Vplyv vody na geotechnické konštrukcie, Bratislava 03.- 04.

júna 2013

ACCURACY OF EXTRAPOLATION METHODS

FOR NON-FAILED STATIC LOAD TESTS
Maciej Król1, Jarosław Rybak2

ABSTRACT:
According to recommendations of Eurocode 7, every pile foundation design should be based on
results of static axial compressive load test (directly or as reference test for static calculations or
dynamic testing). The ultimate load (capacity) is being obtained on the basis of those results.
Technically: in order to perform such a test, the pile should be loaded by weighted box or
platform or hydraulic jack acting against anchored reaction frame. In many cases, limitations of
loading force (due to too large sizes of kentledge or limited pull out resistance of anchoring
piles), do not enable us to derive ultimate capacity directly from tests. Brinch-Hansen 80%, Chin-
Kondner and Decourt methods (based on the simplest least squares approximation) are being
examined, in order to obtain a proper value of ultimate pile load from non-failed test. Obtained
results (conclusions) can be used in engineering practice to evaluate the risk linked to the
extrapolation of test results.

1. Introduction

According to Eurocode 7 [1], pile foundations shall be designed on the basis of results
from static load tests. They are used for direct determination of capacity, verification of static
calculations or for calibration of other tests (e.g. dynamic testing) at similar soil conditions. The
tests consist in applying increasing loads stepwise at specified time intervals and in measuring
simultaneously the settlement for each step. As a result, a load v. settlement curve is obtained
which subject to further interpretation, In general, load should increase until the pile reaches the
limit state (unhindered increment in settlements with respect to increase of load applied), which
would allow to determine the ultimate capacity. In practice, due to limited strength of testing
station, insufficient capacity of anchoring piles or too small kentledge (see Fig. 1), it is often
impossible to create the loads which would cause that pile operates in soil within plastic range.
Even if the force applied during the test is larger than design capacity and the structure safety is
confirmed, the range of test provides no information about pile ultimate capacity and would not
allow to calibrate the calculation model. When the ultimate capacity is unknown, we do not make
use of the margin which would allow, for instance, to reduce the number of pile in foundation, or
to reduce their length.

1
Maciej Król Eng, MSc student, Adress: Wroclaw University of Technology, Wyb. Wyspianskiego 27, 50-370
Wroclaw, Poland, phone No.: +48713204096, E-mail: 176969@student.pwr.wroc.pl
2
Jarosław Rybak PhD, Adress: Wroclaw University of Technology, Wyb. Wyspianskiego 27, 50-370 Wroclaw,
Poland, phone No.: +48713204096, E-mail: jaroslaw.rybak@pwr.wroc.pl
 Figure 1 Static test with kentledge

Due to uncertainty of extrapolation methods, the limitation consisting in assuming the

maximum load effected in field testing as the ultimate capacity is often used in designing
practice. Such recommendation was given by Fellenius (2001). Currently, economic reasons are
of utmost importance, hence while making static tests (which are in themselves costly and time-
consuming), attempts are made to get as many information about capacity as possible. This paper
outlines how the number of measurements (for the range of test load effected) affects the
accuracy of determining the ultimate capacity- Qu. The review was based on data from 30 site
logs of test loads applied to prefabricated driven piles. Examinations were run within supervision
of Wrocław University of Technology on construction and modernization of road infrastructure
in 2010-2012.

2. Basic methods of load test extrapolation

There are various methods to determine the ultimate capacity with extrapolating the
results gained before pile reaches the state of ultimate load bearing capacity. Polish Standard PN-
83/B-02482 (1983) allows for determining Qu from the auxiliary waveform dQ/ds. The condition
dQ/ds=0 means unhindered increment in settlement without increase of load. The extrapolation
methods most popular in worldwide biography for foundation piles, described in detail by
Fellenius 2001, were tested for their accuracy (possible error of estimating Qu). Each
extrapolation method used in these analyses consists in fitting the results of settlement
measurements in successive pile loading steps into assumed functional form. In order to facilitate
the fitting of extrapolating function parameters, the co-ordinate system is re-scaled (ordinates and
abscissae), so thanks to it the settling-strain relation is roughly linear in the final phase of pile
operation. This allows to fit the function extrapolating the test load curve using the simplest
linear regression in spread sheet. To estimate the ultimate capacity, the slope of a straight line and
the free term need to be fitted.
 The last points on settlement-strain curve, which represent elastic/plastic operation of pile
prior to reaching the ultimate capacity are of key importance. As the criterion of selecting the
range of point fitting, the method shown in Fig. 2 proposed by De Beer (1968) to evaluate the
pile capacity may be helpful.

