Computers and Geotechnics 30 (2003) 81107 www.elsevier.

com/locate/compgeo

Three-dimensional pile-soil interaction in soldier-piled excavations

S.H. Honga,*, F.H. Leeb, K.Y. Yongc
a

ST Architects and Engineers Pte Ltd, 60 Admiralty Road West #03-01, Singapore 759947, Singapore b Department of Civil Engineering, 1 Engineering Drive 2, Singapore 119260, Singapore c Department of Civil Engineering, 1 Engineering Drive 2, Singapore 117576, Singapore Received 14 June 2001; received in revised form 29 March 2002; accepted 27 April 2002

Abstract Strutted excavations, including soldier-piled excavations, are often analysed using twodimensional (2D) nite element (FE) analyses with properties which are averaged over a certain span of the wall. In this paper, the eects of smearing the stiness of the soldier piles and timber laggings into an equivalent uniform stiness are examined, based on comparison between the results of 2D and 3D analyses. The ability of the 3D analyses to model the exural behaviour of the soldier piles and timber laggings is established by comparing the exural behaviour of various FE beam representations to the corresponding theoretical solutions, followed by a reality check with an actual case study. Finally, the results of 2D and 3D analyses on an idealized soldier piled excavation are compared. The ndings show that modelling errors can arise in several ways. Firstly, a 2D analysis tends to over-represent the coupling to pile to the soil below excavation level. Secondly, the deection of the timber lagging, which is usually larger than that of the soldier piles, is often underestimated. For this reason, the overall volume of ground loss is, in reality, larger than that given by a 2D analysis. Thirdly, a 2D analysis cannot replicate the swelling, and therefore softening, of the soil face just behind the timber lagging. Increasing the inter-pile spacing will tend to accentuate the eects of these modelling errors. # 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Soldier piles; Timber lagging; Excavation; Arching; Pile deection; Preloading

* Corresponding author. Tel.: +65-6765-9247; fax: +65-6769-6756. E-mail address: hongsh@sembcorpenc.com (S.H. Hong). 0266-352X/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved. PII: S0266-352X(02)00028-9

82

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

1. Introduction Soldier piles with timber lags have been used extensively as an excavation support system, particularly in sti soil conditions and where ground water ingress into the excavated area is not problematic [13]. The main advantage of using soldier piles is their relatively low cost and ease of installation, compared to other forms of support systems such as diaphragm walls and bored piles. Furthermore, since the soldier piles are not contiguous, much fewer soldier piles often need to be driven in comparison to sheet piles, thereby yielding signicant savings in time and cost of installation and thus allowing excavation to commence with a minimum of lead time. Soldier piles are often designed and analysed as contiguous wall systems even though they are, in reality, not so [4]. Nonetheless, designers have long recognized that the soldier pile retaining system is not only often more exible than other systems but the stiness of the system is non-uniform in that the piles are much stier than the timber lags. To address this characteristic, Trada [5] and BS8002 [6] recommended a 2025% reduction of Pecks earth pressure for the soldier piles and 50% reduction for the timber lagging. Armento [7] also proposed a similar reduction based on measurements from the excavations for the 12th and 19th Street Stations of the BART system. Empirical guidelines for the passive earth pressure mobilized in the soil in front of the soldier piles have also been suggested. For instance, Broms [8], Teng [9], NAVFAC [10] and the CGS [11] proposed that a soldier pile of width B can be assumed to mobilize the passive earth pressure imposed by a lateral extent of ground of span 3B (i.e. 1.5B on both sides of the pile). The fact that the soldier piles and timber laggings are of vastly dierent construction and have very dierent stinesses means that the interaction between the retaining system and the soil is three-dimensional (3D) in nature. However, most of the nite element (FE) analyses on soldier pile walls to date are two-dimensional (2D) and based on the assumption of plane strain [3,1214]. The dierences in pile and timber lagging stinesses cannot be modelled by these analyses, which necessarily assumes a smeared uniform stiness for the entire span of the retaining wall. Briaud and Lim [15] used 3D FE analyses to study tieback soldier pile walls. However, their study focuses more on the parameters of the tiebacks and depth of soldier pile embedment rather the eects of dierent types of analyses. Furthermore, Briaud and Lim [15] modelled the soldier piles with beam elements. As will be discussed later, beam elements may not realistically represent the interaction between the soldier piles and surrounding soil, owing to their inability to replicate the nite crosssectional dimensions of real soldier piles. In this paper, the eects of smearing the stiness of the soldier piles and timber laggings into an equivalent uniform stiness are examined, based on comparison between the results of 2D and 3D analyses. Smeared uniform stiness is used for the retaining systems in the 2D analyses. On the other hand, in the 3D analyses, dierent elements and material types represent the soldier piles and timber laggings. In the rst part of the paper, the ability of the 3D analyses to model the exural behaviour of the soldier piles and timber laggings is examined. This is accomplished

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

83

by comparing the exural behaviour of various FE beam representations to the corresponding theoretical solutions. A reality check is then conducted using ORourkes study on the G Street test section as a reference. Finally, the results of a comparative study between 2D and 3D analyses on an idealized soldier piled excavation will be discussed to highlight the eects of using smeared uniform stinesses in 2D analyses. All of the analyses reported herein were conducted using an in-house version of CRISP (Britto and Gunn [16]), with 3D beam and reducedintegration brick elements incorporated.

