Engineering Failure Analysis 149 (2023) 107261

Contents lists available at ScienceDirect

Engineering Failure Analysis

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

Calibration and validation of a 3D homogenised model to simulate

the damage progression of pultruded GFRP composites
Damião Costa a, José A. Gonilha b, c, *, Nuno Silvestre a
a
IDMEC, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal
b
CERIS, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal
c
Instituto Superior de Educação e Ciências (ISEC Lisboa), Alameda das Linhas de Torres 174, 1750-142 Lisboa, Portugal

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

Keywords: A damage progression failure model for tridimensional finite element (FE) analyses has recently
Pultruded GFRP been proposed, allowing the simulation of composites as homogenized materials, greatly reducing
Failure computational costs. It is able to predict the damage and post-failure behaviour of composites,
Finite elements
while its calibration can be performed with data derived from the standardized experimental
Damage progression
Laminates
characterization of the composite materials. This paper presents the calibration of that model for
5 new glass-FRP pultruded profiles, followed by the simulation of different application/design
cases: (i) wide compact tension; (ii) compact compression; (iii) web-crippling; and (iv) bolted
double-lap connection tests. While the results show that further investigation is needed to address
the experimental characterization of the in-plane shear properties and the numerical simulation
of bearing failure, overall the models were well able to predict the experimental strength, post-
failure behaviour, failure modes and damage patterns, showing that the feasibility of using this
damage progression model as a design tool.

Introduction

Since the introduction of fibre reinforced polymers (FRP) as structural materials, in the second half of the 20th century, the study
and prediction of their failure has been the focus of various research [1], as the understanding of their failure behaviour is essential for
a safe and economic design.
The initial research focused on the prediction of failure initiation, the model proposed by Hill [2] for anisotropic materials
eventually led to the, well known, formulations of the Tsai-Hill [3], for orthotropic materials, and the Tsai-Wu [4], for anisotropic
materials, models. Later, a set of failure criteria for unidirectional laminates (transversely isotropic), associating each criterion with the
compressive and tensile failure in the fibres and matrix, was proposed by Hashin [5]. Several other failure models have been proposed,
including, the more recent and accurate LaRC04 [6] and Puck [7], which on the other hand, are more difficult to implement, requiring
more experimental data. Additionally, using these models in finite element (FE) analysis often requires ply discretization for multi-
layered laminates [8], and, therefore, high computational costs, rendering them unpractical for most design situations. In this
context, owing to their easier implementation, the earlier models, namely the Tsai-Hill [3], Tsai-Wu [4] and Hashin [5] models, have
been the most widely used, although there are many situations for which they are not valid. In particular, for pultruded FRPs –
relatively thick laminates with fiber reinforcement in the longitudinal and transverse directions – the orthotropic condition for the

* Corresponding author.
E-mail address: [email protected] (J. A. Gonilha).

https://doi.org/10.1016/j.engfailanal.2023.107261
Received 8 November 2022; Received in revised form 16 January 2023; Accepted 13 April 2023
Available online 17 April 2023
1350-6307/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Tsai-Hill [3] and Hashin [5] models is not verified; whereas the parameters needed to establish the Tsai-Wu [4] model for the analysis
of anisotropic materials are rarely available, as they are very difficult to determine experimentally [9].
On the other hand, the design of FRP structures cannot rely only on damage initiation models only, since the structural strength is
often much larger than the onset of material failure, demanding the adoption of damage progression models. Two of the most widely
used damage progression models in composites are those proposed by (i) Matzenmiller et al. [10] and (ii) Camanho and Dávila [11].
The former, known as MLT, proposes an exponential damage evolution, for which the maximum damage and the damage “velocity”
can be calibrated against experimental results. The latter, on the other hand, proposes a progressive strain-softening base on the
concept of fracture energy, similarly to that proposed by Bazant [12] for concrete materials, presenting a higher physical significant
but requiring the experimental determination of the fracture energies for all modes considered.
Based on the damage initiation and progression models proposed by Hashin [5], Matzenmiller et al. [10] and Camanho and Dávila
[11], Lapczyk and Hurtado [13] developed a damage formulation that was built-in the ABAQUS software package [14], quickly
gaining popularity, although it should be stressed that that formulation should not be used for multidirectional laminates, including
pultruded profiles, and can only be used in in-plane analysis, greatly limiting the development of complex tri-dimensional models for
design, as demonstrated by Mohammed and Tarfaoui [15].
On the other hand, most of the recent research has relied on the use of modelling the individual plies of the laminates, prescribing
cohesive failure laws between the plies (e.g. [15–19]). In addition to the high computational costs required in this type of approach, it
also requires more experimental data for the definition of the cohesive failure laws, rendering it cumbersome for many design situ­
ations. Furthermore, this modelling approach cannot be used when using off-the-shelve composites, which is very common in Civil
Engineering where pultruded laminates are the most widely used, for which the fibre architecture and stacking are not known (and are
proprietary information).
To overcome these limitations, Gonilha et al. [20] proposed a new damage progression model allowing the tridimensionality
modelling of FRP composites as homogenized materials. The authors proposed two failure initiation criteria, one for in-plane and other
for out-of-plane failure, which are combined by an adaptation of the MLT exponential damage law, allowing to calibrate different
parameters for every direction, followed by a constant stress stage, which typically occurs after fracture. The proposed model has
proven to accurately model the mechanical behaviour of composites, including complex phenomena and design scenarios such as
interlaminar shear failure [20], web-crippling [21], and beam-to-column connections [22]. However, those studies used the same
pultruded glass fibre reinforced polymer (GFRP) material, therefore, further research is needed to assess the ability of the damage
progression model in predicting the behaviour of different composite materials and under different load/design scenarios. This is
particularly relevant because, as highlighted by Kaddour et al. [23], when facing the same design scenarios, different existing models
can provide widely different results, stressing the need to experimentally validate these models.
The present paper aims at reducing that gap in the knowledge, contributing to show that the damage progression model is an
accurate and viable solution for the design of composite structures. To that end, the model parameters were calibrated according to
mechanical properties of five new pultruded GFRP profiles, followed by the modelling of different application cases, namely wide
compact tension tests, compact compression tests, web-crippling tests and double-lap bolted connection tests. The results presented in
this paper allowed to assess the accuracy of the damage progression model for different materials and load/design scenarios.

2. Brief overview of the model

The damage progression model used in this paper [20] is first established by its failure initiation criteria, governed by two 3D
envelopes (forming ellipsoids), which signal in-plane and out-of-plane failure, according to Eqs. (1) and (2), respectively [20].
( ) ( ) ( )2
σ 11 − C11 2 σ 22 − C22 2 τ12
Fpl (σ) = + + =1 (1)
A11 A22 A12
( )2 ( )2 ( )2
σ 33 − C33 τ13 τ23
Fop (σ ) = + + =1 (2)
A33 A13 A23

where σii and τij represent the normal and shear stresses in direction ii and plane ij, respectively, and 1 corresponds to the longitudinal
direction, 2 to the transverse direction and 3 to the thickness direction. Factors Cii , Aii and Aij are determined as a function of the
material tensile, compressive and shear strengths for the various directions/planes, and can be found in Appendix A.
The damage progression model, on the other hand, can be divided in two stages: (i) first, damage progresses with MLT exponential
laws [10], which are established for each direction, i.e. each of the 3 elastic and 3 distortion moduli composing the 3D stiffness matrix,
according to Eqs. (3) and (4) [20].
( m±
fpl (σ̂ ) i
)
(3)
−
± m± •e
Di = di,max 1− e i i = 1, 2, 4

( m±
fop (σ̂) i
)
(4)
−
± m± •e
Di = di,max 1− e i i = 3, 5, 6

2
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

where Di is the damage parameter and the i index represents the relevant direction of the stiffness matrix, i.e. i = 1, 2, 3 respect to the
elastic moduli in directions 1, 2 and 3, respectively, while i = 4, 5, 6 respect to the distortion moduli in planes 12, 13 and 23,
respectively. The MLT law exponential parameters are given by d± i,max and mi , representing the maximum damage allowed at this stage
±

and the exponential progression parameter, respectively. Both parameters are set for each direction and, for the elastic moduli,
different values can be set for the compressive and tensile failure behaviours. Finally, fpl (̂
σ ) and fop (̂
σ ) are the failure indexes for in-
plane and out-of-plane failure, for a given effective stress tensor, which can be derived from Eqs. (1) and (2), as follows:
( ) ( )
F ̂σ = 1 and F ̂σ = 1 (4a and b)
pl fpl op fop

The (ii) second stage of the damage progression model can be described as a constant stress stage, which ensues when a limit strain
(ε±
jj,u or γ jk,u ) is reached, and is governed by the post-failure strength retention percentage in each of the stiffness matrix directions (ri ),
±

as expressed in Eqs. (5) and (6):

r±
i • Sjj
±
Di = 1 − i = 1, 2, 3 jj = 11, 22, 33 (5)
Ejj εjj

ri • Sjk
Di = 1 − i = 4, 5, 6 jk = 12, 13, 23 (6)
Gjk γjk

where S± jj is the axial compressive or tensile (depending on the superscript signal) strength in direction jj, and Sjk is the shear
strength in plane jk. Ejj and Gjk are the initial elastic and distortion moduli, respectively, and εjj and γ jk are the axial and distortion
strains in direction jj and in plane jk, respectively.
It should also be mentioned that, in order to use this damage progression model in Finite Element (FE) analysis, a smoother
transition between the (i) damage progression and (ii) constant stress stages must be provided, as a function of the mesh size, similarly
to what was proposed by Bazant [12]. This regularization stage can be defined by setting a regularization strain (ε± ii,u,eq or γ jk,u,eq ) as a
function of an equivalent element length (α) and the characteristic length of each finite element (Lc ):
/ /
ε±ii,u,eq = ε±ii,u × α Lc γ±jk,u,eq = γjk,u × α Lc (7)

The model considers a single mesh regularization parameter for each material, independent of the direction or failure mode. Up to
this point, there has been no evidence suggesting the need to consider otherwise, however future research should consider this
possibility.
Overall, the stress–strain relationship of the proposed model can be depicted, for a simplified uni-directional case, from Fig. 1.
Additional details can be found in the original research [20], including aspects such as viscous regularization schemes to improve
numerical convergence.