Figure 2 De Beer double logarithmic method

The settlement-loading relation is given in logarithmic scale, hence it is roughly linear in

intervals. The operating ranges of pile are presented in Figure 2. The intersection point of curves
drawn on the diagram represents approximate value of load above of which the pile operates in
plastic range. It should be emphasized that the concept of De Beer (1968) was not aimed at
determining the value of Qu but merely the load above which permanent pile displacement
occurred. Similar nature (however for another system of co-ordinates) had the interpretation of
test loads according to Polish Standard (1983).
In actual fact, the ultimate capacity can be determined by means of points on the straight
line representing the plastic range of pile operation. The more points is recorded in this range, the
higher precision of ultimate capacity estimation is guaranteed. If the number of measurements is
too small, errors are generated making the extrapolation method useless. Below, three methods of
estimating the ultimate capacity are presented on the basis of data from 30 site logs of static tests.
Errors generated by each method are compared and the effect of the number of measurements in
elastic/plastic range and in plastic range of pile operation on extrapolation accuracy is verified.

2.1. Brinch-Hansen 80% method

According to Brinch-Hansen (1963), in so called 80% method, the ultimate capacity Qu is

understood as such loading for which the settlement su exceeded four times the value s measured
after 80% of the load Qu has been applied. The settlement-loading relation (s – Q) is then
transformed into the coordinate system: s - abscissae and s / Q - ordinates (Fig. 3). For the last
points of the performed test at which the settlement reached stabilization the following linear
dependence is found (by way of approximation): s / Q = A ⋅ s + B .
 Figure 3 Brinch-Hansen 80% method

The ultimate capacity can be then derived from formula (1)

1
Qu = . (1)
2 A⋅ B
The settlement-loading relation can be transferred to equation (2)
s
Q=
A⋅ s + B (2)
where: A – slope of regression straight line;
B – intersection point between regression straight line and OY axis.

Figure 4 Relative difference between measured capacity and extrapolated ultimate load

The ultimate capacity according to Brinch-Hansen procedure was determined and compared with
results of static loads. The standard value of ultimate capacity was taken as the result of the last
test of each trial before unhindered increase of settlements is observed. Accuracy of determining
Qu depending on the number of points used in regression is shown below. The term „error in
ultimate load estimation“ is understood as the result of the formula (3):
QuT − Qmax
ER = (3)
Qmax
where: QuT – theoretical ultimate capacity calculated on the basis of all known loading steps
Qmax – load applied in the last step

When all available points (i.e. those „to the right“ of the curve bend in de Beer
interpretation) are used, the determined ultimate capacity is similar to that resulting from tests
(the difference is not higher than 15%).

Figure 5 Fitted theoretical curves (underestimated capacity)

Table 1. Comparison of static test with theoretical results (according to Fig. 5)

Load [kN] Load in
Settlement static
[cm] 7 points 6 points 5 points 4 points 3 points 2 points test
0 0 0 0 0 0 0 0
0,06 215,234 220,6001 225,6159 230,0088 239,3052 245,7665 188,4
0,17 358,66 367,38 375,49 382,55 397,38 407,60 376,80
0,30 476,44 487,68 498,05 507,03 525,68 538,40 565,20
0,51 619,05 632,94 645,58 656,40 678,50 693,28 753,60
0,93 830,42 847,13 861,89 874,20 898,31 913,66 942,00
1,58 1062,53 1080,32 1095,14 1106,86 1127,76 1139,59 1130,40
2,31 1263,81 1280,28 1292,69 1301,53 1314,09 1318,80 1318,80
3,44 1501,91 1513,35 1519,21 1521,17 1516,20 1507,20 1507,20
4,90 1735,25 1737,10 1731,75 1722,66 1691,48 1663,33 1695,60
6,30 1907,78 1898,77 1881,40 1860,92 1804,10 1758,20 1884,00
8,57 2120,94 2092,78 2055,13 2016,13 1919,66 1847,77 2072,40
10,65 2267,47 2221,33 2165,41 2110,27 1980,96 1888,75 2260,80
14,11 2443,12 2368,08 2283,97 2204,84 2029,18 1909,99 2449,20
15,71 2503,67 2416,06 2320,09 2231,16 2037,40 1908,21 2543,40
 The method provides no unambiguous statement whether the ultimate capacity Qu
determined according to Brinch-Hansen is over- or under-estimated. Figures 5 and 6 present
waveforms for different piles for which 7 last points were selected for linear regression. In the
first case, along with reducing the number of points, the capacity is considerably underestimated,
while in the other one – it is overestimated.