2. Modelling of soldier piles in 2D and 3D analyses In this section, the modelling of soldier piles in 2D and 3D analyses is rst addressed. Soldier piles act essentially in exure, and it is well-known that 4-noded quadrilaterals do not model exural behaviour well, owing to the use of incomplete polynomials as shape functions, which causes shear-locking [17,18]. The 8-noded quadrilateral element suers from the same problem, albeit to a lesser extent [18]. It has been shown [19,20] that, by using reduced-integration eight-noded quadrilaterals, behaviour in 2D can be readily modelled. The same may presumably be true for the 20-noded brick element. In order to assess the ability of various elements to represent the soldier pile, the idealized problem of a long cantilever beam subjected to a triangular load distribution, the latter representing an idealized earth pressure prole, was studied using various 2D and 3D

Fig. 1. Comparison of deection proles under distributed load for 2-D 43 m-cantilever beam. (a) 32-bit precision. (b) 64-bit precision.

84

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Fig. 2. Comparison of bending moments under distributed load for 2-D 43 m-cantilever beam. (a) 32-bit precision. (b) 64-bit precision.

Fig. 3. Eect of aspect ratios for 2-D 43 m-cantilever beam in 64-bit precision using RIQUAD elements. (a) Deection. (b) Bending moment.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

85

Fig. 4. Comparison of deection proles under distributed load for 3-D 43 m-cantilever beam. (a) 32-bit precision. (b) 64-bit precision.

Fig. 5. Comparison of bending moments under distributed load for 3-D 43 m-cantilever beam. (a) 32-bit precision. (b) 64-bit precision.

86

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

representations, as shown in Figs. 15. The 2D representations were achieved using beam element as well as full- and reduced-integration 8-noded quadrilateral elements, hereafter termed FIQUAD and RIQUAD, respectively. The 3D representations were achieved using beam element as well as full- and reduced-integration 20noded brick elements, hereafter termed FIBRICK and RIBRICK, respectively. For each type of element, two dierent cases were studied. In the rst case, one element was used to model the whole length of cantilever beam. In the second, the cantilever was modelled using 18 elements of equal length. For the quadrilateral elements, the element aspect ratios for the 1- and 18-element representations are 1:70.5 and 1:3.9, respectively. Fig. 1a compares the deected shape computed using 32-bit precision with the theoretical value from EulerBernoulli beam theory. As can be seen, all of the 8noded quadrilateral representations show signicantly smaller deection than the theoretical value. The only representations that come close to the theoretical value are predictably, those using beam elements. Fig. 2a shows the corresponding bending moment proles obtained from the various representations. For beam elements, the bending moment prole was interpolated from the integration point values using a spline function. For quadrilateral elements, the bending moments were evaluated using the method suggested by Rahim and Gunn [21]. As can be seen, the discrepancy in the deection is also reected in the bending moment proles. Figs. 1b and 2b show the corresponding deected shapes and bending moment proles computed using 64-bit precision. As can be seen, a dramatic improvement in accuracy is now achieved in deection and bending moment in the 18-element quadrilateral representations. On the other hand, both 1-element quadrilateral representations show little or no improvement in accuracy. In summary, exural behaviour of the cantilever is well represented by single and multiple beam elements, in both 32- and 64-bit precision. The 18-element quadrilateral representations adequately represent the cantilever behaviour if they are used with 64-bit precision. In practice, this is unlikely to be a major constraint since 64-bit precision may be needed, in any case, to capture the large dierence in stiness between soil and pile [16]. Between the two 18-element quadrilateral representations tested, the RIQUAD representation produces a smoother bending moment prole whereas the FIQUAD representation produces a rather jagged bending moment prole. The 1-element quadrilateral representations are unable to adequately replicate the deection and bending moment proles, regardless of whether 32- or 64-bit precision is used. In other words, the representations which model cantilever exure well are the beam element and the RIQUAD element with 64-bit precision. Fig. 3 shows the eect of using dierent number of RIQUAD elements, and therefore aspect ratios. As can be seen, reasonably good bending moment representation can be achieved for aspect ratio less than about 14. Figs. 4 and 5 shows the corresponding deection and bending moment plots for the 3D representations. As can be seen, the performance trends of the various 3D representations closely mirror those of their 2D counterparts, with the beam elements being the most reliable and the FIBRICK element being the least.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

