseur de battage de pieux. Ce program@e utilise des corrCla-
tions entre les donnCes fournies par l'analyseur de battage de pieux et les rCsultats du programme CAPWAP. I1
contient aussi un module d'intelligence artificielle entrain6 h prCdire la capacitC portante estimCe par le programme
CAPWAP. I1 n'a pas CtC c o n p pour remplacer le programme CAPWAP, mais plut8t pourindiquer qu'une analyse
For personal use only.

CAPWAP additionnelle pourrait Stre requise.


Mots clks : essai de chargement dynamique, programme d'ordinateur, intelligence artificielle, CAPWAP, mCthode Case.

Can. Geotech. J. 31, 246-253 (1994)

Introduction estimate the mobilized static bearing capacity of the test


High-strain dynamic testing of piles (ASTM 1990) has pile, as shown in Fig. 1. The Case method uses the force
been available for about 20 years. The first pile driving ana- and velocity values corresponding to times TI and T, that
lyzers (PDA) were simple analog computers without any are separated by a time interval equal to twice the pile
storage capability. In 20 years, the PDA has evolved con- length L (below the gauges) divided by the stress wave
siderably and is now a flexible digital microcomputer with. speed C, i.e., 2LIC. For a pile with a uniform cross sec-
graphic display and internal storage. However, even though tion, [I]-[4] below can be used to summarize the Case
some new parameters have been added, the basic equations method. It should be pointed out that these equations assume
used to calculate the bearing capacity of the test piles are still linear elastic pile behavior, only axial stresses in the pile, no
the same. internal pile damping, and a rigid plastic soil resistance
Basically, the testing system consists of two pairs of model. T h e latter of these a s s u m p t i o n s i s of c o u r s e a
gauges attached to the opposite sides of the test pile, at the simplification of the actual soil behavior.
same distance from its toe. Each pair is composed of a strain (FT, + F T ~ ) E A ( V T , + V T ~ )
transducer and an accelerometer. During a hammer blow, [I] RTL = +-
2 C 2
the measured strain is converted to force using the pile
cross-sectional area and the dynamic modulus of elasticity at
the measuring location. The velocity of the pile, at the loca-
tion of the gauges, is calculated by integrating the acceler-
ation signal with respect to time.
The force and the velocity signal are then used by the
PDA to compute the soil resistance mobilized during the
blow. If the ultimate resistance of the soil has not been
mobilized (very little or no permanent pile penetration), a
larger blow, if possible, can be given to the pile. where RTL is the total resistance (static plus dynamic resis-
To be used by the PDA, the force and velocity signals tance), E is the elastic modulus, A is the cross section of
are digitized into fixed values (usually 1024 samples for the pile, C is the stress wave speed, VB is the computed pile
each signal) that are separated by fixed time intervals of toe velocity at time ( T I + TJ2, D is the damping force
typically 100 k s . (increase in resistance caused by the high loading rate), J is
The Case method (Rausche et al. 1985) is then used to a dimensionless damping coefficient, and RS is the com-
Prinlcd in Canada I lmprimc au Canada
ROBERT 247

--
LIC
- - -.
4
-
,-,

*j--VELOCITY
x (EAIC)
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20

FIG. 1. Bearing capacity calculation (Case method).

puted mobilized static capacity. Time T , is usually the time STATIC LOAD TEST CAPACITY (kN)
of maximum velocity T p or the time of impact. The mobilized
soil resistance is then RS1. However, large soil quakes FIG.2. CAPWAP capacity vs. static load test results. Data sources:
(deformation needed to go from an elastic to a plastic soil 1, Hannigan and Webster 1987; 2, Cheng and Ahmad 1988;
behavior) will sometimes delay the time when the full resis- 3, Thompson and Goble 1988; 4, Seidel et al. 1988; 5, Abe et al.
tance is activated. In these cases, the PDA can vary T , one 1990; 6, Hussein and Likins 1991; 7, Federal Highway
Administration 1991; 8, Morgano 19885-
time increment at a time, from T p to T p + 3 0 ms, and cal-
culate a new RS. It then searches for the maximum calculated :.
RS value or RMX. and static load test capacity. ~ h e s edata were obtained
Of all the parameters needed to evaluate RS1 or RMX, through a survey of a limited database. The results shown in
the only unknown is the constant J. It should be pointed Fig. 2 are from load tests carried out since 1985. They cor-
For personal use only.