2.1. Calibration methods

As established before, the failure initiation model is set based on the material resistances in each direction and accounting for the
differences in tensile and compressive strength, which can be determined from material characterization tests (standardized tests,
except for the axial strength of the through thickness direction, 33).
On the other hand, the parameters of the damage progression model must be calibrated. This process may be performed by curve-

Fig. 1. Stress–strain relationship (for one directional case) with four stages: (i) undamaged; (ii) damaged progression; (iii) mesh regularization; and
(iv) constant stress (σbls) beyond a limit strain [20].

3
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

fitting/adjusting the numerical results to the results of the experimental material characterization tests, namely the stress vs. strain
and/or load vs. displacement curves [20]. Fig. 2 depicts the integration of the damage progression model, via a user material (UMAT)
subroutine – which can be downloaded from https://coregroup.tecnico.ulisboa.pt/downloads/abaqus-umat/ –, with the ABAQUS
software, and the relation between the models and the calibration process.
The MLT exponential law parameters, d± i,max and mi can be calibrated so the numerical stress vs. strain curves match the shape of
±

their experimental counterparts in the initial and damage progression stages. Those curves should also be used to determine the limit
strains (εjj and γ jk ) after which the model enters its constant stress stage. It should be mentioned that when the behaviour is linear until
failure, typically in the longitudinal direction, both in tension and compression, for common FRP composites and, in particular,
pultruded laminates, d± i,max and mi can be readily set to very low (~0) and high (~10) values, respectively, signalling little damage in
±

the damage progression stage, while the limit strain can be defined as the ratio between the respective strength and the elastic
modulus, derived from the experimental mechanical characterization.
The mechanical behaviour of the model is also governed by the strength retention percentage (ri± ). Since in the experimental testing
it is often difficult to obtain accurate strain readings in the vicinity of and after failure, this parameter can be taken as the ratio between
the maximum load and the load after failure, taken directly from the load vs. displacement curves.
Finally, the mesh regularization parameter (α) could, in principle, be calibrated with any of the material characterization test
results, as long as fracture occurs in the tests [20]. However, it was found that the calibration with experimental tests showing stable
fracture development is more reliable. Thereafter, in this paper, compact tension (CT) tests performed in the transverse direction were
used in the calibration of this parameter.

3. Pultruded GFRP materials

In the present work, 6 different pultruded shapes were considered, from 4 different suppliers, namely Alto, Perfis Pultrudidos, S.A.
(A), a Portuguese manufacturer; Creative Pultrusions (C), a US manufacturer; Fiberline Composites (F), a Danish manufacturer; and STEP,
Sociedade Técnica de Estruturas Pultrudidas (S), a Portuguese supplier. All profiles comprised E-glass fibres and polyester resin matrix,
the most common composition for structural pultruded components, and presented constant plate thickness (except in the web-
junction fillets). Table 1 presents the designation adopted for each material, their manufacturer, and their geometric properties.

3.1. Mechanical characterization

All the pultruded profiles used in this study were duly characterized by means of experimental coupon testing, performed in
previous research, with the exception of the in-plane shear characterization by means of 10◦ -off axis tensile tests [20] of the webs of
profile I150-A, which were performed within the scope of the present research. Nevertheless, for this paper, the raw data of the
previous experiments were collected and treated independently. Table 2 summarizes the mechanical properties of the profiles,
including the references of the experimental data collected, and number of specimens tested in each series. For all profiles, the material
characterization tests were performed in coupons retrieved from the web (W) and flanges (F), except for SHS120-A, for which it is not
possible to distinguish between webs and flanges.
While the results obtained were consistent, presenting acceptable scatter within specimens retrieved from the same laminate, they
show that the different GFRP pultruded materials considered in this study present a considerable range regarding their mechanical
properties. In the longitudinal direction, the average tensile strength (S+ 11 ) ranges from ~300 MPa to ~450 MPa, while the elastic
modulus (E11 ) ranges from ~25 GPa to ~45 GPa. It is interesting to notice that, from these results, there is no apparent correlation
between the tensile strength and elastic modulus. Moreover, the laminates that were also tested in longitudinal compression show very
similar elastic moduli in longitudinal compression and tension, well within the CoV, and present higher strength in longitudinal
compression than in tension.
In the transverse direction, owing to geometric restraints, the mechanical characterization results are available for the web lam­
inates only. Excluding the material of profiles I152-C, the transverse tensile strength (S+ 22 ) ranges between ~35 MPa and 70 MPa, i.e.,
the top value about doubles the lower figure. A similar observation can also be made for the transverse tensile elastic moduli (E22 ),
ranging from ~8 GPa to ~18 GPa. Unlike what occurs for the longitudinal direction, there seems to be a correlation between the
transverse tensile strength and elastic moduli, which is also likely related to the amount of fibres disposed in the transverse direction. In
terms of transverse compression (S−22 ), the strength also shows a wide range, between ~60 MPa and ~120 MPa, while the elastic
moduli (E22 ) varying between ~8 GPa and ~13 GPa.1 The results show that the transverse compressive strength of these materials is
significantly higher than its tensile counterpart, however the same is not observed for the elastic moduli, which, in general present a
similar magnitude to their tensile counterparts. Moreover, it does not seem possible to establish a correlation between the transverse
tensile and compressive strengths, signalling that if the former is governed by fibre content, the latter must be governed by the resin
properties.
The web laminates of profile I152-C were the only laminates presenting higher transverse tensile strength than compressive
strength, with a transverse tensile strength ~1.7 to ~3.5 times higher than that of the other laminates. These are somewhat unexpected

1
The compressive elastic modulus of SHS120-A was determined following ASTM D 695 [27], which is known to provide lower bounds of this
parameter [32], and, therefore, is not considered in this comparison.

4
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Fig. 2. Flow chart of the integration of the damage model (via UMAT) with ABAQUS [14], and of the models with the calibration process.

Table 1
Pultruded profiles used in the present study.
Designation Source Shape Height Width Thickness

(mm) (mm) (mm)

I150-A Alto, Perfis Pultrudidos I section 150 75 8

SHS120-A Square hollow section 120 120 10
I152-C Creative Pultrusions I section 152 76 6
I200-F Fiberline Composites I section 200 100 10
I150-S STEP, Sociedade Técnica de Estruturas Pultrudidas I section 150 75 8
U150-S U section 150 45 8

results, especially given the fact that the transverse compressive strength is well within the range obtained for the other laminates, that
may be related to the higher fibre content disposed at 45◦ and 90◦ [31].
Regarding the in-plane shear properties, the results were more consistent for all laminates, with the in-plane shear strength (S12 )
ranging from ~40 MPa to ~70 MPa. Furthermore, it should be noticed that all lower values were obtained for the profiles from
producer A, with all the remaining laminates presenting shear strength over 60 MPa. Once again, a correlation cannot be established
between the in-plane shear strength and in-plane distortion modulus. Additionally, the properties of two laminates (I150-A-W and
SHS120-A) were obtained with two different methods, the V-notch beam test (ASTM-D5379 [28]) and 10◦ off-axis tensile test
(Hodgkinson [24]), and the result suggest that the latter method underestimates the in-plane shear strength, while the former un­
derestimates the distortion modulus. This can have a significant influence in the modelling of FRP composites, as will be discussed in
the following sections.
For the interlaminar shear properties, results were only available for three laminates, all from producer A, presenting very similar
strengths for both the longitudinal (S13 , ~30 MPa) and transverse directions (S23 , ranging from ~6 to ~8 MPa).

5
 Table 2

D. Costa et al.
Mechanical properties of the pultruded materials.
Material Test Tensile longitudinal Compressive longitudinal Tensile transverse Compressive transverse In-plane shear Interlaminar shear

Standard EN ISO 527 ASTM-D 6641 ASTM D 695 EN ISO 527 ASTM-D 6641 [26] ASTM D 695 ASTM-D5379 [28] Hodgkinson [24] ASTM-D2344 [29]
[data
origin]
[25] [26] [27] [25] [27]

Property S+
11 (MPa) E11 (GPa) υ12 (–) S11 (MPa) E11 (GPa) υ12 (–) S11 (MPa) E11 (GPa) S22 (MPa) E22 (GPa) S22,y S22 (MPa) E22 (GPa) S22 (MPa) E22 (GPa) S12 (MPa) G12 (GPa) S12 (MPa) G12 (GPa) S13 (MPa) S23 (MPa
− − + + − −

I150-A- No. 6 5 0 3 4 0 5 4 5 3
W Avg. 388 43.4 0.23 462 44.9 0.30 – – 44.7 9.4 15.2 64.2 8.1 – – 46.8 3.0 20.4 4.7 27.0 8.1
CoV 6% 2% 8% 7% 4% 24% – – 4% 12% 13% 3% 8% – – 7% 8% 10% 3% 5% 8%
[20]
I150-A-F No. 4 0 0 0 0 0 5 0 5 4
Avg. 354 39.6 0.29 – – – – – – – – – – – – 47.9 3.7 – – 31.2 6.3
[20] CoV 9% 3% 5% – – – – – – – – – – – – 6% 8% – – 3% 19%
SHS120- No. 4 0 11 0 0 8 8 5 7 0
A Avg. 326 25.2 – – – – 435 21.2 – – – – – 88.9 4.8 41.4 – 16.8 3.5 30.6 –
CoV 5% 6% – – – – 12% 15% – – – – – 18% 18% 15% – 20% 13% 7% –
[30]
I152-C- No. 6 6 0 5 6 0 4 0 0
W Avg. 426 25.2 – 437 24.6 – – – 121 10.9 69.6 104 11.3 – – 65.3 4.2 – – – –
CoV 3% 2% – 6% 3% – – – 7% 7% 11% 11% 7% – – 3.3% 15% – – – –
[31]
I152-C-F No. 2 0 0 0 0 0 0 0 0
Avg. 467 26.8 – – – – – – – – – – – – – – – – – – –
[31] CoV – – – – – – – – – – – – – – – – – – – – –
6

