Documente Academic
Documente Profesional
Documente Cultură
Abstract: Although the pile-raft foundation is preferred for supporting a tall wind turbine, the geotechnical design and selection of suitable
design parameters are based on a complex procedure. Except the foundation, all the other aboveground components are precast members that
are assembled at the project site to build a wind turbine. Therefore, it is necessary to consider the possible variations in soil properties and wind
speed in the design of the foundation. In this paper, a reliability-based robust design procedure for a pile-raft foundation that supports a 130-m-
tall wind turbine on a layered clayey soil is presented. Upon completion of the geotechnical design for the mean wind speed and undrained shear
strength, a parametric study and Monte Carlo simulation were conducted by varying the wind speed and the undrained cohesion of each layer to
establish a relationship among the design variables (number and length of piles and radius of the raft) and the random variables (wind speed and
undrained cohesion). Finally, a reliability-based robust design was created considering the total cost and robustness as the objectives. The stand-
ard deviation of the response of concern, which is the differential settlement, was considered the measure of robustness. The optimization
yielded a set of preferred designs known as the Pareto front, and the suitable design was selected for a given cost limitation and performance
requirement using the Pareto front. DOI: 10.1061/(ASCE)GM.1943-5622.0001061. © 2017 American Society of Civil Engineers.
Author keywords: Pile raft; Geotechnical design; Design optimization; Wind turbine; Hybrid foundation.
Introduction higher energy at a small additional cost. Lewin (2010) found that an
increase in the turbine height from 80 to 100 m results in a 4.6%
Although it is one of the fastest growing clean and renewable ener- greater wind speed and a 14% increase in power output, and that an
gies in the world, wind energy accounts for only 3.3% of the total increased height from 80 to 120 m results in an 8.5% greater wind
electricity generated worldwide. Nonetheless, according to the speed and a 28% increase in power production. Because the initial
2015 Global Wind Report (GEWC), the global cumulative installed construction cost of a wind farm covers the highest percentage of
wind capacity greatly increased by the end of 2015, up 17% from the total cost of the project, it is logical and cost-efficient to increase
the preceding year (GEWC 2015). By the end of 2015, the United the wind energy production by building taller towers that can gener-
States added 4,000 new wind turbines contributing approximately ate additional power from the same number of wind turbines.
8,598 MW of energy, which increased the total installed capacity, Although it increases the wind energy production of a single
from the end of 2014, by 13% (GWEC 2015). Although a signifi- wind turbine, building taller towers poses significant challenges to
cant number have been installed in the United States, wind turbines the geotechnical engineer in designing and selecting the most cost-
account for only 4.7% of the total electricity produced nationwide. efficient foundation for the given subsurface and wind conditions.
The selection of suitable locations for onshore wind farms depends A taller wind turbine tower not only increases the vertical dead load
on factors such as wind speed, soil condition, availability of con- but also significantly increases the lateral load and bending moment
struction material, environmental impacts, and other limitations at the base of the tower. Larger design loads, especially the moment,
imposed by local and federal agencies. not only make the foundation design more complex but also make it
The energy output of individual wind turbines can be increased larger, demanding that a significant amount of resources be allotted
by building taller towers to access higher and steadier wind. It has into foundation design and construction to meet safety and service-
been shown that wind energy is directly proportional to the third
ability requirements. Because a significant percentage of the total
power of wind speed, so taller towers can produce significantly
cost of installing a wind turbine is allocated for the design and
construction of the foundation, it is necessary to develop new
1
Graduate Student, Glenn Dept. of Civil Engineering, Clemson methodologies to design and select the most cost-efficient foun-
Univ., 123 Lowry Hall, Clemson, SC 29634. E-mail: shwetas@ dation for a given set of geotechnical and wind conditions.
clemson.edu Typically, the mat (raft) foundation, the pile group foundation,
2
Associate Professor, Glenn Dept. of Civil Engineering, Clemson and the pile-raft foundation are used to support wind turbines,
Univ., 202 Lowry Hall, Clemson, SC 29634 (corresponding author). depending on the subsurface condition, tower height, and wind
E-mail: nravic@clemson.edu speed at the site. The raft foundation is easy to construct and pro-
3
Graduate Student, Glenn Dept. of Civil Engineering, Clemson Univ., vides significant bearing capacity because of its larger footprint,
123 Lowry Hall, Clemson, SC 29634. E-mail: prahbar@clemson.edu
but its design is controlled by differential and total settlements,
Note. This manuscript was submitted on March 15, 2017; approved on
August 18, 2017; published online on November 28, 2017. Discussion pe- especially when subjected to larger loads. In large-load situations,
riod open until April 28, 2018; separate discussions must be submitted for deep foundations are added to the raft foundation to create what is
individual papers. This paper is part of the International Journal of known as a hybrid or pile-raft foundation, which is economical
Geomechanics, © ASCE, ISSN 1532-3641. for supporting tall wind turbines.