out, however, that the damping force is assumed to be respond to concrete and steeI piles driven into either sand,
directly proportional to the pile toe velocity. Therefore, silt, or clay. To be included in this figure, the ultimate sta-
J will have little influence on the determination of the sta- tic capacity of the piles had to be mobilized in both the
tic bearing capacity of the test pile when VB is small, as is dynamic and static tests. In all the other cases, the test
the case when the pile is resting on a very hard soil layer. For (either static or dynamic) that mobilized the ultimate capacity
the same reason, RMX is typically less sensitive than RS1 to of the pile overpredicted the other.
the variations of J, since RMX is activated later than R S l , Figure 2 is only presented to show that the application
and therefore VB tends to be smaller in this case. of the Case method with CAPWAP analyses can provide sat-
The constant J usually ranges from 0 to 1 and is nor- isfactory results when compared with carefully carried out
mally higher for fine-grained soil (Rausche et al. 1985). It can static load tests.
be evaluated quite easily if the static bearing capacity of Most of the static test results shown in Fig. 2 are based on
the test pile is known. This can be achieved by performing the Davisson's method (Fellenius 1980) of interpretation of
a static load test on the same pile or a CAPWAP (Case pile failure load. This method is widely used for the interpreta-
wave analysis program) analysis on the data recorded by tion of static load test results and it generally provides a
the PDA during the test (see below). lower bound bearing capacity for the test pile. The CAPWAP
A static load test is very expensive to perform. It takes results can therefore be regarded as conservative with respect
time and if a load cell is not used to measure the applied to other failure criteria, since they generally correlate well
load, as is usually the case, the error on the measured bear- with the Davisson's method.
ing capacity can be quite large (Fellenius 1984). It should be pointed out, however, that both types of pile
CAPWAP is a computer program designed to create a model load tests have limitations.
of the interaction between t h e pile a n d t h e soil ( P i l e Static load tests are usually carried out without the use
Dynamics Inc. 1990). This is achieved by using the theory of a load cell, and the error on the maximum load applied to
of wave propagation, the known information about the pile the piles can be quite large, as mentioned earlier. Their
and one signal (velocity, for instance) recorded during a interpretation is often difficult, since the bearing capacity
hammer blow on the pile to evaluate what the other recorded of the test piles will be a function of the failure criteria
signal (force for instance) should look like. Using an itera- used in the interpretation (Fellenius 1980). Furthermore,
tive process, the resistance distribution along the pile, the soil the duration of the static load test will sometimes influence
quakes, damping, and other dynamic characteristics of the the results.
soils are varied and the calculated curve is compared with the Dynamic load tests are based on a simplified soil model.
measured curve until a good match is achieved. It is then They generally give good results, but should only be car-
possible to obtain a good model and therefore determine ried out by experienced personnel. They are also sometimes
the mobilized static bearing capacity. It should be pointed out, influenced by the available driving equipment (uneven driv-
however, that since CAPWAP uses a simplified soil model, a ing cushion for drop hammers, limited driving energy for
good match is sometimes difficult to achieve and the analysis diesel hammer, etc.).
can then take a long time to carry out. However, dynamic load tests are generally considered a
Figure 2 shows a comparison between CAPWAP capacity good alternative to static load tests because they can solve
248 CAN. GEOTECH. J. VOL. 31, 1994 \

0 1000 2000 3000 4000 5000 6000


Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20

CAPWAP CAPACITY (kN)


0.0 1
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
SKIN FRICTION ICAPWAP CAPACITY

FIG.4. Influence of skin friction on RAU.

