You are on page 1of 14

Geotechnical Engineering Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

Proceedings of the Institution of Civil Engineers Paper 900086 Received 12/11/2009 Accepted 09/02/2011 Keywords: computational mechanics/embankments/failures

ICE Publishing: All rights reserved

Deformation and strength of embankments on soft Dutch soil

1 j Evert den Haan PhD 2 j Antoine Feddema BEng

Geotechnical Researcher, Deltares, Delft, The Netherlands

Senior Consultant/Manager, Deltares, Delft, The Netherlands

1 j

2 j

Embankment and dyke stability in the Netherlands has always been evaluated by effective stress analysis. The subsoil of most of these structures is organic, weak and soft, but the internal friction angle of these soils is surprisingly high. Empirical methods are used to obtain acceptable, reduced values of friction angle from triaxial tests for use in stability analyses. It appears possible, however, to do full justice to the peculiar combination of low strength and stiffness and high friction angle by means of the nite-element method using a viscous version of the Cam-clay model. All parameters of the model are found from a single test in a constant-rate-of-strain K0 oedometer. The approach is illustrated by two case histories, after rst providing insight into the peculiar properties of the Dutch soils, and the manner in which they are dealt with.

C c9 e h h0 K0 K0,nc k M p9 p90 c p9q e p9 0 q su su = p 9 t tage tcreep nat vol k* * * ax 9 p 9 rad 9 creep factor cohesion voids ratio height initial height lateral earth pressure at rest virgin compression value of K0 permeability critical-state strength parameter isotropic effective stress initial equivalent yield stress equivalent isotropic effective stress initial value of equivalent stress deviatoric stress undrained strength undrained strength ratio time age of deposit creep duration natural strain volumetric strain cam-clay swelling factor cam-clay compressibility factor creep factor Poisson ratio axial effective stress vertical yield stress radial effective stress

v0 9 9

initial vertical effective stress internal friction angle



Dutch organic soils and peat, although weak and soft, have surprisingly high values of the effective strength parameters and undrained strength ratios. Their internal friction angles, 9, can far exceed the 30358 range that is common in sands, and generally become higher as the organic content increases. For example, peat has 9 values of up to 908, and in organic clays values of 40608 are common. 9 is an important parameter in the Netherlands, as the effective stress approach is used almost exclusively in stability analyses. To obtain reasonable factors of safety, 9 is determined at strains far below failure, and the true strength of the material is not accounted for. For a realistic assessment it is necessary to combine (low) stiffness and (high) strength parameters in one analysis, and the nite-element method is the obvious means to achieve this. Good results have recently been obtained with nite-element calculations of the deformation of embankments constructed on soft Dutch soils. This will be illustrated by two case histories. The constitutive model used was a viscous version of the modied Cam-clay model. The necessary soil stiffness and strength parameters were determined from oedometer tests in which a constant rate of strain was applied, and lateral stress was measured. The paper starts off by giving the background to the high strength 1

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

parameters of Dutch organic soils and peat, and the way in which Dutch engineers have dealt with this in the past. Then the constitutive model and the parameter determination are described. Finally the two case histories are presented.

n.c. o.c.
200 Peat High-organic clay Low-organic clay Dense and loose sand 150 Lateral tension boundary


Strength of Dutch soft soils

Dutch organic soils and peat, although weak and soft, have surprisingly high values of the effective strength parameters and undrained strength ratios. Figure 1 shows stress paths from undrained triaxial compression tests on samples covering a wide range of organic contents. For comparison, test results are also shown for a loose and a dense sand. The organic soils were sampled below the crest and the toe of the Lek river dyke, and were reconsolidated to the in situ stresses. The heavily compressed crest samples are normally consolidated, and Table 1 provides the characteristics of these tests. The internal friction angles 9 far exceed the 30358 that is common in sands, and are higher as the soil becomes softer and more organic. Peat, for example, has values up to 908, and organic clay from 408 to 608. The application of such values in Bishop slip circle analysis would yield unrealistically high factors of safety, but the methods of parameter determination developed by Dutch engineers for their effective stress slip analyses yield much lower values. The high triaxial strengths and the positive effect of organic content remained largely unnoticed up to the 1990s. The undrained strength ratios su = p are also high: 0.44 and 0.40 9 for the organic clays in Figure 1, and rather than failing, the peat dilates and continues to increase its shear resistance. The lightly stressed toe material also dilates during shear and is clearly overconsolidated. Data for Queenborough organic clay from Jardine et al. (2003) are included in Table 1. The values of 9 and su = p are 9 somewhat smaller than for the Dutch organic soils, but the 9 value is also quite high. Figure 2 plots 9 of normally consolidated organic Dutch soils against their bulk density. Each bar in the gure is based on numerous tests. A consistent trend exists. Bulk density correlates

q: kPa
100 50 0 0 50 p : kPa 100

Figure 1. Effective stress paths of Dutch soft soils in undrained triaxial compression illustrating the high angle of internal friction. (Reconsolidation to eld stresses. n.c., normally consolidated material under dyke crest; o.c., overconsolidated material adjacent to dyke)

Bulk density: kN/m3 Peat High-organic clay Low-organic clay Queenborough clay

Water content: % 309 170 76 85

9: degrees 83 54 44 3538

su = v0 9 0.74 0.53 0.49 0.53

su = p 9 0.62a 0.44a 0.40a .300.33 0

10.4 12.8 14.7 14.6

Assuming OCR 1.2.

