Contents lists available at ScienceDirect

Composites Science and Technology

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

Bending failure analysis and modeling of thin fiber reinforced shells

Georgios A. Pappas *, 1, Arthur Schlothauer 1, Paolo Ermanni
Laboratory of Composite Materials and Adaptive Structures, ETH Zurich Leonhardstrasse 21, 8092, Zürich, Switzerland

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

Keywords: In this work, the impact of high stress gradients, found in bending of thin unidirectional fiber reinforced shells
Carbon fibers A. (~0.1 to 1 mm), on compressive micro-buckling failure, was analyzed. Such thin shells show increased resistance
Flexible composites A. to compressive failure under high curvatures, which may even allow tensile fiber damage to drive ultimate
Computational mechanics C.
failure for very low thickness (e.g. <0.5 mm). The main scope of this work is to analyze this increased resistance
Failure criterion C.
Non-linear material
to compressive failure and propose a robust modeling scheme. The mechanical failure response was captured by
a shell-buckling experimental campaign. The origins of the increased compressive failure resistance were initially
attributed to the reduction of shear stresses acting on the most susceptible domain of a representative wavy fiber.
This effect was effectively described by an analytically derived, stress-gradient-dependent parameter. The hy­
pothesis for the establishment of this parameter was corroborated by a numerical micromechanical model
adopting the embedded cell approach. This model also revealed important micromechanical interactions which
were incorporated by simple stress and strain factors. The derived failure prediction scheme was further
extended to include the non-negligible, non-linear elastic material behavior of carbon fibers by means of a nu­
merical algorithm. The validity of the failure prediction model was demonstrated by the successful comparison
with results acquired from the shell-buckling experiments on a unidirectional carbon-fiber reinforced epoxy
system. To this end, the validity of the initial hypothesis of stress-gradient-dependence on compressive failure
was corroborated. Major effect on the overall behavior modeling has carbon fiber’s material non-linearity, as
well as micromechanical interactions.

1. Introduction & background in compressive failure of FRPs e. micro-buckling, was acknowledged as the reason for early failure of
composites in compression. The adopted energetic approach provided
Unidirectional (UD), fiber reinforced polymers (FRPs) are known to closed form solutions to describe the two failure modes illustrated in
have lower longitudinal compressive than longitudinal tensile strength. Fig. 1a & b, namely, in-phase micro-buckling and out-of-phase micro-­
Consequently, the bending strength is controlled by compressive failure. buckling. The former case is causing shear damage to the matrix (shear
By contrast, thin shells (thinner than ~0.5 mm) subjected to bending, failure), while the latter causes a mixed damage mode (extensional
demonstrate remarkable resistance to compressive failure and experi­ mode) that is only expected for low fiber volume fractions, Vf , of
mental evidence shows that the tensile side fails first [1–3]. The origins ~20–30% [4]. In well aligned FRPs, or composites of very brittle
of this unique behavior have not been identified for the time being. The fiber/matrix interface (polymer or ceramic), transverse cracking due to
subject of this work is the investigation of the sources of this particular Poisson’s effect may occur (see Fig. 1c), also leading to micro-buckling.
behavior, using experimental evidence and mechanical modeling of the Rarely, brittle fibers may fail in the maximum shear stress plane (Fig. 1d)
failure response. at even lower strains.
The compressive failure behavior of UD FRPs of conventional According to Ref. [4], when neglecting terms of higher order, the
thickness has been thoroughly analyzed with various experimental, compressive strength of a UD lamina due to shear failure mode, σ c,y , can
analytical and numerical tools. The first widely accepted work that be estimated by means of Vf and matrix’s shear modulus, Gm , as: σc,y =
provided a predictive analytical scheme based on constituent properties, Gm /(1 − Vf ). However, as indicated by Argon [5] and later thoroughly
was done by Dow and Rosen [4]. In that work, buckling on fiber level, i. investigated by Kyriakides et al. [6], this approach stands as a very

* Corresponding author.
E-mail address: gpappas@ethz.ch (G.A. Pappas).
1
G.P. and A.S. contributed equally.

https://doi.org/10.1016/j.compscitech.2021.108979
Received 25 May 2021; Received in revised form 2 August 2021; Accepted 2 August 2021
G.A. Pappas et al.

optimistic upper bound (~4 times the experimental values [5]), since it [ ( )]
is established for perfectly aligned fiber micro-structure. In reality, an
’
y’ (x’ ) = y0 1 − cos π xλ
(2a & b)
existing initial misalignment causes important shear stress on the matrix &(
even under pure compressive normal stresses. The shear failure of the π)
ϕmax ≃ tanϕmax = y0 1 +
matrix will initially lead to instability, and consequently to fiber bundle λ
micro-buckling resulting in high bending curvatures of the fibers. The Here, 2λ is the wavelength of the waveform and y0 is its amplitude.
failure of the fibers forms the so-called, kink-bands [5,7]. Kink-band The maximum angle, ϕmax , lies at λ/2 location and can be evaluated by
formation leads to significant reduction of the load bearing ability [6]. Eq. (2b). Thus, characteristic values of λ and y0 are needed to define a
According to Argon [5], the shear stress on a misaligned composite representative ϕmax .
bundle is τ = σc ⋅ϕ0 (for small angles: ϕ0 ≃ sin ϕ0 ), thus, a critical The thorough mechanical analysis in Ref. [10], reveals that the
strength can be approximated as σc,y = τc,y /ϕ0 , where τc,y , is composite’s transverse shear stress, τ (Fig. 1e & f), that triggers failure, results from
shear yield stress and ϕ0 is a representative misalignment angle. About a the micromechanical equilibrium on a misaligned band. Thus, apart
decade after Argon, Budiansky [7] included the effect of local matrix from the dominant σc and ϕ0 , it also depends, on the sliding shear stress
yielding. His input is described in Eq. (1) and can be interpreted as τS , the local transverse & longitudinal normal stresses σ T & σ L , and the
follows: the locally misaligned domain will sustain shear until the shear created coupled stress due to moment, m, as illustrated in the detail
yield strain of the composite, γy,c , is met at τc,y , causing a further included in Fig. 1f. Such analysis can increase the accuracy on the pre­
misalignment ϕ’ = γ c,y , before instability (see Fig. 1e). dicted compressive strength, but leads to non-trivial differential equa­
tions that can be resolved using material constitutive laws. Moreover,
τc,y existing analytical micromechanical schemes [7–11] contribute on the
σ c,y = , with ϕ’ = γ c,y (1)
ϕ0 + ϕ’ post peak-load response and the formation of kink-bands assuming an
The thorough analysis presented in Ref. [6] shows that the initial elasto-plastic matrix, while FE-based numerical schemes verify the
failure event can be practically captured by Eq. (1). However, it is very sequence of failure events on the formation and damage progression of
sensitive to the choice of initial misalignment distribution. Fibers’ kink-bands [6,12].
alignment disturbance can have random forms. Nonetheless, it can be All aforementioned schemes that consider some initial fiber
approximated by a sinusoidal wave [6,8,9], such as the one depicted in misalignment, are able to approximate the experimentally acquired
Fig. 1f, described by the following equation: compressive strength values, which are usually close to the bending
ones. A common compressive strain to failure due to micro-buckling is
~1 to 1.5% for FRPs [13], while the tensile ones are restricted by fiber’s
maximum tensile strain limit. By contrast, thin shells (thinner than ~0.5
mm) subjected to bending, demonstrate remarkable resistance to

Fig. 1. UD lamina longitudinal compression failure modes: a) In-phase micro-buckling / shear failure. b) Out-of-phase micro-buckling / extensional mode
(Vf <∼ 20–30%). c) Transverse matrix/interface failure due to Poisson’s effect. d) Shear fiber fracture. e) Detail ① of a): Shear stress and strain induced due to
existing misalignment under homogenized compressive stress. Also schematized are the additional (actual and homogenized) shear strain & rotation ϕ’ that τ in­
duces. f) Stress state around a representative ‘wavy’ fiber at the domain of the maximum waviness angle, ϕmax , for remote uniform compressive strain field. Not
to scale.
G.A. Pappas et al.