I200-F- No. 5 5 0 5 6 0 7 0 0
W Avg. 323 29.6 – 442 29.9 – – – 70.7 17.6 41.3 122 10.9 – – 67.1 2.9 – – – –
CoV 3% 5% – 7% 6% – – – 3% 18% 5% 14% 11% – – 3% 12% – – – –
[31]
I200-F-F No. 6 0 0 0 0 0 4 0 0
Avg. 317 35.3 – – – – – – – – – – – – – 69.3 3.1 – –
[31] CoV 6% 6% – – – – – – – – – – – – – 4% 10%- – –
I150-S- No. 5 6 0 4 8 0 5 0 0
W Avg. 376 29.4 – 551 28.1 – – – 34 8.4 20.0 123 12.9 – – 69.8 3.2 – – – –
CoV 5% 8% – 13% 12% – – – 18% 13% 6% 6% 15% – – 7% 11% – – – –
[31]
I150-S-F No. 5 0 0 0 0 0 2 0 0
Avg. 385 33.9 66.2 3.5

Engineering Failure Analysis 149 (2023) 107261

– – – – – – – – – – – – – – – – –
[31] CoV 5% 8% – – – – – – – – – – – – – – – – – – –
U150-S- No. 4 6 0 5 6 0 4 0 0
W Avg. 347 26.6 – 451 25.8 – – – 69.8 8.7 21.6 83.5 7.0 – – 70.8 4.2 – – – –
CoV 3% 5% – 5% 10% – – – 9% 15% 7% 9% 7% – – 9% 14% – – – –
[31]
U150-S- No. 2 0 0 0 0 0 0 0 0
F Avg. 374 27.3 – – – – – – – – – – – – – – – – – – –
CoV – – – – – – – – – – – – – – – – – – – – –
[31]
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Additionally, Table 2 also presents the end of the proportionality limit for tension in the transverse direction (S+
22,y ), which,
recognizing the tensile non-linear behaviour of FRP composites in the transverse direction, is useful to define the on-set of damage in
that direction [20].
The data presented in Table 2 does not cover the out-of-plane compressive and tensile properties of the laminates, namely the out-
of-plane compressive and tensile strengths and elastic modulus (S−33 , S+ 33 and E33 , respectively). Standardized mechanical tests to
determine these properties are not yet available, and there are few results present in literature, which can have significant variations,
particularly for compression. In this context, for those properties, the same values were considered for all materials in this paper,
namely the tensile strength (S+ 33 ) and elastic modulus (E33 ) retrieved from Vallée [33], who characterized pultruded laminates made
with E-glass fibres and polyester resin, similar to those considered in the present study. Regarding the compressive strength (S−33 ), since
the few values reported in literature vary widely, ranging from ~140 MPa to ~900 MPa [34–36], for a different range of composites –
different fibres, resins and thicknesses – a conservative value of 90 MPa was adopted in the present study.
Table 3 presents the mechanical properties used in the modelling of each material, identifying, owing to the previously mentioned
limitations of the material characterization test data, where the values were taken from literature or assumed to be the same as those
obtained for one of the other laminates. In particular, when data was available for the web laminate of a given profile, but not for the
flange, the values of the web were adapted. In cases where the properties were not determined for any of the laminates of a given
profile, but they were available for I150-A-W (the laminate with a more complete mechanical calibration), the values used for the latter
laminate were adopted. When no data was available for any of the profiles considered in this study, literature values were adopted. In
all cases, the elastic moduli adopted were those obtained through the tensile tests. On the other hand, it should be mentioned that, in
order to better characterize the non-linear behaviour observed in the tensile tests on the transverse direction (cf. Fig. 4c), the input
tensile strength on that direction corresponded to the end of the proportionality limit (S+ 22,y ), and not the corresponding ultimate
strength (S+
22 ), as suggested in [20].

3.2. Calibration models and results

The web and flange material of one of the profiles (I150-A, cf. Table 1) had been previously calibrated in the research that
introduced the damage progression model [20]. The remaining materials were calibrated following the same procedures, in particular,
by comparing the stress vs. strain and/or load vs. displacement curves obtained with the FE models to their experimental counterparts.
In this study, the following material characterization tests were considered for the calibration process: (i) tensile and (ii) compressive
tests for both longitudinal and transverse directions; and (iii) in-plane shear tests.
The tensile tests were simulated by modelling the free length of the specimens (i.e., length in-between the grips), as depicted in
Fig. 3a, considering triple symmetry boundary conditions, to reduce computational costs, and a “weak region”, with a 1 mm2 area,
throughout the thickness, for which the limit strains (εjj and γ jk ) were reduced in 10%, to promote failure initiation. The adoption of
this “weak region”, which has a negligible effect in the predictions of on-set and failure loads, is a common method in models
replicating quasi-uniform stress state specimens, in order to avoid failure at the boundary regions, where stress peaks tend to occur,

Table 3
Input data for the FE models: mechanical properties for each laminate.
Parameter Unit I150-A- I150-A- SHS120- I152-C- I152-C- I200-F- I200-F- I150-S- I150-S- U150-S- U150-S-
W F A W F W F W F W F

E11 GPa 43.4 39.6 32.7 25.2 26.9 29.6 35.3 29.4 33.9 26.6 27.3
E22 9.4 9.4† 4.8 10.9 10.9† 17.6 17.3 8.4 8.4† 8.7 8.7†
E33 3.5* 3.5* 3.5* 3.5* 3.5* 3.5* 3.5* 3.5* 3.5* 3.5* 3.5*
G12 3.0 3.3 3.5 4.2 4.2† 2.9 3.1 3.2 3.5 4.2 4.2†
G13 3.0** 3.3** 3.5** 4.2** 4.2† 2.9** 3.1** 3.2** 3.5** 4.2** 4.2**
G23 3.0** 3.3** 3.5** 4.2** 4.2† 2.9** 3.1** 3.2** 3.5** 4.2** 4.2**
υ12 – 0.23 0.29 0.30* 0.23†† 0.23†† 0.24 0.24† 0.23†† 0.23†† 0.23†† 0.23††
υ13 0.23** 0.29** 0.30* 0.23†† 0.23†† 0.24** 0.29† 0.23†† 0.23†† 0.23†† 0.23††
υ23 0.30* 0.30* 0.30* 0.30* 0.30* 0.30* 0.30* 0.30* 0.30* 0.30* 0.30*
S+11
MPa 388 354 326 426 467 323 317 376 385 347 374†
S−11 462 462† 435 437 437† 442 442† 551 551† 451 451†
#
S+22 15.2 15.2† 15.2 69.6 69.6† 41.3 41.3† 20.0 20.2† 21.6 21.6†
S−22 64.2 64.2† 88.9 104 104† 122 122† 123 123† 83.5 83.5
S+33 8.0* 8.0* 8.0* 8.0* 8.0* 8.0* 8.0* 8.0* 8.0* 8.0* 8.0*
S−33 90.0* 90.0* 90.0* 90.0* 90.0* 90.0* 90.0* 90.0* 90.0* 90.0* 90.0*
S12 46.8 47.9 41.4 65.3 65.3† 67.1 69.3 69.8 66.2† 70.8† 70.8†
S13 27.0 31.2 30.6 27.0†† 27.0†† 27.0†† 27.0†† 27.0†† 27.0†† 27.0†† 27.0††
S23 8.1 6.3 8.1†† 8.1†† 8.1†† 8.1†† 8.1†† 8.1†† 8.1†† 8.1†† 8.1††

*Values estimated based on literature.

**Values obtained for plane 12 were adopted.
#
The end of proportionality limit stress was used as input (S+
22,y ).
†
Values adopted from the corresponding web laminate.
††
Values adopted from I150-A-W.

7
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Fig. 3. Overview of the calibration models’ boundary conditions: (a) tensile; (b) compressive; (c) v-notch beam; and (d) compact tension specimens.