Table 1. Characteristics of triaxial tests on Dutch (Figure 1) and Queenborough organic soils

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

to some extent with organic content, and indications of the latter are shown. The very high 9 value of peats with organic contents above 20% is due to the effects of bre tensioning. The bres in peat are orientated predominantly in the horizontal plane, and during triaxial compression the bres extend and reinforce the matrix of granular and humied organic material in which the bres are embedded. Landva and La Rochelle (1983) describe this mechanism at length, and the 9 908 condition in peat has often been reported. It occurs when the pore pressures during shear increase to equal the cell pressure. Coutinho and Lacerda (1989) show similar relationships as in Figure 2 for Brazilian (Juturnaiba) organic soil. However, the organic contentbulk density relationship of those soils differs from the average Dutch relationship, and points to more precompression of the former. Bulk density and water content are the mainstays of correlations with compressibility and strength parameters in Dutch organic soil practice. These usually sufce, and only rarely are the organic content, degree of humication or Atterberg limits determined. Most of the relatively uncompressed peats are brous and pseudo-brous, whereas deep, buried peat can be amorphous. A puzzling feature of Figure 2 is that, despite the very different structures of peat and organic clay, the trend is quite uniform across such a large range of organic contents. Jardine et al. (2003) in their study of the Queenborough clay explain the increase of the shear strength parameters by colloidal organic material affecting the surface behaviour of clay particles. Colloidal activity is also expected in amorphous, humied peat, but less so in brous peat, and so the uniform trend in the gure is surprising. The current Dutch test and design methods and the evaluation of stability are briey discussed below, before turning to the alternative method using nite-element calculations with parameters from a constant rate of strain (CRS) K0 -oedometer.


Dutch methods

Keverling Buisman (1934) developed the Dutch cell apparatus (Figure 3) from earlier somewhat similar devices in use by Ehrenberg, and by Terzaghi (see also De Boer, 2005, from which it appears that the rst triaxial tests ever, by von Karman in 1910 on marble and sandstone (Vasarhelyi, 2010), went unnoticed by the soil mechanics pioneers). The Dutch cell differs from the triaxial apparatus mainly by the cross-section of the piston being equal to that of the sample. Samples were 15 cm high and 6.6 cm in diameter, and free drainage occurred through porous platens. Radial drainage also occurred through folds in the loose-tting rubber membrane. The membranes were tailor-made, and included rubber anges that were clamped into the upper and lower plates; corrections were made for the uplift on the sample from the annular gap around the piston. Pore pressures were not measured, and back-pressure was not used. The test was performed in multiple stages, in each of which the vertical load was increased, and the cell pressure was rst allowed to equilibrate and was then lowered by draining off droplets of water from the cell. This increased the deviator stress and induced shear straining, the lateral component of which caused the cell pressure to increase gradually and to arrest the shear strains. In older procedures the cell pressure was lowered to near failure, but later, when large numbers of tests had to be performed, a standard of 3 kPa was adopted. A stage was considered complete when the rate of vertical strain had subsided to 10 m per hour. The horizontal strains in the sample were quite limited, and in essence a near-K0 condition was imposed. The c9 and 9 values that were deduced from four consecutive stages were low, but were satisfactory in the sense that factors of safety near 1 were obtained for dykes considered to be on the verge of failure. Effective stress analysis was nearly always used

80 60 40 20 0 10 35% 60% 20% 12 10%

From Figure 1 Betuwe railway, Alblasserwaard High-speed railway, Rijpwetering

: degrees

Organic contents 18 20

14 16 Bulk density: kN/m3

Figure 2. Internal friction angle 9 of Dutch organic soils as function of bulk density

Figure 3. The Dutch cell apparatus

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

in embankment analysis, although a quick cell test procedure existed in which consolidation was not allowed, and which basically yielded undrained strength. The cell test has now been replaced by multi-stage triaxial testing, using isotropic consolidation and undrained shear. To avoid the high 9 values, the stages are terminated at axial strains between 2% and 5%, and the stress envelopes at these strains are taken to represent failure. The exact strain value to use has become a matter of debate, and even confusion, and Den Haan and Kruse (2006) explain how the disturbance during preceding stages produces increased values of c9 and reduced values of 9. The undrained approach to strength is now being investigated as an alternative for limit equilibrium methods. The possible merits of the simple shear test are also under investigation, as this test closely mimics failures in peat, which tend to follow horizontal shear planes. The effective stress approach and the reduced friction angles are also used in nite-element calculations. The soft and viscous

nature of organic soils has been modelled by specialised creep models, for example in Plaxis by the soft soil creep model (Vermeer and Neher, 1999), and in a similar model that is included in the Imperial College nite-element program (Bodas Freitas, 2008). The former has led to poor matches with eld measurements of embankment deformations when using the reduced friction angles. An alternative method of parameter determination has been developed that seems to give much better matches. Before discussing this, it is necessary rst to describe briey the creep model used in the nite-element calculations.