micro-buckling and are able to sustain compressive strains as high as 2 to shell composites. The obtained laminates are of very high quality with
3%, allowing the tensile side to fail first when fiber’s strain to failure is practically no voids (very low porosity, below 0.5%), and very uniform
met [1–3]. through-thickness fiber distribution as shown in Fig. 2a. The resulted
The objective of the current study is to reveal the origin of the thicknesses were 130, 164, 335, 420 (13a), 589 and 800 μm with a mean
important resistance to compressive micro-buckling failure of thin shells standard deviations within each family of ±3–7 μm and a nominal cured
under bending, compared to shells of conventional thickness, and pro­ ply thickness tply ≃ 32.5 ± 0.5 μm, measured using digital optical mi­
pose a predictive scheme for their bending strength/failure curvature. croscopy (Keyence VHX6000). The corresponding Vf , defined by the
The focus is applied on the compressive shear failure since transverse fiber mass content and the final thickness was 58.5 ± 0.5% for all, except
cracking (fiber splitting as in Fig. 1c) is shielded due to the limited for 13b-with mean thickness of 403 μm (tply ≃ 31 μm) which was ~61%.
development of Poisson’s effect at high curvatures [2]. Initially, a The nominal Vf was verified using optical analysis (via Matlab®) and
linear-elastic fiber response is considered to investigate the local stress thermogravimetric analysis (Pyris TGA). The composite plates were cut
interactions. Subsequently, this scheme is extended to incorporate the into L × B = 100 or, 150 × 40 mm2 specimens. The width, B, was chosen
important non-linear behavior of fibers such as the carbon ones [2,14, as such to ensure uniform strain in that direction. The choice of length, L,
15]. The predictions from the modeling scheme are compared with the is related to plate tangency attainment (see Figs. 2b and 3c), buckling
results from conducted shell buckling experiments. The UD nature of the curvature at failure and image-acquisition-equipment’s depth of field
shells allows a clear isolation of the failure mechanism and material that was used for the Digital Image Correlation (DIC). Consequently, the
behavior. The overall analysis, the developed modeling scheme and 18- and 25- ply specimens were longer. A fine airbrushed white paint
corresponding results contribute to the robust design of foldable com­ layer was applied in all specimens on both sides, before applying a
posites structures. The proposed failure modeling can be used for either special stamp to randomly introduce black speckles (~170 μm2) that
purely UD FRPs or within first ply failure criteria. UD FRPs are expected enable the DIC strain measurements.
to reach the maximum non-complex curvatures [16], nevertheless, they
will not be able to sustain high complex curvatures due to the reduced
2.2. Testing and strain acquisition
transverse strength.

The thin shell-buckling configuration is illustrated in Fig. 2b. This

2. Experimental campaign shell buckling configuration allows for very high curvatures without
complicated testing rigs, while the compressive forces are negligible
Thin shell flexure experiments, of UD carbon-epoxy thin coupons, compared to the dominant buckling induced bending [2]. Two
were conducted using the shell-buckling test configuration [1,17]. The aluminum plates were mounted in a Zwick Roell universal testing ma­
testing process followed is also meticulously described in Ref. [2], chine equipped with a 5 kN load-cell to measure the force, F. The ex­
nevertheless, an outline of the essentials is included herein for periments were conducted at a constant displacement rate of 10
completeness. mm/min until coupons’ ultimate failure. In order to evaluate the strains
and curvature on coupons’ surfaces, two dual camera setups (4 × Pro­
2.1. Material and specimens Silica GT3400) were employed for the image acquisition (2 Hz framer­
ate). A symmetrically inclined positioning enabled the essential 3D DIC
Six different specimen series were produced from seven different scheme that can cope with the occurring large displacements. The ac­
plates with 4-, 5-, 10-, 13-(two plates: a & b), 18- & 25- plies, made of 30 quired images were post processed using VIC-3D v7, to evaluate
gsm thin ply Toray T700S carbon fiber prepreg, with TP402 epoxy resin compressive and tensile strains on x-axis, εxx,c & εxx,t respectively, as
(NTPT Switzerland), manufactured following the recommended auto­ well as the corresponding curvatures, κc & κt enabling the evaluation of
clave cycle (initial dwell at 80 ◦ C for 30 min and 2 h curing at 135 ◦ C the midline curvature, κ (see also [2]). All in all, the 3D DIC technique
with 5 bar pressure) [18]. To assure full control of Vf , the curing was allows strain acquisition and curvatures of very high range in a remote
done between two polished steel plates, while fine, hardened manner (e.g. strain gages add thickness comparable to the specimens).
stainless-steel, precision gauge strips prevented over-compaction and
controlled the thickness. A smooth surface finish, was provided by the 2.3. Characteristics of experimental response
use of 50 μm thick Polyimide release film (Airtech Thermalimide RCBS)
[19]. Breather cloth or peel-ply were absent in this process as they would Pictures of representative 10-ply and 18-ply specimens, acquired
introduce considerable surface and thickness imperfections on the thin with the image acquisition setup, are shown in Fig. 3. Specimens up to

Fig. 2. a) Representative cross-section micrographs on fabricated specimens of 5 & 13 plies. A fine layer of acrylic paint used for DIC analysis is indicated. b)
Schematic of the experimental shell-buckling configuration with the camera setup for the DIC analysis, illustrated. A photograph of an actual experiment is shown on
the bottom-left corner.
G.A. Pappas et al.

Fig. 3. a) & b) Images from the camera setup illustrating the concurrent state from tension and compression side of a representative 10- (top) and 18-ply (bottom)
specimen.
c) F − κ plot of the 10- and 18-ply specimen shown in a) & b). The exact points that the snapshots in a) & b) were captured, are also indicated. The part before the
linearity domain corresponds to the initial buckling and the loading until specimen tangency obtained.
d) Detail of the bottom-right corner focusing on a buckled fiber bundle (indicated by a red arrow) that forms at an angle that the imaging setup can capture well.
The 10-ply specimen shows no discernible damage on the compressive side when damage is pronounced on the tensile side. On the contrary, the 18 ply specimen
shows discernible damage, in the form of fiber-microbuckling on the compressive side, before any damage occurs on the tension one. Some tensile damage occurs
only after the appearance of several buckled fiber bundles (‘fiber-blisters’) on the compression side, shortly before ultimate failure. For further interpretation of the
behavior and the sequence of events the reader is referred to the SM (document and videos). (For interpretation of the references to colour in this figure legend, the
reader is referred to the Web version of this article.)

