Composite Structures 176 (2017) 812–822

Contents lists available at ScienceDirect

Composite Structures
journal homepage: www.elsevier.com/locate/compstruct

Experimental and theoretical similitude analysis for flexural bending of

scaled-down laminated I-beams
M.E. Asl ⇑, C. Niezrecki, J. Sherwood, P. Avitabile
Department of Mechanical Engineering, University of Massachusetts Lowell, One University Avenue, Lowell, MA 01854, United States

a r t i c l e i n f o a b s t r a c t

Article history: In this paper, similitude theory is presented as a methodology on how to design a meaningful scaled-
Received 3 January 2017 down subcomponent that emulates the strain field experienced in the full-scale component. The scaling
Revised 9 May 2017 laws for the flap-wise bending of a laminated composite I-beam with shear deformability are derived
Accepted 7 June 2017
based on the similitude analysis. The set of scaling laws is treated as a criterion to design a scaled-
Available online 9 June 2017
down laminated I-beam that is partially similar to a full-scale I-beam that is representative of what is
used in a utility-scale wind turbine rotor blade. The designed I-beam is manufactured and loaded using
Keywords:
a three-point bending test while digital image correlation is used to measure the strain and displacement
Similitude
Sub-component testing
fields. The analytical solution is examined with the measured strain and displacement values and a finite
Laminated I-beam element simulation for validation. With the derived scaling laws shown to be credible, they are then
Scaling law applied to the measured strain and displacement values of the scaled-down I-beam to predict the corre-
Wind turbine blade sponding values of the full-scale I-beam. The scaled model is shown to replicate the strain field of its full-
scale parent I-beam. Moreover, applicability of the scaled beam in predicting the displacement field of the
full-scale I-beam using experimental data is demonstrated.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction inated beams emulating the cross section geometry of a wind tur-
bine rotor blade and loaded the designed beams in a four-point
The certification of a new material system to be used in the bending test. The measured deflection of the beams was in agree-
manufacture of a wind turbine blade begins with coupon testing ment with the simple beam theory prediction and the finite ele-
of the materials and lead up to the full-scale testing of a blade ment model. However, the correlation between the designed
made with this new material system. Coupon testing does not nec- specimens and the parent I-beam geometry was not studied. Cairns
essarily represent the performance of the new material systems et al. [2] designed a pull-out test for the root stud section of the
when used in blades, and full-scale blade testing is laborious and blade. Mandell et al. [3] pursued an experimental study on the
costly, e.g. on the order of $300 k to $500 k per full-blade test. transition angle of the sandwich panels to the spar cap region of
The coupon testing and the full-scale testing are both required the wind turbine blade. The designed subcomponents were aimed
for certification of the new materials (e.g. bio-based and recyclable to determine the optimal transition angle between the sandwich
resin systems) and design variations incorporated to utility-scale panel and the spar cap using the fracture loads in the tension tests.
wind turbine blades. The costs associated with the manufacture Laustsen et al. [4] pursued an experimental and numerical study to
of a new blade type and the time associated with the full-scale investigate the load response and failure behavior of a grid-scored
testing of the blade type, are significant barriers to use new mate- composite sandwich panel when subjected to quasi-static multiax-
rial systems in utility-scale blades. Subcomponent testing has the ial loading conditions. A multiaxial test fixture was developed and
potential to overcome the gap between the full-scale and coupon two criteria for predicting the initiation of the failure were pro-
tests and to reduce the overall time and expense to certify a new posed. Sayer et al. [5] proposed the Henkel beam design to inves-
material system. tigate the performance of the bonding paste under combined
The design of subcomponent tests of the wind turbine blade has loading and later used this subcomponent to study the effect of
been pursued in several studies. Mandell et al. [1] developed lam- the bond-line thickness on the flexural stiffness of the beam [6].
Kleiner et al. [7] developed the modified designs for Henkel beam
for fatigue testing by adding reinforcement and increasing the
⇑ Corresponding author. length of the beam. Two symmetric and asymmetric laminated I-
E-mail address: [email protected] (M.E. Asl).

