You are on page 1of 13

Geotextiles and Geomembranes xxx (xxxx) xxx

Contents lists available at ScienceDirect

Geotextiles and Geomembranes


journal homepage: www.elsevier.com/locate/geotexmem

Deterministic and probabilistic assessment of margins of safety for internal


stability of as-built PET strap reinforced soil walls
Richard J. Bathurst a, *, Yoshihisa Miyata b, Tony M. Allen c
a
Department of Civil Engineering, GeoEngineering Center at Queen’s-RMC, Royal Military College of Canada, Kingston, ON, K7K 7B4, Canada
b
Department of Civil and Environmental Engineering, National Defense Academy, 1-10-20 Hashirimizu, Yokosuka, 239-8686, Japan
c
State Geotechnical Engineer, Washington State Department of Transportation, State Materials Laboratory, P.O. Box 47365, Olympia, WA, 98504-736, USA

A R T I C L E I N F O A B S T R A C T

Keywords: The paper demonstrates deterministic and reliability-based assessment of strength limit states (tensile resistance
Mechanically stabilized earth (MSE) walls and pullout) and the service limit state for soil failure for mechanically stabilized earth (MSE) walls constructed
Polyester (PET) strap reinforcement with polyester (PET) strap reinforcement. The general approach considers the accuracy of the load and resistance
Tensile strength
models that appear in each limit state equation plus uncertainty in the estimate of nominal load and resistance
Pullout strength
Soil failure (service) limit state
values at time of design. Reliability index is computed using a closed-form solution that is easily implemented in
Reliability-based design a spreadsheet. Three PET strap MSE wall case studies are used to demonstrate the reliability-based assessment
approach and to compare margins of safety using different load and resistance model combinations. In some
walls using the Coherent Gravity Method to compute loads, the recommended nominal factors of safety for
tensile strength and pullout limit states were not satisfied. However, reliability analyses showed that the walls
satisfy recommended minimum target reliability index values for the limit states investigated, usually by large
amounts. The most critical limit state is the soil failure limit state which is used in the Simplified Stiffness Method
to keep the reinforced soil zone at working stress conditions assumed for geosynthetic MSE walls under oper­
ational conditions.

1. Introduction recommended for extensible geosynthetic reinforced MSE wall systems.


Only the Simplified Stiffness Method (Allen and Bathurst, 2015, 2018)
Mechanically stabilized earth (MSE) walls constructed with polyester provides a seamless transition between inextensible and extensible
(PET) strap soil reinforcement are becoming more common to perform reinforcement classifications for the internal stability design of MSE
the earth retaining wall function. Nevertheless, at the time of this paper walls. Using the global reinforcement stiffness value that appears in the
only the French code (AFNOR, 2009) and the recent AASHTO (2020) Simplified Stiffness Method, Miyata et al. (2018) concluded that PET
specifications in the USA offer guidance on the design of PET strap walls. strap walls fall into the extensible (geosynthetic) wall category along
Current design codes in Canada (CSA, 2019), the UK (BSI, 2010), Japan with geogrids and geotextiles (or at least at the stiffer end of these walls)
(PWRC, 2013) and Hong Kong (Geoguide 6, 2002) remain silent. rather than MSE walls with relatively inextensible (steel) reinforcement.
Miyata et al. (2018) collected reinforcement load measurements Miyata et al. (2018) assessed the accuracy of load predictions using
from eight instrumented PET strap walls at end of construction. The the Coherent Gravity Method, the Simplified Method and the Simplified
reinforcement was identified as relatively inextensible in some case Stiffness Method by comparing predicted loads with those deduced from
study sources, and thus treated similar to steel reinforcement products, reinforcement strain measurements from the eight wall cases in their
while in the other cases the PET strap reinforcement was classified as an study. The Coherent Gravity Method and the Simplified Method resulted
extensible geosynthetic type. This distinction is important for the in­ in conservative (safe) estimates of reinforcement loads under opera­
ternal stability design of MSE walls because the Coherent Gravity tional conditions on average. However, the Simplified Stiffness Method
Method (e.g., AASHTO, 2020; BSI, 2010; CSA, 2019) and variants are was the most accurate and did not result in an excessive level of
recommended in design codes for relatively inextensible reinforcement, conservativeness.
while the AASHTO (2017) Simplified Method and variants are Miyata et al. (2019) carried out a complementary investigation of the

* Corresponding author.
E-mail addresses: bathurst-r@rmc.ca (R.J. Bathurst), miyamiya@nda.ac.jp (Y. Miyata), allent@wsdot.wa.gov (T.M. Allen).

https://doi.org/10.1016/j.geotexmem.2020.06.001
Received 29 February 2020; Received in revised form 26 May 2020; Accepted 8 June 2020
0266-1144/© 2020 Elsevier Ltd. All rights reserved.

Please cite this article as: Richard J. Bathurst, Geotextiles and Geomembranes, https://doi.org/10.1016/j.geotexmem.2020.06.001
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

accuracy of different pullout models for the ultimate (failure) pullout nominal factor of safety and the reliability index (or probability of
capacity for PET strap reinforcement. The models were taken from the failure) that are computed using different load and pullout model
research literature and default models specified in design codes for steel combinations. The design limit state for the connection between the
and geosynthetic reinforcement materials. The accuracy of these models reinforcement straps and the concrete facing panels is not considered in
and new models developed by the authors was investigated by this investigation.
comparing pullout capacity predictions with measured values in the The assessment of margins of safety in terms of probability that an
database of pullout tests compiled by the authors from many sources. internal limit state is not satisfied (i.e., failure) provides the designer
The accuracy of the reinforcement load and pullout models in the with a more nuanced appreciation of margins of safety during design
two studies described above was determined from analysis of bias sta­ than can be gained from allowable (working) stress (factor of safety)
tistics where bias is the ratio of measured (observed) value to predicted design used in past practice and current load and resistance factor design
(calculated) value. The same collections of bias values can be used to (LRFD).
carry out reliability-based (probabilistic) assessment of internal stability Reliability analyses of the case study walls in this study are based on
tensile and pullout strength limit states and the service limit state for soil the problem geometry and parameters shown in Fig. 1. The details of
failure of PET strap MSE walls at the time of design or for as-built these figures are described later in the paper. The simple geometry and
structures. reinforcement arrangement for the example walls have the advantage of
This paper demonstrates for the first time a rigorous reliability a vertical face, horizontal back-slope and a single reinforcement type.
assessment approach for the limit states identified above using three The current study is timely because the latest edition of the AASHTO
examples of as-built PET strap walls taken from the study of Miyata et al. (LRFD) Bridge Design Specifications in the USA (AASHTO, 2020),
(2018) and the collections of load and resistance bias values mentioned adopts the Simplified Stiffness Method as the primary method for in­
above. The assessment of margin of safety is based on the deterministic ternal stability design of geosynthetic MSE walls, including those

Fig. 1. Geometry and parameter definitions for tensile strength, pullout and soil failure limit states.

2
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

constructed with PET strap reinforcement. In the new AASHTO speci­ summarized in Table 1. The CSA (2019) specifications in Canada follow
fications, the Simplified Stiffness Method is called the “Stiffness Method” the AASHTO (2017, 2020) recommendations for the Coherent Gravity
for brevity and the polymeric straps are called “geostrips”. Method.