Layer Soil Depth (m) Unit weight (kN/m3) cu (kPa) f 0 (degrees) E (kPa) Poisson’s ratio
1 Medium dense sand 0–1.22 17.28 — 50.1a
2.75 10 4a
0.4
2 Soft to firm clay 1.22–9.15 16.50 98.81 0 1.48 104 0.5
3 Cooper marl 9.15–25.00 19.64 106.66 0 3.20 104 0.5
a
Calculated using the CPT data but not used in the design because the bottom of the raft rests on the second layer.
The basic idea behind the use of a pile raft is to increase the bearing moment capacity of a block containing the piles, raft, and soil. The
capacity of the foundation with the use of a raft and to decrease the ultimate moment capacity of the raft, pile group, and block of the
total and differential settlements with the use of deep foundations. pile raft were determined using the method presented in Hemsley
However, the quantification of the exact percentages of total loads (2000). The key equations are summarized in the following subsec-
carried by the raft and by the piles is the most challenging aspect in tions for the sake of completeness.
the design of a pile-raft foundation. This difficulty is mainly due to
a lack of understanding of the complex interaction among the soil, Case 1: Ultimate Moment Capacity of the Pile Raft
raft, and piles and the mobilized strengths along the interface at Considering Individual Capacity
given total and differential settlement values. Thus, a reliable design The ultimate moment capacity of the raft, Mu-R, was calculated
guideline is not yet available, especially for the foundation sub- using Eq. (2) (Hemsley 2000) as follows:
jected to combined moment, lateral, and vertical loads. "
In this study, a preliminary geotechnical design of the pile-raft 1=2 #
Mu-R 27 P P
foundation was performed following the procedure outlined by ¼ 1 (2)
Mm 4 Pu Pu
Hemsley (2000) in which the procedures proposed by Poulos and
Davis (1980) and Randolph (1994) were incorporated. The factors
considered in the preliminary design were the ultimate vertical, where Mm = maximum possible moment that the soil can support;
moment, and lateral geotechnical capacities, total elastic and differ- P = applied vertical load; Pu = ultimate centric load on the raft when
ential settlements, the rotation of the tower due to wind load, and no moment is applied. For this method, the maximum moment for
the lateral movement of the foundation. The size of the raft and the circular raft, Mm is given by
size and number of the piles required to satisfy the design require-
ments were determined in the preliminary design stage. The qu D3 p 1
Mm ¼ (3)
capacity of the pile-raft foundation was checked for vertical load, 4 4 3
lateral load, bending moment, total and differential settlements, and
rotation. A minimum factor of safety of 2 was considered safe for where qu = ultimate bearing capacity of raft; and D = diameter of
the vertical load, lateral load, and bending moment as suggested by circular raft.
Hemsley (2000), and a vertical misalignment of 3 mm/m was con- The ultimate moment of all the piles in the foundation system,
sidered safe for the rotational stability of the tower (Grünberg and Mu-P, was estimated using Eq. (4) (Hemsley 2000) as follows:
Göhlmann 2013). A spreadsheet (not shown) was prepared to auto-
mate the iterative calculations and to perform parametric studies. X
Np
Mu-P ¼ Puui jxi j (4)
i¼1
Design for Vertical Load
The vertical capacity of the pile raft was calculated as the lesser of where Puui = ultimate uplift capacity of ith pile; jxi j = absolute dis-
the (1) sum of the ultimate capacities of the raft, Pu-R, and all the tance of ith pile from the center of group; and Np = number of piles.