10-ply-thickness do show any discernible damage on the compressive load increases, shown in Fig. 2 (see also [2]). The failure events (tensile
side at the moment that pronounced failure occurs on the tensile side, fiber breakage or compressive microbuckling) are indicated by the
evidenced by fiber tensile breakage, which later leads to distinct fiber sudden load drops.
split, attributed to fiber-matrix failure and pure matrix cracking. This For better interpretation of the experimental characteristics and
can be seen with simple visual observation of the acquired pictures sequence of failure events, representative videos are included as sup­
(Fig. 3) but also from the strain redistribution that can be evaluated plementary material (SM), along with a SM document with further
during the DIC analysis (see also [2]). experimental results and discussion. The critical strains measurements,
On the contrary, the thicker specimen families such as 13 or 18-ply as well as curvature at failure for all tested specimen families are re­
and above show a response similar to what is expected for specimens ported in §4, along with the comparison of the developed modeling
of conventional thickness (e.g. 1–2 mm according to norms [20,21]). In scheme.
detail, in these thicker specimen families, compressive damage precedes
and leads the failure. The damage comprises compressive fiber micro­ 3. Mechanical analysis and modeling
buckling as also manifested in Fig. 3b and d. The microbuckled fiber
bundles, form characteristic ‘fiber blisters’ on the surface, comprised by In order to identify the reasons why compressive micro-buckling/
fiber bundles separated from the central domain of the material Fig. 3d). shear failure is suppressed in very thin shells, the mechanical analysis
Since the specimens are loaded predominantly under bending in this initially focused on the homogenized stress field around the φmax domain
shell buckling configuration (see also [2]), the fibers close to the (Fig. 1f), acknowledging that thicker specimens followed the expected
compressive surface are expected to suffer the highest stresses. Notably, shear induced fiber microbuckling explained in §1. A linear-elastic
transverse cracking (fiber splitting) is not observed as leading damage approach allowed an analytical evaluation. FE-numerical micro-me­
event and this is attributed to the limited development of Poisson’s ef­ chanics depicted the local stress field. Finally, the derived model was
fect analyzed in Ref. [2] (refer also to Section 1). implemented in a numerical algorithm to incorporate the carbon fiber’s
Compared to pure compression load cases where failure of the elastic, material non-linearity. The estimated stresses are compared with
microbuckled fiber-bundles leads to the characteristic kink-band for­ the strength of the matrix to predict failure.
mation, discernible kink-bands will not form in such thin shell bending
load case. This is attributed to the fact that compressive damage forms 3.1. Properties of constituents and homogenization
shortly before ultimate failure, and originates mainly on the surface,
where the buckled bundles can freely deform, resulting in discernible During the analysis, the matrix (TP402 epoxy resin see §2) was
blisters. Moreover, the bending stress/strain gradients through- considered isotropic with a modulus Em = 3.44 GPa [18], while a shear
thickness allow for high compressive strains/stresses on the fibers modulus Gm ≃ 1.27 GPa was evaluated for a typical Poisson’s ratio νm =
close to the surface leading to a limited zone with compressive failure. 0.35. At this stage of the analysis, a nominal tensile fiber modulus is
The snapshots illustrated in Fig. 3a and b are captured shortly before considered (chord-modulus), E11,t = 230 GPa, and the nominal longi­
ultimate failure, as seen in the F − κ plot in Fig. 3c. The F− κ plot depicts tudinal tensile strain to failure is εf,y− t = 2.1% [22]. Nominal fiber
an initial increase in load with increasing curvature once contact be­ diameter df , is 7 μm. The longitudinal composite’s modulus E11 , was
tween plate and specimen is established. This is followed by the buckling evaluated using conventional rule of mixtures (ROM), when
( )
of the specimen (relatively flat part), which is then followed by the linear-elastic response was considered, i.e. E11 = Vf ⋅E11,t + 1 − Vf ⋅Em .
domain where the shell is tangent to the two aluminum plates where the The remaining engineering constants were approximated using
G.A. Pappas et al.

Halpin-Tsai micromechanical corrections [13] (ξ1 = 2 for E22 , ξ2 = 1 for Hence, Eq. (1) can be reformulated to predict the bending strength of
G12 ), with fiber properties by Ref. [23]. The corresponding calculated a UD composite as:
engineering constants for Vf = 60% are: E22 ,E33 = 8.2 GPa, G12 ,G13 = τc,y
σb,y = (ϕ ,
4.26 GPa, G23 = 2.68 GPa, ν12 , ν13 = 0.31 & ν23 = 0.43. 4)
’ )⋅ξ
max +ϕ
/
2y0 ( )
with ξ = 1 − & y0 = df + tm 2 + y0
3.2. Meso-/micro-mechanics of shear failure micro-buckling under t
bending The effect of thickness in Eq. (4) is illustrated in Fig. 4b. The insert in
Fig. 4b provides a schematic representation of the stress gradient vari­
In order to explore the load-state at the maximum shear region of a ation, due to thickness, at ϕmax region. To produce Fig. 4b, representative
typical sinusoidal fiber misalignment, i.e. at ϕmax , a ‘wavy’ fiber of values of fiber waviness were assumed with y0 = 15 μm & λ = 750 μm
diameter df is considered, along with a matrix-rich zone of thickness tm . [9] resulting in ϕmax ≃ 0.078 rad (4.45◦ ). The interfiber, matrix-rich
Under pure compression, the domain around ϕmax is exposed to a zone was evaluated for a typical hexagonal fiber packing by:
compressive stress field σc (homogenized), as in Fig. 1e and f. When the (√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ )
domain around ϕmax is exposed to pure bending conditions (see Fig. 4a), π
tm = df √̅̅̅ − 1 (5)
according to Euler-Bernouli beam-theory for a linear elastic material, a 2⋅ 3⋅Vf
linear through-thickness stress distribution, σ(y), occurs, of a slope α =
t /(σmin − σmax ), with σ max = |σ min | and α = − 1/(κ ⋅E11 ), with κ denoting For Vf = 60% and df = 7 μm (for T700S), tm = 1.6 μm, thus, y0 =
the implemented curvature. Here, t is the thickness & σmax , σmin are the 19.3 μm. Composite’s τc,y was assumed equal to the ultimate shear
maximum tensile and compressive stresses respectively. strength of the matrix, τm,max . This approximation accounts for a fiber-
When the bent section is thick enough, it can be considered that the matrix interface stronger than the pure matrix, under homogeneous
compressive stress at ϕmax is practically σ min explaining why the flexural shear. The strength τm,max was estimated as 71 MPa, based on the
strength is controlled by composite’s compressive failure in much experimental data in Ref. [24], for an epoxy system with very similar
thicker coupons (e.g. 1 or 2 mm as in norms [20,21]). Nonetheless, ϕmax properties to TP402 (tensile strength 62 MPa & flexural strength 148
appears at a distance y0 = (df + tm )/2 + y0 , from the free surface MPa, according to Ref. [18]). The adopted τc,y value is in the range of
(Fig. 4a & Eq. (2)). Thus, the compressive stress at ϕmax shall not be measured shear strengths for such CF-epoxy systems [13]. It should be
considered equal to σ min when y0 is comparable to the distance from the noted that the ultimate shear strength can be much higher due to fibers’
neutral axis, yna (= t/2 for linear elastic case). Typical values for y0 vary realignment when the ±45 tensile test value is considered [25]. Finally,
from 1 to 10 × df , or ~7 to 70 μm for df ≃7 μm [6,8,12], hence are ϕ’ was considered equal to the shear strain γ c,y , which the composite will
important for thin shells of roughly t ≤0.5 mm, while λ lies between 100 show at τc,y = τm,max . At the ultimate shear stress (strength point),
and 400 × df . matrix’s secant shear modulus due to plastic deformation is ∼ 0.5Gm
Considering the very first misaligned fiber from the surface in the 2D [24]. The use of the latter leads to a homogenized secant shear modulus
scheme illustrated in Fig. 4a, it is anticipated that the shear stress (that β⋅G12 , with β ≃ 0.55. This approach renders ϕ’ = γ y,c =
will cause shear yielding) around a fiber on the matrix-rich zone is τy /(β⋅G12 ) ≃ 0.03. The evaluated β-ratio that depicts the modulus’
proportional to the compressive stress, as in Ref. [5]. Thus, the shear reduction due to yield, can be compared with the 0.7 proposed in
stress gradient should be equal to the gradient of the compressive ones. Ref. [10].
Hence, the effective compressive bending stress that will cause shear As depicted in Fig. 4b, the bending strength due to shear failure on
yield at ϕmax for a linear stress distribution (y-axis) is: σb (y0 ) = (σ1 +σ 2 )/ the compressive side by Eq. (4), tends to infinity as ξ tends to zero, i.e.
2 (Fig. 4a), while the Mean Value Theorem allows to estimate this for very small thicknesses where y0 approaches the bending neutral axis.
effective stress as σ b (y0 ) = 1/α⋅(yna − y0 ), with 1/α, the stress gradient. This increase of the resistance to compressive failure for low thickness
This local compressive bending stress can be evaluated by the minimum allows the stresses on the tensile side to drive the failure when the
stress due to bending (on the surface), reduced through the parameter: critical strain (here 2.1%) is reached, as shown in Fig. 4b. Thus, a
ξ = 1 − 2y0 /t to incorporate the stress gradient. Therefore, the critical bifurcation point/critical thickness can be identified on the bending
shear is defined as: strength vs. thickness diagram, at which bending failure switches from
tension-driven to compression/shear-driven. Such a behavior is
τ = ξ⋅σ min ⋅ϕmax (3)