87

In soldier-piled excavations, the soldier piles not only act in exure but also attract earth pressures from the retained soil. Although this phenomenon is not modelled, and thus irrelevant, in 2D analyses, it is relevant to a 3D analysis wherein soldier piles and timber laggings are modelled explicitly. In this section, the ability of the beam and brick representations of the soldier piles to capture the interaction between pile and soil in a 3D analysis will be studied using a typical section of an idealized unstrutted, soldier-piled excavation shown in Fig. 6. As can be seen, the chosen section consists of one half-soldier pile and the ground in front of and behind it. The soldier piles are modelled using 64-bit precision and are spaced at an interval of six times the width of the soldier pile, center-to-center. In this analysis, the outer cross-sectional dimensions and exural rigidity EI of the soldier are based on the properties of the 610 mm305 mm149 kg/m H-pile [22]. The outer cross-sectional dimensions are only simulated in the RIBRICK elements, not beam element. Nodes on the four vertical faces of the mesh are only allowed to move in-plane. The bottom face is xed in all directions. Timber laggings are not modelled, as the objective of this exercise is to study the ability of various soldier pile representation to model the pile-soil interaction. Soil Layer 1 is underlain by Layer 2 at 22.5 m depth to model the increasing stiness of real ground. Each soil layer is assumed to be elastic and uniform. The soldier pile is socketed 1.5 m deep into Layer 2. A Poissons ratio of 0.02 is used for RIBRICK soldier pile elements as pure bending behaviour is only

Fig. 6. Locations of beams and linear strain bricks for modelling soldier piles in the nite element model of an idealistic excavation. (a) Plan view of beam element as soldier pile in section AA. (b) Plan view of RIBRICK element as soldier pile in section AA. (c) Cross-sectional view in xy plane of the FEM model. (d) Soil layers, struts and excavation levels.

88

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Table 1 Idealised soil and soldier pile properties for beam and brick soldier piles study Soil layer 1 (depth: 022.5 m) Elastic s=19.5 kN/m3 E=50 000 kN/m2 =0.3 K0=0.94 kx=1.0109 m/s ky=1.0109 m/s s=19.5 kN/m3 E=120 000 kN/m2 =0.3 kx=1.0109 m/s ky=1.0109 m/s E=1108 kN/m2 Ixx=1.259103 mm4 Iyy=9.308105 mm4 =0.29 A=0.019 m2 E=4.36107 kN/m2 =0.02 G=2.14107 kN/m2

Soil layer (depth: 22.537 m) Elastic

Half soldier pile modelled with beam at 6B pile spacing

Half-soldier pile modelled with RIBRICKs at 6B pile spacing (610152.5 mm)

achieved if Poissons ratio is zero [23]. The properties of the soldier piles and soil are shown in Table 1. The soldier pile is assumed in-place at the start of the analysis. The excavation process is assumed to occur in three lifts with pore pressure dissipation being modelled using the Biots [24] consolidation formulation. The nal depth of excavation is 11.5 m. As shown in Fig. 7a, the solid and beam elements give similar results and the differences appear to be minor. The dierences are more clearly reected in Fig. 7b, where the bending moment of the solid elements is clearly larger than that of the beam element, although the dierences are still not substantial. This can be attributed to the nite cross-sectional dimensions of the RIBRICK element, which enables it to attract a larger proportion of the earth pressure through the arching process than the beam elements. Fig. 8 shows the local deection of soldier pile and soil face, in plan view, at 9 m depth below ground surface. As can be seen, the dierential movement between pile and soil is clearly larger with the beam representation than the RIBRICK representation of the soldier pile. Furthermore, as shown in Figs. 9a and b, although the development of stress arch in the soil is evident in both representations, the local stress distribution is very dierent. The brick representation results in a much higher concentration of stresses in the vicinity of the soldier pile than the beam representation, owing to its nite dimensions. In addition, the stress arch spans over a shorter distance but extends deeper into the retained soil in the case of the brick representation. This allows the soil movement to be better coupled to the pile

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

89

Fig. 7. (a) Deection prole of soldier pile and soil at the center between adjacent piles. (b) Bending moment prole of soldier piles.

movement than the beam representation. Thus, although a beam element is slightly more accurate in replicating the exural behaviour of the soldier pile, local pile-soil interaction is better represented using solid elements. For this reason, subsequent 3D analyses discussed herein will model the soldier piles using reduced-integration 20-noded bricks with 64-bit precision, unless otherwise stated.