2. Load models 2.3. Simplified Method

2.1. General The Simplified Method (AASHTO, 2017; CSA, 2019) and a similar
approach in Japan (PWRC, 2013) are used for (extensible) geotextiles
The three load models introduced in the previous section are briefly and geogrids (Fig. 1b). In these formulations, eccentricity is not
reviewed here for the case of frictional (cohesionless) soils only. All considered, hence:
three load models are empirically based using loads estimated from
Tmax = Sv Kr σ v (2)
reinforcement strain measurements taken from instrumented structures
under operational conditions. The Coherent Gravity and Simplified The coefficient term is computed as Kr = Ka. The soil friction angle is
Methods estimate loads considering soil strength and reinforcement capped at ϕ = 40◦ in the North American code, but not in the Japanese
type, whereas the Simplified Stiffness Method is largely based on the code. In this study we use the capped friction angle approach and any
reinforcement stiffness to compute reinforcement loads. Table 1 gives a cohesive shear strength component is ignored in accordance with North
summary of the load models and variants used in this study and the American practice (i.e., c = 0).
treatment of soil strength components and vertical pressure eccentricity
adopted in this study to match design code recommendations. 2.4. Simplified Stiffness Method

2.2. Coherent Gravity Method The Simplified Stiffness Method is applicable to both extensible and
inextensible reinforcement products (Allen and Bathurst, 2015, 2018).
The Coherent Gravity Method (Fig. 1a) was originally developed for The earliest variant of the method was the “K-stiffness Method” (Allen
relatively inextensible steel strip soil reinforcement (e.g., Schlosser, et al., 2003, 2004; Bathurst et al., 2005; Miyata and Bathurst, 2007a)
1978; Schlosser and Segrestin, 1979). The Coherent Gravity Method can and was developed for MSE walls constructed with frictional backfill
be described by general equations of the following form: soils. The current method now includes cohesive-frictional (c− ϕ)
[ ] backfill soils but the general approach remains unchanged. As the name
Tmax = Sv Kr σ v
L
(1) of the method implies, a key parameter in the method is the stiffness of
L − 2e the reinforcement that has proven to greatly influence the magnitude of
reinforcement loads observed in field walls.
where, Sv = tributary area of each reinforcement layer, Kr = coefficient The calculation of the maximum tensile load (Tmax) in a reinforce­
of earth pressure, σv = vertical stress acting at the elevation of the ment layer for the simple geometry and conditions described earlier is a
reinforcement layer including any uniformly distributed surcharge function of the parameters shown below:
pressure (q), L = reinforcement length. Parameter e = vertical pressure ( )
eccentricity due to the retained soil acting against the reinforced soil Tmax = f H, Φfs , Sv , z, ϕ, c, σ v , J (3)
zone.
In AASHTO (2017, 2020), BSI (2010) and PWRC (2014) guidelines, where, Φfs is the non-dimensional facing stiffness factor and J is the
Kr varies linearly with depth below the top of the internal soil wedge reinforcement secant creep stiffness based on 1% strain and 1000 h as
from K0 = 1 − sin ϕ to the active earth pressure value Ka = (1 – sin ϕ)/(1 recommended by Miyata et al. (2018) for PET straps. All other param­
+ sin ϕ) at a depth of 6 m, and remains constant thereafter; here, ϕ is the eters have been defined earlier. Eccentricity is not considered. The
peak friction angle of the soil. In Japan, eccentricity is ignored (PWRC, reader is directed to the papers by Allen and Bathurst (2015, 2018) for a
2014). In this paper we use the AASHTO (2017, 2020) and BSI (2010) detailed description of Equation (3). The relationship between secant
interpretations for the Coherent Gravity Method in which eccentricity is creep stiffness and ultimate strength of geosynthetic reinforcement
considered (Equation (1)). In AASHTO/PWRC specifications, any soil materials, including polyester strap reinforcement, is explained in the
cohesion is ignored while in the BSI code cohesion is considered, pro­ paper by Allen and Bathurst (2019).
vided it is persistent and dependable over the life of the structure. The At the time the Simplified Stiffness Method was developed there was
treatment of peak friction angle (unrestricted or capped at 40◦ ), cohe­ only one instrumented PET strap wall available in the literature
sion (available or ignored), and eccentricity (considered or ignored) (Schlosser et al., 1993). The load predictions for this wall using the
using the AASHTO/CSA and BSI Coherent Gravity Methods are Simplified Stiffness Method (i.e., K-stiffness Method) were nevertheless
judged to be reasonable. Miyata et al. (2018) collected tensile load
Table 1 measurement from more recent PET strap walls and determined that the
Load models and calculation details. reinforcement used in the first instrumented wall was very much stiffer
than the products used today. They recommended the use of 1% strain
Load model Calculation details
for the calculation of the secant creep stiffness to improve load pre­
ϕ ≤ 40◦ Cohesion Vertical pressure dictions for some walls rather than 2% strain at 1000-hours recom­
(capped) considered? eccentricity considered?
Yes/No Yes/No
mended in the Simplified Stiffness Method (Allen and Bathurst, 2015).
Nevertheless, the improvement was modest and for most case studies the
Coherent Gravity Yes No Yes
difference in stiffness at these two strain levels is not detectable. Both
AASHTO (2020) Equation 1
Coherent Gravity No Yes Yes stiffness values are permitted in AASHTO (2020) for PET straps.
BSI (2010) Equation 1
AASHTO Yes No No 3. Pullout models
Simplified Equation 2
AASHTO (2017)
Simplified Yes Yes No
The general form of the PET strap pullout models in this study is:
Stiffness Equation 2
Allen and
Pc = 2f*σv Le Rc (4)
Bathurst (2015)
Here, Pc = peak pullout capacity (e.g., units of kN/m), f* =