and to assess mesh sensitivity [12,13], a study already performed for the current model [20].
The compressive tests were also simulated by modelling the free length of the specimens, both for the combined lateral compression
(CLC, ASTM-D6641 [26], depicted in Fig. 3b) and direct compression (ASTM D 695 [27], SHS120-A only) test methods. In these
models, a “weak region” was also implemented, but only double symmetry boundaries were used, for the width and thickness.
Regarding the in-plane shear tests, following the experimental characterization, two types of models were developed: (i) v-notch
beams (Iosipescu test) and (ii) 10◦ off-axis tensile tests. The v-notch beam specimens were modelled considering a through thickness
symmetry boundary, as shown in Fig. 3c. In these models, the external “compressive” boundaries (bottom left and top right in Fig. 3c)
were imposed directly on the GFRP specimens, while the “tensile” boundaries (top left and bottom right in Fig. 3c) were modelled with
steel sections, allowing contact nodes to change during the simulations. In all models presented in this paper, the steel elements were
modelled as linear-elastic isotropic with an elastic modulus of 200 GPa and a Poisson coefficient of 0.3; and contact between elements
was defined as compressive only “Hard contact”, for the normal direction, and isotropic penalty, with a 0.1 friction coefficient, for the
tangential behaviour, using surface-to-surface “Finite Sliding” contact formulations. It should be mentioned that preliminary tests
showed that friction coefficients ranging from 0 to 0.4 had little influence on the results.
The models to simulate the 10◦ off-axis tensile specimens were very similar to those used for the tensile tests (cf. Fig. 3a), with the
internal axes of the materials rotated by 10◦ with respect to the longitudinal direction. However, since, in this case, a non-uniform
stress state in under analysis, the “weak region” was not included; symmetry boundary conditions were used only in the thickness
direction, to respect the kinematic boundaries of the test.
Additionally, as mentioned earlier (cf. Section 2.1), and unlike what was made in the original research [20], compact tension (CT)
tests were also considered as part of the calibration, in particular, to determine the mesh regularization parameter (α), as the much
more stable post-failure region, in comparison with the tensile tests, allows for a better fitting of this parameter. The CT specimens
were modelled with a mid-thickness symmetry boundary condition, as shown in Fig. 3d, while the external boundaries were modelled
by means of two steel pins, in contact with the GFRP specimens. Moreover, to reduce the computational costs, the region aligned with
the crack of the specimens was finely meshed, while the remaining geometry a coarser mesh was used.
All calibration models used linear solid brick finite elements with reduced integration (C3D8R) and enhanced hourglass control.
Implicit, geometrically linear analyses were conducted, allowing for a maximum residual flux norm of 1% for normal iterations, and
5% for severe discontinuity iterations.
Before calibrating all materials, mesh sensitivity analysis were performed for each type of model, testing mesh sizes of 1 mm, 0.5
mm and 0.25 mm for the tensile and compressive tests; 2 mm, 1 mm and 0.5 mm for the Iosipescu tests; 3 mm, 1.5 mm and 0.75 mm for
the 10◦ off-axis tests; and 0.6 mm, 0.3 mm and 0.15 mm for the CT tests (in this case, only for the region of the crack, the remaining
specimen was modelled with a 2.5 mm mesh). The results showed that all meshes tested presented stable stiffness, but, in some cases,
the coarser meshes led instability in the ultimate load obtained. Considering these results, the calibration of the materials was per­
formed using meshes with global size of 1 mm for the tensile, compressive and Iosipescu tests, 1.5 mm for the 10◦ off-axis test, and 0.6
mm for the CT test.
Table 4 presents the final calibration parameters defined for each of the materials considered, while Table 5 summarizes the main
results obtained from the modelling of the material characterization tests and the relative difference with respect to their experimental

8
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Fig. 4. Numerical and experimental stress vs. strain curves under longitudinal (a) tension and (b) compression, and transverse (c) tension and (d)
compression.

counterparts. Additionally, Table 6 presents a similar comparison between the numerical and experimental results of the CT tests,
considering notch (crack) lengths of 30 mm and 35 mm.

3.3. Discussion

The FE models, after calibrations, were well able to reproduce the experimental behaviour of all materials considered. In particular,
Fig. 4 shows, as an example, numerical the stress vs. strain behaviour of web (W) specimens under longitudinal (a) tensile and (b)
compressive, and transverse (c) tensile and (d) compressive stresses, against the experimental scatter, while Fig. 5a and b depict the
same curves for the v-notch beam and 10◦ off-axis tests. These results show that the numerical results are well within the experimental
scatter, reproducing the shape of the curves obtained in the experiments.
The results on Table 5 present, with few exceptions, excellent to good predictions of the stiffness and strength observed in the tests.
The largest relative differences between the numerical and experimental results were observed for: (i) longitudinal compression of
SHS120-A; (ii) transverse compression of I200-F-W and I150-S-W; and (iii) the distortion moduli obtained with the v-notch beam tests.
Regarding the longitudinal compression of SHS120-A, the experimental tests were performed following ASTM D 695 [27], which, as
mentioned earlier, is known to underestimate the elastic modulus and material strength [32], leading to the poorer results. Regarding
transverse compression of I200-F-W and I150-S-W, the elastic modulus obtained was significantly lower than that determined in the
experiments. This was due to the fact that, for these laminates, the tensile and compressive elastic modulus are significantly different
(cf. Table 2), and the input moduli used were former. Finally, for the v-notch beam tests, there were significant discrepancies between
the distortion moduli measured experimentally, used as input, and those derived from the numerical results. In both cases, the same

9
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Table 4
Calibration parameters used for each laminate for each laminate.
Parameter I150-A- I150-A-F SHS120- I152-C-W I152-C-F I200-F-W I200-F-F I150-S-W I150-S-F U150-S- U150-S-F
W A W
[20]
[20]

d+
1,max 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001
d−1,max 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001
d+
2,max
0.55 0.55 0.55 0.55 0.55 0.8 0.8 0.8 0.8 0.6 0.6
d−2,max 0.5 0.5 0.5 0.5 0.5 0.4 0.4 0.4 0.4 0.4 0.4
d+
3,max
0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9 0.9
d−3,max 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001
d4,max 0.95 0.95 0.97 0.95 0.95 0.95 0.95 0.99 0.98 0.95 0.95
d5,max 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
d6,max 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
m+1
10 10 10 10 10 10 10 10 10 10 10
m−1 10 10 10 10 10 10 10 10 10 10 10
m+2 1 1 1 1 1 0.8 0.8 0.8 0.8 0.8 0.8
m−2 10 10 10 5 5 0.7 0.7 2 2 2 2
m+3 100 100 100 100 100 100 100 100 100 100 100
m−3 100 100 100 100 100 100 100 100 100 100 100
m4 5 5 0.3 5 5 5 5 5 5 5 5
m5 1 1 1 1 1 1 1 1 1 1 1
m6 100 100 100 100 100 100 100 100 100 100 100
ε+11,u
9.0 × 103 9.0 × 103 1.1 × 102 1.7 × 102 1.8 × 102 1.1 × 102 9.0 × 103 1.3 × 102 1.2 × 102 1.3 × 102 1.4 × 102
ε−11,u 1.1 × 102 1.2 × 102 2.1 × 102 1.8 × 102 1.8 × 102 1.5 × 102 1.5 × 102 2.0 × 102 2.0 × 102 1.7 × 102 1.7 × 102
2 2 2 − 2 2 2 2 2 2 2
ε+22,u
1.0 × 10 1.0 × 10 1.0 × 10 1.9 × 10 1.9 × 10 1.1 × 10 1.1 × 10 1.1 × 10 1.1 × 10 1.7 × 10 1.7 × 102
ε−22,u 1.0 × 102 1.0 × 10 1.8 × 102 1.0 × 102 1.0 × 102 1.0 × 102 1.0 × 102 1.5 × 102 1.5 × 102 1.3 × 102 1.3 × 102
ε+33,u 1 1 1 1 1 1 1 1 1 1 1
ε−33,u 1 1 1 1 1 1 1 1 1 1 1
γ12,u 1 1 0.3 × 102 1 1 1 1 1 1 1 1
γ13,u 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103 8.5 × 103
γ23,u 6.5 × 103 4.1 × 103 6.5 × 103 6.5 × 103 6.5 × 103 6.5 × 103 6.5 × 103 6.5 × 103 6.5 × 103 6.5 × 103 6.5 × 103
r+1 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.02 0.01 0.01 0.01
r−1 0.05 0.05 0.05 0.04 0.04 0.05 0.05 0.05 0.05 0.04 0.04
r+2
0.05 0.05 0.05 0.03 0.03 0.02 0.02 0.01 0.01 0.05 0.05
r−2 0.20 0.20 0.20 0.25 0.25 0.50 0.50 0.20 0.20 0.32 0.32
r+3 1 1 1 1 1 1 1 1 1 1 1
r−3 1 1 1 1 1 1 1 1 1 1 1
r4 1 1 0.01 1 1 1 1 1 1 1 1
r5 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40
r6 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40 0.40
α(mm) 15 15 15 50 50 30 30 40 40 30 30

procedure was used, the strains were measured between the vertices of a 10 mm side square centred with the notch of the specimens –
experimentally the measures were made with a video-extensometer and numerically by measuring the relative differences between the
external nodes. This underestimation is likely related with the measuring procedures, namely because the measuring points may be too
far from the centre of the specimens (where ideally the strains would be measured) and are significantly influenced by the compression
of the supports – which means that the distortion moduli determined experimentally are also, likely, underestimated.
In this context, it is important to highlight that in on order to simulate both the Iosipescu and 10◦ off-axis tests, the calibration of
I150-A-W was changed. In the first case, the calibration used, in [20], was that presented in Table 4, while for the latter the following
parameters were changed m4 = 0.4, γ12,u = 2.4 × 10− 3 and r12 = 0.01, essentially making the damage evolution much more pro­
gressive and non-linear, and imposing a limit to that evolution. These differences underscore the uncertainty related with the me­
chanical characterization tests adopted, in particular the method to measure strains in the Iosipescu tests. This problem will be further
discussed in Section 4.5, when dealing with shear-out failure of bolted connections, which depends heavily on the in-plane shear
damage.
Regarding the CT tests, Fig. 6 presents the load vs. crack mouth open displacement (CMOD) curves obtained for both 30 mm and 35
mm notch lengths. The numerical results presented an excellent agreement with their experimental counterparts in terms of maximum
load, with the highest relative difference of 18% (cf. Table 6). This larger difference is likely due to variations of the real notch length
from its nominal value [31,37], corroborated by the high relative difference in terms of stiffness. The models were also able to
reproduce the overall shape of the experimental curves, being generally close to the experimental scatter. It should be mentioned that
the simulation of the tests on the I152-C-W proved to be the most difficult, with the 30 mm notch model loosing numerical convergence
well before the expected failure, which is likely related with the unusual high tensile strength of this material in the transverse di­
rection. In fact, for the 35 mm notch model, bearing damage around the bolts was observed, which contributed to the lower stiffness

10
 D. Costa et al.
Table 5
Results obtained with the mechanical characterization FE models and relative difference to experimental results (Δ).
Material Test Tensile longitudinal Compressive longitudinal Tensile transverse Compressive transverse In-plane shear

Standard EN ISO 527 [25] ASTM-D6641 [26] ASTM D 695 [27] EN ISO 527 [25] ASTM-D6641 [26] ASTM D 695 [27] ASTM-D5379 [28] Hodgkinson [24]