This program is used to provide an estimated value of bear-


ing capacity which can be comparqd against a value com-
puted using the Case method and an appropriate J constant
(either a conservative value used:during the test or a value
provided by CAPWAP analyses carried out for one or more
piles tested on the same site). Whenever the two estimated
bearing capacities are significantly different, a detailed
For personal use only.

CAPWAP analysis is probably required.


0 1000 2000 3000 4000 5000 6000
Linear correlations
CAPWAP CAPACITY (kN)
The PDA can provide many parameters to help the expe-
FIG.3. Upper bound (a) and lower bound (b) for CAPWAP rienced operator to estimate the J constant and make a pre-
capacity prediction. liminary estimate of the bearing capacity during the test.
An upper bound capacity can be calculated using the
problems more quickly and make it possible to test more Hiley formula (Hiley 1930) as follows:
piles for a lower cost and thus provide more information
on the overall quality of the foundation piles; they also give 2 EMX
[5] Rup=
information on the structural integrity of the tests piles. D,,, + DMX
However, it is important to point out that, like a static
pile test, a dynamic pile test gives the bearing capacity of a . where EMX and DMX are, respectively, the maximum trans-
pile at the time it is carried out. Changes in bearing capac- ferred energy to the pile and the maximum displacement of
ity with time (relaxation, soil setup) cannot be taken into the pile at the location of the gauges, as calculated by the
account. Resting periods of up to 48 h in the case of relax- PDA; and D,,, is the final set of the pile, as measured
ation and longer in the case of soil setup are therefore rec- directly on the pile by the contractor. As shown in Fig. 3a,
ommended before carrying out any pile load test whenever Rup generally provides a good upper bound capacity.
the bearing capacity of a pile might change with time. It is The data shown in this figure and in the following fig-
also important to carry out both static and dynamic load ures presented in this paper have been gathered in eastern
tests after the same resting period. Canada since 1987. The test piles are uniform steel H and
pipe (open and closed bottom) piles driven into sand, silt,
New approach clay, till, shale, or a combination of these types of soil.
CAPWAP analysis takes anywhere from half an hour to sev- Another parameter provided by the PDA is RAU, which
eral hours to perform, depending on the complexity of the is the total mobilized capacity calculated when V , is equal
soil-pile model. Therefore, it is not usually possible to per- to zero or the total mobilized capacity calculated when RTL
form a CAPWAP analysis for every pile tested. On small becomes equal to RS for the first time after the impact. As
projects, one or two CAPWAP analyses are normally enough shown in Fig. 3b, RAU is a very good lower bound capacity.
to define the value of the J factor to b e used with the T h e RAU computation assumes that the resistance is
Case method. For large projects where soil conditions and located at the pile toe, and it generally works well if there is
pile type vary, a large number of CAPWAP analyses may be relatively little skin resistance as shown in Fig. 4.
required to achieve good results (Riker and Fellenius 1988). RAU and Rup are the basis of the preliminary analysis
To facilitate this new approach, a computer program, of the test results provided by the PDA, since they are unaf-
based on empirical relationships and an artificial intelli- fected by J. However, they show a large scattering and there
gence technique called neural network (Lawrence 1988), is often a large difference between them. Nevertheless, RAU
has been written to improve the quality of the test results. is very useful because it generally underestimates CAPWAP
ROBERT
> \
249
5000

4000

,3000
z2
LT
2000

1000 AVERAGE : 1,019

STANDARD DEVIATION :
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20

n
" A
0 1000 2000 3000 4000 5000
CAPWAP CAPACITY (kN)

FIG.5. RA2 vs. CAPWAP capacity. FIG.7.Influence of total skin friction on RA2.
5000

4000

,3000
B
For personal use only.