3
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

dimensionless interaction coefficient (the equivalent symbol F* is used 3.2. Bi-linear pullout models
in US design practice), σ v = vertical stress acting at the elevation of the
reinforcement layer in the resistant zone, Le = anchorage length of the Models in this category are described by f* that is a bi-linear function
reinforcement element, and Rc = coverage ratio = bw/Sh where bw = of depth as follows:
width of the reinforcement element and Sh = horizontal centre-to-centre
f * = f0 (1 − z/z0 ) + f1 (z/z0 )for z ≤ z0 = 6 m (5a)
spacing between elements perpendicular to the strap length. In some
cases the straps are arranged in closely spaced parallel pairs in which
f* = f1 for z > z0 = 6 m (5b)
case the width bw is taken as the distance between outside edges. More
frequently the reinforcement is placed in a splayed configuration start­ Equations of this type appear in North American specifications and
ing at the connection with the facing panels. In this study we assume that guidelines (AASHTO, 2020; FHWA, 2009; CSA, 2019), in the Japanese
the reinforcement lengths are long enough that each individual strap code (PWRC, 2014) for steel strips, and in the French code (AFNOR,
acts independently over the anchorage length. Miyata et al. (2019) 2009) for “geosynthetic strips”. Coefficients f0 and f1 vary linearly with
examined the utility and accuracy of different pullout capacity equa­ tan ϕ (Table 2b).
tions according to the classifications in the next section. In this study all
straps are placed flat. Hence, three-dimensional (3D) U-shaped far-end
configurations investigated by Razzazan et al. (2019) are not 3.3. Non-linear pullout models
considered.
In the pullout capacity calculations performed later in the paper, the The non-linear pullout equations considered in this study have been
anchorage length (Le) is computed using the geometry shown in Fig. 1a proposed by Miyata and Bathurst (2012) and Miyata et al. (2019) and
assuming a bi-linear failure geometry (Coherent Gravity Method), and have the following exponential form:
Fig. 1b assuming a linear failure geometry (Simplified and Simplified f0 − f1
Stiffness Method). The former is associated with inextensible rein­ f * = f1 + ( ) (6)
forcement and the latter with extensible reinforcement (e.g., AASHTO, exp ζ σpav
2020).
Parameter ζ is a dimensionless coefficient and pa = 101 kPa (atmo­
3.1. Linear pullout models spheric pressure) is used to make the vertical stress term dimensionless;
for σv → ∞ (i.e., increasing z), f* → f1. This equation has the same
Linear pullout models are described by a single value of f* that is number of fixed terms as Equation (5a), but has the advantage of being
constant with depth z and is proportional to tan ϕ (Table 2a). The linear smoothly continuous. The friction angle of the soil does not appear in
model with f* = 0.67 tan ϕ is the default value in AASHTO (2020) and this model formulation because the soil strength parameters for the
CSA (2019) design codes for pullout of geogrids and geotextiles, and cohesionless granular soils in the database of pullout box tests available
single ribbed steel strips placed at z > 6 m, and in AASHTO (2020), CSA to perform the model calibration fell within a narrow range. Values of
(2019) and the French code (AFNOR, 2009) for single smooth strips coefficient terms are summarized in Table 2c.
regardless of depth. Miyata et al. (2019) adopted the same formulation
for single PET straps for the pullout limit state. However, better agree­ 4. Tensile strength
ment between measured pullout load and predicted capacity was ach­
ieved for a pair of straps using f* = tan ϕ. The two other models that In this investigation the ultimate tensile strength (rupture) capacity
follow below did not require a different formulation for f* depending on of the reinforcement is taken as the nominal long-term tensile strength
the single or double strap pullout configurations used in wall (AASHTO, 2020; CSA, 2019) computed as:
construction. Tult Tult
Tal = = (7)
RF RFID × RFCR × RFD
Table 2
Pullout models (Miyata et al., 2019). The numerator is a reference laboratory ultimate tensile strength
(Tult), typically a minimum average roll value (MARV). This reference
(a) Linear model
tensile strength is reduced by factors that account for loss of strength
Wall Strap configuration Model designation Model coefficients over the design life of the reinforcement due to installation damage
f* (RFID), creep (RFCR) and degradation (durability) mechanisms (RFD).
Delaware single L-1 0.67 tan ϕ Parameter RF is the combined reduction factor. A detailed explanation
Nagasaki double L-4 tan ϕ of how Tult is determined in North American practice and the relation­
Sao Paulo single L-1 0.67 tan ϕ ship between Tult and secant creep stiffness for different geosynthetic
(b) Bi-linear model product types can be found in the paper by Allen and Bathurst (2019). In
this study, RF = RFID × RFCR × RFD = 1.06 × 1.36 × 1.30 ≈ 1.85 based
Wall Strap configuration Model designation Model coefficients
(Equations (5a) and (5b)) on product data for the PET straps in this investigation and North
American practice AASHTO (2010, 2015).
f0 f1

Delaware single BL-2 1.2 tan ϕ 0.8 tan ϕ


5. Soil failure (Simplified Stiffness Method)
Nagasaki double BL-4 1.2 tan ϕ 0.8 tan ϕ
Sao Paulo single BL-2 1.2 tan ϕ 0.8 tan ϕ
A key feature of the Simplified Stiffness Method is the soil failure
(c) Non-linear model
limit state which applies to MSE walls constructed with extensible
Wall Strap configuration Model designation Model coefficients geosynthetic reinforcement materials. Satisfying this limit state ensures
(Equation (6))
that the strain in each reinforcement layer under operational conditions
f0 f1 ζ does not exceed a value that will cause a contiguous zone of shear stress
Delaware single NL-1 1.5 0.5 2.3 to develop in the reinforced soil zone that is equal to the soil peak
Nagasaki double NL-3 1.5 0.5 2.3 strength. This constraint is required if working stress conditions
Sao Paulo single NL-1 1.5 0.5 2.3 consistent with the assumption of operational conditions in the
Simplified Stiffness Method apply. In this paper the corresponding

4
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

maximum nominal tensile resistance TJ is calculated as: linked quantitatively to the classical notion of factor of safety as shown
next.
TJ ​ = ​ J × εmax (8)
The parameters introduced thus far are assumed to be random var­
where, εmax is the maximum tensile strain in the reinforcement to satisfy iables described by smoothly continuous frequency distributions that
the soil failure limit state. Allen and Bathurst (2018) also recommended can be used in Monte Carlo simulations to compute the probability of
that the maximum strain in any layer should not exceed 2.5% and 3.0% failure Pf = P(g < 0). The objective of reliability-based design is to ensure
for stiff and flexible faced walls, respectively, in any individual layer, that Pf is less than a maximum acceptable value, or equivalently, the
and the average maximum strain from all layers should not exceed εmax reliability index β is greater than an acceptable target value. Fortu­
= 2% and 2.5% for stiff and flexible faced walls, respectively. In the nately, distributions for the bias values (λR and λQ) for the PET strap
current AASHTO (2020) code, these average strain limits have been reinforcement load and pullout models described in the previous section
removed and the individual layer strain criteria reduced to 2% and 2.5% can be computed from the data reported by Miyata et al. (2018, 2019).
for stiff and flexible faced walls, respectively, to simplify the code. These Bias values for the estimation of the secant creep stiffness at 2% strain
changes are on the safe side for design. and 1000 h are available from Allen and Bathurst (2019). In these prior
works, bias values were shown to be lognormally distributed for prac­
6. General approach tical purposes. If uncertainties in the estimates of nominal load and
resistance terms (Rn or Qn) are also assumed to follow a lognormal dis­
The general form of the limit state design equations in this investi­ tribution at the time of design, then the following closed-form solution is
gation is: available to compute reliability index β (Bathurst and Javankhoshdel,
2017):
Rm λR Rn
g​ = ​ − ​1​ = ​ − ​1 (9) β ​ = ​ A × ln(Fn ) ​ + ​ B (11)
Qm λQ Qn

Here Rm and Qm are the measured (observed) resistance and where, the ratio of the mean estimate of nominal resistance (μRn) and
measured (observed) load, respectively. Parameters λR and λQ are mean estimate of nominal load (μQn) for a layer is taken as the factor of
resistance and load method bias (or bias values for brevity), respectively, safety for that layer, hence:
and are used to transform nominal values (Rn and Qn) to corresponding
μRn Rn
measured values, hence: Fn = = (12)
μQn Qn
λR = Rm/Rn (10a)
In US LRFD practice (AASHTO, 2020) for PET strap walls and
λQ = Qm/Qn (10b) considering only dead load due to soil self-weight, the nominal factor of
safety for the tensile strength limit state can be taken as Fn ≥ γQ/φ =
Bias values can be understood to account for over- or under- 1.35/0.55 ≈ 2.5 where γ Q = load factor and φ = resistance factor. For the
estimation of observed load and resistance values when nominal load pullout limit state Fn ≥ γ Q/φ = 1.35/0.70 ≈ 1.9, and for the soil failure
and resistance models are used in a limit state design equation. The ratio limit state, Fn ≥ γQ/φ = 1.20/1.00 = 1.2.
of each pair of measured (observed) resistance and load values is the The remaining terms in Equation (11) are collections of statistical
operational factor of safety (OFS = Rm/Qm). This value is most often characteristics of bias and nominal values, specifically:
different from the nominal factor of safety Fn = Rn/Qn that is familiar to