The creep model for soft viscous soil

The creep model combines modied Cam-clay (Roscoe and Burland, 1968) with the isotache description of soil compressibility (e.g. Den Haan, 1996). Figure 4 illustrates the model for triaxial conditions, and makes use of the parameters in Table 2 for Sliedrecht peat. Figure 4(a) shows the well-known modied Cam-clay ellipse drawn through the present (p9, q) stress state in A, p9 being the isotropic effective stress ( ax =3 2 rad =3) and q 9 9
p eq: kPa (log 10 scale)

1 0 14 12 10 A M v 02


p c0


100 Stress decreases



1/2x B


Volume strain


Creep rate decreases

q: kPa

8 6 4

1 ln(10)* C




08 2 0 0 2 4 6 8 p : kPa (a) 10 p eq 12 14 10

Reference isotache Creep rate */(day)


Figure 4. Illustration of the nite-element creep model: (a) p9q plot with modied Cam-clay ellipse as plastic potential surface through state point (A, present stress state; h, rate of volumetric creep strain; , rate of shear creep strain); (b) isotaches in p9vol space with p9q and vol determining rate of volumetric creep e

Depth: m-G.L.

Bulk density: kN/m3 10.5

Water content: % 500


* 0.29

* 0.027


9: degrees 58.6








Table 2. Characteristics of Sliedrecht peat, Betuwe railway, km 16.7

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

the deviatoric stress ( ax rad ). However, the ellipse is now 9 9 used as a plastic potential surface rather than as yield surface. The outward normal at any point on the ellipse denes the direction of the rate of change of the viscoplastic or creep strain vector, with the horizontal component forming the volumetric part and the vertical component forming the distortional part. The height of the ellipse is determined by the Cam-clay Mparameter, and the cut-off on the p9 axis is the equivalent stress p9q : The latter is used in the isotache graph (Figure 4(b)) to e determine the magnitude of the volumetric creep strain rate. Isotaches are simply a collection of lines in stressstrain space on which the rate of strain is constant. The slope of the lines, * vol /ln(p9), is similar to the Cam-clay parameter e/ln(p9). The volumetric creep rate @vol,vp /@t is constant on each line. The viscoplastic nature of soil compression is such that, at constant p9, volume decreases with time, but at an ever-decreasing rate. This is reected in Figure 4(b) by lower lines having a lower rate of strain. The vertical spacing of these lines is determined by the creep parameter * vol / ln(@vol,vp /@t ). This parameter is similar to the well-known C vol /log(t ) creep parameter. The creep strain rate is read off at the present values of p9q and volumetric strain. Elastic e strains are given by the Cam-clay swelling factor k* and Poisson ratio . Elastic strain rates are calculated and added to the creep strain rates, and by integration in time and space the strains and deformations are obtained. A reference isotache is dened on which the rate of volumetric creep strain is equal to */(1 day). The initial yield equivalent stress p90 is on this line as shown, and an OCR c value is given by the ratio of p90 and the initial value of the c equivalent stress p9 : 0 A loading increment during a multi-stage test on a laboratory sample is shown schematically in the isotache graph. The initial position A is on the reference isotache, and because consolidation is rapid, strains are at rst essentially elastic, which brings the state to point B on an isotache where rates of strain are high. Creep then occurs along BC, rapidly at rst but quickly diminishing with time. At any point on BC the rate of creep is by approximation equal to */tcreep where tcreep is the creep duration. So the rate at B would be innite (or, rather, very high), and after one day C is reached, where the rate is */(1 day). This is why the reference isotache corresponds to the common one-day laboratory compression curve, and its cut-off equals the usual preconsolidation pressure. For the same reason, the use of OCR 1 in calculations produces unrealistically high rates of strain, equal to those after 1 day of loading on laboratory-sized samples. In situ stressstrain develops along lower isotaches than can be measured in the laboratory, owing to the larger timescale and larger drainage distances, and therefore OCR values should be well above 1, even for normally consolidated soil. In the latter case, OCR can be derived from the isotache on which creep rate equals */tage , where tage is the age of the deposit under the in situ value of p9.