piles, Pu-P, such that Pu-PR = Pu-R þ Pu-P; and (2) the ultimate The ultimate moment capacity of the pile raft, Mu-PR, system
capacity of a block, Pu-B, that consists of the piles and the raft plus considering the individual capacity is given by
that of the portion of the raft outside the periphery of the pile group
(Hemsley 2000). The ultimate capacity of raft, Pu-R, was calculated Mu-PR ¼ Mu-R þ Mu-P (5)
using the general bearing capacity equation by Vesic (1973, 1975),
and the ultimate capacity of all the piles, Pu-P, was calculated as the
sum of the ultimate downward capacity of all the piles; that is, Case 2: Ultimate Moment Capacity of the Pile Raft
Pu-P = NpPult-dn, where Np is the number of piles and Pult-dn is the Considered as a Single Block
ultimate downward capacity of a single pile, which is the sum of The ultimate moment capacity of the block, a single unit consisting
the ultimate skin resistance, Ps, and toe resistance, Pt. In this study, of the raft and the piles, MuB was estimated using Eq. (6), (Hemsley
the a and Meyerhof’s methods, provided in Das (2011), were used 2000), as follows:
to calculate the ultimate skin and toe resistance of a single pile,
respectively. Finally, the factor of safety for the vertical load was Mu-B ¼ aB
p u BB D2B (6)
calculated using Eq. (1), as follows:
where BB and DB = width and depth of the block, respectively; pu =
minðPu-PR ; Pu-B Þ average lateral resistance of soil along the block; and aB = factor
FSP ¼ (1)
P depending upon the distribution of the ultimate lateral pressure with
depth (0.25 for a constant distribution of
p u and 0.2 for a linearly
where P = design vertical load. increasing
p u with depth from zero at the surface). It is noteworthy
is fully utilized
160
where Kr = stiffness of the raft; Kp = stiffness of the pile group; and
arp = raft-pile interaction factor. The raft-pile interaction factor was 120
assumed to be 0.8 because as the number of piles in the group 80
increases, the interaction factor increases and tends toward a con- Raft and piles Pile capacity is fully mobilized
40 are functioning (only raft is working)
stant value of 0.8 as reported by Randolph (1994). Among the vari-
ous methods for estimating the raft stiffness, the method outlined by 0
Randolph (1994) was used. To estimate the stiffness of the pile 0 100 200 300 400
Settlement (mm)
group, the method proposed by Poulos (2001b) was used. First, the
target stiffness of the pile raft was determined by dividing the total
Fig. 1. Calculated load versus total elastic settlement curve for a pile-
vertical load by the assumed allowable settlement. Then, Eq. (9)
raft foundation
was then solved to determine the stiffness of the pile group with the
Simulations Lp (m) Np Rr (m) Standard deviation of response (mm) Total cost (US$)
1,000 29.87 52 7.93 5.57 369,742
10,000 31.25 52 7.91 5.09 392,929
M
MRa Mpile
Downloaded from ascelibrary.org by National Taiwan University of Sci and Tech on 01/29/18. Copyright ASCE. For personal use only; all rights reserved.
Rotaon θRa
Fig. 2. Conceptual differential settlement calculation diagram: (a) pile raft; (b) rotation of raft; (c) rotation of piles
ratio, eo, was calculated using the correlation between Cc and eo pro-
posed by Hough (1957). These consolidation parameters are listed
in Table 2. Finally, the consolidation settlement, Sc, of the second
and the third layer was calculated to be 37.62 and 16.72 mm, respec-
tively, resulting in a total consolidation settlement of 54.34 mm.
The total settlement (elastic plus differential settlements) was
calculated to be 96.76 mm, which is within the tolerable limit of
100 mm for tall structures (Raju 2015). Nevertheless, it is notewor-
thy that the uniform vertical settlement of the entire system may not
be hazardous, especially for wind turbines located away from criti-
cal infrastructure.
where MRaft = fixed-end moment at the soil-structure interface (the was used to determine the differential settlement of the raft, assum-
percentage of moment shared by the raft to yield an equal differen- ing that the raft rotates about its center line.
tial settlement of the piles in this study); cs = foundation modulus;
IRaft = second moment of inertia; Es = modulus of elasticity of soil; Differential Settlement of Piles
f 0 = shape factor for overturning (0.25); and ARaft = area of the foun- The differential settlement of the pile group was estimated on the basis
dation. After calculating u Raft, a simple trigonometric relationship of the individual pile-settlement profile due to the resultant vertical
undrained cohesion were also studied. The total cost of the founda-
differential settlements of the raft and the piles were equal. The corre- tion includes material, labor, and equipment costs, and it was calcu-
sponding final values were considered as the moment carried by the lated as a sum of the total costs of the raft and piles using the unit
piles and the raft, with the final differential settlement deemed to be cost of each. The unit cost for the raft and for the prestressed con-
the differential settlement of the pile-raft foundation. This exercise crete pile used in this study was $342.19/m3 and $192.19/m, respec-
resulted in a differential settlement of 44.23 mm and a rotation of tively, as obtained from RS Means data (Phillip and Adrian 2013).