2LT
2000

CORR. RA2 ICAPWAP CAPACITY


1000
AVERAGE : 0.999
0.2
STANDARD DEVIATION : 0.115 %

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
CAPWAP SKIN FRICTION 1 CAPWAP TOTAL CAPACITY CAPWAP CAPACITY (kN)

FIG.6.Influence of CAPWAP skin friction on RA2. FIG.8. RA2 corrected using total skin friction from the pile
driving analyzer.
capacity (Fig. 4) and leads therefore to a conservative pre-
be done, however, if the amount of skin resistance with
diction of static pile capacity.
respect to the pile bearing capacity is known and, of course,
Another very useful parameter provided by the PDA is
if a CAPWAP analysis is available, since it is not possible to
RA2. RA2 can be defined as the total mobilized capacity
get the mobilized static skin resistance from the PDA.
of the pile when it becomes equal to twice the pile velocity
However, the PDA provides SFT, which is an estimate
(at the location of the gauges) multiplied by EAIC, or
of the total skin resistance (static and dynamic). Figure 7
[6] R A 2 = R T L ( T ) shows the relation between R A ~ / C A P W Acapacity
P and
if SFTlRA2. A linear regression analysis performed on the
data of Fig. 7 gives
RA2
[a] -= 0 . 1 0 6 ~ + 0 . 9 7 4
where T is the time at which RTL is being calculated, and CAPWAP RA2
V(T) is the velocity at the location of the gauges at time T A s shown in Fig. 7 , R A 2 overestimates the CAPWAP
after the beginning of the blow. RA2 is also unaffected by capacity when SFTlRA2 is greater than 0.25 and under-
J. It shows less scatter than RAU and Rup and, even though estimates it when SFTlRA2 is smaller than 0.25. If the data
it is not always a very good estimate of the mobilized static points are rotated clockwise with respect to a point located
bearing capacity, it is a good starting point as shown in Fig. 5. at SFTIRA2 = 0.25 and RA~ICAPWAP capacity = 1, a slight
RA2 has a tendency to overestimate the bearing capac- improvement in bearing capacity prediction should be
ity of piles with high skin resistance and to underestimate it possible. This operation is carried out using the mirror
for piles with high toe resistance, as shown in Fig. 6. Thus, image of [8] with respect to an axis of symmetry corre-
if RA2 is reduced for high skin resistance piles and increased sponding to a R A ~ I C A P W Acapacity
P ratio of 1 (eq. [a]).
for high toe resistance piles using the results presented in Rup and RAU are also used as upper and lower bounds to
Fig. 6, its predictions should be improved. This can only reduce the scatter.
250 CAN. GEOTECH. J . VQL. 3 1 . 1994
\ \

AVERAGE : 0.994

STANDARD DEVIATION : 0.081


Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20

CAPWAP CAPACITY (kN)

FIG.9. Predicted CAPWAP capacity vs. CAPWAP capacity (after


10 linear corrections).

Figure 8 shows the predicted capacity (RAP) obtained as


follows:

[9]
[
R A P = R A 2 -0.106-+ SFT
RA2
1.027
For personal use only.

A slight improvement over Fig. 5 is noticeable.