Figure 6 Fitted theoretical curves (overestimated capacity)

Tab 2 Comparison static test with theoretical results (according Fig 6)

Capacity [kN] Capacity
from
Settlement static
[cm] 7 points 6 points 5 points 4 points 3 points 2 points test
0,00 0,00 0,00 0,00 0,00 0,00 0,00 0,00
0,08 228,52 227,73 227,42 226,53 225,68 225,55 188,40
0,23 388,24 386,96 386,46 385,08 383,78 383,59 376,80
0,49 560,37 558,70 558,06 556,35 554,83 554,62 565,20
0,92 761,51 759,60 758,93 757,26 755,98 755,86 753,60
1,46 946,73 944,92 944,37 943,30 942,91 942,98 942,00
2,12 1125,16 1123,81 1123,57 1123,73 1125,01 1125,43 1130,40
2,96 1305,61 1305,17 1305,49 1307,75 1311,74 1312,71 1318,80
4,09 1497,48 1498,66 1499,90 1505,55 1513,89 1515,73 1507,20
5,34 1668,44 1671,74 1674,16 1684,13 1697,96 1700,89 1695,60
7,27 1873,60 1880,55 1884,96 1902,19 1925,28 1930,05 1884,00
9,82 2075,40 2087,48 2094,68 2122,05 2158,15 2165,53 2072,40
12,94 2252,10 2270,44 2281,04 2320,83 2373,06 2383,67 2260,80
19,05 2468,53 2498,30 2515,11 2577,80 2660,37 2677,16 2449,20
22,57 2545,25 2580,90 2600,91 2675,51 2774,21 2794,33 2543,40

2.2 Chin-Kondner Method

According to Chin-Kondner method, the settlement-loading relation (s – Q) is

transformed into the coordinate system: s - abscissae and s / Q - ordinates (Fig. 7).
 For the last points of the performed test at which the settlement reached stabilization, the
following linear dependence is found (by way of approximation): s / Q = A ⋅ s + B . The ultimate
capacity can be then derived from formula (4) and the settlement-load relation can be transferred
to equation (5)
Qu = 1 / A . (4)
s
Q= (5)
A⋅ s + B
where: A – slope of regression straight line;
B –– intersection point between regression straight line and 0Y axis.

Figure 7 Chin-Kondner method

Figure 8 Relative difference between measured capacity and extrapolated ultimate load

The Chin-Kondner method is applicable for measurements which provide no clear

direction change for straight lines of de Beer method. It also allows to use more measurement
points as the waveform is along longer linear interval, hence it is less „sensitive“ to small amount
of measurement points. Its clear drawback is the fact that the results are strongly overestimated
(in all cases tested, the capacity from extrapolation was much higher than that determined in field
testing). Thus, the ultimate capacity from calculations should be reduced by means of safety
factor depending on the number of points used for its determination. The above diagram clearly
shows higher discrepancy of the results. The points for which the straight line of linear regression
is determined (despite the fact they roughly lie on a single straight line – the coefficient R2 is c.
99%), would not always guarantee that the capacity is safely determined.

2.3 Decourt method

According to Decourt method, the settlement-loading relation (s – Q) is transformed into

the coordinate system: Q - abscissae and Q / s - ordinates (Fig. 9). For the last points of the
performed test at which the settlement reached stabilization the following linear dependence is
found (by way of approximation): Q / s = A ⋅ Q + B .

Fig 9 Decourt's method

Figure 10 Relative difference between measured capacity and extrapolated ultimate load
 The ultimate capacity can be than derived from formula (6) and the settlement-load
relation can be transferred to equation (7)
Qu = − B / A . (6)
B⋅s
Q= (7)
1− A⋅ s
where: A – slope of regression straight line; B – intersection point between regression straight
line and 0Y axis.