http://dx.doi.org/10.1016/j.compstruct.2017.06.017
0263-8223/Ó 2017 Elsevier Ltd. All rights reserved.
 M.E. Asl et al. / Composite Structures 176 (2017) 812–822 813

beams were designed by Zarouchas et al. [8] emulating the spar To conclude structural similarity, similitude analysis can pro-
caps and the sandwich shear web of the blade to determine the vide the necessary and sufficient requirements between model
failure mechanisms using the digital image correlation and acous- and prototype. These similarity conditions can be extracted using
tic emission techniques [9]. Jensen [10] investigated a box girder to either dimensional analysis or direct use of the governing equa-
examine the dominant failure modes of a flap-wise loaded wind tions, in which the latter results in specific scaling laws. Some of
turbine blade. The length of the tested box girder was 35 m and the best examples of the application of Similitude Analysis has
the dominant failure modes were characterized using different been done by Simitses and Rezaeepazhand [13]. They demon-
measurement techniques. Tippmann et al. [11] tested a scaled- strated the credibility of similitude analysis when they derived
down 9-m wind turbine blade that was constructed with several the similarity conditions for the buckling of orthotropic and sym-
embedded defects representing the most common manufacturing metric cross-ply laminated plates using the direct approach. They
defects typically found in a utility-scale wind turbine blade. White applied the same approach to extract the scaling laws for the vibra-
et al. [12] developed a dual-axis fatigue test on a truncated wind tion of scaled laminated rectangular plates [14] and to determine
turbine blade using the resonance excitation with forced hydraulic the stability condtions for the scaled-down composite plates under
loading. These previous efforts on subcomponent testing of wind shear and axial loads [15,16]. The same approach was extended
turbine blades cover a wide range of design in size and detail and further developed in their work to investigate the applicability
and provide a means to study material behavior without pursuing of predicting the vibration frequencies of the laminated shells
a full blade test. However, none of the studies used a defined math- using scaled models [17,18]. Based on their analytical results, the
ematical methodology based on the governing equations and ply-level scaling technique may be used to design scaled models
structural similarity of the subcomponent and its parent compo- that accurately predict the vibration frequencies of their full-
nent to directly derive the configuration of the subcomponent from scale prototype. The designed models using ply-level scaling accu-
the configuration of the parent component. rately predict the response of their full-scale prototype when the
An important aspect to address in the design of a subcompo- governing equations used in the derivation of the scaling laws
nent is for the subcomponent to be an effective representation of are valid and applicable (i.e. within the linear elastic range of the
its full-scale parent component. If a subcomponent is not an accu- response). The scaled models designed using ply-level scaling can
rate scaled-down representation of its full-scale parent compo- replicate the stress and strain fields of their prototype in a fatigue
nent, then the results from subcomponent testing will not be a test as long as the loading condition is within the linear elastic
reflection of the mechanical behavior of the full-scale parent com- range of the structural response of the structure. The similarity
ponent. Therefore, it is critical to understand how the experimental of the failure modes may also be investigated only if the relevant
data from the testing of a subcomponent can be extrapolated to its governing equations and scaling laws that predict the failure of
full-scale component. The relationship between the designed sub- the structure are included in the design procedure. However, use
component (referred to as the ‘‘model”) and its full-scale parent of ply-level scaling is limited to specific lamination schemes and
component (referred to as the ‘‘prototype”) must be derived from no systematic approach was suggested to design the scaled models
the governing equations that describe the response of the proto- when ply-level scaling requirements cannot be completely
type and its scaled models for different loading conditions satisfied.
(Fig. 1). The structural parameters that describe the relationship Numerous experimental studies have been pursued to deter-
between the prototype and the scaled models are referred to as mine the effect of scaling on strength and stiffness of the compos-
the scaling laws which can be derived directly from the governing ite laminated structures which is referred to as the size effects in
equations and are descriptor of the structural similarity between literature. The ultimate question in size effect studies is to deter-
the model and the prototype. mine whether strength and stiffness are affected by scaling. From
the conclusions of the most studies on size effects, one can say that
the size effect on stiffness is almost nonexistent [19–21]. However,
the effect of size on strength of composite structures may not be
negligible. Jackson et al. [19] pursued an experimental study on
composite laminated beams and concluded that there is consider-
able size effect on strength of the beams. Grimes et al. [22] con-
cluded that for bonded laminated structures, the largest size
effect on static strength is less than 4.5% which was shown to be
a result of poor quality tooling and differences in environmental
exposure. O’Brien [23] used a fracture mechanic approach in con-
junction with Weibull statistics and suggested that the effect of
scaling on strength comes from different damage sequences which
occur in different scales. According to these studies, one can say
with confidence that effect of size on stiffness is negligible while
more work on strength needs to be done.
Several experimental studies have been pursued to investigate
the creditability of structural similitude and accuracy of the scaling
laws. Meruane et al. [24] carried out an experimental modal anal-
ysis to verify the derived scaling laws for flexural vibration of rect-
angular plates. Luo et al. [25] derived the scaling laws for vibration
of a thin-walled annular plate and performed the impact hammer
test on the plates with different scales to compare the experimen-
tal frequencies with the predicted frequencies using the scaling
laws. Torkamani et al. [26] derived the scaling laws for free vibra-
tions of orthogonally stiffened cylindrical shells and experimen-
Fig. 1. Schematic geometry of the prototype (top) and its scaled model with tally evaluated the accuracy of the derived scaling laws. The
different layup than that of the prototype (bottom). predicated frequencies using the similitude analysis were in good
814 M.E. Asl et al. / Composite Structures 176 (2017) 812–822