The same process can also be used with other parameters
provided by the PDA, such as forces, velocities and dis- PREDICTED CAPACITY I CAPWAP CAPACITY
placements. The exact equations to be used will vary with the
FIG. 10. Distribution of predicted capacity with respect to
data available, the order in which each linear correlation is CAPWAP capacity (after 10 linear corrections).
used and possibly the type of pile used, (steel, concrete, or
timber; uniform or nonuniform; driven or cast in place). Of
course, since many of these parameters are interrelated, the put layer provides the answer. The hidden layer or layers
improvements provided by each new correlation should be are needed to store the model created during training.
less significant. Furthermore, as the database increases in Each neuron in a hidden layer is connected to every neu-
size, the influence of a particular soil o r pile type will ron in the adjacent layers but not to the other neurons of
become less significant and some of the correlations used its own layer. Each connection has a strength or weight that
to produce Fig. 9 may no longer be required. is used to modify the output of the neurons. The weights
Figure 9 shows what can be achieved with 10 linear cor- can be positive, which will tend to make the neuron g o
relations. The predicted capacity (RAP) has less scattering high, or negative, which will tend to make the neuron go
than RA2. In fact, 95% of the RAP values are within +15% low. The training process changes these weights to get the
of the CAPWAP capacity, as shown in Fig. 10. correct answers.
It should be pointed out that each particular value of RAP All the inputs of a neuron (either the data used for the
will be a function of the correlations used but, on average, prediction or the output from the neurons of a previous
an improved bearing capacity prediction should be possible layer) are combined into a single value (X), which is then
when the above-mentioned procedure is applied to PDA data. applied to a transfer function. The transfer function con-
trols what the output of a neuron looks like.
Neural network A transfer function often used is the sigmoid function
A back-propagation neural network is a form of artificial described by
intelligence. It is not programmed to follow rules but learns (HIGH - L O W ) +LOW
[lo' T F ( X ) = 1+ e , - GAIN ( x-CENTRE)]
by example and repetition. When a pattern is presented to the
network, it tries to predict the right answer. The feedback, or by [ l l ] when LOW, CENTRE, HIGH, and GAIN are,
which is then sent to the network, will be used to modify it respectively, equal to 0, 0, 1, and 1, as shown in Fig. l l b :
to improve the quality of the predictions. In other words,
a neural network learns from its mistakes. Neural networks
are good at finding patterns in a set of data and then creating
a model that can be used to obtain answers from new data. This transfer function provides a nonlinear output to the
A neural network is composed of neurons that are divided neurons and ensures that this output is always between
into layers. A minimum of three layers are needed: one 0 and 1.
input layer, one hidden layer, and one output layer, as shown The natural range of input for a neuron is the range of
in Fig. l l n . The input layer gets the input data, and the out- its transfer function, in this case 0 and 1. All the data pre-
ROBERT

/ DEFINETHE PROBLEM 1- , '


THE 1
I CHOSE THE
i INFORMATION TRAIN WELL ,
1

1 GATHER DATA

1 CREATE THE NETWORK


1
c
FILES
INPUT HIDDEN OUTPUT
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20

LAYER LAYER LAYER

TRAIN THE NETWORK -

I
TEST THE NETWORK
THE
I

RUN THE NETWORK ), -


1 NETWORK
DOES NOT
TEST WELL

FIG.12. Neural network design.


For personal use only.

-8 -6 -4 -2 0 2 4 6 8
NEURON INPUT
FIG.11. ( a ) Neural network. ( b ) Sigmoid transfer function.

sented to the neural network are renormalized using minimum


and maximum values, which are different for each input
line. These minimum and maximum values are representa-
tive of the actual lower and upper bounds of the training u
0 500
AVERAGE
STD DEV.
CORR.
0.98
0.067 %

1000 1500 2000 2500 3000


NETWORK
101
0.060 %
~
data. The output of the neural network must then be con-
verted from values ranging between 0 and 1 to values cor- CAPWAP CAPACITY (kN)
responding to the training pattern using a similar procedure.
Neural networks perform well in finding complex pat-
terns. They have some limitations, though.
They will not use values that are outside the range of
their training data. They will automatically bring any out
of range (either smaller or larger than the actual lower and
upper bounds of the training data) input value into range.
Therefore, they cannot give answers that are outside their
training pattern lower and upper bounds. AVERAGE 1.00
All the input data cannot be used to train the network,
because some of the data must be set aside to test the trained L