0.17°, a rotation that induced the horizontal displacement of The details of the parametric study and the results are presented the
383.7 mm at the top of the tower, which is within the acceptable limit. following sections.
Fig. 4. Effect of variation in wind speed on (a) number of piles and (b) total cost of pile raft
Fig. 5. Effect of variation in wind speed on (a) length of pile and (b) total cost of pile raft
tion for different numbers and lengths of piles revealed that for the Although the parametric study results presented in the preceding
higher wind speeds, the design requirement is met more economi- section shows the effect of variations in soil properties and wind
cally by increasing either Np or Lp rather than by increasing Rr. speed on the design outcomes, they do not consider the change in
more than one variable at the same time. Moreover, they do not give
Effect of Undrained Cohesion on the Design Variables a clear indication on how to select the most cost-efficient and robust
design for a given performance criterion. In such a situation, a
Considering the medium site variability, a coefficient of variation reliability-based robust design optimization can be used to develop
(COV) of 25% was assumed to determine the variation in undrained
a criterion to select the most suitable design for the given perform-
cohesion, cu, for the second and third layers of soil. For the second
ance and cost criteria. One such framework is presented in this
layer, the standard deviation of the undrained cohesion (cu), s cu,
paper.
was calculated as 24.70 kPa using 25% COV and the mean of the
In reliability methods, risk or reliability is calculated for a given
cu, m cu, was 98.81 kPa. Similarly, the standard deviation of the third
performance criterion or a performance function. The computational
layer was calculated as 26.66 kPa using the same COV as for the
second layer and a mean cu of 106.66 kPa. The parametric study
was conducted by varying cu by 62s cu above and below the mean
value for both layers. The variation of cu used in this parametric
study is also shown in Fig. 7.
As shown in Fig. 7, for each case in the parametric study, the cu
values of the second and third layer were changed simultaneously
while the wind speed was kept constant. Although only the varia-
tions up to 25 m in depth are shown in Fig. 7, the cu for a depth
greater than that was assumed as the same for the parametric study.
The Np, Lp, and Rr required to meet all the design requirements at
five levels undrained cohesion and the corresponding total costs of
the foundation are presented in Figs. 8, 9, and 10, respectively. The
results indicate that Np, Lp, and Rr decreased with an increase in cu
along with the total cost of the foundation. For the lowest cu, in
Fig. 8(a), piles were arranged in three circumferences to meet all the
design requirements without facing a group effect. In Fig. 10(a), it
can be seen that Rr remains the same even with an increase in cu
because it is the minimum radius requirement according to the bot-
tom diameter of the tower. Similar to the results for the variations in
wind speed, adjusting Np or Lp, but not Rr, is the most economical Fig. 7. Soil profile showing variations in undrained cohesion
method for meeting all of the design requirements.
Fig. 6. Effect of variation in wind speed on (a) radius of raft and (b) total cost of pile raft
Fig. 8. Effect of variation in undrained cohesion on (a) number of piles and (b) total cost of pile raft
Fig. 9. Effect of variation in undrained cohesion on (a) length of pile and (b) total cost of pile raft
Fig. 10. Effect of variation in undrained cohesion on (a) radius of raft and (b) total cost of pile raft
480,000 480,000
Total cost ($)
440,000 440,000
400,000 400,000
360,000 360,000
320,000 320,000
3 4 5 6 7 8 3 4 5 6 7 8
Std. dev. of differential settlement (mm) Std. dev. of differential settlement (mm)
(a) (b)
Fig. 12. Pareto front optimized to both total cost and standard deviation: (a) 1,000 simulations; (b) 10,000 simulations
Std. dev. of differential settlement (mm) Sdiff ¼ exp f11:53 þ 3:55lnðV Þ 0:13ln½cuð2Þ 1:05ln½cuð3Þ
*Pareto front used is for 10,000 simulations.