1
A = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎤̅ (13a)
√ ⎡
√ 2
√ ⎣(1+COVQn )(1+COVλQ )(1+COVRn )(1+COVλR )(1+ρR COVRn COVλR ) (1+ρQ COVQn COVλQ ) ⎦
2 2 2 2 2
√ln 2
(1+ρn COVRn COVQn )

geotechnical engineers as the starting point in conventional allowable


stress design (ASD). The operational factor of safety can be broadly and
understood to be the “true” margin of safety and hence of interest in both
deterministic and reliability-based analysis and design. Finally, it may

⎡ ⎤
( ) √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
⎢ μλR (1+COV2Qn )(1+COV2λQ ) ⎥
ln⎣ μ
λQ (1+COV2 )(1+COV2 ) ⎦ Rn λR

B=√ ̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√ ⎡ ⎤̅ (13b)
√ 2
√ ⎣(1+COVQn )(1+COVλQ )(1+COVRn )(1+COVλR )(1+ρR COVRn COVλR ) (1+ρQ COVQn COVλQ ) ⎦
2 2 2 2 2
√ln 2
(1+ρn COVRn COVQn )

be noted that the limit state function could also be written as the dif­ Equation (11) and component terms follow from probability theory.
ference between Rm and Qm. However, Equation (9) is preferred because All assumptions and full details of its derivation can be found in the
the nominal factor of safety is present and allows reliability index to be appendix to the paper by Bathurst and Javankhoshdel (2017).

5
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Parameters μλR and μλQ are mean values of resistance and load method the limit state equation. If there is no model error then the mean esti­
bias values (λR and λQ), introduced earlier. The nominal resistance value mates of the factor of safety are equal to the nominal factor of safety and
(Rn) and nominal load value (Qn) used at design time in the limit state OFS = Fn. This is an unlikely occurrence for geotechnical soil-structure
design equations (i.e., factor of safety - Equation (12)) are equivalent to interaction problems. If OFS diverges from the nominal factor of
μRn and μQn in the above equation. Their corresponding coefficients of safety, this is a quantitative indication of the underlying accuracy of the
variation (COV) for the random variables introduced thus far are resistance and load models that appear in the limit state equation and
denoted as COVRn for nominal resistance, COVQn for nominal load, reveals whether the nominal factor of safety (Fn) is conservative or non-
COVλR for resistance bias, and COVλQ for load bias. Parameters ρR and ρQ conservative on average.
are Pearson’s correlation coefficients between variables Rn and λR, and
between Qn and λQ, respectively, and represent bias dependencies with 7. Example walls
nominal values. Parameter ρn is the correlation coefficient between Rn
and Qn and is called nominal correlation following the terminology Three well-documented case histories of full-scale instrumented PET
introduced by Lin and Bathurst (2018). In the analyses carried out later strap MSE walls are used in this study to demonstrate reliability analyses
in this paper, ρn = 0 for the tensile strength and soil failure limit states for the three limit states introduced earlier. The Delaware wall was
because nominal load values and nominal resistance values are sampled constructed in the USA in 2012. The Nagasaki wall was constructed in
from independent populations. Soil material properties and their sta­ Japan in 2015 and the Sao Paulo wall in Brazil in 2011. These three
tistical characteristics for the pullout limit state are the same for the load examples were selected to provide a wide geographical coverage and a
equation associated with the active zone in Fig. 1a and b and the pullout range of soil properties and reinforcement arrangement. Wall geometry
equation associated with the passive zone. Hence, ρn ∕ = 0 and will vary and component properties for these structures are given in Tables 3 and
with changes in the frequency distributions for friction angle and unit 4. Additional details of these walls based on original source materials
weight assumed at the location of each reinforcement layer as demon­ appear in the paper by Miyata et al. (2018).
strated by Lin and Bathurst (2018).
The magnitudes of the COV of the nominal values in Equations (13a) 7.1. Delaware wall
and (13b) are taken as 0, 0.1, 0.2 and 0.3. The larger the value of COVRn
and COVQn, the greater the uncertainty in the choice of the nominal A description of this wall appears in papers by Luo et al. (2015) and
value for Rn and Qn, respectively, at the time of design. Bathurst et al. Rimoldi et al. (2013). Fig. 2 shows a cross-section of the structure. Two
(2019a,b) selected the values of 0.1, 0.2 and 0.3 to match notions of different strap materials were used in its construction. The reinforce­
high, typical and low project “level of understanding”, respectively, that ment straps were 50 mm wide and were placed in a single strap
are adopted in Canadian foundation engineering LRFD practice (CSA, arrangement with a splayed configuration. The splay angle was 3◦ which
2019; Fenton et al., 2016). is small enough to be ignored in pullout calculations. The spacing be­
The advantage of expressing β using Equation (11) is to highlight the tween reinforcing straps in the horizontal direction was Sh = 0.75 m at
log-linear relationship between the deterministic notion of margin of the facing connection.
safety (i.e., factor of safety Fn) and margin of safety in probabilistic terms
(i.e., reliability index). This linearity is demonstrated later in the paper.
7.2. Nagasaki wall
Equation (11) gives the same outcomes as Monte Carlo simulation
provided a sufficient number of simulations are carried out. However, it
Primary information on this wall was gathered from Kimura et al.
has the advantage that it can be implemented in an Excel spreadsheet
(2016). A cross-section of this wall appears in Fig. 3. The wall was
and is thus convenient for sensitivity analyses. Furthermore, Monte
constructed and instrumented as a demonstration project to investigate
Carlo simulations can be tedious, particularly when calculations are run
the performance of the PET strap MSE wall technology using PET strap
out to very small probabilities of failure. In fact, Monte Carlo solutions
reinforcement products recently introduced to the Japanese MSE wall
are not practically possible for reinforcement layers with excessively low
market. The reinforcement straps were 50 mm wide and were placed in a
probabilities of failure (very large reliability index) that appear later in
double strap configuration aligned perpendicular to the wall face. The
this paper.
gap between parallel straps was 50 mm. The wall was surcharged
Finally, reconstructing Equation (11) into a single expression with all
following the end of construction. In this paper only the non-surcharged
terms in A and B, reveals that the operational factor of safety introduced
case is analyzed for brevity.
earlier can be expressed as:
μλR μRn μλR 7.3. Sao Paulo wall
OFS ​ = Fn = (14)
μλQ μQn μλQ
This value can be interpreted as the true average factor of safety for A pair of instrumented walls differing only with respect to the
backfill soil were constructed in Brazil and reported by Jayakrishnan

Table 3
Summary of PET strap MSE wall cases.
Wall designation ( Wall Construction Wall Equivalent soil Soil unit Fines Soil peak Cohesion c Reference
Miyata et al., 2018) name date height H surcharge height Sa weight γ content friction (kN/m2)
(m) (m) (kN/m3) <0.074 angleb,c
mm (%) ϕ (o)

PSW3 Delaware 2012 8.5 0.5 18.7 NA 45 32 Luo et al. (2015)


Rimoldi et al. (2013)
PSW4 Sao Paulo 2011 6.4 0 17.0 5 35 0 Capilleri et al. (2019)
Jayakrishnan (2013)
PSW6 Nagasaki 2015 6.6 0 17.3 10 41 50 Kimura et al. (2016)

Notes: NA = not available.


a
S = surcharge pressure q divided by soil unit weight γ.
b
From conventional triaxial compression tests or direct shear box tests.
c
Calculated as ϕsecant from ϕ and c using the method of Miyata and Bathurst, 2007b for the Simplified Stiffness Method only.