3. Comparison with case history In this section, a comparison is made with ORourkes [3] eld measurement of a 18.3 m-deep braced excavation in the G Street test section (Panel 158). Fig. 10 shows

Fig. 8. Plan view of lateral deection retaining system 9 m depth after nal excavation stage.

90

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Fig. 9. Lateral stress contours for at 9.3 m below ground level for excavation to 11.5 m. (a) Beam soldier pile elements. (b) RIBRICK soldier pile elements.

Fig. 10. Plan view of retaining system after ORourke (1975).

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

91

Fig. 11. Finite element mesh used for reality check with ORourke (1975).

the plan view of the retaining system. The modelled geometry and excavation sequence follows closely that of ORourkes. Fig. 11 shows the 2D and 3D FE meshes used in the current analyses. The 2D mesh follows that of ORourkes, whilst the 3D mesh is merely an extrusion of the 2D mesh in the third dimension. Fig. 11 shows the soil parameters from ORourke [3], which were adopted in this comparative study. Since ORourke only presented total stress parameters, only a total stress analysis was performed. Furthermore, following ORourke, the DruckerPrager soil model is used for all soil layers. In the 2D analysis, the soldier pile wall was modelled using beam elements with a smeared wall stiness derived by scaling the bending stiness of individual soldier piles over the distance separating them and neglecting the eects of the timber lags; this also follows ORourkes [3] procedure. In the 3D analyses, the soldier piles and timber lags are modelled using RIBRICK elements, the properties of which are shown in Table 2. The soldier pile properties were derived from the section properties of the 24 WF 100 H-piles used in the excavation. Less is known about the timber lags apart from the fact that they were made from oak wood. Back-calculation using BS8002 [6] guidelines and AITC [25] suggest that the required thickness of the timber lags is likely to be about 100 mm, which falls within the common range of timber lagging thickness [2]. Because of the uncertainty over the thickness of the timber lagging, the analysis was repeated with timber lagging thickness of 50 and 100 mm. Since timber lagging is normally installed in the form of long, narrow planks slotted

92

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Table 2 Idealised soil and soldier pile properties for ORourkes case study Soldier piles E (kPa) 5.0910 Struts Level 1 (36 WF 260) Level 2 Level 3 Level 4 Level 5 Timber lagging E (kPa) 1108 1108 1108 1108 1108 Eh (kPa) 12103 Ev (kPa) 1
7

 0.02 Ixx (m4) 0.0072 1.19106 0.55106 1.19106 1.19106 hh 0.02 h 31013 Iyy (m4) 4.54104 1.19106 0.55106 1.19106 1.19106

Ghv (kPa) 2.5107 Area (m2) 0.0623 0.0296 0.0238 0.0296 0.0296 Gh (kPa) 0.3676

horizontally in between the anges of the soldier piles [2,6], the exural rigidity of the timber lagging to bending in the vertical direction is likely to be very small. To model this behaviour, an anisotropic elastic material model is used for the timber lagging, with the Youngs modulus in the vertical direction being set much lower than that in the horizontal direction, Table 2. This diers from Briaud and Lims [15] analyses which used isotropic shell elements to model the timber lagging. The

Fig. 12. Comparison of a series of 2-D and 3-D analyses from ORourke (1975). (a) Soldier pile deection. (b) Bending moments.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

93

braces and walers were modelled using beam elements, with properties adjusted to replicate those in the eld. Fig. 12a compares the computed wall deection to the ORourkes measured nal deection. As can be seen, all the computed deections agree reasonably well with one another and with the eld measurements. The same can be said of the bending moment diagrams in Fig. 12b. This indicates that the results of the FE analyses are reasonably realistic. Nonetheless, it is interesting to note that the 3D FE analysis predicts a slightly larger maximum soldier pile deection than the 2D analyses; in spite of the fact that the smeared stiness of the soldier pile wall in the 2D analysis does not take into account the contribution from the timber lagging. In other words, if the 3D analysis is taken as a reasonably accurate representation of reality, then the wall deection from the 2D analysis represents an optimistic, rather than average, estimate of the soldier pile and timber lagging deection. The reason for this will be examined in greater detail later. Fig. 13ac show the deection proles leading to at various excavation levels. As can be seen, more substantial dierences between measured and predicted wall deection are present in the earlier stages. This may be attributed to the fact that soil behaviour is not truly linearly elastic prior to yielding. The lower strain levels which are incurred in the early stages could have allowed higher soil stiness to be mobilised than in the nal stages. The timber lagging thickness of 50 and 100 mm leads to nearly the same result, thus indicating that the results are not heavily contingent upon the assumed thickness of timber lagging. In both cases, the computed deection of the timber lagging is signicantly larger that that of the soldier pile deection. This is to be expected, since the timber lagging is much more exible than

Fig. 13. Deection of soldier pile and soil at midspan with respect to excavation sequence. (a) Excavation depth at 6.1 m. (b) Excavation depth at 11.9 m. (c) Excavation depth at 18.3 m.