Property S_ E_ S_ E_ S_ E_ S_ E_ S_ E_ S_ E_ S_{{12}} G_ S_{{12}} G_

{{11}} {{11}}^ {{11}} {{11}}^ {{11}} {{11}}^ {{22}} {{22}}^ {{22}} {{22}}^ {{22}} {{22}}^ ^{\hskip {{12}}^ ^{\hskip {{12}}^
^{+} {\hskip ^{-} {\hskip ^{-} {\hskip ^{+} {\hskip ^{-} {\hskip ^{-} {\hskip 5.0pt} {\hskip 5.0pt} {\hskip
(MPa) 5.0pt} (MPa) 5.0pt} (MPa) 5.0pt} (MPa) 5.0pt} (MPa) 5.0pt} (MPa) 5.0pt} (MPa) 5.0pt} (MPa) 5.0pt}
(GPa) (GPa) (GPa) (GPa) (GPa) (GPa) (GPa) (GPa)

SHS120- FE 350 32.7 – – 665 32.8 – – – – 85.1 4.9 – – 17 3.5

A Δ 7% 0% – – 53%* 55%* – – – – − 4% 1% – – − 1% 2%
I152-C- FE 421 25.2 441 25.3 – – 123 10.5 105 11 – – 66 2.8 – –
W − 1% 0% 1% 3% 2% − 3% 1% − 3% 1% − 34
11

Δ – – – – – –
I152-C- FE 473 26.8 – – – – – – – – – – – – – –
F Δ 1% 0% – – – – – – – – – – – – – –
I200-F- FE 320 29.5 431 29.8 – – 70 16.7 138 15.3 – – 67 2.3 – –
W Δ − 1% 0% − 2% 0% – – − 1% − 5% 14% 41% – – − 1% − 20% – –
I200-F-F FE 311 35.3 – – – – – – – – – – 69 2.44 – –
Δ − 2% 0% – – – – – – – – – – 0% − 20% – –
I150-S- FE 374 29.4 572 29.5 – – 34 7.3 116 8.4 – – 73 2.3 – –
W Δ − 1% 0% 4% 5% – – 0% − 13% − 6% − 35% – – 5% − 29% – –
I150-S-F FE 393 33.9 – – – – – – – – – – 72 2.5 – –
Δ 3% 0% – – – – – – – – – – 9% − 30 – –
U150-S- FE 339 26.6 438 26.7 – – 69 8.2 86 7.4 – – 70 2.7 – –
W Δ − 2% 0% − 3% 4% – – 0% − 7% 3% 5% – – − 1% − 36% – –

Engineering Failure Analysis 149 (2023) 107261

U150-S- FE 374 27.3 – – – – – – – – – – – – – –
F Δ 0% 0% – – – – – – – – – – – – – –
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Table 6
Numerical and experimental results of the compact tension tests.
Material Notch length = 30 mm Notch length = 35 mm

K(kN/mm) Fmax (kN) K(kN/mm) Fmax (kN)

FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ

I152-C-W 3.9 4.1 (2%) − 4% – 4.7 (4%) – 2.3 2.6 (10%) − 9% 2.6 2.9 (3%) − 11%
I200-F-W 6.1 5.8 (7%) 6% 3.1 3.1 (6%) 1% 4.0 2.7 (4%) 46% 2.3 2.0 (7%) 18%
I150-S-W 4.1 4.2 (2%) − 2% 1.6 1.6 (2%) 1% 2.3 2.6 (5%) − 13% 1.1 1.2 (6%) − 4%
U150-S-W 4.1 3.8 (8%) 8% 2.7 2.6 (4%) 5% 2.6 2.4 (8%) 5% 1.9 1.8 (2%) 4%

Experimental results taken from [31,37].

Fig. 5. Numerical and experimental stress vs. strain curves under in-plane shear: (a) v-notched beam and (b) 10◦ off-axis tensile tests.

Fig. 6. Numerical and experimental load vs. CMOD obtained in the compact tension tests.

12
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

(− 9%) and, ultimately, strength (− 11%) in comparison to their experimental counterparts. Nevertheless, these are relatively small
differences, particularly considering the aforementioned uncertainties regarding the notch lengths of the experimental specimens.

4. Application cases

4.1. General description of the cases and models

Four different application cases were considered to assess the quality of the predictions of the damage progression model: (i) wide
compact tension (WCT); (ii) compact compression (CC); (iii) web-crippling; and (iv) bolted double-lap connection tests.
The WCT tests are similar to the CT tests, used in calibration, but present a longer and more stable post-failure response, allowing
for a better assessment of the model performance [31], allowing to assess the calibration of the α parameter. The available test results
correspond to tests performed in laminates I152-C-W, I200-F-W, I150-S-W and U150-S-W, considering 30 mm and 35 mm length
notches [31,38]. The CC tests have a similar configuration to that of the CT tests, but with a compressive load, for which the progressive
damage model has not been tested. Results are available for the same laminates as the WCT tests, for a 40 mm notch.
The web-crippling tests were, once again, used to assess the quality of the model due to the importance numerical methods in
design of profiles under web-crippling. In fact, the new European pre-standard for design of pultruded structures [38], demands the
strength under web-crippling phenomena to be determined by experimental testing or numerical methods. In this paper, the web-
crippling tests of profiles I152-C, I200-F and I150-S, under internal two flanges (ITF) and external two flanges (ETF) loading condi­
tions, were considered [39].
Finally, although the damage progression model has already proved to be able to predict the behaviour of complex connections,
such as beam-to-column connections [22], the quality of the predictions at the local bolt/laminate interface has not been assessed.
Therefore, double-lap tests were simulated, for the I150-A-W and SHS120-A laminates, considering edge distances from 15 mm to 70
mm.
All models were developed using C3D8R elements, with enhanced hourglass control, allowing, when needed – typically in the fillet
at the web-flange junction of the profiles – the use of C3D6 linear wedge elements. Implicit, geometrically linear (unless stated
otherwise), analyses were conducted, allowing for a maximum residual flux norm of 1% for normal iterations, and 5% for severe
discontinuity iterations. The external boundary conditions were modelled by steel elements – linear-elastic isotropic with an elastic
modulus of 200 GPa and a Poisson coefficient of 0.3 – using surface-to-surface “Finite Sliding” contact formulations, considering
compressive only “Hard contact”, for the normal direction, and isotropic penalty, with a 0.1 friction coefficient, for the tangential
behaviour.

4.2. Wide compact tension tests

Similarly to the CT models, WCT specimens were modelled using a mid-thickness symmetry boundary, and refining the mesh on the
alignment of the notch only, as depicted from Fig. 7. The external boundaries were modelled with steel pins, fixing the bottom one and
applying the displacement to the top one. A mesh sensitivity analysis was performed, considering I200-F-W laminates, considering 1
mm, 0.5 mm and 0.25 mm global size meshes at the notch. Outside the notch, cuboid elements with a quarter of the laminates thickness

Fig. 7. Overview of the wide compact tension models: (a) boundary conditions and (b) mesh details.

13
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

were considered. The results stabilized for the 0.5 mm mesh, which was used in the remaining simulations.
Fig. 8 compares the numerical load vs. CMOD curves against the experimental scatter, while Table 7 summarizes the main results.
The FE models were well able to predict the experimental behaviour, namely the shape of the load vs. CMOD curves in the load and
unloading stages, which are generally within the experimental scatter. For all laminates, except I152-C-W, the maximum loads (Fmax )
presented a good agreement with their experimental counterparts, while the initial stiffness presented larger relative errors. It should
be mentioned that the latter parameter is more sensitive to small variations on the notch length, which could explain the larger
differences. For laminates I152-C-W with a 30 mm notch the FE model lost numerical convergence before the unloading stage, for a
load higher (+13%) than the experimental average, while with a 40 mm notch convergence was lost at the beginning of the unloading
stage, also attaining a maximum load higher (+22%) than the experimental average. Although these predictions are not grossly
overestimated, they underscore the difficulties in the modelling of this particular material, which may stem from the unusually high
transverse tensile strength (cf. Table 2). Additionally, it was reported that, in the experimental tests, instability phenomena may have
been at play [31], which was not considered in the numerical models.
The models were also well able to capture the material damage propagation, which, as expected, evolved from the crack tip. Fig. 9
presents an example of the evolution of the transverse elastic modulus (E22 ) and stresses (σ 22 ).

4.3. Compact compression tests

Compact compression (CC) models, depicted in Fig. 10, with a notch length of 40 mm, were very similar to CT and WCT, using mid-
thickness symmetry boundary conditions, steel pins to apply the external boundaries and varying the mesh size only on the alignment
of the crack, namely, testing global mesh sizes of 2 mm, 1 mm and 0.5 mm. The test sensitivity analysis, performed for laminates I200-
F-W, showed that the results were stable for a mesh size of 1 mm, therefore that was the mesh size used in the remaining tests.
Additionally, preliminary tests were conducted to understand the influence of the post-failure contact between crack surfaces, which
showed that, for the relevant displacement range, this effect can be neglected.
Fig. 11 presents the numerical load vs. crack mouth close displacements (CMCD) against the experimental scatter, while Table 8
summarizes the main results. In this case, the results are not compared against the maximum load, which in the experiments often
occurred in the crack progression stage, with significant scatter, but with the failure onset load (Fonset ), taken as the first load peak.
The results show a reasonable agreement in terms of failure onset load and stiffness, except for I200-F-W laminates, for which these
parameters were overestimated. These differences can be attributed to two factors: (i) in the transverse direction, this material presents
an elastic modulus in tension significantly higher than that in compression (~+60%, cf. Table 2), and the former was used as input (cf.
Table 3); and (ii) the strength retention for compressive failure in the transverse direction (r2− = 0.50, cf. Table 4) is unusually high.
Moreover, this parameter was calibrated with results of tests performed following recommendations from ASTM D 6641 [26], which,
as discussed in previous work [20], may overestimate the strength retention.
To address these concerns, an additional simulation of the I200-F-W CC tests, labelled in Fig. 11b as I200-F-W Alt., was performed,
using the transverse elastic modulus in compression as input for E22 , instead of its tensile counterpart, and a strength retention for
compressive failure in the transverse direction (r2− ) of 0.40, instead of the original 0.50. The results (cf. Table 8), with these modifi­
cations, show an excellent agreement with the experimental data. In this regard, it should be stressed that the high difference registered
between the tensile and compressive elastic moduli of the I200-F-W, in which the latter is higher than the former, is not common, with
pultruded materials generally presenting elastic moduli of the same magnitude for tension and compression (cf. Table 2). The model

Fig. 8. Numerical and experimental load vs. CMOD obtained in the wide compact tension tests: (a) 30 mm and (b) 40 mm notch length.