6
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Table 4
Summary of PET strap properties (data from Miyata et al., 2018).
Wall Connection Strap Individual Strap Strap Tensile testb Coverage Reinforcement Global
designation type configuration strap width element thickness t ratiod secant stiffness per stiffnesse
bs (mm) widtha bw (mm) Rc m length of wall Sglobal
(mm) at 1000 h (kN/m2)
J (kN/m)

Tensile Tensile 1% strain level 1% strain


strength stiffness per level
Tult per strap at 1%
strap (kN) strain (kN/
m)c

PSW3f Steel loop single 83 and 90 83 and 90 1.8 and 2.5 41.8 and 4800 and 0.22 and 600 and 800 1020
51.6 6700 0.24
PSW4 HDPE saddle single 50 50 3.3 46.3 7900 0.13 750 940
loop
PSW6 HDPE saddle double 49.5 150 4.5 56.1 14400 0.074 and 820 and 1200 1540
loop 0.11

Notes: PET strap sheath material is polyethylene (PE).


a
For single strap configurations bw = bs. For double strap configuration bw = 2 × bs + bg where bg = 50 mm is the gap between parallel straps.
b
Values from tensile tests at 10% strain minute.
c
Width of individual strap is bs.
d
For load and pullout calculations and single strap configuration Rc = bs/Sh and Sh = centre-to-centre strap spacing in the running length of wall direction. For load
calculations and double strap configuration Rc = 2 × bs/Sh and for pullout calculations Rc = (2 × bs + bg)/Sh. For double strap configurations Sh = centre-to-centre
spacing of each element in the running length of wall direction.
e 1 ∑n
Sglobal = J where n = number of reinforcement layers (Allen et al., 2003).
1 i
f H
Strap properties taken from or interpolated from National Testing and Product Evaluation Program (NTPEP) report (AASHTO, 2010).

Fig. 2. Delaware wall (from Miyata et al., 2018).

(2013) and Capilleri et al. (2019). In this study we focus on the wall respectively.
constructed with a sand backfill classified as SW using the Unified Soil Two sets of load bias value statistics are shown in Table S1 for each
Classification System and having the cross-section shown in Fig. 4. The load model; one set of bias values was computed for walls constructed
reinforcement straps were 50 mm wide and were placed in a single strap with c− ϕ backfill soils (ϕ > 0, c > 0), and the other for walls with
arrangement with a splayed configuration of 4 or 8◦ . These splay angles frictional backfill soils only (ϕ > 0, c = 0). For example, the bias statistics
were ignored in pullout computations. The calculations in the current for c− ϕ soils are applicable to the as-built Delaware and Nagasaki walls,
study correspond to the end of construction prior to placement of a while bias statistics for cohesionless soils apply to the Sao Paulo wall. In
surcharge. North American practice, only cohesionless soils are specified for MSE
walls (AASHTO, 2020; CSA, 2019) but this constraint is relaxed with
8. Bias statistics caveats in the UK (BSI, 2010) and the Japanese (PWRC, 2014) technical
manual. In North American design practice for MSE walls the peak
Bias statistics for load and resistance models were collected (or friction angle is capped at 40◦ , and this interpretation was respected for
computed) from data reported in previous studies by the authors the AASHTO and Simplified Stiffness Method load model calculations in
(Miyata et al., 2018, 2019; Allen and Bathurst, 2019). They are sum­ this paper.
marized in Tables S1, S2 and S3 in the Supplemental Material for the Bias values with and without outlier removal are also shown in
maximum tensile load, tensile strength and maximum pullout capacity, Table S1. For practical purposes the differences in these bias values and

7
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Fig. 3. Nagasaki wall (from Miyata et al., 2018).

Fig. 4. Sao Paulo wall (from Miyata et al., 2018).

their influence on numerical outcomes are negligible. Tables S1, S2 and 9.1. Strength limit state – Tensile resistance
S3 also record the magnitude of bias dependencies (correlation) between
observed bias values and predicted nominal values (ρQ and ρR). For cases Fig. 5 shows the results of calculations for reliability index for each
where the null hypothesis that correlation between bias and calculated reinforcement layer in each wall. A range of reliability index values and
nominal values are uncorrelated at level of significance of 5% cannot be matching probabilities of failure (Pf) are identified in each plot. A
rejected, values of ρQ and ρR were taken as zero in reliability minimum target β = 2.33 is recommended for the tensile resistance limit
calculations. state (Pf = 1/100) because the reinforcement layers are highly strength-
redundant (Allen et al., 2005; Bathurst et al., 2008). If one layer fails, the
9. Results other layers can compensate. Larger reliability index values of β = 3.09,
3.40 and 3.54 (Pf = 1/1000, 1/3000 and 1/5000) are also shown for
Selected results from all three walls are presented here to demon­ comparison. A value of Pf = 1/5000 has been used to compute resistance
strate the main outcomes. The results of all analyses can be found in the factors for LRFD of bridges and other structures, and Pf = 1/1000 has
Supplemental Material for this paper. Unless noted otherwise, the plots been used for foundations in general. A value of Pf = 1/100 has been
in the figures to follow are computed taking COVQn = COVRn = 0, used for highly strength-redundant foundations such as pile groups
implying that there is no uncertainty in the “level of understanding” of (Allen et al., 2005).
the limit states, project conditions and material properties at the time of The plots show that for all three walls the margin of safety is above
design. Nevertheless, differences in computed β values using nominal the recommended target β = 2.33 and in most cases above β = 3.54. For
COV values of 0.1, 0.2 and 0.3 were detectable, but they did not change each wall, β values are very similar for loads calculated using the
probabilistic estimates of margins of safety in any practical way. This is AASHTO and BSI Coherent Gravity Methods, and the AASHTO Simpli­
because the structures are excessively safe as demonstrated in the next fied Method. The safest outcomes correspond to loads computed using
sections. the Simplified Stiffness Method. The least safe outcomes can be seen for
the Sao Paolo wall (Fig. 5c) using AASHTO and BSI Coherent Gravity

8
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Methods, and the AASHTO Simplified Method to compute tensile loads.


The first three methods in each figure show a general trend of
decreasing margins of safety with depth. This may be expected since
these methods predict increasing reinforcement loads with depth for
simple reinforcement arrangements that follow notions of classical earth
pressure theory and from which these methods are adapted. The trends
are interrupted for the Delaware and Nagasaki walls because of local
changes in reinforcement coverage ratio (Rc). A distinctive feature of the
trend in β values computed using the Simplified Stiffness Method is a
limiting constant β value for layers located closest to the bottom of the
wall. This outcome is expected because the Simplified Stiffness Method
predicts constant tensile load under operational conditions once a crit­
ical depth below the crest of the wall is reached (Allen and Bathurst,
2015).
The data plots for each wall in this figure were computed with
COVQn = 0, which gave the lowest β value for all load calculation
methods with the exception of the Simplified Stiffness Method. For the
Simplified Stiffness Method, the lowest margin of safety occurs using
COVQn = 0.3. This outcome for the Simplified Stiffness Method may be
expected since the margin of safety decreases with greater uncertainty in
the estimate of nominal load (i.e., larger COVQn values). Calculations
with COVQn ∕ = 0 for the other load models are not shown to avoid visual
clutter, but β actually increased slightly with increasing COVQn value.
This counter-intuitive trend for the other load models is a consequence
of the large negative correlation between nominal load values and bias
values (ρQ) that can be seen in Table S1. Inspection of the parenthetical
terms (1 + ρQ COVQn COVλQ) in A and B components of Equation (11),
reveals that β will increase with increasing negative bias dependency
with predicted nominal load. This effect is not present when ρQ = 0,
which is the case for load calculations using the more accurate Simpli­
fied Stiffness Method together with bias statistics for cohesive-frictional
backfill soils (ϕ > 0, c > 0).
An important practical lesson from the results of tensile resistance
limit state calculations for the three walls in this study is that the margin
of safety against exceeding the tensile strength of any reinforcement
layer is greater than the recommended target value of β = 2.33
regardless of the load model used to compute nominal maximum rein­
forcement tensile loads and the magnitude of uncertainty (level of un­
derstanding) assumed for nominal values.
Fig. 6 shows plots of reliability index against nominal factor of safety
for the tensile resistance limit state for the Delaware wall using Equation
(11). The data present as straight lines using log-linear axes. The loca­
tion of the curves vary depending on the load model used to compute
nominal load (Qn) because each load model has different bias statistics.
The Simplified Stiffness Method gives the largest margins of safety both