2:07lnðLp Þ 3:02lnðNp Þ 1:10lnðRr Þg (12)
Fig. 13. Application of Pareto front for design selection
System Random variables Robustness measure Safety constraint Optimization method Reference
Drilled shafts Soil properties, construction variations, Weighted sensitivity Target reliability Simplified RBRGD in Khoshnevisan
in clay loading index of response index spreadsheet et al. (2016)
Monopile Undrained shear strength, friction angle, lat- Total cost Failure probability RBDO by coupling SS Overgård et al.
foundation eral load method with SA stochastic (2016)
optimization algorithm.
Drilled shaft Friction angle, coefficient of lateral earth Variation in failure Failure probability RGD with NSGA-II Juang et al.
pressure probability, feasibility (2013)
robustness
Shallow Undrained shear strength, loads (moment, Volume of concrete Target probability d-RBD with MCS Ben-Hassine
foundation vertical, horizontal) of failure and Griffiths
(2012)
Shallow Undrained shear strength, effective friction Variation in failure Failure probability RGD with NSGA-II Juang and Wang
foundation angle, COV compressibility, vertical central probability, feasibility (2013)
load robustness
Shallow Geotechnical parameters (unit weight, fric- SI based on gradient Safety margin (dif- RGD with NSGA-II Gong et al.
foundation tion angle, Young’s modulus, Poisson’s ra- of system response to ference between (2014)
tio), loading parameters (dead load, live noise factors resistance and
load), construction tolerance (width, length, load)
depth), model error (ULS and SLS solution)
Spread Unit weight, effective friction angle, opera- Construction cost ULS and SLS RBD Wang (2009)
foundation tive horizontal stress coefficient requirement
Note: d-RBD = direct reliability-based design; MCS = Monte Carlo simulation; RBDO = reliability-based design optimization; RGD = robust geotechnical
design; SA = simulated annealing; SLS = serviceability limit state; SS = subset simulation; ULS = ultimate limit state.
from 7.0 to 3.7 mm, increased the total foundation cost from in Engineering and Scientific Applications, E. Dávila, G. Uzcátegui,
$360,000 to $460,000, which is a $10,000 difference between the and M. Cerrolaza, eds., Venezuelan Society of Numerical Methods in
two amounts. Engineering, Caracas, Venezuela, 17–23.
Coduto, D. P. (2001). Foundation design: Principles and practices, 2nd
Design Selection Ed., Prentice Hall, Upper Saddle River, NJ.
Das, B. M. (2011). Principles of foundation engineering, 7th Ed., Cengage
The resulting Pareto front can be judged by designers, and the final Learning, Stamford, CT.
design can be selected on the basis of the performance requirements Deb, K., Pratap, A., Agarwal, S., and Meyarivan, T. (2002). “A fast and elit-
and available funds for the construction of the foundation. The dif- ist multi objective genetic algorithm: NSGA-II.” IEEE Trans. Evol.
ferent possible optimum design outcomes that can be extracted Comput., 6(2), 182–197.
from the Pareto front are presented in Fig. 13. In this section, the Gong, W., Khoshnevisan, S., and Juang, C. H. (2014). “Gradient-based
Pareto front for 10,000 simulations is used for demonstration. design robustness measure for robust geotechnical design.” Can.
Generally, to compare the concept of a Pareto front with a conven- Geotech. J., 51(11), 1331–1342.
Grünberg, J., and Göhlmann, J. (2013). Concrete structures for wind tur-
tional design, it should be noted that the least costly design, which
bines, K. Bergmeister, F. Fingerloos, and J.-D. Wörner, eds., Ernst &
is the most sensitive design on a Pareto front (marked as B in Sohn, Berlin.
Fig. 13), is usually considered as the final design in conventional Gudmundsdottir, B. (1981). “Laterally loaded piles.” M.S. thesis, Univ. of
practices where uncertainties are not involved. Similarly, the design Alberta, Edmonton, AB, Canada.
that is the least sensitive but the most expensive of all on the Pareto GWEC (Global Wind Energy Council) (2015). “Global wind report:
front (marked as A in Fig. 13) can also be obtained if the client Annual market update 2015.” Brussels.
desires to have the most robust design. Nevertheless, the most opti- Hemsley, J. A. (2000). Design applications of raft foundations, Thomas
mal design that meets the given performance and cost requirements Telford Ltd., London,.