14
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Table 7
Wide compact tension tests, summary of results and relative difference to experimental results (Δ).
Material Notch length = 30 mm Notch length = 40 mm*

K(kN/mm) Fmax (kN) K(kN/mm) Fmax (kN)

FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ

I152-C-W 5.4 5.5 (–) − 2% 5.5 4.8 (–) 13% 2.8 2.6 (13%) 8% 5.9 4.9 (7%) 22%
I200-F-W 8.4 7.2 (10%) 17% 5.8 5.5 (5%) − 7% 5.0 4.2 (10%) 19% 4.7 4.8 (2%) − 2%
I150-S-W 5.8 5.2 (12%) 12% 3.0 2.8 (8%) 7% 3.3 3.1 (8%) 5% 2.6 2.3 (3%) 14%
U150-S-W 6.4 5.0 (3%) 27% 4.5 4.7 (3%) − 3% 3.4 2.9 (5%) − 18% 4.1 3.8 (7%) 9%

Experimental results taken from [31,37].

*
For the I200-F-W specimens, the notch length was 38 mm.

Fig. 9. Damage progression in the simulation of the wide compact tension test of laminate I150-S-W, with 30 mm notch length: transverse (a) elastic
modulus (E22 ), and (b) stress (σ 22 ).

Fig. 10. Overview of the compact compression models: (a) boundary conditions and (b) mesh details.

15
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Fig. 11. Numerical and experimental load vs. CMCD obtained in the compact compression tests for laminates: (a) I152-C-W and U-150-S-W; and (b)
I200-F-W and I150-S-W.

Table 8
Compact compression tests, summary of results and relative difference to experimental results (Δ).
Material K(kN/mm) Fonset (kN)

FE Exp. (CoV) Δ FE Exp. (CoV) Δ

I152-C-W 9.8 9.7 (10%) 1% 8.7 7.8 (4%) 12%

I200-F-W 15.5 12.9 (7%) 20% 14.7 11.8 (7%) 25%
Alt.* 13.3 3% 11.8 0%
I150-S-W 10.3 10.3 (4%) 0% 11.2 11.7 (2%) − 4%
Alt.** 10.3 0% 12.4 6%
U150-S-W 11.1 10.2 (6%) 8% 9.3 8.3 (3%) 11%

Experimental results taken from [40].

*
Using the compressive transverse elastic moduli (cf. Table 1) and r2− = 0.4.
**
Using a linear elastic material outside the crack alignment.

does not allow the input of different moduli in tension and compression, which is a limitation, because that could lead to numerical
convergence problems. In the few cases where this issue might be relevant, the best course of action would be to run simulations with
both the tensile and compressive moduli, taking into account both results in design.
Regarding the shape of the load vs. CMCD curves, with the exception of laminate I200-F-W, the models were well able to predict the
overall shape of the curve, although, in all cases, the experimental softening stage was much steeper than predicted by the models. This
difference was likely due to the evolution of (bearing) damage around the holes, preventing stresses to be properly transferred and
hindering the crack propagation. In the experimental tests, this type of damage was also observed, but failure always occurred in the
crack. It should be mentioned that the post-peak behaviour of this type of bearing failure cannot be correctly capture without element
deletion, as further discussed in Section 4.5. To avoid this problem, for laminate I150-S-W, the simulations of which presented the
highest concentration of bearing damage (which was its ultimate failure mode) leading to a poorer shape of the load vs. CMCD curves, a
new analysis was performed, labelled I150-S-W Alt. in Fig. 11b, considering the material outside the crack alignment as linear-elastic,
leading to an excellent prediction of the behaviour. Nevertheless, it should be stressed that this modelling analysis is not ideal,
especially when the failure modes are not known a priori, thus special caution is to be taken when modelling structural cases which
could, potentially, present bearing failure.
Regarding the failure modes, with the exception of laminate I150-S-W, as mentioned earlier, the models were well able to
reproduce the damage patterns observed in the experimental tests, as exemplified in Fig. 12, including the compressive damage on the
crack tip and the, later, tensile crack on the far-end of the specimens.

4.4. Web-crippling tests

As mentioned earlier, the performance of the model in simulating web-crippling phenomena was performed for profiles I152-C,
I200F and I150-C. As in previous research [21], the present study considered ITF and ETF load configurations, particularly consid­
ering 15 mm and 100 mm bearing lengths (b), respectively. For profiles I152-C both load configurations were tested with b = 100 mm,

16
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Fig. 12. Damage pattern in the simulation of the compact compression tests: (a) numerical model (E22 ), and (b) experimental specimen.

because the corresponding experimental failure mode was buckling, which has not been assessed with the current damage progression
model.
The models were developed considering the properties of web and flange laminates, according to Tables 3 and 4, including the web-
flange junction, which, in a simplifying assumption, was considered as part of the web, since no mechanical properties have been
reported for this potentially weaker zone [41–43] of these profiles. ETF models were developed with two symmetry boundaries – mid-
thickness and mid-height –, while ITF models added a third symmetry boundary (mid-length). Exceptionally, the tests of profile I152-C

Fig. 13. Example of the web-crippling models, I152-C under ITF loading with a bearing length of 100 mm: (a) boundary conditions and (b)
mesh details.

17
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

were simulated with a complete geometry because the experimental specimens failed due to buckling, as mentioned earlier. There­
after, for these cases, geometrical non-linear analyses were conducted, in which an initial imperfection – with the shape of the first
buckling mode, derived from preceding linear buckling analyses – of 0.01 mm was considered. In this regard, it should be mentioned
that Almeida-Fernandes et al. [44] found that imperfections under 0.1 mm had a negligible effect in the strength predictions of this
profile.
As for the previous models, a mesh sensitivity analysis was conducted, focusing on the area of interest, in this case the bearing
length and a side projection of ~1/2 the height of the profiles, as exemplified in Fig. 13 for profile I152-C. In that area, global mesh
sizes corresponding to 1/2, 1/4 and 1/8 of each profiles thickness was considered. This analysis showed that the numerical results
stabilized for a global mesh size of a 1/4 of the thickness (1.5 mm for I152-C, 2 mm for I150-S and 2.5 mm for I200-F). Outside the area
of interest, a global mesh size of 1/2 the profiles’ thickness was considered.
Fig. 14 presents the numerical load vs. vertical displacement curves against the experimental scatter, while Table 9 summarizes the
main results. It should be stressed that the experimental curves correspond to the overall displacements measured by the universal test
machine in which the specimens were tested, while the stiffness in Table 9 corresponds to the relative vertical displacements of the
webs, measured with a video-extensometer system (which is unable to measure the displacement in the vicinity of and after failure).
Therefore, after properly measuring and comparing the experimental and numerical stiffness, the stiffness of the curves in Fig. 14 was
adjusted for a better comparison and understanding of the results, as previously done for the I150-A profile [21].
The results show that the shape of the load vs. vertical displacement curves are very well captured by the models. The results show a
good agreement in terms of web-crippling stiffness, except for profile I200-F, which is related to the use of the transverse tensile elastic
modulus as input, much higher than its compressive counterpart. The ITF strength was overestimated for profiles I200-F and I150-S-W,
while it presented an excellent agreement for the ETF configuration. This is also likely related to the large differences between the
input (tensile) and the compressive transverse elastic moduli (cf. Table 2). In fact, as shown in previous research [21], failure in ITF
configuration is governed by the transverse compressive stresses, while for the ETF configuration in-plane shear failure dominates the
damage progression. The models presented excellent strength prediction for profile I152-C but, in this case, the failure mode associated
was the buckling of the webs, which was also very well captured, as shown in Fig. 15.

4.5. Double-lap tests

The double-lap test setup [45] was significantly simplified for modelling purposes, modelling the GFRP plates and the bolts only,
fixing the latter and applying the displacement at the former. Fig. 16 presents an overview of the model boundaries, including the
double symmetry boundaries used, and a detailed view of the various meshes tested, namely with global mesh sizes of 2 mm, 1 mm and
0.5 mm. In this respect, it should be mentioned that the mesh was refined in the relevant part of the model – from the free edge to the
bolt centre, plus 15 mm –, while a coarser mesh with 4 mm was used in the remaining part of the GFRP plate. The mesh sensitivity
analysis showed that the maximum load level was stabilized for a 1 mm mesh, but the post-peak behaviour could be better captured
with a 0.5 mm mesh, thus the finer global mesh size was chosen for the remaining tests.
For the bolted double-lap connections, two laminates were considered, I150-A-W2 and SHS120-A. For the former, edge distances (e)
of 15 mm, 35 mm and 70 mm were considered, as well as two bolts with 35 mm of edge and pitch distance; while for the latter, data was

Fig. 14. Numerical and experimental load vs. vertical displacement obtained in the web-crippling tests: (a) ITF and (b) ETF load configurations.

2
The tests were conducted on 40×8 mm2 plate section pultruded profiles, ordered with the same fibre arquitecture and matrix composition.

18
 D. Costa et al.
Table 9
Web-crippling tests, summary of results and relative difference to experimental results (Δ).
Material ITF, 100 mm bearing ETF, 15 mm bearing ETF, 100 mm bearing

K(kN/mm) Fmax (kN) K(kN/mm) Fmax (kN) K(kN/mm) Fmax (kN)

19

FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp (CoV). Δ FE Exp. (CoV) Δ

I152-C-W 77.1 85.1 (4%) − 9% 72.0 72.3 (4%) 0% – – – – – – 66.8 66.4 (12%) 1% 51.2 51.7 (21%) − 1%
I200-F-W 148 123 (3%) 20% 160 131 (2%) 22% 38.4 32.9 (8%) 17% 32.3 32.0 (4%) 0% – – – – – –
I150-S-W 82.1 91.0 (2%) − 10% 110 91.4 (10%) 21% 21.0 21.6 (4%) − 3% 24.1 24.3 (2%) − 1% – – – – – –

Experimental results taken from [39].