94

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Fig. 14. Eect of smearing on rotational xity below excavation level.

the soldier piles and thus expected to deect more. In other words, signicant dierences can arise between the soldier pile deection and that of the intervening soil. Fig. 13 also shows that, just below excavation level, there is a short segment of the soldier pile which moves ahead of the soil. This is to be expected; the soldier pile wall is not a continuous wall below excavation level. The presence of soil between adjacent soldier piles allows the piles to move relative to the intervening soil; this movement being accentuated if the soil yields. In 2D analysis, the discontinuous nature of the wall cannot be modelled since a uniform wall with a reduced smeared stiness is assumed instead. This suppresses the relative pile-soil movement, thus exaggerating the degree of rotational xity aorded by the pile embedment. As shown in Fig. 14, using a smeared stiness for the soldier pile and the intervening soil in a 3D analysis leads to a deection prole that is virtually identical with that from a 2D analysis, thus indicating that the dierence in deection between 2D and 3D analyses noted earlier arises from the representation of the nonuniform stiness of the soldier pile and intervening soil. It is interesting to note that the S-prole present in ORourkes eld data was not predicted in the FE analyses. One possible reason for this is the high values of cohesion and friction angle prescribed by ORourke [3] for the top layers of soil. ORourke reported signicant variation in the modulus of the Upper Brown sand

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

95

from about 41 to about 97 MPa. This is suggestive of signicant variation in soil properties with location, which is not considered in the analysis wherein uniform soil properties are assumed. As shown in Fig. 15, progressive reduction in c and  values does lead to an S-shape soldier pile deection prole.

4. 3D and 2D comparisons 4.1. Idealised excavation The complexity of the soil prole and multi-level bracing at the G Street excavation renders an in-depth study of pile-soil interaction dicult. For this reason, the idealized excavation involving two levels of bracing in a single soil type is studied, as shown in Fig. 6. To enhance the long-term stability of the soil face, 3-inch thick timber laggings were modelled. As can be seen, the two levels of bracing were prescribed at 4.5 and 7.5 m below the original ground level. Boundary conditions are similar to those used in the back-analysis of the G Street excavation. Initial groundwater table is assumed to be 2 m below original ground level, this being a typical groundwater table depth in Singapore conditions. The corresponding 2D mesh is not shown since it is merely a vertical section of the 3D mesh.

Fig. 15. Modelling the S-shaped deection prole of the soldier pile in 3D.

96

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Table 3 Idealised soil and soldier pile properties for idealised study Soil layer 1 (depth: 022.5 m) Modied Cam-clay S=19.5 kN/m3 cc=0.1 cS=0.01 M=1.2 ecs=1.47 =0.3 K0=0.94 S=19.5 kN/m3 E=120 000 kN/m2 =0.3 Eh=16.3103 kN/m2 E=1 kN/m2 hh=0.02 vh=2.2105 Ghv=0.37 kN/m2 E=1.09108 kN/m2 Ixx=1.259103 mm4 =0.29 A=0.019 m2 E=2.18107 kN/m2 =0.02 G=1.07107 kN/m2 E=4.36107 kN/m2 =0.02 G=2.14107 kN/m2 E=2.67107 kN/m2 Ixx=1.259103 mm4 =0.29 A=0.019 m2 E=1.22107 kN/m2 Ixx=1.259103 mm4 Iyy=9.308105 mm4 =0.29 A=0.019 m2 E=2.44107 kN/m2 Ixx=1.259103 mm4 Iyy=9.308105 mm4 =0.29 A=0.019 m2

Soil layer 2 (depth: 22.537 m) Elastic

Timber lagging at 3 in thickness Elastic

Soldier pile modelled with 2D beam element

Half soldier pile modelled with RIBRICK at 3B pile spacing (610152.5 mm)

Half soldier pile modelled with RIBRICK at 6B pile spacing (610152.5 mm)

Strut at 3.75 m spacing modelled with beams Level 1 at 165.0 kN/m Level 2 at 441.1 kN/m

Half strut with stiness equivalent to 3.75 m spacing modelled with beams for cases of 3B pile spacing Level 1 at 75.5 kN Level 2 at 201.8 kN Half strut with stiness equivalent to 3.75 m spacing modelled with beams for cases of 6B pile spacing Level 1 at 151.0 kN (3060%) Level 2 at 403.6 kN (4070%)