agreement with experimental data from the modal test. Ilbeigi of a new material on a smaller scale than on a full-scale part with-
et al. [41] proposed a method to design reduced scale models for out any loss in the fidelity of the testing. Similitude analysis is
an euler-bernoulli beam and later they extended the proposed applied to the I-beam geometry of a commercial-scale wind tur-
technique to study reduced scale models for systems with dis- bine rotor blade [37] to create a second scaled-down model.
parate spatial and temporal scales [42]. Because a typical wind-blade I-beam is twisted and tapered, these
In two previous works by the authors [27,28], a clear methodol- ‘‘distortions” present a challenge in developing a sub-component
ogy was proposed to design the partially similar models for a sim- design based on a closed-form set of governing equations. As a con-
ply supported laminated plate. The degree of similarity between sequence, this second scaled-down model is designed using the
the model and the prototype was evaluated based on the derived Distorted Layup Scaling technique [29]. Because the Distorted
scaling laws from governing equations. The fidelity of the devel- Layup Scaling technique is used, the model is only partially similar
oped models in predicting the structural response of the parent to its full-scale prototype in the blade. To achieve the complete
prototype was shown to be proportional to the discrepancy of similarity between model and the prototype, the structural simi-
the models in satisfying the requirements of the similarity. The larity conditioned derived from the governing equations must be
same approach was improved in the later studies [29] to develop fully satisfied. For a laminated structure, the structural similarity
scaled-down laminated beams that emulate the spars and shear between the scaled model and the prototype is fully satisfied only
web section of a commercial-scale wind turbine blade. They if ply-level scaling technique is applied. However, ply-level scaling
derived the scaling laws for similarity of the strain fields [43,44] is only applicable to specific lamination schemes. When complete
and developed structural similarity conditions for vibration predic- similarity is not applicable, the partial similarity is sought between
tion [45] and dynamic characterization [46] in composite I-beams. the model and the prototype where some of the structural similar-
These spar caps and the shear web of the wind turbine blade ity conditions are relaxed or partially satisfied to design a model
carry the majority of the aerodynamic loads applied to the rotor that is partially similar to its prototype. In this study, the partial
blade and are structurally critical as they form the backbone of similarity between the scaled subcomponent and the full-scale
the blade and make an interesting case study to develop a subcom- prototype is sought using the Distorted Layup Scaling technique
ponent test. Kensche et al. [30] pursued a review on fatigue aspects as ply-level scaling technique is not applicable due to the proto-
of fiber reinforced plastics used in wind turbine and identified the type lamination scheme. The designed model is manufactured
shear web and the spar caps as the fatigue-critical region of the and loaded in a three-point bending test, and the normal strain dis-
blade. Raman et al. [31] conducted a numerical study and sug- tribution and transverse deflection along the span of the bottom
gested that bonded joints, root section and trailing-edge of the flange of the I-beam are measured using the digital image correla-
blade are the critical zones in the blade. Haselbach et al. [32] con- tion (DIC) technique for the different prescribed loads. The loading
ducted a numerical study to investigate the ultimate strength of a condition was selected such that the strain level applied to the
blade in a trailing-edge failure dominated load direction. Branner scaled subcomponent is comparable to that of the blade near its
et al. [33] proposed a static subcomponent test to mimic the com- maximum chord region during the fatigue testing. The designed
pressive loading of the trailing edge panels and bond-lines. Hasel- model replicates a specific segment of the blade and the three