Engineering Failure Analysis 149 (2023) 107261

D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Fig. 15. Damage pattern and deformed shape for the I152-C profile under web-crippling load configurations: (a) ITF (scale factor of 3 for lateral
displacement), and (b) ETF.

Fig. 16. Overview of the double-lap bolted connection models: (a) boundary conditions; (b) mesh details; and (c) geometry detail of the model with
two bolts.

available for 37 mm and 70 mm of edge distance.

As discussed in Section 3.3, there are reasonable doubts regarding the ability of the results of both Iosipescu and 10◦ off-axis tests to
fully convey the in-plane shear behaviour of the laminates. In order to address this uncertainty, the double-lap test of laminate I150-A-
W, which has the fullest experimental mechanical characterization, with an edge distance of 15 mm (more prone to shear failure) was

20
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Table 10
Input and calibration parameters used to model the double-lap tests of laminate I150-A-W.
Parameter set G12 S12 d4,max m4 γ12,u r4

(GPa) (MPa) (–) (–) (–) (–)

Iosipescu 3.0 46.8 0.95 5 1 1

2
10◦ off-axis 4.7 20.4 0.95 0.4 2.4 × 10− 0.01
2
Mixed 4.7 46.8 0.95 5 2.4 × 10− 0.01

Fig. 17. Numerical and experimental load vs. displacement curves for the double-lap tests of laminate I150-A-W, with 15 mm of edge distance,
considering the set of properties derived from: (i) Iosipescu tests, (ii) 10◦ off-axis tensile (OAT) test, and a (iii) mix of both.

Fig. 18. Numerical and experimental load vs. displacement curves for the double-lap tests of laminate I150-A-W with edge distances of (a) 35 mm,
(b) 70 m, and (c) 35 mm + 35 mm (2 bolts).

21
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

modelled considering the inputs and calibrations, regarding in-plane shear, derived from both the Iosipescu and 10◦ off-axis tests.
Additionally, Table 10 also presents a mixed input/calibration set, using the most reliable properties obtained with each test (cf.
Section 3.3), namely the strength and pre-failure curve shape of the Iosipescu test and the elastic modulus, ultimate strain and strength
retention obtained with the 10◦ off-axis tensile test.
Fig. 17 presents the numerical load vs. displacement curves obtained for the three sets of input/calibration parameters. In the
experimental setup, relative displacements between two points – one at the specimen and another at the testing device – were
measured [45], therefore those results encompassed the deformations of several additional steel plates that composed the experi­
mental setup, not the deformations of the specimens alone. In that context, to allow for the comparison of the overall shape of the
numerical and experimental curves, the numerical displacements were scaled so that the experimental average stiffness matched the
initial (scaled) numerical stiffness. In this respect, it should be mentioned that the ability of the models to replicate the real stiffness has
been well-established in the previous tests, and that this scaling, while helpful for a quick comparison of the overall behaviour, has
little relevance for the purposes of the current analysis. The results show that all calibrations present a good prediction of the first peak
load. However, after failure, the original calibration, stemming from the Iosipescu tests, leads to an increase of the load, which was not
observed in the experiments. This can be attributed by the lack of ultimate strain (γ12,u ) in that calibration, because the Iosipescu tests
did not present fracture due to the limited stroke of the apparatus. On the other hand, the calibration derived from the 10◦ off-axis
tensile tests led to a decrease of the stiffness prior to failure, owing to the very low exponential evolution parameter (m4 ), which
was not registered in the experimental curves. Finally, as expected, the model with the mixed set of parameters presented the best
simulation of the experimental results, with a quasi-linear behaviour up to failure and a long and stable (slightly underestimated) post-
failure load plateau. Considering these results, the following analysis were performed using the mixed set of parameters for laminate
I150-A-W. As for laminate SHS120-A, since there is no calibration retrieved from the Iosipescu tests – no data was available regarding
the strain measurements – the original calibration already considered the properties retrieved from the 10◦ off-axis tensile tests, except
for the in-plane shear strength, which was taken from the Iosipescu test.
Fig. 18 presents the load vs. displacement curves for the double-lap tests of laminate I150-A-W, with edge distances (e) of 35 mm,
70 mm and with two bolts (35 mm + 35 mm); while Fig. 19 presents the same curves for laminates SHS-A, with edge distances of 37
mm and 70 mm. Table 11 compares the maximum loads obtained with the numerical models and the respective experimental tests.
The results show that when specimens failed due to shear-out (for edge distances under 40 mm) the models were well-able to
capture the behaviour, with a good agreement with the experimental data in terms of maximum loads, well within the experimental
scatter, and correctly predicting the post-failure behaviour, although underestimating the residual strength in the case of laminate
SHS120-A. Unlike the experimental results, the modelling of the SHS120-A tests also showed a significantly non-linear behaviour
before failure, which is related to the low exponential evolution parameter (m4 ), derived from the calibration of the 10◦ off-axis test,
that, as discussed earlier, is likely to be underestimated. The shear-out failure mode was also very well captured by the models, as
exemplified in Fig. 20a for the I150-A-W laminate with 15 mm of edge distance, with damage concentrating in a narrow band, which,
eventually, slips off the laminate. It should be stressed that the local damage around the bolt, the onset of which can start in the early
stages of the tests, has no practical influence in the global behaviour of the specimens, until the compressive capacity of that area is
reached, in which case bearing failure occurs.
On the other hand, for large edge distances (70 mm), although the models presented a good agreement with experimental data in
terms of the failure (first peak) load, the failure mode is not well captured. While in the experimental test the bearing failure causes the

Fig. 19. Numerical and experimental load vs. displacement curves for the double-lap tests of laminate SHS120-A with edge distances of (a) 37 mm,
and (b) 70 m.

22
 D. Costa et al.
Table 11
Double-lap tests, maximum load and relative difference to experimental results (Δ).
Material F1st peak (kN)

e = 15mm e = 35mm e = 37mm e = 70mm 2 bolts

23

FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ FE Exp. (CoV) Δ

I150-A-W 5.9 5.3 (17%) 11% 12.8 13.7 (12%) − 7% – – – 12.7 14.9 (7%) − 15% 22.9 19.9 (6%) 19%
SHS120-A – – – – – – 16.9 19.3 (10%) − 12% 21.4 18.1 (8%) − 18% – – –

Experimental results taken from [45].

Engineering Failure Analysis 149 (2023) 107261

D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

Fig. 20. Damage pattern for the I150-A-W laminates subjected to double-lap bolt tests with edge distances (e) of (a) 35 mm, and (b) 70 mm.

ovalization of the holes, as the GFRP material is progressively crushed by the bolts, eventually leading to shear-out, the models show no
signs of hole ovalization due to bearing, with damage quickly progressing to shear-out, as shown in Fig. 20b. This is essentially due to
the contact formulation: the contact surfaces do not change during the analysis even if the elements are heavily distorted, i.e., the bolt
shank is always in contact with the elements immediately around the hole, even if these elements have lost all bearing capacity. In
order to overcome this issue and correctly model the post-failure behaviour of GFRP laminates under bearing failure, the elimination of
heavily distorted elements must be implemented, allowing the contact between the bolt shank and the next row of laminate elements,
which, in the ABAQUS software [14], requires explicit analysis.

6. Conclusions and future developments

The present paper presented the calibration of a recently proposed damage progression model for 5 different new pultruded GFRP
materials, assessing the accuracy of the model with 4 different application cases. The following main conclusions can be derived from
this work:

• The damage progression model can be calibrated for different materials using the results of standardized mechanical character­
ization coupon tests, although, whenever possible, compact tension tests should be used to calibrate the mesh regularization
parameter (α).
• The analysis of the experimental results showed remarkable differences in the in-plane shear properties determine by Iosipescu or
10◦ off-axis tensile tests, leading to added uncertainty in the calibration of the respective parameters.
• The application of the damage progression model in the simulation of wide compact tension and compact compression tests showed
its ability to predict the behaviour of the different materials under almost opposite solicitations.
• The simulation of web-crippling tests shows the damage progression model can be a useful tool in design.
• The results also show that the damage progression model can be successfully coupled with geometrical non-linear analysis,
correctly predicting buckling failure modes and progression.
• The simulation of bolted double-lap tests showed the importance of the in-plane shear parameters in the modelling of shear-out
failure, highlighting the need to improve the experimental methods to determine those properties.
• Those results show that the damage progression model is able to predict the shear-out and bearing strength with very reasonable
accuracy, although the simulations presented were not able to predict the bearing post-failure behaviour.
• Future research should focus on implementing the damage progression model in explicit analyses, in order to use element deletion
in the simulation of bearing failure.

Overall, the results presented in this paper showed that the damage progression model is able to simulate the behaviour of various
materials, under very different solicitations, confirming the feasibility of its use in structural design.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

The code with the damage progression model is available (link included in the paper); the remaining data will be made available
upon reasonable request.

Acknowledgements

The authors wish to thank the financial support of Fundação para a Ciência e a Tecnologia (FCT) through the funding of the
research unit CERIS (UIDB/04625/2020). This work was supported by FCT, through IDMEC, under LAETA, project UIDB/50022/

24
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

2020. The authors are very grateful to Dr. Lourenço Almeida-Fernandes for kindly sharing the extensive data of his investigation.

Data availability statement

All data and models that support the findings of this study are available from the corresponding author upon reasonable request.
The UMAT used for implementing the GFRP damage progression model is available at the CORE Group website: https://coregroup.
tecnico.ulisboa.pt/downloads/abaqus-umat/.