network. Neural networks tend to remember the training 0 500 1000 1500 2000 2500 3000
facts. Thus, network testing must use data that were not CAPWAP CAPACITY (kN)
involved in the training process to verify its ability to
generalize and predict. FIG.13. ( a ) Neural network (0) and linear correlation (+)
The procedure used to design a neural network is shown in test results. ( b ) Average of the linear correlation and neural net-
work test results.
Fig. 12. The first step is to define the problem or what you
want to know. Then the input parameters must be chosen,
the input data must be gathered, and a first attempt is made Computer program
to create a network. If the neural network does not train well A computer program has been written to include linear
(does not produce answers that are within the required margin correlations and a neural network run-time module that uses
of error with respect to the training data), new parameters all the parameters required to calculate RAP, the minimum
are needed. If it trains well but does not test well (does not and maximum values of the training data and the weight
produce answers that are within the required margin of error matrices produced during training.
with respect to the testing data), then more data are required. The user only needs to input the required parameters com-
252 CAN. GEOTECH. J. VOL. 3 1, 1994 \ \

puted by the PDA and pile data in a full-screen editor to Cheng. S.S M., and ~ h m a d , S.A.
' i988. Comparison of pile
obtain an answer. The process is quick and the result is capacity evaluated by dynamic m&asure&ents and static load-
completely independent of the operator's judgement. ing tests-Five case studies in Ontario, Canada. In Proceedings
A comparison of the linear correlation results and of the of the 3rd International Conference on the Application of
neural network results is shown in Fig. 13a. These results are Stress Wave Theory to Piles, Ottawa, Canada. pp. 477-489.
Federal Highway Administration. 1991. Dynamic pile monitor-
for the data used to test the neural network (25% of the ing and pile load test report. Demonstration project no. 66.
total database). They have been chosen at random and another Federal Highway Administration, Washington, D.C., U.S.A.
set of data will give different results. For instance, Fig. 9 Fellenius, B.H. 1980. The analysis of results from routine pile
shows that the RAP values predicted by linear correlations loading tests. Ground Engineering, 13(6): 19-3 1.
have a standard deviation of 0.081, when the whole data- Fellenius, B.H. 1984. Ignorance is bliss-And that is why we
base is considered, instead of the 0.067 value shown in sleep so well. Geotechnical News, 2(4). pp. 14-15.
Fig. 13a. The neural network portion of the program gives Hannigan, P.J., and Webster S.D. 1987. Comparison of static
results that can vary within the same range. load test and dynamic pile testing results. In Proceedings of the
A neural network that trains and tests well will, on aver- 2nd International Symposium of the Deep Foundation Institute,
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20

age, slightly overestimate the CAPWAP capacity, and the stan- Luxembourg.
Hiley, A. 1930. Pile driving calculations with notes on driving
dard deviation of its results will be close to the standard forces and ground resistances. Structural Engineer, 8: 246-259,
deviation of linear correlation results. 278-288.
The average of the results obtained from both types of Hussein, M., and Likins, G.E. 1991. Static pile capacity by
correlations will often improve the capacity prediction, espe- dynamic methods. In Proceedings of the 1st Geotechnical
cially if the neural network result is higher than RAP. Engineering Conference, Cairo University.
Figure 13b shows the average values computed from the Lawrence, J. 1988. Introduction to neural networks. 2nd ed.
data presented in Fig. 13a. California Scientific Software, Nevada City, Calif.
It should be pointed out that this computer program is Morgano, C.M. 1988. CAPWAPC correlrition-A case study. 1988
not used to predict the bearing capacity of piles but is rather PDA users day. Pile Dynamics Inc., Cleveland, Ohio.
used as an effective indicator to help pinpoint piles for Pile Dynamics Inc. 1990. CAPWAP refgrence manual. Pile Dynamics
Inc., Cleveland, Ohio.
which a CAPWAP analysis might be required. Such CAPWAP Rausche, F., Goble, G., and Likins, G.E. 1985. Dynamic deter-
analysis may be required due to variations in soil condi- mination of pile capacity. ASCE Journal of the Geotechnical
For personal use only.