Fig. 4. a) Homogenized compressive and shear stress around a representative ‘wavy’ fiber with the domain of the maximum waviness angle, ϕmax , indicated, for pure
bending conditions. b) Bending yield strength of a UD composite laminate as a function of thickness by Eq. (4), i.e. linear elastic material. In the dashed part (beyond
~2930 MPa) tensile failure precedes micro-buckling. The insert shows a schematic of the stress profile around a wavy fiber domain, as a function of thickness, for a
given maximum stress on the surface (not to scale).
G.A. Pappas et al.

captured in thin CF reinforced shells in Refs. [1–3] and herein for a convergence study. The model was generated using μm & N units, to
specimens up to 10-ply (~335 μm), where the failure is driven by fiber’s increase the accuracy of the numerical solver.
tensile strength. Use of Eq. (4) along with the condition for tensile failure
provides a bending failure vs. thickness model for linear material 3.3.2. Modelling results and interpretation
approximation. The resulted εxx and γ12 profiles on the matrix, for three different
modeled thicknesses are shown in Fig. 6a & b. As illustrated, the higher
the thickness, the higher the normal strains and consequently the shear
3.3. Numerical micromechanics ones on the ϕmax region. The depicted strain concentrations relate with
the tendency for additional fiber misalignment (see ϕ’ in Fig. 6c).
3.3.1. Embedded cell model description
In order to evaluate the exact stress-state on the critical matrix-rich
In order to verify the validity of the assumptions considered to form
zone, the region around fiber ① (indicated in Fig. 6b) was considered
the stress gradient dependent parameter ξ, a 3D numerical finite element
as reference. This fiber is not the most critical one, yet it is surrounded
(FE) model of the area of interest around ϕmax , as illustrated in Fig. 5,
by a perfect hexagonal packing and not influenced by boundary effects
was created in Abaqus Standard v6.14. This model adopts the embedded
of the embedded cell (contrary to the top-most one). For location ① , the
cell approach [26]. A cylindrical, wavy (y0 = 15 μm, λ = 750 μm)
equivalent distance from the surface is y0,① ≃ 1.5⋅(df + tm ) + y0 . The
fiber-bundle (7 fibers of df = 7 μm) of a perfect hexagonal packing (Vf =
shear stress profile in local coordinates (1-2-3 in Fig. 5), τ12,m , within the
60%) was modeled explicitly and embedded in a homogenized com­
matrix around fiber ① is shown in Fig. 7a-Top as a function of central
posite section with a perfect interface. The modeled domain around φmax
angle (Fig. 6b) and depicts a clear thickness effect. Remarkably, for very
had a length (x − axis) equal to λ/5, a 75 μm height (y-axis) and a 250
low thicknesses, the peak stresses around the fiber are significantly
μm width (z-axis), large enough to provide uniform strain.
reduced, even for a dy = df (see also Fig. 6a & b), while for thicker
For this model, linear elastic material and small deformations were
specimens this difference vanishes, demonstrating that the stress
considered to allow direct comparison with Eq. (3). The homogenized
gradient effect becomes negligible after certain thickness.
domain and the fibers were modeled as orthotropic materials with the
Since typical polymer matrices are expected to have an elastoplastic
properties mentioned in §3.1 and [23] respectively, while the matrix
behavior [24] under pure shear loading, it can be assumed that insta­
was isotropic with Em , vm (see §3.1). The embedded cell/fiber-bundle is
bility and fiber micro-buckling are triggered when most of the matrix on
inclined by the ϕmax angle (i.e. z- & 3- axis coincide in Fig. 5). Different
fibers’ circumference has entered the yielding regime. As seen in
linear strain distributions with a common minimum nominal strain on
Ref. [27], ~80% of the material around a fiber bears damage before
the surface, εb, min = − 2% and slopes/curvatures κ = − 2⋅εb,min /t , for
instability. In this framework, the 2D approximation by means of Eq. (3),
t = 75, 100, 150, 200, 500, 1000 & 2000 μm, were imposed on the ( )
modeled domain. These nominal strains were imposed via symmetric τ12,c = σ b,min ⋅ϕmax ⋅ξ y0,① (for homogenized shear stress field) is
compressive displacements on the x-direction, ux , applied on faces A & B compared in Fig. 7b with the effective shear stress around fiber ① from
∫Γ
(face normals also in x-direction, see Fig. 5) providing homogenized the embedded cell model, calculated as τ12,m− eff = 1/Γ⋅ 0 τ12,m (Γ) dΓ,
engineering strains as εb = 2ux /(λ /5). This allowed for the imple­ where Γ = π df , i.e. the circumference of the fiber. Fig. 7b shows that the
mentation of an equivalent strain profile that conforms to the ones ex­ parameter ξ can capture the thickness-effect tendency, but shows that
pected for this bending case. This framework allowed to evaluate the τ12,c is way higher than the τ12,m− eff . This shows that, even-though the
load-state of the φmax -region for different thicknesses. For complete­
local shear on the matrix is driven by the compressive stress, is also
ness the modeled domain was also imposed to pure compression con­
highly affected by the complementary stresses that contribute to the
ditions with εb, min = εc = − 2% (see Fig. 5), corresponding to t→ ∞.
local equilibrium [10] (see also Fig. 1f). The latter is highly affected by
No further constrains were implemented on the remaining faces to the orthotropic nature of the fibers as depicted in Fig. 6d where τ12 and
allow the full development of the Poisson’s effect. The selection of εb, min τxy are shown.
is indicative and cannot affect the validity of the results, since, small
Detailed stress analysis on the domain around ① , included in the SM
deformations (linear geometry) were assumed. The models were dis­
document, showed that the maximum shear stress on this matrix zone, is
cretized in ~1,800,000 linear, reduced integration brick elements
about 30% higher than the minimum principal (compressive) stresses
(Abaqus C3D8R) with the embedded cell having a finer mesh with
and more than 5 × higher than the maximum principal (tensile) ones.
typical element edge of ~0.4 μm. This meshing pattern was chosen after

Fig. 5. Embedded cell approach applied to investigate the effect of bending stress & strain gradient at the
‘ϕmax - domain’ of a sinusoidal fiber misalignment; maximum homogenized applied strain: εb,min = − 2%. The strains were imposed via symmetric compressive
displacements on the x-direction, applied on faces A & B.
G.A. Pappas et al.

Fig. 6. Results from the embedded cell model at the ‘Reference area’ indicated in Fig. 5:
a) Normal strains, ε12 & b) Engineering shear strains, γ12, results on the matrix for t = 100, 200 & 2000 μm. c
) Shear deformation γ12 for pure compression, (t→∞). d) Resulted τxy & τ12 distribution, for t = 200 μm.

Fig. 7. a) Shear stress on matrix τ12,m & τmax,m defined by the principal stresses (from embedded cell model, see also text) around fiber ① shown in Fig. 6b vs. center
angle. b
) Shear stress based on Eq. (3) for εb,min = − 2% and y0 = 19.1 μm at ① vs. τ12,m & τmax,m . The corresponding χ 1 factor is also shown with reference to the secondary
plotting axis. The FE model is linear elastic with small deformations, thus ϕ’ due to yielding is considered zero.