Appendix A. . Failure envelope parameters

As defined in Gonilha et al. [20], Cii are the coordinate of the centre of the failure envelopes over axis ii, which can be computed by,
⃒ ⃒
S+ − ⃒Sii− ⃒
Cii = ii (A1)
2
where S+
ii and Sii represent the tensile and compressive strengths in direction i, respectively. On the other hand, Aij represent the
−

maximum distance between the centre of the corresponding ellipsoid and the envelope surface over axis ij, and are determined by,
+
S11 − C11
A11 = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) (A2)
2
C22
1− A22

√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√(
√ + )2 ( C C )2
√ S22 − C22 − S+11− C2211
√
A22 = √ ( )112 (A3)
1 − S+C− 11C11
11

S12
A12 = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) ( ) (A4)
2 2
C11 C22
1− A11
− A22

⃒ − ⃒
+
A33 = S33 + ⃒S33 ⃒ (A5)

S13
A13 = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) (A6)
2
C33
1− A33

S23
A23 = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( ) (A7)
2
C33
1− A33

where Sij represents the shear strength in the plane ij.

References

[1] R.E. Rowlands, Strength (failure) theories and their experimental correlation, in: G.C. Sih, A.M. Skudra (Eds.), Failure mechanics of composites, handbook of
composites, vol. 3, North-Holland, New York, 1985, pp. 71–128.
[2] R. Hill, The mathematical theory of plasticity, Oxford University Press, Oxford, 1950.
[3] S.W. Tsai, Strength theories of filamentary structures, in: R.T. Schwartz, H.S. Schwartz (Eds.), Fundamental Aspects of Fiber Reinforced Plastic Composites
(1966), Conference proceedings, Wiley Interscience, New York, 1968, pp. 3–11.
[4] S.W. Tsai, E.M. Wu, A general theory of strength for anisotropic materials, J Compos Mater 5 (1971) 58–80.
[5] Z. Hashin, Failure criteria for unidirectional fiber composites, J Appl Mech 47 (1980) 329–334.
[6] Pinho ST, Dávila CG, Camanho PP, Iannucci L, Robinson P. Failure models and criteria for FRP under in-plane or three-dimensional stress states including shear
non-linearity. NASA/TM-2005-213530; 2005.
[7] A. Puck, H. Schürmann, Failure analysis of FRP laminates by means of physically based phenomenological models, Compos Sci Technol 62 (2002) 1633–1662.
[8] R. Talreja, Damage and fatigue in composites – a personal account, Compos Sci Technol 68 (2008) 2585–2591.
[9] A. Makris, C. Ramault, D. Van Hemelrijck, A. Clarke, C. Williamson, M. Gower, R. Shaw, R. Mera, E. Lamkanfi, W. Van Paepegem, A review of biaxial test
methods for composites. International Conference on Experimental Techniques, 2007.
[10] A. Matzenmiller, J. Lubliner, R.L. Taylor, A constitutive model for anisotropic damage in fiber-composites, Mech Mater 20 (1995) 125–152.
[11] P.P. Camanho, C.G. Dávila, Mixed-mode decohesion finite elements for the simulation of delamination in composite materials, NASA/TM-2002-211737 (2002).
[12] Z.P. Bazant, Imbricate continuum and progressive fracturing of concrete and geomaterials, Meccanica 19 (1984) 86–93.
[13] I. Lapczyk, J.A. Hurtado, Progressive damage modeling in fiber-reinforced materials, Composites: Part A 38 (2017) 2333–2341.
[14] SIMULIA. ABAQUS 2018 Documentation Collection. Dassault Systèmes; 2017.
[15] M.A. Mohammed, M. Tarfaoui, A progressive damage modelling of glass/epoxy cylindrical structure subjected to low-velocity impact, Eng Fail Anal 134 (2022),
106036.

25
D. Costa et al. Engineering Failure Analysis 149 (2023) 107261

[16] K.C. Warren, R.A. Lopez-Anido, S.S. Vel, H.H. Bayraktar, Progressive failure analysis of three-dimensional woven carbon composites in single-bolt, double-shear
bearing, Composites Part B 84 (2016) 266–276.
[17] A.M.G. Coelho, J.T. Mottram, Numerical evaluation of pin-bearing strength for the design of bolted connections of pultruded FRP material, J Compos Construct
21 (5) (2017).
[18] M. Akterskaia, P.P. Camanho, E. Jansen, A. Arteiro, R. Rolfes, Progressive delamination analysis through two-way global-local coupling approach preserving
energy dissipation for single-mode and mixed-mode loading, Compos Struct 223 (2019).
[19] Zhuang F, Arteiro A, Furtado C, Chen P, Camanho PP. Mesoscale modelling of damage in single- and double-shear composite bolted joints. Compos Struct 2019;
226.
[20] J.A. Gonilha, N. Silvestre, J.R. Correia, V. Tita, D. Martins, Novel progressive failure model for quasi-orthotropic pultruded FRP structures: formulation and
calibration of parameters (Part I), Compos Struct 255 (2021), 112974.
[21] J.A. Gonilha, N. Silvestre, J.R. Correia, V. Tita, L. Almeida-Fernandes, Novel progressive failure model for quasi-orthotropic pultruded FRP structures:
application to compact tension and web-crippling case studies (Part II), Compos Struct 255 (2021), 112973.
[22] J.A. Gonilha, D. Martins, Numerical simulation of the damage progression of pultruded GFRP beam-to-column connections under monotonic and cyclic loads,
Compos Struct 300 (2022), 116180.
[23] A.S. Kaddour, M.J. Hinton, P.A. Smith, The world-wide failure exercises: how can composites design and manufacture communities build their strength. 16th
European Conference on Composite Materials (ECCM16), 2014.
[24] J.M. Hodgkinson, Mechanical testing of advanced fiber composites, CRC Press, Boca Raton, California, 2000.
[25] EN ISO 527–1, Determination of tensile properties – Part 1: general principles, Brussels (1997).
[26] ASTM D 6641/D 6641M-16e1. Standard test method for compressive properties of polymer matrix composite materials using a combined loading compression
(CLC) test fixture. West Conshohocken; 2016.
[27] ASTM D 695. Standard test method for compressive properties of rigid plastics. West Conshohocken (PA): American Society for Testing and Materials (ASTM);
2002.
[28] ASTM D 5379/D 5379M. Standard test method for shear properties of composites materials by the V-notched beam method. West Conshohocken (PA); 2005.
[29] ASTM D 2344/D 2344M-00. Standard test method for short-beam strength of polymer matrix composites materials and their laminates. West Conshohocken
(PA); 2000.
[30] D. Martins, M. Proença, J.A. Gonilha, M.F. Sá, J.R. Correia, N. Silvestre, Experimental and numerical analysis of GFRP frame structures. Part 1: cyclic behaviour
at the connection level, Compos Struct 220 (2019) 304–317.
[31] L. Almeida-Fernandes, N. Silvestre, J.R. Correia, Compressive transverse fracture behaviour of pultruded GFRP materials: experimental study and numerical
calibration, J Compos Construct 24 (4) (2020) 04020019.
[32] Wegner PM, Adams DF. Verification of the combined load compression (CLC) test method. Report No. DOT/FAA/AR-00/26. Washington (DC): Federal Aviation
Administration; 2000.
[33] T. Vallée, Adhesively bonded lap joints of pultruded GFRP shapes, PhD Thesis in Sciences, École Polytechnique Fédérale de Lausanne (2004).
[34] X. Qian, H. Wang, D. Zhang, G. Wen, High strain rate out-of-plane compression properties of aramid fabric reinforced polyamide composite, Polym Test 53
(2016) 314–322.
[35] R.A. Govender, L.A. Louca, A. Pullen, A.S. Fallah, G.N. Nurick, Determining the through-thickness properties of thick glass fiber reinforced polymers at high
strain rates, J Compos Mater 46 (10) (2011) 1219–1228.
[36] P.M. Schubel, J.J. Luo, I.M. Daniel, Through-thickness characterization of thick composite laminates, in: M.I. Saint Louis (Ed.), SEM annual conference and
exposition on experimental and applied mechanics 2006, conference proceedings, 2006, pp. 1793–1800.
[37] L. Almeida-Fernandes, N. Silvestre, J.R. Correia, M.R.T. Arruda, Fracture toughness-based models for damage simulation of pultruded GFRP materials,
Composites Part B 186 (2020), 107818.
[38] CEN/TS 19101: 2022, Design of fibre-polymer composite structures. Brussels: CEN; 2022.
[39] L. Almeida-Fernandes, J.R. Correia, N. Silvestre, Effect of fibre layup and bearing length on the web-crippling behaviour of pultruded GFRP profiles, Compos
Struct 267 (2021), 113884.
[40] L. Almeida-Fernandes, J.R. Correia, N. Silvestre, Compressive transverse fracture behaviour of pultruded GFRP materials: experimental study and numerical
calibration, Compos Struct 247 (2020), 112453.
[41] G.J. Turvey, Y. Zhang, A computational and experimental analysis of the buckling, postbuckling and initial failure of pultruded GRP columns, Comp Struct 84
(22) (2006) 1527–1537.
[42] L. Feo, A.S. Mosallam, R. Penna, Mechanical behavior of web–flange junctions of thin-walled pultruded I-profiles: an experimental and numerical evaluation,
Compos Part B: Eng 48 (2013) 18–39.
[43] A. Quadrino, R. Penna, L. Feo, N. Nisticò, Mechanical characterization of pultruded elements: Fiber orientation influence vs web-flange junction local problem.
Experimental and numerical tests, Compos Part B: Eng 142 (2018) 68–84.
[44] L. Almeida-Fernandes, N. Silvestre, J.R. Correia, Fracture toughness-based models for web-crippling of pultruded GFRP profiles, Composites Part B 230 (2022),
109541.
[45] D. Martins, J. Gonilha, J.R. Correia, N. Silvestre, Exterior beam-to-column bolted connections between GFRP I-shaped pultruded profiles using stainless steel
cleats. Part 1: experimental study, Thin-Wall Struct 163 (107719) (2021).

26