Fig. 5. Reliability index for tensile resistance limit state: a) Delaware wall, b)
Nagasaki wall, c) Sao Paulo wall (margin of safety against Tmax ≥ Tal) (COVQn
= COVRn = 0 unless noted otherwise).
Fig. 6. Reliability index versus nominal factor of safety for reinforcement
tensile resistance limit state for Delaware wall (margin of safety against Tmax ≥
Tal) (COVQn = COVRn = 0).

9
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Fig. 7. Operational factor of safety versus nominal factor of safety for rein­
forcement tensile resistance limit state for Delaware wall (margin of safety
against Tmax ≥ Tal) (COVQn = COVRn = 0).

in deterministic (nominal factor of safety) and probabilistic terms


(reliability index). For the two Coherent Gravity Method load cases, the
minimum nominal factor of safety falls below a value of 2.5 for the
reinforcement layers closest to the bottom of the wall, but margins of
safety in probabilistic terms are well above the minimum target reli­
ability index of 2.33.
Plots of operational factor of safety (OFS) versus nominal factor of
safety (Fn) are presented in Fig. 7 for the tensile resistance limit state
using the load models and variants in this investigation. The data curves
plot as straight lines using Equation (14). Recall that OFS can be un­
derstood to be the true factor of safety on average. The plots show that
OFS > Fn by factors of 3.24 to 1.13. Hence, the most conservative (i.e.,
safest for design) is the AASHTO (2020) Coherent Gravity Method, while
the Simplified Stiffness Method is the least conservative, but safe as
demonstrated earlier.
Similar qualitative outcomes for the tensile resistance limit state for
the other two walls in this study can be found in the Supplemental
Material (Figures S1 and S2).

9.2. Strength limit state - Pullout

Reliability analysis outcomes for the pullout limit state are shown in
Fig. 8. Three different pullout models were used in these calculations.
The AASHTO Coherent Gravity Method was used for the calculation of
load because it typically gave the largest Tmax values as seen in
Tables S4, S5 and S6. The plots in Fig. 8 show that the lowest reliability
index for each reinforcement layer is most often the limit state calcu­
lation using the linear pullout model introduced earlier. This model can
be found in current design codes (e.g., AASHTO, 2020; CSA, 2019).
Fig. 8c shows that the topmost reinforcement layer for the Sao Paulo
wall is just below but close to β = 2.33. Thus this layer is judged to satisfy
the target margin of safety against pullout failure from a practical point
of view. The margins of safety are detectably greater using the bi-linear
and non-linear models but the differences are not of practical impor­
tance since all outcomes satisfy the minimum target reliability index β =
2.33. For the same reasons discussed in the previous section, the reli­
ability index values using the linear and bi-linear pullout models are
Fig. 8. Reliability index for pullout limit state: a) Delaware wall, b) Nagasaki
least using COVRn = 0 because of the large negative correlation between
wall, c) Sao Paulo wall using AASHTO Coherent Gravity Method for maximum
pullout bias and nominal pullout capacity (Table S3). For the non-linear
reinforcement load (Tmax) and three different pullout models for pullout ca­
pullout model, ρR = 0. Differences in calculated β values using this pacity (Pc) (margin of safety against Tmax ≥ Pc) (COVQn = COVRn = 0).
pullout model with COVQn = COVRn = 0.1, 0.2 and 0.3 are not shown,
but are small and not of practical concern.

10
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Fig. 9. Reliability index versus nominal factor of safety for pullout limit state
for Sao Paulo wall (margin of safety against Tmax ≥ Pc) (COVQn = COVRn = 0).

Fig. 11. Results of reliability analyses for the soil failure limit state at end of
Fig. 10. Operational factor of safety versus nominal factor of safety for pullout construction (margin of safety against Tmax ≥ TJ at maximum allowable rein­
limit state for Sao Paulo wall using AASHTO Coherent Gravity Method for forcement strain εmax = 2%) (COVRn = COVQn = 0).
maximum reinforcement load (Tmax) and three different pullout models for
pullout capacity (Pc) (margin of safety against Tmax ≥ Pc) (COVQn = COVRn using the same load and pullout models and these outcomes can be seen
= 0).
in Figures S3c and S4c in the Supplemental Material.

Fig. 9 shows the data for the Sao Paulo wall plotted as reliability
index versus nominal factor of safety. This plot shows that while the 9.3. Service limit state - Soil failure
target reliability index of β ≥ 2.33 is satisfied for all layers but the
topmost layer using the linear pullout model, the nominal factor of Recall that the soil failure limit state is considered only for the
safety criterion Fn ≥ 1.9 is not. Similar plots for the other two walls Simplified Stiffness Method. If this limit state is satisfied, then the tensile
showed that the same two criteria were satisfied for the pullout limit loads (or strains) in the reinforcement layers are low enough that the
states using the same load and pullout models by large margins (see shear stress in the soil within the reinforced soil zone will not exceed the
Supplemental Material – Figures S3b and S4b). soil peak strength. Therefore, the wall is assumed to be under working
Finally, Fig. 10 presents plots of OFS versus Fn for the Sao Paulo wall stress conditions consistent with operational conditions.
and demonstrates that the true factor of safety on average is much greater The consequence of not meeting this limit state is typically excessive
than the nominal value regardless of the choice of pullout models. For deformation of the wall which becomes apparent well before wall
the combination of AASHTO Coherent Gravity Method to compute load collapse. Hence, AASHTO (2020) addresses the soil failure limit as a
and the linear pullout model, the operational factor of safety is 3.69 service limit state with a relatively low reliability index value of β = 1
times the nominal value. The conservative (excessively safe) margins of (or Pf = 1/6) as recommended by Allen and Bathurst (2018). This target
safety against pullout for all pullout limit state models can be largely β value is consistent with service limit states in structural engineering
attributed to the long reinforcement lengths (Le) which are selected to (Azizinamini et al., 2014).
meet a minimum factor of safety against external sliding, but are not The plots in Fig. 11a show that the margin of safety (reliability index)
necessary to meet minimum margins of safety against pullout failure. with respect to the soil failure limit state decreases in the order of the
Large ratios of OFS/Fn > 1 were also computed for the other two walls Nagasaki wall, Delaware wall and Sao Paulo wall. In all cases the reli­
ability index is in excess of β = 1 (or Pf = 1/6).