Note: Walers are assumed to be suciently sti and able to distribute the load evenly to all soldier piles and thus not modelled in the idealized analyses.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

97

Table 3 shows the properties of soil and structural components used in the analyses. The analyses conducted are eective stress analyses and the soil properties prescribed are representative of granitic sapprolites. Such geologic material occurs widely in Singapore [26] and can be described as a silt of intermediate or high plasticity [27]. Soldier piles have been used as a retaining system in such soils [28]. The modied Cam Clay model is used to describe the behaviour of the soil skeleton since it allows soils with dierent over-consolidation ratios (OCRs) to be modelled while requiring relatively few parameters compared to other, more complex models. The lack of detailed information about the soil to be modelled does not justify the use of more sophisticated models in such an idealized study. Fig. 16 shows the excavation and wall installation sequences of the idealized excavation. For the non-preloaded cases, the preloading stage is simply omitted in the analyses. As shown in Table 4, the eects of preloading and centre-to-centre inter-pile spacing on the discrepancy between 2D and 3D analyses are examined.

Fig. 16. Construction sequence for idealized excavation from stage A through G.

Table 4 Summary of eective stress analyses for the idealised excavation at OCR=3 and k=1108 m/s Analysis 2D-1 2D-2 3D-1 3D-2 3D-3 3D-4 Geometric representation Plane strain Plane strain 3B 6B 3B 6B Preloading No Yes No No Yes Yes

Note: B is the width of the soldier pile.

98

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

4.2. Wall deection Fig. 17 compares the 2D and 3D computed wall deection proles for the nonpreloaded soldier piles at the nal stage of construction. As can be seen, for interpile spacings of 3 and 6 times the soldier pile width B, the 3D analyses predict slightly larger maximum soldier pile deection than the 2D analysis, this being consistent with the trend indicated for the G-Street excavation. The deection prole at the mid-span of the timber lagging is more variable. For an inter-pile spacing of 3B, the timber lagging deection remains reasonably well coupled to the soldier pile deection. For an inter-pile spacing of 6B, the timber lagging deection becomes much larger than the soldier pile deection. It should be noted that an inter-pile spacing of 6B is not unrealistic; in the G-Street excavation [3], the inter-pile spacing was approximately 6.5. Comparison of the soldier pile and timber lagging deection proles shows that, above excavation level, the mid-span deection is larger than the soldier pile deection, which is consistent with the fact that the soldier pile is supporting the soil face through the timber lagging and the soil arch spanning across adjacent piles. On the other hand, the reverse is the case just below excavation level. This occurs because, whereas the soil face at the mid-span can adopt rather sharp

Fig. 17. Comparison of wall deection between 2D and dierent pile spacing for cases without preloading at nal stage of construction.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

99

Fig. 18. Comparison of wall deection between 2D and dierent pile spacing for cases with preloading at nal stage of construction.

curvature, the soldier pile cannot, owing to its much higher exural stiness. Thus, soldier pile deection can only decrease gradually below excavation level thereby resulting in higher deection in the pile than the mid-span soil. This is similar to the 3D FE results on the deection of pile groups adjacent to surcharge, reported by Bransby and Springman [29].

Fig. 19. Comparison of eects of preloading on wall deection for 3D-4 at 4.5 m below ground level with stages B, C and D shown in Fig. 16.

100

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Fig. 20. Location of stress paths plotted for soil behind rst level of strut at 4.22 m below ground level in xz plane.

Fig. 21. Stress path plot for 2D-1 without preload. Stages AG are shown in Fig. 16. Units of stresses in kPa.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

101

Fig. 18 shows the wall deection proles of the preloaded cases at the nal stage of construction. As can be seen, preloading pushes the soldier pile backwards into the retained soil, this combining with the excavation below strut level to produce the S-shape prole that is also present in ORourkes measurement, but not in the idealised non-preloaded case discussed in the previous section. Fig. 19 compares the dierences in deection at the excavation face 4.5 m below ground level before and after preloading. As shown in Fig. 16, in stage B, excavation just reaches a depth of 4.5 m. At this point, the deection of the soldier pile is larger than that of the intervening soil, owing to the large exural stiness of the wall, as discussed earlier. In stage C, the process of preloading pushes soldier pile and the soil rearwards, but with the soldier pile displacing more. Subsequent excavation (stage D) allows the soil to move forward more than the soldier pile as the latter is restrained by the strut and waler system. 4.3. Stress changes This interaction between soil and soldier pile is illustrated in Figs. 2126, which show the stress paths for two points in the retained soil at the upper strut level,

Fig. 22. Stress path plot for 2D-2 with preload. Stages A to G are shown in Fig. 16. Units of stresses in kPa.