bach et al. [34] pursued a numerical study on the delaminations in point loading was applied to the scaled subcomponent to induce
the spar cap of a 8.65 m blade subjected to the flap-wise bending the same strain field the spar caps of the blade experience at that
moment and concluded that delamination induced buckling modes specific region (i.e. the maximum chord) in a flap-wise bending
in the spar caps have to be considered critical. Branner et al. [35] loading condition. Measured strain and displacement values are
investigated the compressive strength of the thick panels used in compared to the finite element simulation and the analytical
the spar caps of the blade in an experimental study and showed model [38,39]. Derived scaling laws are applied to the measured
that large delaminations in the spar caps may cause instant failure full-field normal strain to predict the strain distribution of the pro-
of the component. Chen et al. [36] pursued an investigation on the totype under adjusted loads and to demonstrate the similarity of
failure of a 52.3 m composite blade and identified the delamination strain fields between the two scales, i.e. the model and the proto-
of unidirectional laminates in the spar cap as the root cause of the type. The similarity of the strain fields between the designed
catastrophic failure of the blade. The design of a subcomponent to scaled-down beams and the full-scale beam suggests that the
mimic the spar caps and the shear web of the blade bonded structural performance of the new materials (e.g. bio-based) can
together using the material systems used in the utility-scale blade be tested using a scaled-down model. Because the designed
allows investigating the structural performance and fatigue life of small-scale beams replicate the strain field of the full-scale beam,
the critical regions of the blade in an expedited and less expensive fatigue testing of the spar caps and shear web geometry of the
way compared to the full-scale testing. The spar caps and the shear blade can be expedited and facilitated by use of the designed
web are typically combined to form an I-beam, C-beam or scaled-down beams. Structural similarity ensures that the strain
Box beam. field of the scaled model is similar to its full-scale counterpart
Because wind turbine rotor blades are made of laminated mate- under adjusted loads calculated based on the derived scaling laws.
rials, design of a scaled-down model can be a challenge because the If the scaled model withstands the same strain levels applied to its
thickness of the plies in a composite laminate cannot be scaled full-scale counterpart during the high-cycle fatigue testing of the
down to any desired ratio. Although structural similarity between blade (i.e. two million cycles or 20 years of the accelerated dam-
the scaled subcomponent and the full-scale component is critical, age), then the full-scale counterpart may also survive during the
any subcomponent design must satisfy the manufacturability test because the scaled model and full-scale component have the
requirements as well. The structural similarity requirements are same material system and experience the same strain levels. The
derived from the governing equations and the manufacturability transverse deflection of the prototype is also predicted by using
limitations are also considered to achieve a design for the scaled the measured deflection of the model with different scaling laws.
subcomponent that is practical to be manufactured. Estimating the deflection of the prototype from the measured
In the current research, the I-beam configuration has been deflection of the scaled model has significant importance for frac-
selected with the objective to design a scaled-down composite I- ture analyses of the adhesive connections in a wind turbine blade
beam that has a strain distribution that is the same as its full- [40]. Predicted strain and deflection values are compared to the
scale counterpart. The similarity of the strain fields between the nominal analytical values to investigate the accuracy of the
two scales can be used to characterize the mechanical behavior designed model in replicating the strain field of the prototype
 M.E. Asl et al. / Composite Structures 176 (2017) 812–822 815