11
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Fig. 11b shows that the nominal factor of safety for the most critical generate the load bias statistics used in this study.
layers in the Delaware wall could be reduced from Fn = 1.9 to about Fn = The general approach is equally applicable to MSE walls constructed
1.2 and still satisfy the minimum target reliability index value (β = 1.0). with other relatively extensible polymeric materials (geotextiles and
However, there are many other factors to consider that may make this geogrids) (e.g., Bathurst et al., 2019a) and relatively inextensible ma­
adjustment to reinforcement type and arrangement impractical. For terials (steel strips and steel grids) (e.g., Bozorgzadeh et al., 2020a, b)
example, PET straps with lower stiffness in the same product line may but using load and resistance models and matching bias statistics
not be available, or increasing the horizontal spacing of the connections applicable to the soils and reinforcement materials collected for these
with the concrete facing could compromise the structural stability of walls.
these units. In this study the large margins of safety for each limit state were
great enough to remove the practical influence of uncertainty in the
10. Conclusions estimate of nominal load and resistance values on calculated reliability
index (i.e., level of understanding). This may not be case for other PET
This paper has focused on the calculation of margins of safety for strap MSE walls or MSE walls constructed with other reinforcement
three internal stability limit states for three full-scale instrumented walls materials and arrangements.
constructed with PET strap reinforcement products. The connection The paper evaluates the soil failure limit state that is used in the
limit state was not investigated in this study. Margins of safety for the Simplified Stiffness Method to ensure that geosynthetic reinforcement
reinforcement layers in these structures were expressed by (determin­ strains are low enough to keep the reinforced soil at working stress
istic) nominal factor of safety, operational factor of safety and reliability conditions assumed for walls under operational conditions. Load
index (or matching probability of failure). These calculations were methods that are based solely on the strength of the soil and reinforce­
performed using the as-built materials and wall geometry for each ment type cannot guarantee that this condition is met. The results of
structure as found in the related source documents. Different load calculations in this study are in agreement with the experience of the
models and pullout models reported in design codes and improved authors that the soil limit state is typically the most critical limit state for
models proposed by the authors in earlier works were used to carry out geosynthetic reinforced MSE walls and most often controls internal
the calculations. Load data from a larger collection of instrumented PET stability design when the Simplified Stiffness Method is used to compute
strap walls compiled by the authors were used to generate load bias data reinforcement loads (Allen and Bathurst, 2018, 2019).
that are necessary to compute probabilistic estimates of margins of Regardless of the load and resistance models used by the designer,
safety expressed by reliability index. Similar collections of bias statistics the probabilistic assessment of margins of safety is more informative
for reinforcement strength, stiffness and pullout capacity in related work than the calculation of the nominal factor of safety alone. Nevertheless,
by the authors were used for the resistance side in reliability limit state the closed-form equation for reliability index (β) includes the nominal
calculations. factor of safety which is typically the first calculation made in the design
The main conclusions from this study are: process; thus the closed-form solution for β originally proposed by
Bathurst and Javankhoshdel (2017) provides a quantitative link to
1. Recommended minimum reliability index values for tensile strength allowable (working) stress design from past practice.
(β = 2.33), pullout (β = 2.33) and soil failure (for Simplified Stiffness
Method only) (β = 1.0) limit states were satisfied for all three walls Acknowledgements
regardless of the load and resistance models used in the calculations.
In most cases, margins of safety were orders of magnitude higher The authors are grateful for financial support from the Natural Sci­
than the target minimum values. ences and Engineering Research Council of Canada (NSERC) (Grant
2. For the Delaware wall in this study, the minimum target determin­ Number: 94344-2013), the Ministry of Transportation of Ontario (MTO)
istic factor of safety Fn = 2.5 for the tensile resistance limit state Highway Infrastructure Innovations Funding Program (Grant Number:
(using the AASHTO and BSI Coherent Gravity Methods to compute 9017-R-0030) and the Japan Ministry of Education, Culture, Sports,
loads) was not satisfied for the two reinforcement layers closest to Science and Technology (Grant-in-Aid for Scientific Research (B) No.
the bottom of the wall. However, the margins of safety expressed as 17H03309).
reliability index were well above the target value of β = 2.33 and the
true operational factor of safety was greater than 5; thus the wall is Appendix A. Supplementary data
judged to be safe against insufficient tensile strength under opera­
tional conditions. Supplementary data to this article can be found online at https://doi.
3. For the Sao Paulo wall in this study, the minimum target determin­ org/10.1016/j.geotexmem.2020.06.001.
istic nominal factor of safety Fn = 1.9 for the pullout limit state was
not satisfied for any layer using the current AASHTO Coherent References
Gravity Method and the simple linear pullout model recommended
Allen, T.M., Bathurst, R.J., 2015. An improved simplified method for prediction of loads
in North American design codes (AASHTO, 2020; CSA, 2019).
in reinforced soil walls. ASCE J. Geotech. Geoenviron. Eng. 141 (11), 04015049.
Nevertheless, the true operational factor of safety was shown to be in Allen, T.M., Bathurst, R.J., 2018. Application of the Simplified Stiffness Method to design
excess of 4 for all layers and the computed reliability index was close of reinforced soil walls. ASCE J. Geotech. Geoenviron. Eng. 144 (5), 04018024.
to or above β = 2.33; thus, pullout is judged not to be a practical Allen, T.M., Bathurst, R.J., 2019. Geosynthetic reinforcement stiffness characterization
for MSE wall design. Geosynth. Int. 26 (6), 592–610.
concern for this wall (or any other wall in this study). Allen, T.M., Bathurst, R.J., Holtz, R.D., Walters, D.L., Lee, W.F., 2003. A new working
4. The minimum computed margins of safety for each wall described stress method for prediction of reinforcement loads in geosynthetic walls. Can.
deterministically and probabilistically were for the soil failure limit Geotech. J. 40 (5), 976–994.
Allen, T.M., Bathurst, R.J., Holtz, R.D., Lee, W.F., Walters, D.L., 2004. A new working
state. However, this limit state applies only to reinforcement loads stress method for prediction of loads in steel reinforced soil walls. ASCE J. Geotech.
calculated using the Simplified Stiffness Method. Geoenviron. Eng. 130 (11), 1109–1120.
Allen, T.M., Nowak, A.S., Bathurst, R.J., 2005. Calibration to Determine Load and
Resistance Factors for Geotechnical and Structural Design. Transportation Research
This paper provides a methodology for MSE wall designers to eval­ Board Circular E-C079, Washington, DC, p. 93.
uate the margins of safety for three internal stability limit states for PET AASHTO, 2010. NTPEP Report 8508.1. Final Product Qualification Report for Linear
strap walls. An important caveat when applying this approach to any Composites Paraweb/Paralink and Paragrid Product Line. National Testing and
Product Evaluation Program (NTPEP) Report Series, Laboratory Evaluation of
other wall is that the wall must fall within the range of project condi­
tions, geometry and types of materials for the walls that were used to