The extrapolation method proposed by Decourt is accessible, however if the number of

significant points is too low, the results are definitely overestimated. Even for larger points taken
to determine the regression straight line, the ultimate capacity should be verified by means of
appropriate safety factor. The method is well applicable for rough estimations. The diagram
above illustrates considerable discrepancies of Decourt method of extrapolation. Substantial
errors can be observed even for large number of pile loading steps. Determination of regression
line for less than 5 points is generally useless (the error generated amounts to about 40% of the
load measured in the last step; this provides no engineering profit from such estimation).

3. Summary and conclusions

Each method addressed in this paper was evaluated with respect to its accuracy of
estimating the ultimate capacity. Below, on Fig. 11, there are diagrams of average estimation
error and standard deviation for ultimate capacity depending on the number of points used in
linear regression. Error was calculated as percentage difference between theoretical ultimate
capacity determined from all points and the last recorded load applied during the static test.

Fig 11 Average inaccuracy and standard deviation (method comparison)

The review of 30 site logs of pile static testing proves that the Brinch-Hansen 80%
method is the most accurate of all extrapolation methods. However, to apply it, relatively large
number of load steps is necessary. The method is totally useless for extrapolating the points from
the elastic range of pile operation. Calculations made with proposals of Chin-Kondner and
Decourt show that estimated capacities are to the larger degree different from those determined in
 static tests. However, the features of applied approximating functions allow to use them for loads
performed in narrower range. An advantage of possible extrapolation for points from
elastic/plastic interval of pile operation is compensated by the loss in accuracy in determining the
capacity. The doubt about the sense of using them is reflected in recommendations of Fellenius
(2001) and in International Building Code (2006) where Chin-Kondner and Decourt methods
were recognized as insufficiently conservative.

Tab 3 Method comparison

Brinch-Hansen 80% Chin-Kondner Decourt
Points Standard Standard Standard
used Mean deviation Mean deviation Mean deviation
2 24,86% 9,76% 50,71% 10,67% 45,13% 10,48%
3 16,84% 10,12% 42,93% 9,99% 37,42% 11,43%
4 11,09% 9,47% 36,20% 8,03% 31,18% 10,39%
5 10,18% 7,04% 29,09% 7,78% 25,24% 8,66%
6 6,46% 4,75% 20,83% 7,14% 18,93% 8,24%
7 4,43% 2,92% 9,59% 4,73% 14,60% 7,84%

It is also an essential information that measurements from the first steps of pile loading
have no effect on the result. Theoretical curves fit well in plastic range, while large discrepancies
are observed in the elastic range of pile operation. Hence, attempts to extrapolate the ultimate
capacity from the initial measuring points is completely useless as the result would not be even
slightly similar to real value.

References

Brinch Hansen J. (1963): Discussion, Hyperbolic Stress-Strain response, Cohesive soil. Journal
of soil mechanics and foundation engineering division, ASCE, 89, pp. 241-242
Chin, F.K (1970): “Estimation of the Ultimate Load of Piles Not Carried to Failure”, Proc. 2nd
Southeast Asia Conference on soil Engineering, pp. 81-90.
DeBeer, E. E., (1968): Proefondervindlijke bijdrage tot de studie van het grensdraag vermogen
van zand onder funderingen op staal. Tijdshift der Openbar Verken van Belgie, No. 6, 1967
and No. 4, 5, and 6, 1968.
Decourt, L., (1999): Behavior of foundations under working load conditions. Proceedings of the
11th Pan-American Conference on Soil Mechanics and Geotechnical Engineering, Foz
DoIguassu, Brazil, August 1999, Vol. 4, pp. 453 - 488.
Fellenius B. H., (1975): Test loading of piles. Methods, interpretation, and new proof testing
procedure. ASCE, Vol. 101, GT9, pp. 855 - 869.
Fellenius B. H., (1980): The analysis of results from routine pile loading tests. Ground
Engineering, London, Vol. 13, No. 6, pp. 19 - 31.
Fellenius B.H.,(2001): What capacity value to choose from the results a static loading test. We
have determined the capacity, then what? Deep Foundation Institute, Fulcrum Winter, pp. 19
– 22 and Fall, pp. 23 – 26
PN-83/B-02482 (1983): Foundations. The capacity of piles and pile foundations.
PN-EN 12699 (2003): The execution of special geotechnical works. Displacement piles.

View publication stats