Parameter determination

All parameters of this creep model can be determined from the constant rate of strain (CRS) K0 oedometer test (Den Haan and Kamao, 2003). This concerns not only the compressibility parameters k*, *, * and , but also the critical-state strength parameter M. Figure 5 is a schematic diagram of the CRS K0 oedometer, which is placed in a triaxial cell to make use of piston control, logging and back-pressure facilities. Horizontal stress is measured by strain gauges placed on the back of a section of the oedometer ring, which has been turned down to membrane thickness. Pore pressure is measured underneath the sample, and drainage is to the triaxial cell space. Sample diameter is 63 mm, and sample height is 2035 mm. Correction for wall friction is possible by measuring the vertical load at both the top and the bottom. Figure 6 shows the result of such a test on peat from Sliedrecht, the rst of the cases to be presented further on. By measuring not only vertical stress and strain but also the horizontal stress, the complete stressstrainstrain-rate relationship is known, so that the creep model can be tted to the results. The test procedure

Load cell

Triaxial cell space


Horizontal stress Load cell

Pore pressure

Figure 5. Schematic diagram of constant rate of strain K0 oedometer

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema



Measured Calculated

Vertical strain

Relaxation Unloading


q: kPa




0 0 50 100 Time: h (a) 150

100 p : kPa (b)


500 400 Measured Calculated v h

01 0

v: kPa 10



v, h : kPa

200 100 0 0 50 100 Time: h (c)

Vertical strain



Measured Calculated



06 (d)

10 08 06 K0 04 02 0 Measured Calculated 0 50 Time: h (e) 100 150

Figure 6. Test result and t, CRS K0 oedometer test on Sliedrecht peat

includes an unloadreload loop to assist in determining k* and , and a relaxation phase to assist in determining *. The interaction of the various model parameters is such that, for normally consolidated states and constant rate of strain,

so that M can also be obtained using reasonable estimates of K0,nc , */k* and . Table 2 provides the characteristics and parameters of the test on Sliedrecht peat. The strains are quite high, up to 65%, and in this


s 1 K 0,nc 2 1 K 0,nc 1 2( =k 1) M 3 =k 1 K 2 1 2K 0,nc 1 2K 0,nc 1 2 0,nc 1

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

case natural strains were used in the tting exercise. Natural strain is dened by the integration of innitesimal increments of compression relative to the present height,



limited to triaxial tests on material taken from under the crest of the dyke, and which was reconsolidated at approximately the in situ stresses. It therefore concerns normally to slightly overconsolidated behaviour. It includes the relevant data of Figure 1. The results are plotted against bulk density, and cover a wide range of soft Dutch organic soils. Plotted are M and 9 obtained from the triaxial test, and by tting the CRS K0 oedometer test. Measured K0,nc values from the CRS K0 oedometer test are also shown. The high values of M and 9 are again conspicuous, just as in Figure 1 (sin 9 3M/(6 + M); by approximation 9 % 25M for 0.5 , M , 2.5). The agreement between triaxial M and M from the CRS K0 oedometer test is reasonable. The very high strength of the peat and the highly organic Gorcum light clay (GL), where triaxial compression values tend to M 3 and 9 908, is underestimated, however, and this could be an indication of the inuence of organic bres on strength, which would not be fully developed in oedometric conditions. The measured K0,nc values are also given in the gure. Their low values in the peats will be noted; these indicate that partial bre tension develops in the horizontal plane. The agreement in Figure 7 is considered sufcient to apply the CRS K0 oedometer parameters, including M, in the creep model. CRS K0 oedometer parameter sets have been determined for two embankment construction projects, and applied in nite-element calculations. The values of M and 9 from the tests are used


ln h=h0

ln 1

At small strains, the natural strain is very similar in magnitude to the normal (linear) strain . However, at larger strain levels, natural strain is numerically increasingly larger than linear strain, and this has the effect of straightening the concave stressstrain curve, which often develops at higher stresses well in excess of the yield stress (as in Figure 6(d)). The friction angle found from the t is quite high, and the question is whether this value coincides with the triaxial compression value. The answer to this question is obtained from a project where numerous CRS K0 oedometer tests and triaxial tests were performed. In this project, Ground-breaking methods of dyke safety evaluation (Van Duinen, 2008), which was performed for Rijkswaterstaat (the national public works authority), three dyke cross-sections were thoroughly investigated. Samples were obtained using the Begemann continuous stocking sampler (Begemann, 1971). The triaxial and CRS K0 oedometer tests performed for this project have been used to compare M values: see Figure 7. The gure is

30 25 20 15 10 05 0 10 Peat 12 Organic clay 14 16 18 20 GL GL GL

900 619 486 369 254


Bulk density: kN/m3 M, triaxial tests M, fitted to oedometer tests K0,nc measured

Figure 7. Comparison of M from triaxial compression tests and M obtained from tting CRS K0 oedometer tests by the creep model, both as function of sample bulk density. Measured K0,nc is also shown (GL: Gorcum light clay)

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

without any reduction, and are much higher than is usual in soft soil. On the other hand c9 0 is taken. The Cam-clay model expects zero cohesion in normally consolidated conditions, and in triaxial tests very little cohesion is measured.
Settlement: m