12
R.J. Bathurst et al. Geotextiles and Geomembranes xxx (xxxx) xxx

Geosynthetic Reinforcement. American Association of State Highway and Jayakrishnan, P.V., 2013. Report on Full Scale Instrumented MacRES Wall at Jundiai,
Transportation Officials (AASHTO), Washington, DC, USA. Brazil. Officine Maccaferri S. Bologna, Italy.
AASHTO, 2015. Standard Practice R69-15. Determination of Long-Term Strength for Kimura, T., Ichikawa, S., Aoki, S., Ohta, H., 2016. The influence of different stiffness of
Geosynthetic Reinforcement. American Association of State Highway and strip reinforcements showed to the characteristics of tension value and distribution
Transportation Officials (AASHTO), Washington, DC, USA. displayed in strips in actual reinforced earth structure. In: Proceedings of 51th
AASHTO, 2017. LRFD Bridge Design Specifications, eighth ed. American Association of Japanese Geotechnical Society Annual Meeting, Okayama, Japan. Japanese
State Highway and Transportation Officials (AASHTO), Washington, DC, USA. Geotechnical Society, Tokyo, Japan, pp. 601–602 (in Japanese).
AASHTO, 2020. LRFD Bridge Design Specifications, ninth ed. American Association of Lin, P., Bathurst, R.J., 2018. Influence of cross-correlation between nominal load and
State Highway and Transportation Officials (AASHTO), Washington, DC, USA. resistance on reliability-based design for simple linear soil-structure limit states. Can.
AFNOR, 2009. NF P94-270: Renforcement des sols. Ouvrages en sol rapporté renforcé Geotech. J. 55 (2), 279–295.
par armatures ou nappes extensibles et souples. Dimensionnement (Association Luo, Y., Leshchinsky, D., Rimoldi, P., Lugli, G., Xu, C., 2015. Instrumented mechanically
Française de Normalisation). La Plaine Saint Denis cedex – France (in French). stabilized earth wall reinforced with polyester straps. Transport. Res. Rec. 2511,
Azizinamini, A., Power, E.H., Myers, G.F., Ozyildirim, H.C., 2014. Bridges for Service Life 9–17.
beyond 100 Years: Innovative Systems, Subsystems, and Components. (No. SHRP 2 Miyata, Y., Bathurst, R.J., 2007a. Evaluation of K-Stiffness method for vertical
Report S2-R19a-RW-1) Transportation Research Board, Washington, DC. geosynthetic reinforced granular soil walls in Japan. Soils Found. 47 (2), 319–335.
Bathurst, R.J., Allen, T.M., Lin, P., Bozorgzadeh, N., 2019a. LRFD calibration of internal Miyata, Y., Bathurst, R.J., 2007b. Development of K-stiffness method for geosynthetic
limit states for geogrid MSE walls. ASCE J. Geotech. Geoenviron. Eng. 145 (11), reinforced soil walls constructed with c-ϕ soils. Canadian Geotechnical Journal 44
04019087. (12), 1391–1416.
Bathurst, R.J., Allen, T.M., Nowak, A.S., 2008. Calibration concepts for load and Miyata, Y., Bathurst, R.J., 2012. Analysis and calibration of default steel strip pullout
resistance factor design (LRFD) of reinforced soil walls. Can. Geotech. J. 45 (10), models used in Japan. Soils Found. 52 (3), 481–497.
1377–1392. Miyata, Y., Bathurst, R.J., Allen, T.M., 2018. Evaluation of tensile load model accuracy
Bathurst, R.J., Allen, T.M., Walters, D.L., 2005. Reinforcement loads in geosynthetic for PET strap MSE walls. Geosynth. Int. 25 (6), 656–671.
walls and the case for a new working stress design method. Geotext. Geomembranes Miyata, Y., Bathurst, R.J., Allen, T.M., 2019. Calibration of PET strap pullout models
23 (4), 287–322. using a statistical approach. Geosynth. Int. 26 (4), 413–427.
Bathurst, R.J., Javankhoshdel, S., 2017. Influence of model type, bias and input Public Works Research Center (PWRC), 2013. Design and Construction Manual of
parameter variability on reliability analysis for simple limit states in soil-structure Geosynthetics Reinforced Soil, (Revised Version). Public Works Research Center,
interaction problems. Georisk 11 (1), 42–54. Tsukuba, Japan, p. 305 (in Japanese).
Bathurst, R.J., Lin, P., Allen, T.M., 2019b. Reliability-based design of internal limit states Public Works Research Center (PWRC), 2014. Design Method, Construction Manual and
for mechanically stabilized earth walls using geosynthetic reinforcement. Can. Specifications for Steel Strip Reinforced Retaining Walls, fourth ed. Public Works
Geotech. J. 56 (6), 774–788. Research Center, Tsukuba, Ibaraki, Japan (in Japanese).
Bozorgzadeh, N., Bathurst, R.J., Allen, T.M., Miyata, A., 2020a. Reliability-based analysis Razzazan, S., Keshavarz, A., Mosallanezhad, M., 2019. Large-scale pullout testing and
of internal limit states for MSE walls using steel strip reinforcement. ASCE J. numerical evaluation of U-shape polymeric straps. Geosynth. Int. 26 (3), 237–250.
Geotech. Geoenviron. Eng. 146 (1), 0401911. Rimoldi, P., Leshchinsky, D., Arrigoni, M., Bortolussi, A., 2013. Vertical wall with
Bozorgzadeh, N., Bathurst, R.J., Allen, T.M., 2020b. Influence of corrosion on reliability- concrete panels facing and geostrips reinforcement: instrumentation and data
based design of steel grid MSE walls. Struct. Saf. 84, 101914. reduction. In: Ling, H.I., Gottardi, G., Cazzuffi, D., Han, J., Tatsuoka, F. (Eds.),
British Standards Institution (BSI), 2010. BS8006-1:2010+A1:2016: Code of Practice for Proceedings of the Design and Practice of Geosynthetic-Reinforced Soil Structures,
Strengthened/reinforced Soil and Other Fills. British Standards Institution, Milton Bologna, Italy. DEStech Publications Inc., Lancaster, PA, USA, pp. 243–260.
Keynes, UK. Schlosser, F., 1978. History, current development and future developments of reinforced
CSA, 2019. Canadian Standards Association (CSA). Canadian Highway Bridge Design earth. In: Proceedings of the Symposium on Soil Reinforcing and Stabilizing
Code. CAN/CSA-S6-19. Mississauga, Ontario, Canada. Techniques, Sponsored by New South Wales Institute of Technology and the
Capilleri, P.P., Ferraiolo, F., Motta, E., Scotto, M., Todaro, M., 2019. Static and dynamic University of New South Wales, Sydney, Australia (In French).
analysis of two mechanically stabilized earth walls. Geosynth. Int. 26 (1), 26–41. Schlosser, F., Segrestin, P., 1979. Dimensionnement des ouvrages en terre armee par la
FHWA, 2009. Federal Highway Administration (FHWA). Design and Construction of methode de l’equilibre local (local stability analysis method of design of reinforced
Mechanically Stabilized Earth Walls and Reinforced Soil Slopes: Vols. 1 and 2. earth structures). In: International Conference on Soil Reinforcement, Paris, France,
FHWA-NHI-10-024. US Department of Transportation, Washington, DC, USA. 1. ISSMFE/ENPC, Paris, France, pp. 157–162 (in French).
Fenton, G.A., Naghibi, F., Dundas, D., Bathurst, R.J., Griffiths, D.V., 2016. Reliability- Schlosser, F., Hoteit, N., Price, D., 1993. Instrumented full scale Freyssisol-Websol
based geotechnical design in the 2014 Canadian Highway bridge design code. Can. reinforced wall. In: Proceedings of Soil Reinforcement: Full Scale Experiments of the
Geotech. J. 53 (2), 236–251. 80s. ISSMFE/ENPC, Paris, France, pp. 299–320.
Geoguide 6, 2002. Guide to Reinforced Fill Structure and Slope Design. Geotechnical
Engineering Office, Hong Kong, China.

13

You might also like