Hough, B. K. (1957). Basic soils engineering, Ronald Press, New York.
can be obtained from the Pareto front using the knee point concept.
Juang, C. H., and Wang, L. (2013). “Reliability-based robust geotechnical
In this study, the normal boundary intersection (NBI) approach, design of spread foundations using multi-objective genetic algorithm.”
shown in Fig. 14, was used to determine the knee point in the result- Comput. Geotech., 48(Mar), 96–106.
ing Pareto front. In this method, for each point on the Pareto front, Juang, C. H., Wang, L., Liu, Z., Ravichandran, N., Huang, H., and Zhang, J.
the distance from the boundary line, which connects two extreme (2013). “Robust geotechnical design of drilled shafts in sand: New
upper and lower points of the Pareto front, was computed in the nor- design perspective.” J. Geotech. Geoenviron. Eng., 10.1061/(ASCE)GT
malized space of the Pareto front. The knee point on the Pareto .1943-5606.0000956, 2007–2019.
front, which has the greatest distance from the boundary line, was Khoshnevisan, S., Wang, L., and Juang, C. H. (2016). “Simplified proce-
dure for reliability-based robust geotechnical design of drilled shafts in
then determined. The knee point determined using the NBI
clay using spreadsheet.” Georisk Assess. Manage. Risk Eng. Syst.
approach for the Pareto front with 10,000 simulations is also shown Geohazards, 10(2), 121–134.
in Fig. 13 as point C. Kulhawy, F. H., and Mayne, P. W. (1990). “Manual on estimating soil prop-
The optimum designs for 1,000 and 10,000 simulations obtained erties for foundation design.” Rep. EL-6800, Project 1493-6, Cornell
via the NBI approach are given in Table 4. It is observed that the op- Univ., Ithaca, NY.
timum design for both numbers of simulations has a similar value Lewin, T. J. (2010). “An investigation of design alternatives for 328-ft
for both the cost and the standard deviation of the response. (100-m) tall wind turbine towers.” M.S. thesis, Iowa State Univ.,
Ames, IA.
Malhotra, S. (2011). “Chapter 10: Selection, design and construction of off-
Conclusion shore wind turbine foundations.” Wind turbines, I. Al-Bahadly, ed.,
InTech, London.
A reliability-based robust design optimization of the pile-raft foun- Mayne, P. W., and Kemper, J. B. (1988). “Profiling OCR in stiff clays by
CPT and SPT.” Geotech. Test. J., 11(2), 139–147.
dation for a tall wind turbine on clayey soil was presented in this pa-
Overgård, I. E. V., Depina, I., and Eiksund, G. (2016). “Reliability-
per. According to the deterministic geotechnical design outcomes, based design of a monopile foundation for offshore wind turbines
it was found that the final design was controlled by a differential set- based on CPT data.” Proc., 17th Nordic Geotechnical Meeting:
tlement and rotation. The use of the pile raft, which takes advantage Challenges in Nordic Geotechnics, Icelandic Geotechnical Society,
of both the raft and piles to control the bearing capacity and settle- Reykjavik, Iceland, 495–502.
ment, respectively, was found to be the best option for meeting the Phillip, W., and Adrian, C. (2013). RSMeans building construction cost
design requirements. The results of the parametric study showed data, 71 Ed., The Gordian Group, Rockland, MA.
that the design requirements can be met by increasing either the Poulos, H. G., and Davis, E. H. (1980). Pile foundation analysis and design,
T. W. Lambe and R. V. Whitman, eds., Wiley, New York.
number of piles, length of piles, or radius of the raft at high wind
Poulos, H. G. (2001a). “Methods of analysis of piled raft foundations.”
speeds. The results of the Pareto front created from the design opti- Technical Committee TC-18 on Piled Foundations, International
mization results showed a clear trade-off relationship between the Society of Soil Mechanics and Geotechnical Engineering.
total cost of the foundation and the standard deviation of the Poulos, H. G. (2001b). “Piled raft foundation: Design and applications.”
response (differential settlement). Such a relationship is useful for Geotechnique, 51(2), 95–113.