1 0


Time: days 100


10 000

Meas. Calc. 1


Betuwelijn embankment

The measured deformations of the Betuwelijn railway embankment at km 16.7 near Sliedrecht have been analysed by means of nite-element calculations in the project Lateral ground stresses on piles, which was performed for CUR, the Dutch equivalent of the UKs Construction Industry Research and Information Association (CIRIA). The cross-section used in the nite-element calculation was as shown in Figure 8. The nite-element mesh consisted of 997 15-noded elements. An updated mesh analysis was used (updated Lagrange) to account for the large deformations. Seven CRS K0 oedometer tests were performed on samples distributed over the 8.5 m depth of the soft layers. Figure 6 and Table 2 provide data on one of these tests. In accordance with the updated mesh technique, the compressibility parameters k* and * are taken with respect to natural strains. The bulk density of the samples varied from 10.3 to 15 kN/m3 , the water content from 670% to 75%, and the M values found from the soft soil model varied from 2.4 to 1.7. The correlation of M and bulk density agreed with Figure 7, but no triaxial tests were performed in this project. The nite-element calculation with these parameters gave good agreement with the measured horizontal and vertical deformations, as shown in Figure 9. Only the initial deformations after the rst lift are heavily overpredicted, which may be due to some uncertainty regarding the initial loading sequence.
6 4 Settlement plate 5 2 4 3 0 2 1 2 Soft layers (fibrous and pseudo-fibrous peat and high-organic clay)

2 (a) Horizontal displacement: m 0 0 2 01 02 03 04 05 06

Elevation: m NAP

4 6 8 10 12 14 16 (b) Meas. Calc. Stage 1 (71 days) Stage 5 (372 days) 602 days

Figure 9. Deformations, measured and calculated, Betuwelijn railway embankment, km 16.7: (a) centreline settlements; (b) lateral deformations at inclinometer location

Inclinometer g.w.l.

Elevation: m NAP

4 6 8 10 12 14 0

Dense sand






Distance from centreline: m

Figure 8. Cross-section, Betuwelijn railway embankment, km 16.7 near Sliedrecht

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema



The project Macro Stability Experiment of the IJkdijk Foundation was instigated to test innovative dyke stability monitoring techniques (Van et al., 2009). A dyke, 6 m high and 100 m long, was built at Booneschans and brought to failure: see Figure 10. A cross-section of the dyke is given in Figure 11. It consists of a sand core (1) covered by clay (2), and the subsoil consists of a thin crust of clay (3), followed by 12 m of peat (4), a thin Allerd sandlayer (5), (sand with a slight organic content) and a base of stiff Pleistocene sand (6). The dyke was brought to failure by lling the basin behind the dyke and digging a toe ditch

(phase I), then deepening and widening the ditch (phase II), and nally by pumping water into the sand core of the dyke (phase III). For this purpose inltration tubes were installed as shown in Figure 11, and connected to a pumping system. The dyke failed a few hours into the pumping phase. Extra steps were in place to ensure failure (emptying the ditch, and waterlling of a row of containers on the crest), but proved unnecessary. The sand core and clay shell structure of the dyke is typical of river dykes in the Netherlands, and water-lling of the core mimics the effect of high river levels. A full description of the project, including a limit equilibrium analysis of the failure, will be given in Zwanenburg et al. (2012). Here a nite-element analysis will be presented of the deformations occurring during construction and up to the point of failure. The peat layer was the dominant source of deformations. It was modelled by the creep model described earlier. Average parameters were taken from six CRS K0 oedometer tests performed on this peat. Bulk density varied between 9.9 and 11.4 kN/m3 , and water content between 285% and 625%. Table 3 gives the parameters used. The M value is 2.6, and is therefore very high. Small-strain compressibility parameters were used in this case. The OCR in the peat was found by calibrating the calculation to the oedometer tests: the rst construction lift was applied drained, and OCR was adapted by trial and error to obtain the same strain in the peat as the average value in the tests at the calculated vertical effective stress (vert % 18% at v 42 kPa). This 9 approach circumvents the rather large variation in yield stresses found in the tests. The permeability of the peat (k) was determined in the CRS K0 oedometer tests from the hydraulic

Figure 10. IJkdijk macro stability experiment dyke at Booneschans directly after failure

6 4 2 0

Settlement plate Infiltration tubes Pore pressure gauges I 2 III 4 5 1 2 Inclinometer I II g.w.l.

Elevation: m NAP

3 2 4 6 6 8 10 12








Distance from toe: m

Figure 11. Dyke cross-section, IJkdijk macro stability experiment, 2008

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

Bulk density: kN/m3 10.5


* 0.22

* 0.02


9: degrees 65.1



k0 : m/s

Ck 0.35





1.15 3 108

Table 3. Characteristics of IJkdijk peat, Booneschans

gradient set up from the undrained base to the draining upper boundary. It was found that k decreases as strain increases, and this was formulated as