and also the applicability of the derived scaling laws in predicting

the deflection of the prototype using experimental data are consid-
ered. Although a few experimental studies have been pursued on
the scaling of simple laminated geometries such as laminated
plates [13], the applicability of the scaled models for the complex
laminated geometries has not been verified experimentally before
now. This study presents the first experimental work on a scaled
laminated composite I-beam structure with the laminated flanges
and sandwich shear web. The feasibility of using a complex scaled
laminated structure for characterization purposes is demonstrated
and the proposed technique for the design of a scaled laminated
model which is partially similar to its prototype is verified
experimentally.
The proposed approach in this study can be used to design a
generic composite laminated model that is structurally similar to
its full-scale subcomponent. The designed subcomponent helps
the manufacturer to test new material systems (e.g. resin, fiber
and bonding adhesive) in a configuration which replicates the
strain field that the full-scale component experiences during its
working condition. This primary experimental assessment using
a scaled model of the structurally critical subcomponent of the
structure increases the manufacturer assurance for use of the
new materials before making the full-scale structure from the
new material system relying only on the data from coupon tests. Fig. 2. Geometry and coordinates of the laminated I-beam with sandwich shear
web.

2. Mathematical model

The governing equations for flexural bending of a laminated the cross-section geometry of the composites. Assuming the thin
composite I-beam are introduced in this section. The closed form wall does not deform in its own plate and the Kirchhoff-Love
solutions for deflection and normal strain distribution for an I- assumption in classical plate theory remains valid for laminated
beam placed in a three-point bending configuration are derived. composite thin-walled beams, For I-section shapes, the explicit
The scaling laws are developed and their applicability in design expressions may be derived using a variational formulation as
of the scaled models and predicting the response of the prototype described in [20]:
is discussed.
The first step in the application of Similitude Theory for the cur-
ð3Þ
rent research is to derive the equation for the normal strain distri- b3 3
ðEIx Þcom ¼ ½Aa11 yð2Þ a a
a  2B11 ya þ D11 ba þ A ð1cÞ
bution in the prototype. In this study, the prototype is a laminated 12 11
I-beam consisting of two flanges and a sandwich web with sym-
metric layup connected together using the adhesive bond-line with ðGAx Þcom ¼ Aa55 ba þ A366 b3 ð1dÞ
the material systems similar to that of the spar cap and shear web
where A11 ; A66 ; A55 ; B11 and D11 indicate the extensional, coupling
section of a wind turbine blade near its maximum chord. The flap-
and bending stiffness coefficients for a laminated composite layup.
wise bending of the prototype with simply-supported boundary
The repeated indices indicate the summation and the superscript in
conditions in absence of the thermal effects is considered.
the parenthesis () is the power of the exponent. The symbol a is a
Tension-torsion coupling effects do not exist because of the sym-
summation index varying from 1 to 3 where the indices 1 and 2
metrical ply layups of the flanges and the web and the symmetry
indicate the top and bottom flanges, and 3 is for the shear web,
in the geometry. Torsion and its corresponding scaling laws were
respectively as shown in Fig. 2 and ba represents width of the top
not included because a higher-order torsional mode is unlikely to
and bottom flanges and the web.
occur during normal operating conditions of a wind turbine blade.
Eqs. (1c) and (1d) can be written for a geometrically symmetric
Thus, torsion does not significantly contribute to the amplitude of
I-beam with respect to x and y axis including symmetric and bal-
vibration that attributes to fatigue accumulation [37]. In the
anced laminated flanges and the web as:
absence of any coupling effects, the governing equations for flexu-
ral bending of the shear-deformable laminated I-beam with
respect to x-axis are given by [39]: ð3Þ
f f ð2Þ bw w
ðEIx Þcom ¼ ½2A11 y1 þ 2D11 bf þ A ð1eÞ
12 11
0
ðGAx Þcom ðV 00 þ W
xÞ ¼p ð1aÞ f
ðGAx Þcom ¼ 2A55 bf þ 2Aw w
55 bw þ A66 bw ð1fÞ
00 0
ðEIx Þcom W  ðGAx Þcom ðV þ Wx Þ ¼ 0
x ð1bÞ where f and w represent the flange and the web, respectively. The
extensional, coupling and bending stiffness coefficients
where p is the point load, V indicates the displacement in y direc- Aij ; Bij and Dij for a laminated plate are defined as the following:
tion and the symbol Wx is the rotation of the cross-section with
Z h N Z
X zkþ1
respect to x-axis as shown in Fig. 2. The prime (0 ) indicates differen- 2
ðkÞ
ðAij ; Bij ; Dij Þ ¼ Q ij ð1; z; z2 Þdz ¼ Q ij ð1; z; z2 Þdz ð1gÞ
tiation with respect to z-axis. The symbols ðGAx Þcom and ðEIx Þcom h2 zk
k¼1
denote the shear and flexural stiffnesses of the laminated composite
with respect to x-axis, respectively. The terms ðEIx Þcom and ðGAx Þcom where Q ij are the transferred plane stress-reduced stiffnesses, h
are general expressions for arbitrary cross sections and depend on denotes the laminate thickness and N is the total number of the
816 M.E. Asl et al. / Composite Structures 176 (2017) 812–822