Thus, the maximum shear stress shall drive yielding, having accounted effects and its results showed that the non-linearity is well captured
for the typical failure behavior of a polymer [24]. The distribution of the (~2% tolerance) by Budiansky’s hypothesis for the contribution of the
maximum matrix shear stress around ① τmax,m , is shown in Fig. 7a additional misalignment due to loading [7], i.e.
(bottom) and the effective one as a function of thickness, τmax,m− eff , τ/τNL ≃ ϕmax /(ϕmax + ϕ’ ). However, ϕ’ from FE (see Fig. 6c as in Fig. 1e)
calculated via integration similar to τ12,m− eff , in Fig. 7b. The latter fol­ is about 11 × lower than the estimation by τ12,c /G12 , suggesting a 2nd
lows the same trend as the 2D approximation -τ12,c -, but is lower by a micromechanical factor, χ 2 = 11. This difference originates from the
(mainly) constant factor χ 1 = τmax,m− eff /τ12,c ≃ 0.58, also shown in fact that the misaligned fiber domain is not free to shear, but is
Fig. 7b. For very low thickness, χ 1 has lower values since y0,① →yna and constraint by the surrounding continuum composite. This interaction
the boundary of the embedded cell tends to the neutral axis. was also considered in Refs. [10,11]. In summary, the results of the
For completeness, the simulation of the pure compression equivalent embedded cell model can be employed to revise Eq. (4) as follows:
(t→∞) was re-launched accounting for geometrical non-linearity (NL)
G.A. Pappas et al.

τy /χ
σ b,y = (ϕ +ϕ1’ )⋅ξ,
(6)
max
/ ( / ) /
with ϕ’ = γy,c
χ 2 = τy χ 1 [(β⋅G12 )⋅χ 2 ], χ 1 = 0.58, χ 2 = 11
2y0 ( / )
& ξ= 1− , with y0 = df 2 + tm + y0
t

Here, a first order approximation is done for the contribution of the

additional misalignment due to loading and plastic deformation, φ ,
′

which is assumed to be equal to the misalignment caused by plastic

shear deformation of the matrix at the maximum stress, reduced by the
constrains implemented by the surrounding material.
The factors χ 1 (micromechanical stress) & χ 2 (micromechanical
strain) represent the investigated hexagonal packing with Vf = 60%
and are expected to describe the local stress state for material combi­
nations with similar elastic properties (i.e. orthotropic fibers and poly­
mer matrix). Sensitivity analysis of these factors over Vf , fiber-packing
pattern, matrix and fiber type may be the subject of future studies. Fig. 8. Evaluated single CF and CF-epoxy UD composite longitudinal stress-
strain curve by Eq. (7b) and ROM-homogenization (for composite). Note that,
no compressive damage is considered on the homogenized curve while the
3.4. Compressive shear failure for non-linear material & experimental dashed line corresponds to strains beyond 3% that have not been observed
verification experimentally [2].

3.4.1. Non-linear material model implementation gradients due to bending, as well as the position of the neutral axis [2],
Important efforts have been applied in the past to provide a material thus the parameter ξ is not representative any more. For pure bending
modeling scheme able to represent the CF non-linear material behavior conditions, according to Euler-Bernoulli beam-theory, a linear strain
[1,3,14,15]. Recently, Schlothauer et al. [2] demonstrated that the distribution is expected in the form: ε11 (y) = − κ⋅y, as illustrated in
formulation proposed by Northolt et al. [14], based on the load induced Fig. 9a. This is also valid for non-linear material behavior as confirmed
realignment of the graphite layers/planes, can be employed to depict the by the measured κ in the experimental campaign (see also [2]). Hence,
full extent (tension & compression) of the longitudinal elastic response the neutral axis’ shift can be evaluated by the equilibrium condition on
∑ ∫t
of CF and consequently CF-composite. In detail, on fiber level, the lon­ x-axis, i.e. Fx = 0 or Q = 0 σ (̂ y ) d̂
y = 0 with ̂y measuring from the
gitudinal elastic strains ε11f , relate with the applied stress σ 11f , following tensile side as in Fig. 9a. Effectively, the non-linear behavior of CFs
Eq. (7a) by Ref. [14]: causes a neutral axis shift towards the tensile side.
( )[ ( )] In order to define the bending failure point accounting for the
ε11f = σe11,f1 + g⋅ E10 − e11 ⋅ 1 − exp − σ11f , aforementioned material non-linearity, a numerical algorithm was built
(7a)
g
in Matlab® 2019a. Similar to the approach in Ref. [2], the
E0 , e1 , g →const.
through-thickness stress distribution under pure-bending, for a given
Here, E0 is the initial longitudinal fiber modulus, e1 is the graphite thickness t and each κ- increment, is defined by applying ε11 (y) in Eq.
layers/planes’ in-plane modulus and g is the shear modulus between the (7b), to define σ11f (y) and consequently σ 11 (y) by homogenization.
graphitic layers. The inversion of Eq. (7a) can only take place by use of Then, the neutral axis position is defined by minimization of Q→0 (Q is
the Lambert function W(x) as: evaluated by the trapezoidal rule). Consequently, the bending
( ) (compressive) stress at y0 , σ b (y0 ) is compared with the anticipated
σ11f ε11f = E0 ⋅g⋅W(x)+E0 ⋅eE10⋅ε11f +E0 ⋅g− e1 ⋅g, strength, σ b,y from Eq. (6) (without ξ). The matrix yield-dominated
(7b)
(e1 − E0 )⋅e ( 0 1 11f 0
− E ⋅ e ⋅ ε +E ⋅ g− e1 ⋅ g)/(E ⋅ g)
0 properties τy and β in Eq. (6) are evaluated as described in §3.2. If
x= σ b (y0 ) ≥ σ b,y , or σ b− max ≥ σ y,t (σy,t : tensile strength), then a curvature at
E0
yield/failure, κy , is identified for a given t. The flow diagram of the al­
As described in Ref. [2], utilization of single fiber experimental data
gorithm is illustrated in Fig. 9c. In order to have a high sensitivity on κy
[28,29] allows for the evaluation of the constants in Eq. (7) for T700S
per thickness, the thickness was discretized by d̂ y = t/600 and the κ-
fibers, rendering the following values: E0 = 216 GPa, e1 = 228 GPa &
increment was 0.00025 mm− 1. Curvatures with κ ∈ [0, 0.7] mm− 1 were
g = 0.588 GPa. Then, use of Eq. (7b) and ROM provides the longitudinal
implemented based on the experimental values (refer to §4). Indica­
stress of a UD composite as σ 11 = Vf ⋅σ 11f (ε11 ) + (1 − Vf )⋅ Em ⋅ ε11 . The
tively, the through-thickness stress profile for t = 285 μm is depicted in
resulted longitudinal stress-strain response for a single fiber and the
Fig. 9b and compared with the linear hypothesis.
CF-composite are depicted in Fig. 8. Compressive strains up-to 5% are
included in Fig. 8, since according to Refs. [28,29], CF elastic recovery is
4. Verification of the model with experimental results
expected at least up to this strain level and at least up to ~2.6% for the
UD CF-composite as tested in Ref. [2]. Micro-buckling failure is not
4.1. Fiber misalignment analysis
considered in Fig. 8. Note that any effects of compressive plastic yield on
the matrix, that may appear at ~2% strain [24], are masked in the ho­
A fiber misalignment analysis was conducted following the tech­
mogenization, as the stiffness of the fibers dominates in the longitudinal
nique described in Ref. [30]. The details and the parametric study that
direction, as also stated in Ref. [2]. The tensile strains are restricted by
accompanied the analysis are included in the SM document. In brief, this
the tensile failure of CFs.
analysis allowed the evaluation of the fiber angle distribution using
material’s sections (see Fig. 10a) in fiber direction and an image pro­
3.4.2. Implementation of non-linear material response on the failure model
cessing algorithm. These sections were done after the test in the vicinity
Linear elastic material response was considered for the derivation of
of failed areas, but far enough to ensure intact sections. The traced fiber
parameter ξ that incorporates the stress gradients’ effects. However, as
angle θi , allows for the evaluation of the fiber misalignment measure as.
illustrated in Fig. 8, a very important elastic compressive softening is
The statistical analysis showed, as expected, that mean fiber angle is
expected for certain fibers including the T700S, affecting the stress
G.A. Pappas et al.