The horizontal deformations at the inclinometer location in Figure 12(c) depart from the null measurement 37.35 days after the start of construction, at which point the nite-element calculation predicted a maximum of 0.185 m. Phase 0 in Figure 12(c) shows the horizontal deformation that accumulated during the 5.8 days between the null measurement and the beginning of phase I. Reasonable agreement is obtained between the measured and predicted horizontal deformations during phases 0, I and II. The calculation failed in phase II after applying 33% of the phase (deepening the ditch), earlier than in reality, where failure occurred in phase III. This is a satisfactory outcome, as the nite-element plane-strain simulation does not account for the additional resistance provided by the side planes at both ends of the failure surface. With the thickness of the peat layer decreasing to 1 m at the northern end of the failure zone, the end effects are expected to be especially strong. Zwanenburg et al. (in review) calculate an extra 15% of lateral shear resistance from the side planes. It was envisaged that the vertical load on the subsoil in phase III would be increased by gradually saturating the sand core. The inltration process in phase III is shown in Figure 13. A lower inltration tube placed in the top clay was operated separately from the six inltration tubes in the rst sandll layer. The gure shows the inltration pressures measured near the pumps at the southern end of the embankment and the pore pressures measured in the failed section of the sand core approximately at the level of the inltration tubes, and approximately 1 m higher. The inltration sequence was rather complicated, and pressure build-up was higher than expected. Various pumping pauses further complicate affairs. The discharged amount of water totalled about 200 m3 , which is expected to saturate the lower approximately 1.5 m of the sand core, and to increase the vertical load by approximately 1 kPa. In Figure 13 the development of the maximum horizontal deformation of the toe-line inclinometer is also shown. There is a clear correlation between the inltration pressures in the nal phase before failure and the rate of increase of the horizontal deformations. The additional weight of the water is very small, and is not expected to contribute signicantly to the failure. A post-mortem trial pit constructed through the failure surface revealed that core sand had penetrated the soft layers below the crest. It is now postulated that failure was accelerated by both the high inltration pressures and the penetrating sand. The horizontal thrust would be increased by both, especially if the inltration pressures could propagate into the penetrated sand.


k k 0 10

vol =C k

The parameters k0 and C k were again taken as averages from the six tests performed on the peat. A nite-element calculation was performed using a ne mesh of 2054 15-noded elements. A small-strain analysis was applied, and the construction phase deformations were incorporated in the mesh. Figure 12 compares the results of the calculation with measurements of (a) construction-phase settlements below the crest (b) pore pressures measured in the soft soils under the crest and adjacent to the ditch, during construction and the failure phases (c) the horizontal deformations near the toe during the failure phases. The construction phase was modelled in eight lifts of undrained load application and subsequent consolidation. The inclinometer was placed after construction, and horizontal deformations are taken from the null measurement of the inclinometer. The construction phase settlements are reasonably well predicted, as seen in Figure 12(a). Settlements were not measured during the failure phases. Figure 12(b) shows that the construction phase pore pressures in the peat under the crest are predicted quite well. Had it not been possible to let permeability decrease with increasing compression as described above, the t would not have been as good. The pore pressures during the rst lift are overpredicted, because the thickness of this lift was taken as larger than in reality. This was necessary to enable the inltration wells to be included in the nite-element mesh. The pore pressures in the top clay under the crest are poorly predicted, simply because the gauge is located close to the assumed phreatic line. During the consolidation phases the calculated pore pressures take on the phreatic value. The pore pressures in the peat near the ditch are predicted reasonably well. 10

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

0 01 02 03 04 05 06 7 6 5

Days after start of construction (13 August 2008 06:00) 15 45 10 20 25 30 35 40


Settlement under crest: m

Settlement plate Calculated


Meas. Calc.
In peat under crest In top clay under crest In peat near ditch

Total head: m NAP

4 3 2 1 0 1


2 3 0 5

I II III 45 50

15 10 20 25 30 35 40 Days after start of construction (13 August 2008 06:00) (b) Horizontal displacement, inclinometer no. 53: m 005 010 015

2 0 2 4 6 8 10

0 0 I




IIIB Clay shell Top clay Peat

Depth: m NAP

Full line: measurement Symbols: calculation

Allerd 0: Start of phase I I: End of phase I: filling basin, digging ditch II: End of phase II: ditch deeper PleistoIII: Phase III: infiltration in sand core cene IIIA: During pause in infiltration sand IIIB: Infiltration ends; failure imminent


Figure 12. Results of nite-element calculation, IJkdijk macrostability experiment, 2008: (a) construction phase settlements; (b) pore pressures; (c) horizontal deformations during failure phase


Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

Lower infiltration tube (A) 80 70 60 Infiltration tubes in sandfill (B) Pore pressure at base of sandfill (C) Pore pressure 1 m above base of sandfill (D) Horizontal deformation A 40 80 30 20 10 0 10 09:00:00 B C D 40 20 0 27 Sept. 2008 60 100 160 140 120

(Pore) pressure: kPa





Figure 13. Development of inltration pressures and pore pressures in sand core and of maximum lateral inclinometer deformation, phase III failure stage, IJkdijk macro-stability experiment, 2008