102

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

which is located at a depth of 4.22 m below ground surface. As shown in Fig. 20, one point (marked I) is located just behind a soldier pile whereas the other (marked II) is located just behind the mid-span of the timber lagging. Results for a corresponding point at the same depth from a corresponding 2D analysis are also plotted for comparison. The deviator stress q is strictly a positive quantity. In Fig. 21, a ^ modied quantity q is instead plotted, such that y x q ^  q 1 y x  ^ where  x0 and  y0 are the horizontal and vertical eective stresses. The use of q in place of q allows an active stress state ( y0 >  x0 ) to be distinguished from a passive stress state ( x0 >  y0 ). As Fig. 21 shows, for the 2D analysis, initial excavation leads to an increase in mean eective stress due to the drag from the soil ow above the monitored point. However, this quickly faded away as the formation level approaches the monitored level (end point B) and lateral eective stress starts to decrease. Meanwhile, q continues to increase as shear stresses builds up in the retained soil. Further excavation

Fig. 23. Stress path plots for 3D-1 and 3D-2 at point I without preload. Stages AG are shown in Fig. 16. Units of stresses in kPa.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

103

^ below the formation level leads to a reversal in the sign of q (end point D) as the lateral stress falls below the vertical stress, leading to an active stress state. Comparison of Figs. 21 and 22 shows that preloading causes the stress state to remain in a passive, rather than active, state throughout the entire construction sequence, thereby eliminating the passiveactive reversal predicted for the unpreloaded case. As Figs. 23 and 24 show, the initial stress changes behind the soldier pile (point I) and the timber lagging (point II) are similar to that of the 2D prediction in Fig. 21. At both locations, initial excavation leads to an increase in mean eective stress and a passive stress state due to the drag from the soil ow above the monitored point. As excavation approaches the monitored level, the release in lateral stress due to soil removal in front of the wall leads to an active stress state (end points B and C). However, further excavation below the rst strut level produces very dierent stress paths at the two monitored locations. As shown in Fig. 23, the build-up of lateral stress at the base of the soil arch between adjacent soldier piles results in a passive stress state, which persists up to the end of the construction sequence. On the other hand, as negative pore pressure in front of the soil arch dissipates, the eective stress in the soil behind the timber lagging decreases as the soil approaches a state of incipient active failure at the end of the construction sequence.

Fig. 24. Stress path plots for 3D-1 and 3D-2 at point II without preload. Stages AG are shown in Fig. 16. Units of stresses in kPa.

104

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

In physical terms, this suggests signicant softening of the soil behind the timber lagging. These stress path dierences are accentuated as the soldier pile spacing increases from 3B to 6B. Comparison of Figs. 25 and 26 with Figs. 23 and 24 shows that preloading does not have the signicant eect that it has on the 2D analysis. The only signicance changes that arise from preloading occur behind the soldier pile and consist of an earlier reversal back to a passive stress state (before end point C) and a slightly larger mean eective stress at the end of the construction sequence. Both of these are directly attributable to the preload. There is no signicant dierence at location II. Comparison of Fig. 21 with Figs. 23 and 24 as well as Fig. 22 with Figs. 25 and 26 shows that, at any given end point, the stress state predicted by the 2D analysis appears to lie somewhere between those of locations I and II in the 3D analysis. This is not surprising in view of the fact the retaining eect of the soldier pile and timber lagging is smeared in a 2D analysis. The large changes in the 2D stress paths between Figs. 21 and 22 appears to be due to the fact that smeared stress path of the 2D analysis is aected by the small variations in the strongly divergent stress states of the two locations.

Fig. 25. Stress paths plots for 3D-3 and 3D-4 at point I with preload. Stages AG are shown in Fig. 16. Units of stresses in kPa.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

105

Fig. 26. Stress paths plots for 3D-3 and 3D-4 at point II with preload. Stages AG are shown in Fig. 16. Units of stresses in kPa.

5. Conclusions Strutted excavations, including soldier-piled excavations, are often analysed using 2D FE analyses with properties that are averaged over a certain span of the wall. The foregoing discussion shows that when such an approach is applied to soldierpiled excavations, modelling errors can arise in several ways. Firstly, by modelling the discrete soldier piles as a wall, a 2D analysis tends to over-predict the coupling to pile to the soil below excavation level. In instances where the pile is not embedded into sti soil, this modelling error can give rise to discernible underestimation of soldier pile deection. Secondly, the deection of the timber lagging, which is usually larger than that of the soldier piles, is often underestimated. Thirdly, a 2D analysis cannot replicate the dierences in stress paths taken by the retained soil at dierent locations at the same depth. In particular, it cannot model the softening of the soil face just behind the timber lagging. In instances where the timber lagging is insucient or its installation, this softening can give rise to local collapse, which may exacerbate the ground loss and therefore ground movement. Increasing the inter-pile spacing will tend to accentuate the eects of these modelling errors.