Fig. 9. a) Schematic of through-thickness strain profile in pure bending conditions with non-linear material.
b) Through-thickness stress profile in pure bending conditions for linear (E11,f = E0 = 216 GPa, dashed) and non-linear material, for t = 285 μm. Also indicated is y0
for y0 = 15 μm, λ = 120 × df and Vf = 60%.
c) Yield point tracking algorithm for non-linear material response based on Eq. (7b) and Eq. (6) without ξ, or ξ = 1.

Fig. 10. a) Example of a 10-ply section microscopy used in the analysis based on the technique described in Ref. [30] and the SM. b) Resulting fiber angle and
misalignment distribution for the illustrated micrograph. More details are found in the SM.

practically zero (θ ≃0◦ ) while the mean standard deviation σ θi ≃ 1.95◦ , 4.2. Model vs. experimental results
which is in the range of findings of older studies (0.8◦ to 1.9◦ in
Ref. [31]). The curvature at failure, κy , per specimen family (identified as
The summary of the most important results are included in Table 1. A explained in the SM document) is shown in Fig. 11a. The plotted cur­
mean misalignment angle from specimens of all tested thicknesses is vature values correspond to the curvature of the central region of the
ϕ ≃1.42◦ . As explained in the SM document, the said fiber-waveform buckled shell, where the maximum curvature is expected, considering
characteristics are representative ones that can be traced in micro­ the expected buckled ellipsoidal shape. The corresponding tensile and
scopy on the tested composite system; nonetheless, a statistical analysis compressive strains (by DIC) at failure are shown in Fig. 11b. Table 2 is
of waveforms is a very challenging task, also susceptible to the fiber summarizing the curvature at which failure occurs, the corresponding
tracing method (e.g. methods in Ref. [30] vs [31]). strains on the compressive and tensile surface, as well as the damage
mechanism that triggers the failure. The thickness that compressive
failure starts to precede for this material and Vf = 58.5% is ~420 μm
(13a-ply), as seen in Table 2. Some specimens of this thickness showed
Table 1
almost concurrent tensile and compressive failure. For the 13b-ply
Angle alignment and misalignment values, using microscopy image processing.
The reported ϕmax corresponds to the 95th percentile of angle misalignment. specimen with higher compaction and Vf = 61%, the compressive
The confidence intervals refer to measurements taken from different positions failure preceded in all specimens. In thinner specimens the tensile failure
along the strips cut from the specimens; refer to SM document for more details. precedes, at strains around the tensile limit of the fibers, while in thicker
5-ply 10-ply 13-ply 18-ply 25-ply Mean
ones the compressive microbuckling led to damage.
The corresponding curvature and strain response per thickness using
σθi [◦ ] 1.9 ± 2.05 ± 2.0 ± 2.05 ± 1.85 ± 1.95 ±
the algorithm in §3.4, with T700S/TP402 properties (found in §3.1 &
0.2 0.2 0.15 0.05 0.05 0.1
ϕ [◦ ] 1.33 ± 1.50 ± 1.44 ± 1.49 ± 1.31 ± 1.42 ± §3.2) and for a waviness of y0 = 15 μm, λ = 120 × df that leads to a
0.2 0.2 0.1 0.1 0.3 0.2 ϕmax ≃4.1◦ , are also shown in Fig. 11. This value corresponds to the 95th
φmax estimator 3.95 ± 4.15 ± 4.15 ± 4.35 ± 3.85 ± 4.1 ± percentile of angle misalignment according to the analysis. Although
[◦ ] 0.4 0.45 0.3 0.05 0.2 0.3
this angle appears high, it should correspond to a peak of misalignment
G.A. Pappas et al.

Fig. 11. a) Curvature to failure κy vs. thickness; shell-buckling experiments (mean & Std.) vs. prediction with the failure tracking algorithm depicted in Fig. 9c for
non-liner material. Also shown is the sensitivity to Vf and the linear material approximation (E11,f = E0 = 216 GPa). b) Compressive & tensile strain on the surface at
κy vs. thickness; shell-buckling experiments (mean & Std.) vs. prediction for different Vf .

Table 2
Summary of experimental curvatures to failure & corresponding strains (tensile and compressive), as well as leading failure type. Note that the 13b-ply has a Vf = 61%.
These values are averages of 4–6 specimens per family. The standard deviation is also included.
4-ply 5-ply 10-ply 13a-ply 13b-ply 18-ply 25-ply

t [μm] 130 ± 4 164 ± 3 335 ± 4 420 ± 7 403 ± 7 589 ± 7 800 ± 6

κy [mm− 1] 0.443 ± 0.013 0.359 ± 0.018 0.163 ± 0.005 0.122 ± 0.005 0.125 ± 0.005 0.082 ± 0.002 0.058 ± 0.006
εxx, t [%] 2.02 ± 0.06 2.07 ± 0.06 2.09 ± 0.04 2.17 ± 0.13 1.93 ± 0.07 2.02 ± 0.06 2.01 ± 0.1
εxx, c [%] 3 ± 0.3 2.98 ± 0.1 2.92 ± 0.13 2.9 ± 0.14 2.48 ± 0.14 2.59 ± 0.09 2.60 ± 0.15
Leading failure type Tensile Tensile Tensile Tensile/Compressive Compressive Compressive Compressive