The nite-element calculation failed before these rather complicated processes occurred, after 0.33 of the ditch-deepening phase had been applied. A nite-element calculation with a coarser mesh of some 800 six-noded elements, however, did not fail even in phase III, as detailed in an earlier paper (in Dutch) by Den Haan and Feddema (2009). There, inltration pressures were applied by means of wells, and the sand penetration into the peat was also modelled. A factor of safety (FS) was determined at the end of phase III by undrained reduction of the Cam-clay Mparameter, and FS 1.17 was found, whereas FS 0.85 would be expected if the 15% side plane resistance was accounted for. The coarse mesh has considerably fewer degrees of freedom than the ner mesh, and this appears to offer additional resistance to failure. To be sure mesh neness was sufcient, a very ne mesh of 4297 15-noded elements was also used. It produced essentially the same results as reported in this paper.

The determination of the strength parameters of soft organic clay and peat has long troubled the geotechnical profession in the Netherlands. The very high 9 value of these soils poses problems both in laboratory strength testing and in stability calculations. The procedure used in this paper interpreting constant rate of strain K0 oedometer tests within the framework of a viscous version of modied Cam-clay to produce both strength and compressibility parameters is possibly a viable alternative for Dutch practice. The ability to predict vertical and horizontal embankment deformations has a bearing on ll material consumption, on track or road maintenance, and on deformations of foundations, piles and utilities buried near the toe of the embankment. The ability to faithfully predict pore pressures set up during construction can further reduce the occurrence of failures during construction if adequate surveillance and feedback are performed. The most important function of calculations, however, is to predict failure, and in this respect the adequate indication of failure of the IJkdijk embankment is encouraging. Predicting failure of dykes has become an important aspect of geotechnical engineering in the Netherlands. Dyke authorities are required to evaluate dyke safety every 5 years, and there are some 17 000 km of such dykes in the Netherlands! Evaluation is in terms of the factor of safety and probability indices determined from limit equilibrium analyses. Using the nite-element approach described in this paper, a dyke can rst be built up, from



The two nite-element calculations presented in this paper have applied a creep model that is a viscous version of modied Camclay, to embankments on soft organic clays and peat, with parameters determined from constant rate of strain K0 oedometer tests. The latter include the strength parameter M, which is quite high in these soils, while cohesion is zero, and strain-dependent permeability. The nite-element calculations with this model appear to produce very satisfactory ts to the measured deformations and pore pressures, and the failure of the IJkdijk case is also covered satisfactorily. 12

Max. horizontal deformation: mm

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

history as it were, to the present state, taking advantage of the prolonged creep compression under the dyke and the accompanying increase in shear resistance. Then a factor of safety can be determined by one of various methods in which failure is simulated. The most usual approach is the stepwise, undrained reduction of the Cam-clay strength parameter M until failure occurs, giving the factor of safety as the ratio of the available and the reduced value of M. However, the applied creep model has several limitations that need to be considered. These are (a) volume changes and pore pressures set up by rotation of the stress tensor during construction and loading (b) anisotropy of creep rates (c) overconsolidation effects. More general limitations lie in such matters as not modelling localising deformations along shear surfaces during failure, inadequate knowledge of the shape of the shear stress envelope in principal stress space, the non-uniqueness of solutions when nonassociativity is assumed on the shear stress envelope (e.g. by assuming zero viscoplastic volume change), and the inability to model side plane resistance. Stress tensor rotation occurs during embankment construction as a result of load spreading. At the toe, for example, the initial geostatic state with vertical major principal stress rotates over 908 to the passive state, and locations between crest and toe undergo intermediate amounts of rotation. In soft soils such rotations usually induce volume decrease and pore pressure increase. These effects are not dealt with by the creep model used here. Jardine et al. (1997) describe how, once consolidation has occurred under the rotated stresses, undrained loading without further rotation yields high undrained strength, close to that which is obtained without any rotation. This is due to the soils fabric gradually adapting to the rotated state of stresses and strains. In dyke safety evaluation the failure loads stem mostly from rising water levels, and as these are essentially horizontal, they will induce fresh rotations of the stress tensor, and appear so quickly as to be essentially undrained. This effect is probably small, however: the IJkdijk nite-element model was run with extreme water loading (water level in the basin behind the dyke raised quickly to crest level), and it was found that rotations were less than 108 in the zones in which shear failure is expected to occur. Modied Cam-clay is an isotropic model in the sense that the yield ellipse remains orientated along the isotropic stress axis. Developments are under way in which the ellipse rotates depending on the relative amounts of isotropic and distortional plastic strains. Such anisotropic models appear to improve ts to measured soil behaviour, and have recently been adapted to account for soil viscosity, as well as the specic behaviour of peat (Leoni

et al., 2010). This may provide an avenue for further improvement of the approach presented here. The creep model does not deal adequately with the overconsolidated state. On unloading, rates of creep strain reduce strongly and behaviour becomes essentially elastic, and only critical-state strength is used. Embankments with a signicant passive zone of highly overconsolidated material may therefore be less amenable to the approach described here. The side plane and mesh size effects on the moment of failure in the IJkdijk case have been noted, and in any calculation of failure it will be necessary to take these effects into account. Notwithstanding these limitations, and given that due care is exercised, the procedure described in this paper should allow successful application of the nite-element method to the calculation of deformations and strength of embankments on soft organic soils.