106

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

Acknowledgements The rst author would like to acknowledge the research scholarship and scholarship augmentation provided by the National University of Singapore and Econ Piling Pte. Ltd., respectively. The equipment and software support provided by the Centre for Soft Ground Engineering at the National University of Singapore is also gratefully acknowledged.

References
[1] GCO. Review of design methods for excavations. Hong Kong: Geotech. Control Oce; 1990. [2] Tomlinson MJ. Foundation design and construction, 6th ed. London: English Language Book Society; 1995. [3] ORourke TD. A study of two braced excavations in sands and interbedded sti clay. PhD thesis, University of Illinois; 1975. [4] Peck RB. Deep excavations and tunnelling in soft ground. Proc. Seventh Conf. Soil Mech. and Found. Engrg., Mexico. State of the Art Report; 1969. p. 22590. [5] Trada. Timber in excavations. London: Thomas Telford Ltd; 1990. [6] BS 8002. Code of practice for earth retaining structures. British Standards Institutions; 1994. [7] Armento WJ. Criteria for lateral pressures for braced cuts. In: Proceedings of the specialty conference on performance of earth and earth-supported structures, Vol 1, Part 2. Indiana: Purdue University; 1972. [8] Broms BB. Lateral resistance of piles in cohesive soils. J Soil Mech Found Engrg, ASCE 1964; 90(SM2):2764. [9] Teng WC. Soil stresses for design of drilled caisson foundations subjected to lateral loads. ASCE, Conference on Analysis and Design in Geotechnical Engineering; 1975. [10] NAVFAC. Design manual 7.2: foundations and earth structures. Washington (DC): Department of the Navy, Naval Facilities Engineering Command; 1982. [11] CGS. Canadian foundation engineering manual. Canadian Geotechnical Society, Technical Committee on Foundations; 1992. [12] Clough GW, Weber PR, Lamont J. Design and observation of a tie-back wall. In: Proceedings of the Specialty Conference on Performance of Earth and Earth-Supported Structures, Vol. 1. New York; 1972. p. 13671389. [13] Tsui Y, Clough GW. Plane strain approximations in nite element analyses of temporary walls. In: Proc. Conf. Analysis and Design in Geotech. Engrg., Vol. 1. Austin: University of Texas; 1974. p.173. [14] Gomes Correia A, da Guerra NMC. Performance of three Berline-type retaining walls. In: Proceedings of the 14th International Conference of Soil Mechanics and Foundation Engrg., Vol. 2. Hamburg; 1997. p. 1297300. [15] Briaud J-L, Lim Y. Tieback walls in sand: numerical simulation and design implications. Journal of Geotechnical and Geoenvironmental Engineering 1999;125(2):10110. [16] Britto AM, Gunn MJ. CRISP 90 users & programmers guide; 1990. [17] Livesley RK. Finite element: an introduction for engineers. Cambridge: Cambridge University Press; 1983. [18] Reddy JN. An introduction to the nte element method, 2nd ed.. New York: McGraw Hill, Inc; 1993. [19] Zienkiewicz OC, Taylor RL. The nite element method, 4th ed., Vol. 1. London: McGraw-Hill; 1988. [20] Day RA, Potts DM. Modelling sheet pile retaining walls. Computers and Geotechnics 1993;15:125 43. [21] Rahim A, Gunn MJ. Bending CRISP. CRISP NEWS, No. 4, Sage Engineering Ltd.; 1997. [22] BSI. Steelwork design guide to BS 5950: part 1: 1990, Vol. 1, Section properties, member capacities, 4th ed. British Standards Institutions; 1996.

S.H. Hong et al. / Computers and Geotechnics 30 (2003) 81107

107

[23] MacNeal RH. Finite elements: their design and performance. New York: Marcel Dekker, Inc; 1994. [24] Biot MA. General theory of three dimensional consolidation. Journal of Applied Physics 1941;12: 15564. [25] AITC. Timber construction manual, 4th ed. New York: American Institute of Timber Construction, Wiley; 1994. [26] Dames and Moore. Detailed geotechnical. Study interpretative report. Singapore Mass Rapid Transit System; 1983. [27] BS 5930. Code of practice for site investigations. British Standards Institutions; 1999. [28] Wong IH, Poh TY, Chuah HL. Performance of Excavation for Depressed Expressway in Singapore. J Geotech Geoenv Engrg, Div, ASCE 1997;123(7):61725. [29] Bransby MF, Springman SM. 3-D Finite element modelling of pile groups adjacent to surcharge loads. Computers and Geotechnics 1996;19:30124.