plies. The closed-form solution of the normal strain ezz and trans- The second term of Eq. (1c) vanishes for a symmetric layup for
verse deflection for the uncoupled Eqs. (1a) and (1b) can be derived the flanges and the web (i.e. B11 = 0). Upon applying the similarity
for the concentrated point load and simply-supported boundary transformation to Eq. (1e) and Eq. (1f), the following scaling laws
condition as [20]: can be derived:
pyz
ezz ¼ ð2aÞ
2ðEIx Þcom kðEIx Þcom ¼ kAf k2y1 kbf ¼ kDf kbf ¼ kAw11 k3bw ð6Þ
11 11

3    z3  z
pl z pl kðGAx Þcom ¼ kAf kbf ¼ kAw55 kbw ¼ kAw66 kbw ð7Þ
t0 ðzÞ ¼ 3 4 þ ð2bÞ
48ðEIx Þcom l l 2ðGAx Þcom l 55

In this study, the flanges and the shear web of the models are
In Eq. (2b), the first and second terms represent the deflection
geometrically scaled with the same ratio as the length scale factor
by Euler beam theory and the deflection due to shear deformabil-
(i.e. kbf ¼ kbw ¼ kl ) so the models hold the same geometrical length
ity, respectively.
ratios as their prototype. As the third term in Eq. (1c) is negligible
The scaling laws are derived based on the standard similitude
for a thin-walled beam (i.e. D11 ? 0), Eq. (3) can be written as the
analysis [13]. The parameters of the governing equations of the
following scaling laws using Eqs. (6) and (7).
prototype ðxp Þ are assumed to be connected to those of the model
ðxm Þ by one-to-one mapping and similarity transformation is kAf ¼ kAw11 ¼ kAf ¼ kAw55 ¼ kAw66 ð8Þ
11 55
defined as kx ¼ xp =xm where kx indicates the scale factor for param-
Alternatively, the response scaling laws Eqs. (4a)–(5) may be
eter x. Applying the similarity transformation to Eq. (2) results in
written using Eqs. (6) and (7) as the following:
the following scaling factors:
kðEIx Þcom kp k2l kz kp k2l kz kp k2l kz
k2l ¼ ð3Þ kt0 ¼ ¼ ¼ ð9Þ
kðGAx Þcom kAf k2y1 kbf kDf kbf kAw k3b
11 11 11 w

kp k2l kz kp kz kp kz kp kz
kt0 ¼ ð4aÞ kt0 ¼ ¼ ¼ ð10Þ
kðEIx Þcom kAf kbf kAw55 kbw kAw55 kbw
55

kp kz kp k3z kp k3z kp k3z

kt0 ¼ ð4bÞ kt0 ¼ ¼ ¼ ð11Þ
kðGAx Þcom kAf k2y1 kbf kDf kbf kAw k3b
11 11 11 w

kp k3z kp ky kz kp ky kz kp ky kz
kt0 ¼ ð4cÞ kezz ¼ ¼ ¼ ð12Þ
kðEIx Þcom kDf kbf kAw k3b
kAf k2y1 kbf 11 11 w
11