tions. This computer program does not replace the CAPWAP Engineering Division 111(3): 367-383.
analyses, which are an essential part of dynamic pile testing, Riker, R.E., and Fellenius, B.H. 1988. Case method capacity
but it represent another tool that can be used to improve estimates for piles in glacial soils. In Proceedings of the
the quality of the test results. 3rd International Conference on the Application of Stress
Wave Theory to Piles, Ottawa, Canada. pp. 565-578.
Conclusions Seidel, J.P., Haustorfer, I.J., and Plesiotis, S. 1988. Comparison
Using e m p i r i c a l correlations d e v e l o p e d with o v e r of dynamic and static testing for piles founded into limestone.
In Proceedings of the 3rd International Conference on the
200 CAPWAP results and pattern-recognition techniques
Application of Stress Wave Theory to Piles, Ottawa, Canada.
based on neural networks, a computer program has been pp. 717-723.
written to analyze the results of dynamic pile tests carried out Thompson, C.D., and Goble, G.G. 1988. High case shaft damp-
using the pile driving analyzer. ing constants in sands. In Proceedings of the 3rd International
Even though this program has been developed using a Conference on the Application of Stress Wave Theory to Piles,
limited database, the results are very good and should con- Ottawa, Canada. pp. 555-563.
tinue to improve as more data become available. It has not,
however, been created to replace the CAPWAP program, which List of symbols
can provide a large number of useful parameters, such as
A cross section of the pile
distribution of skin resistance, toe resistance, and soil quake,
C stress wave speed
and its damping constants, but to supplement it.
CENTRE centre value of the combined neuron inputs
The computer program provides three outputs. The first one
D damping force
is RAP, which can be used even if the data are outside the
DHN final set of the pile
range of the database but seem to have little potential for
DMX maximum displacement at the location of the
improvement. The second one is the output of the neural
gauges
network module, which is limited to the range covered by the
E modulus of elasticity
training data. However, since the training patterns cover
EMX maximum energy transferred to the pile during
capacities ranging from 290 to 3100 kN, this limitation gen-
a hammer blow
erally has little influence on the results. It is not, at this
force at the location of the gauges at time T ,
time, better than RAP, but it shows potential for improvement.
after the beginning of the blow
The third one is the average of the other two outputs and
FT? force at the location of the gauges at time T,
often shows the best results, especially if the value from
after the beginning of the blow
the neural network is larger than RAP.
HIGH maximum value of the transfer function
Abe, S., Likins, G.E., and Morgano, C.M. 1990. Three case stud- GAIN sensitivity of the transfer function
ies on static and dynamic testing of piles. Geotechnical News, J damping constant
8(4): 26-28. L length of the pile below the location where the
ASTM. 1990. Standard test method for high-strain dynamic test- gauges are installed
ing of piles (D4549-89). In Annual Book of ASTM Standards, LOW minimum value of the transfer function
sect. 4, vol. 04.08. ASTM, Philadelphia. pp. 1000-1006. RA2 an estimate of the static mobilized capacity
ROBERT \ 253
, I 1

based on the pile velocity and force at the loca- SFT total skin resistance (static + damping) esti-
b*
tion of the gauges mated by the PDA
RAP predicted static pile bearing capacity using lin- T~ time of impact (usually the time when the veloc-
ear correlations ity at the location of the gauges is maximum)
RAU total mobilized capacity when VB becomes equal TF transfer function value (neuron output)
to 0 VB computed pile toe velocity
R ~ P upper bound for the mobilized static capacity V(T) velocity at the location of the gauges at time
RMX maximum value of RS when T, is varied from T after the beginning of the blow
+
Tp to Tp 30 ms VT! velocity at the location of the gauges at time T,
RS mobilized static resistance of the pile at a time T after the beginning of the blow
RS 1 mobilized static capacity when T, is equal to Tp vT2 velocity at the location of the gauges at time TZ
- RTL total mobilized resistance of the pile (static after the beginning of the blow
resistance + damping force) X value of the combined inputs of a neuron
Can. Geotech. J. Downloaded from www.nrcresearchpress.com by 68.148.154.147 on 06/27/20
For personal use only.

Vous aimerez peut-être aussi