expected to lie on the maximum inclination of a fiber waveform. The microbuckling failure occurs when a highly misaligned fiber/fiber band
video of the 18-ply specimen in the SM, illustrates that isolated ‘fiber- (herein Gaussian distribution) is exposed to critical compressive strains.
blisters’ appear in the surface when the first load drop occurs, being For specimens where compressive damage (micro-buckling) leads to
followed by more fiber-microbuckling. Shortly after these events, ulti­ catastrophic failure, blisters become visually discernible the moment the
mate failure follows, since the low thickness of the material cannot load drop occurs. This is in agreement with the findings of [12] which
tolerate even a small amount of damage. The aforementioned sequence show that instability of a critical/misaligned fiber-band, occurs soon
of compressive failure events is affected by the local variation of fiber after matrix shear yield initiates (matrix elasto-plastic response was
waviness, which results in randomly formed blisters along the available modeled in Ref. [12]). Nonetheless, the yielding on the microscale of the
surface. Therefore, the ϕmax estimator that fits the experimental results, critical/misaligned fiber-bands that leads to the first blisters, is not
reflecting only a probability of 5% where fiber misalignment is of captured on the F − κ plots (see Fig. 3 and SM document) and is only
greater value than 4.1◦ , is a reasonable estimator assuming that a limited shown when micro-buckling and blister formation appear.
number of microbuckled fiber can shortly lead to ultimate failure. For completeness, the prediction with the linear material model and
Remarkably, the modeling scheme is successfully predicting the a nominal fiber modulus E11,f = E0 is also shown in Fig. 11a. For small
switch of the failure driving mechanism from tensile fiber rupture to thicknesses with tensile failure, their differences are limited to ~15%.
compressive/shear yielding. This corresponds to the bifurcation points However, when compressive failure becomes the driving failure mech­
on the strain vs. thickness plot in Fig. 11b. In detail, before the bifur­ anism, the differences are much more pronounced. Moreover, the switch
cation point, tensile fiber breakage controls failure and the maximum to compressive failure predicted with the linear model for Vf = 60% is at
strains are thickness independent, while after this point the maximum t ≃100 μm (κy ≃0.10 mm-1), while for the actual, non-linear elastic fiber
strains decrease in a hyperbolic manner, which approaches an asymp­ behavior is at t ≃285 μm (κy ≃0.17 mm-1). The origin of this difference
tote around 1 mm. Moreover, the modeling scheme is able to capture the is depicted in Fig. 9b for t ≃285 μm, where the compressive through-
effect of Vf (assuming unchanged fiber waveforms characteristics) on thickness stress-gradient and stress at y0 can be seen.
the prediction of the failure tendency per thickness, as seen in Fig. 11a
and b. The small difference in the maximum compressive strains at 5. Discussion and broader aspects
failure is attributed to the pure-bending approximation in the model
instead of buckling induced bending (see also [2]). Yet, these small As thoroughly analyzed in §1, the choice of fiber waveform charac­
differences in Fig. 11b, are within the DIC results’ scatter & the confi­ teristics has a vast impact on the compressive/shear yield point pre­
dence bounds of the non-linear material fit with Eq. (7a). The failure diction. To address this, the prediction algorithm was launched with
mechanism switch points are indicated in the κy (t) plot (Fig. 11a). There, different ϕmax values that result from variation of λ and a common y0 =
no discernible curve bifurcation occurs. 15 μm. The results are shown in Fig. 12a. Remarkably, for ϕmax ≤3.9◦ ,
The experimental evidence shows that failure will mainly occur at a fiber tensile rupture is expected to drive failure almost up to t = 0.8 mm.
region around the central domain of the specimens, where the bending In contrast, when a slightly higher waviness is considered i.e.
curvature is the highest. As a result, on one hand, fiber tensile failure is ϕmax ≃4.45◦ (as in the embedded cell mode and in Ref. [9] as well as, in
expected to occur when a weak fiber (e.g. following a Weibull distri­ the range reported in Refs. [6,8]) compressive yield is expected to drive
bution) is subjected to its ultimate tensile strain. On the other hand, failure for t ≥115 μm. It is worth mentioning that for ϕmax ≃4.45◦ and
G.A. Pappas et al.

Fig. 12. a) Compressive & tensile strain on the surface at failure vs. thickness; prediction for different ϕmax (Vf = 60%).
b) Bending stiffness reduction with respect to linear material approximation (E11,f = E0 = 216 GPa) vs. thickness and sustained curvature until κy .

t = 1 mm, the maximum compressive strain at failure is ~2.0%, a value 1 mm thick shells under pure moment, the presented failure prediction
almost identical to the one reported in Ref. [32] for another PAN based scheme would suggest an apparent bending strength of ~2 GPa (based
CF (T800S)-epoxy system. In general a higher fiber ‘representative’ on the total moment and assuming linear stress through thickness).
waviness shall be expected for the conventional fabrication processes
with peel-ply or breather compared to the herein implemented process 6. Concluding remarks
(§2.1) that ensures high flatness (e.g. 4.1◦ vs. 4.45◦ ).
This aforementioned ‘representative’ waviness refers to the through- By revisiting the fiber micro-buckling formation modeling proposed
thickness waviness. The in-plane waviness is not expected to affect the by Argon [5] and refined by Budiansky [7], and by incorporating the
bending behavior significantly, since the ‘wavy’ fibers/fiber bundles will high strain and eventually stress gradients seen in bending of thin
be laterally constrained by their neighbors when the Vf is high enough, fiber-composite shells, the resistance to compressive failure at high
as in out-of-phase micro-buckling in Fig. 1b. Contrary, through- curvatures for low thicknesses, was explained. The key for the me­
thickness ‘wavy’ fibers/fibers and especially close to the surface, have chanical modeling was the consideration of the important stress
no support from the free surface thus, tend to form the experimentally- reduction at ϕmax domain with respect to the maximum one (on the
observed ‘fiber-blisters’ on the surface, once instability occurs (see also surface) for low thicknesses that was described by the stress gradient
SM). This is in-line with the observations in Ref. [30]. parameter ξ. The initial hypothesis for the stress gradient effect was
The ‘representative’ waviness as ϕmax should refer to a critical corroborated by computational micromechanical modeling. The FE
amount of ‘wavy’ fibers that lead to ‘fiber-blister’ formation on the based micromechanical model not only exemplified the validity of the
surface. The spread of the ‘wavy’ fibers is responsible for the experi­ adopted analytical scheme, but also provided micromechanical factors
mental scatter. No particular thickness dependence of the misalignment (χ 1 & χ 2 ) to incorporate the stress and strain variations at the matrix-rich
angle was found for the investigated thickness-families and the adopted domain around a fiber, versus the homogenized field. The effects of the
fabrication process, as seen in Table 1. In general, fiber waviness orig­ microscale have a very high impact on the accurate prediction of the
inates from the intrinsic waviness of the prepreg/fabric and the one behavior, as also discussed in Ref. [10]. The material non-linearity of
induced during the processing. Thus, some increase of the misalignment CFs necessitated a numerical algorithm to evaluate the exact stresses at
level with increasing thickness cannot be excluded, as mentioned in the ϕmax domain, since the closed form solution with parameter ξ for
Ref. [31], for thickness steps of about an order of magnitude higher than linear material (Eq. (4)) is not any more valid. The proposed scheme was
the ones experimented herein (e.g. from 0.8 to 8 mm). However, the able to describe the experimentally observed failure behavior of thin UD
failure model described herein describes a significant strength reduction CF-epoxy shells under high deformation flexion and could explain the
up-to ~1 mm, but the predicted reduction is only ~5% from that switch in failure driving mechanism from compression to tension, found
thickness until 8 mm, assuming a constant ϕmax . A fiber misalignment in very thin shells [1–3]. The proposed failure modeling can be used for
angle increase with thickness, as mentioned in Ref. [31], can explain the either purely UD FRPs or within first ply failure criteria. This failure
bending strength reduction presented in Ref. [32] for thicknesses from 1 prediction scheme along with the non-linear elastic behavior modeling
to 8 mm. described in Ref. [2], can be employed to design durable, composite thin
The fiber misalignment statistics which show, at least for the inves­ shells, subjected to high deformation bending. To do so, material’s
tigated thicknesses, that the fiber angle dispersion is more or less un­ elastic constants are needed (linear and non-linear) along with shear
changed, allows for the use of one representative angle. The fact that the yield strength of the matrix. In the current study, the quasi-static matrix
modeling scheme can capture the experimentally acquired failure with yield strength was considered to model the failure under quasi-static
thickness dependence proves its robustness to capture the main trends. bending. Creep, environmental fatigue or other cyclic effects on the
These trends would have not been captured without the incorporated matrix, can be considered by adjusting matrix’s yield strength (see also
material nonlinearity. The non-linear elastic material behavior leads to [33]). Moreover, knowledge of the ‘representative’ fiber waviness is
reduction of the bending stiffness E11 I with respect to the nominal one required. In summary, all aforementioned material parameters, ‘repre­
evaluated by E11f = E0 . This effect is illustrated in Fig. 12b as a function sentative’ fiber waviness, material non-linearity, as well as micro­
of thickness and κy . This shows that the reduced bending strength of UD mechanical effects are required and need to be incorporated in a
laminates (usually evaluated by the applied moment in standardized thorough modeling scheme to capture the behavior of the tested thin
testing) compared to the tensile one, is due to: i) the compressive micro- shells.
buckling mechanism and ii) the compressive softening of CFs, such as Further experimental campaigns can substantiate the applicability
PAN-based (see also [15]). The latter causes an up-to 40% reduction for range of the proposed scheme and elucidate the sensitivity on the
thicknesses where micro-buckling is absent (Fig. 12b). Indicatively, for essential ‘representative’ waviness. Computational micromechanical
G.A. Pappas et al.