Begemann HKSPh (1971) Soil sampler for taking an undisturbed

sample 66 mm in diameter and with a maximum length of 17 metres. Proceedings of the 4th Asian Conference of the International Society for Soil Mechanics and Foundation Engineering, Bangkok, Thailand, pp. 5457. Bodas Freitas TM (2008) Numerical Modelling of the Time Dependent Behaviour of Clays. PhD thesis, Imperial College London, UK. Coutinho RQ and Lacerda WA (1989) Strength characteristics of Juturnaiba organic clays. Proceedings of the 12th International Conference on Soil Mechanics and Foundation Engineering, Rio de Janeiro, Brazil, vol. 3, pp. 17311734. De Boer R (2005) The Engineer and the Scandal. SpringerVerlag, Berlin, Germany. Den Haan EJ (1996) A compression model for non-brittle soft clays and peat. Geotechnique 46(1): 116. Den Haan EJ and Feddema A (2009) Deformatie en sterkte van ophogingen en dijken op slappe Nederlandse grond. Geotechniek 13(4): 5255. Den Haan EJ and Kamao S (2003) Obtaining isotache parameters from a CRS K0 -Oedometer. Soils and Foundations 43(4): 203214. Den Haan EJ and Kruse G (2006) Characterisation and engineering properties of Dutch peats. In Characterisation and Engineering Properties of Natural Soils (Tan TS, Phoon KK, Hight DW and Leroueil S (eds)). Swets & Zeitlinger, Lisse, Switerland, vol. 3, pp. 21012133. Jardine RJ, Zdravkovic L and Porovic E (1997) Anisotropic consolidation including principal stress axis rotation: Experiments, results and practical implications. Proceedings of the 14th International Conference on Soil Mechanics and Geotechnical Engineering, Hamburg, Germany, vol. 4, pp. 21652168. Jardine RJ, Smith PR and Nicholson DP (2003) Properties of the 13

Geotechnical Engineering

Deformation and strength of embankments on soft Dutch soil Den Haan and Feddema

soft Holocene Thames Estuary Clay from Queenborough, Kent. In Characterisation and Engineering Properties of Natural Soils (Tan TS et al. (eds)). Swets & Zeitlinger, Lisse, Switzerland, vol. 1, pp. 599644. Keverling Buisman AS (1934) Proefondervindelijke bepaling van de grens van inwendig evenwicht van een grondmassa. De Ingenieur, 26 June, 8388 (in Dutch). Landva AO and La Rochelle P (1983) Compressibility and shear characteristics of Radforth peats. In Testing of Peats and Organic Soils (Jarrett PM (ed.)). ASTM International, West Conshohocken, PA, USA, ASTM STP 820, pp. 157 191. Leoni M, Karstunen M and Vermeer PA (2010) Reply to discussion on Anisotropic creep model for soft soils. Geotechnique 60(12): 963966. Roscoe KH and Burland JB (1968) On the generalised stress strain behaviour of an idealised wet clay. In Engineering Plasticity (Heyman J and Leckie FA (eds)). Cambridge University Press, Cambridge, UK, pp. 535609.

Van MA, Zwanenburg C, Koelewijn AR and van Lottum H (2009)

Evaluation of full scale levee stability tests at Booneschans. Proceedings of the 17th International Conference on Soil Mechanics and Geotechnical Engineering, Alexandria, Egypt, vol. 3, pp. 20482051. Van Duinen TA (2008) Grensverleggend onderzoek macrostabiliteit bij opdrijven: Fase 2.C. Deltares, Delft, The Netherlands, Report 419230-0040 (in Dutch). Vasarhelyi B (2010) Tribute to the rst triaxial test performed in 1910. Acta Geodaetica et Geophysica Hungarica 45(2): 227 230. Vermeer PA and Neher H (1999) A soft soil model that accounts for creep. In Beyond 2000 in Computational Geomechanics (Brinkgreve RBJ (ed.)). Balkema, Rotterdam, The Netherlands, pp. 249261. Zwanenburg C, den Haan EJ, Kruse G and Koelewijn A (2012) Failure of a trial embankment on peat in Booneschans, The Netherlands. Geotechnique, geot.9.P.094.


To discuss this paper, please email up to 500 words to the editor at Your contribution will be forwarded to the author(s) for a reply and, if considered appropriate by the editorial panel, will be published as a discussion in a future issue of the journal. Proceedings journals rely entirely on contributions sent in by civil engineering professionals, academics and students. Papers should be 20005000 words long (brieng papers should be 10002000 words long), with adequate illustrations and references. You can submit your paper online via, where you will also nd detailed author guidelines. 14