studies can contribute in expanding the studied effects for different [9] S. Pimenta, R. Gutkin, S. Pinho, P. Robinson, A micromechanical model for kink-
band formation: Part II—analytical modelling, Compos. Sci. Technol. 69 (7) (2009)
microstructures and fiber or matrix types. The shell buckling experiment
956–964.
used for the experimental campaign allowed for essential observations [10] N.A. Fleck, L. Deng, B. Budiansky, Prediction of kink width in compressed fiber
on the free surfaces and also the use of DIC. Other bending setups which composites, J. Appl. Mech. 62 (2) (1995) 329–337.
allow very large displacements, such as the ones proposed in Ref. [3] or [11] B. Budiansky, N. Fleck, Compressive failure of fibre composites, J. Mech. Phys.
Solid. 41 (1) (1993) 183–211.
in Ref. [32], may allow direct moment measurements, nevertheless, [12] S. Pimenta, R. Gutkin, S. Pinho, P. Robinson, A micromechanical model for kink-
particular care shall be devoted in the load transfer areas. band formation: Part I — experimental study and numerical modelling, Compos.
Sci. Technol. 69 (7) (2009) 948–955.
[13] I. Daniel, O. Ishai, Engineering Mechanics of Composite Materials, Oxford
Author statement University Press, New York, 1994.
[14] M. Northolt, L. Veldhuizen, H. Jansen, Tensile deformation of carbon fibers and the
relationship with the modulus for shear between the basal planes, Carbon 29 (8)
Georgios A. Pappas: Conceptualization, Data curation, Methodology, (1991) 1267–1279.
Investigation, Formal analysis, Writing – original draft, Arthur Schlo­ [15] T. Yokozeki, T. Ogasawara, T. Ishikawa, Effects of fiber nonlinear properties on the
thauer: Conceptualization, Data curation, Methodology, Investigation, compressive strength prediction of unidirectional carbon–fiber composites,
Compos. Sci. Technol. 65 (14) (November 2005) 2140–2147.
Formal analysis, Writing – original draft, Paolo Ermanni: Supervision,
[16] A. Schlothauer, D. Cueni, G.A. Pappas, P. Ermanni, Effects of fiber non-linearity
Conceptualization, Methodology, Writing – review & editing. and matrix type on the realization of foldable structures, in: AIAA Scitech 2021
Forum, 2021.
[17] J. Yee, S. Pellegrino, Biaxial bending failure locus for woven-thin-ply carbon fibre
reinforced plastic structures, in: 46th AIAA/ASME/ASCE/AHS/ASC Structures,
Declaration of competing interest
Structural Dynamics, and Materials Conference, Austin, Texas, 2005.
[18] NTPT-Switzerland, NTPT Thinpreg™ 402, Data Sheet," 11, May 2017 [Online].
The authors declare that they have no known competing financial Available: https://www.thinplytechnology.com.
interests or personal relationships that could have appeared to influence [19] A. Schlothauer, N. Schwob, G.A. Pappas, P. Ermanni, Thin-ply thermoplastic
composites for foldable structures, in: AIAA Scitech Forum, American Institute of
the work reported in this paper. Aeronautics and Astronautics, Orlando, FL, 2020.
[20] ASTM Standard D6272− 10, Standard Test Method for Flexural Properties of
Unreinforced and Reinforced Plastics and Electrical Insulating Materials by Four-
Acknowledgements Point Bending, ASTM International, West Conshohocken, PA, 2010.
[21] ISO 14125, Fibre-Reinforced Plastic Composites — Determination of Flexural
This work was supported by the ‘SFA-Advanced Manufacturing’ Properties, first ed., ISO Standard, Geneva, 1998.
[22] Torayca®, "T700S Standard Modulus Carbon Fiber," 13 April 2018. [Online].
grant of the ETH-board as well as the SNF REquip program, SNF206021
Available: https://www.toraycma.com.
150729, for the acquisition of the DIC. [23] R. Kriz, W. Stinchcomb, Elastic moduli of transversely isotropic graphite fibers and
their composites, Exp. Mech. 19 (2) (1979) 41–47.
[24] B. Fiedler, M. Hojo, S. Ochiai, K. Schulte, M. Ando, Failure behavior of an epoxy
Appendix A. Supplementary data matrix under different kinds of static loading, Compos. Sci. Technol. 61 (11) (2001)
1615–1624.
Supplementary data to this article can be found online at https://doi. [25] ASTM Standard D3518/D3518M− 13, Standard Test Method for In-Plane Shear
Response of Polymer Matrix Composite Materials by Tensile Test of a ±45◦
org/10.1016/j.compscitech.2021.108979. Laminate, ASTM International, West Conshohocken, PA, 2013.
[26] L.P. Canal, C. González, J. Segurado, J. LLorca, Intraply fracture of fiber-reinforced
composites: microscopic mechanisms and modeling, Compos. Sci. Technol. 72 (11)
References
(2012) 1223–1232.
[27] F. Naya, M. Herráez, C. Lopes, C. González, S. Van der Veen, F. Pons,
[1] G. Sanford, A. Biskner, T. Murphey, Large strain behavior of thin unidirectional Computational micromechanics of fiber kinking in unidirectional FRP under
composite flexures, in: 51st AIAA/ASME/ASCE/AHS/ASC Structures, Structural different environmental conditions, Compos. Sci. Technol. 144 (2017) 26–35.
Dynamics, and Materials Conference, Orlando, Florida, 2010. [28] T. Langston, The tensile behavior of high-strength carbon fibers, Microsc.
[2] A. Schlothauer, G.A. Pappas, P. Ermanni, Material response and failure of highly Microanal. 22 (4) (2016) 841–844.
deformable carbon fiber composite shells, Compos. Sci. Technol. 199 (2020) [29] M. Ueda, M. Akiyama, Compression test of a single carbon fiber in a scanning
108378. electron microscope and its evaluation via finite element analysis, Adv. Compos.
[3] T.W. Murphey, W. Francis, B. Davis, J.M. Mejia-Ariza, High strain composites, in: Mater. 28 (1) (2019) 57–71.
2nd AIAA Spacecraft Structures Conference, AIAA SciTech Forum, Kissimmee, [30] D. Wilhelmsson, R. Gutkin, F. Edgren, L. Asp, An experimental study of fibre
Florida, 2015. waviness and its effects on compressive properties of unidirectional NCF
[4] N.F. Dow, B.W. Rosen, Evaluations of Filament-Reinforced Composites for composites, Compos. Appl. Sci. Manuf. 107 (2018) 665–674.
Aerospace Structural Applications, National Aeronautics and Space [31] J. Lee, C. Soutis, Thickness effect on the compressive strength of T800/924C
Administration, Washington, D.C., 1965. carbon fibre–epoxy laminates, Compos. Appl. Sci. Manuf. 36 (2) (2005) 213–227.
[5] A. Argon, Fracture of composites, Treatise on Materials Science & Technology 1 [32] M. Wisnom, J. Atkinson, M. Jones, Reduction in compressive strain to failure with
(1972) 79–114. increasing specimen size in pin-ended buckling tests, Compos. Sci. Technol. 57 (9)
[6] S. Kyriakides, R. Arseculeratne, E. Perry, K. Liechti, On the compressive failure of (1997) 1303–1308.
fiber reinforced composites, Int. J. Solid Struct. 32 (6–7) (1995) 689–738. [33] K. Ubamanyu, D. Ghedalia, A.D. Hasanyan, S. Pellegrino, Experimental study of
[7] B. Budiansky, Micromechanics, Comput. Struct. 16 (1) (1983) 3–12. time-dependent failure of high strain composites, in: AIAA Scitech 2020 Forum,
[8] P.S. Steif, A model for kinking in fiber composites- I. Fiber breakage via micro- 2020.
buckling, Int. J. Solid Struct. 26 (5) (1990) 549–561.