kp ky kz The layup and geometry of the prototype I-beam represent the

kezz ¼ ð5Þ
kðEIx Þcom spars and shear web of the Sandia 10 MW blade near its maximum
chord [37]. This blade was chosen because the details for its com-
The derived scaling laws describe the necessary conditions for posite construction are available in the open literature, and this
similarity of two composite I-beams in bending with different specific area of the blade was considered because many failures
scales and the relationship between their deflections and normal in full-scale testing of the commercial-scale blades occur in or near
strain distributions. Eq. (3) denotes that the ratio of the flexural the maximum chord region due to the flap-wise bending. The pro-
to shear stiffnesses has to be equivalent to the square of the length totype is assumed to be a composite laminated I-beam consisting
ratio of the beams to ensure the structural similarity between the of two identical flanges and a sandwich shear web with an overall
prototype and its scaled model. Eqs. (4a)–(4c) describe the rela- dimension of 29.3  2.446  1.5 m3. The flanges have the layup
tionship between the transverse deflection of the prototype and ½452 =046 s and ply thickness of t = 1.36 mm. The shear web is a
its scaled model based on their flexural and shear stiffness ratios sandwich panel with the foam thickness of 160 mm and two face
respectively. Eqs. (4a) and (4c) are both based on the flexural stiff- sheets with layup ½4550 with an overall thickness of 17.5 mm.
ness kðEIx Þcom and essentially predict the same values when geomet- All computations are implemented for the glass/epoxy materials
rical scale factors kz and kl are equal. Eq. (5) denotes the with following material characteristics [37]: E1 ¼ 41:8 GPa, E2 ¼
relationship between the normal strain fields of the model and E3 ¼ 14 GPa, G12 ¼ G13 ¼ 2:63 GPa, G23 ¼ 1:83 GPa, t12 ¼ t13 ¼ 0:28,
the prototype. Eq. (3) is a necessary condition that needs to be sat- t23 ¼ 0:47. To design a scaled-down model for the considered pro-
isfied before using Eqs. (4a)–(4c) for predicting the transverse totype, the Distorted Layup Scaling method [29] was used to find
deflection. However Eq. (5) can be used independently to predict an appropriate layup for the model ensure the structural similarity
the strain field of the different scales. Eq. (3) is a prerequisite for between the model and the prototype. Although, a model made
using Eqs. (4a)–(4c) because the transverse deflection of the beam from isotropic materials would essentially show the same strain
depends on both the shear and flexural rigidity terms. A primary levels under adjusted loads, its failure modes in a high-cycle fati-
observation on Eq. (3) suggests that for design of a scaled-down gue test would be totally different than the materials used in the
I-beam to predict the deflection of its full-scale prototype, flexural spar-cap and shear web of a wind turbine-blade. Therefore, to have
to shear rigidity ratios of the models and the prototype (i.e. a model that shows the same strain levels as its prototype and also
kðEIx Þcom and kðGAx Þcom ) must satisfy Eq. (3). However, to design a its fatigue test results could be incorporated to that of the proto-
model with similar normal strain field such a condition does not type, the material system of the model was selected to be same
exist and load, length and flexural rigidity ratios (i.e. as the materials used in the prototype (i.e. the spar cap and the
kp ; kz and kðEIx Þcom ) are the only required parameters. Based on this shear web in a utility-scale wind turbine blade). The scale of the
observation, the models that are designed to predict the transverse model considered is 1/50 of the prototype (i.e. lp =lm ¼ 50) that
deflection can also be used for predicting the normal strain. How- resulted in an overall dimension of 58.6  4.93  3 cm3 for the
ever, the opposite may not be true. model with two plies for each flange and a ply thickness same as
 M.E. Asl et al. / Composite Structures 176 (2017) 812–822 817

that of the prototype. Eq. (3) was considered to select the layup for the unidirectional glass/epoxy materials, and the shear web was
the model because the model was to predict both transverse constructed from a foam core wrapped in a 45 ply of glass/epoxy.
deflection and normal strain distribution of its prototype. As the The flanges and the shear web were carefully bonded together
layup of the flanges of the prototype ½452 =046 s was symmetric using the adhesive paste and were cured while the adhesive thick-
and balanced and Eq. (3) was derived for the same type of the ness throughout the length of the beams were controlled using a
ply scheme, only symmetric and balanced layups were searched fixture. The manufactured specimens were loaded in a three-
for the scaled model. All the possible layups for the model consid- point bending test configuration to investigate the correlation of
ering its size and number of plies for the flanges were considered. their strain and displacement fields with the analytical solution
Models were assumed to have a combination of cross plies to be and finite element simulation and to demonstrate their accuracy
practical for manufacturing. The candidate layups were selected in predicting the response of their full-scale prototype.
based on their accuracy in satisfying the design scaling law Eq.
(3). To find a layup with the least discrepancy in satisfying Eq. 4.1. Strain field
(3), an error criterion was defined:
   I-beams were loaded in a three-point test setup with incremen-
kEI
e¼ k2l  =k2l  100 ð13Þ tal static loads while DIC cameras were used to measure the full-
kGA
field strain distribution and displacement in the bottom flange of
Models with a layup consisting of two plies and kl ¼ 50 were the I-beams as shown in Fig. 5. The I-beams were loaded in four
searched including 0, 90 and 45 plies. The only possible symmet- quasi-static increments from 222.4 N to 778.5 N (i.e. 50 lbs–
ric and balanced layups turned out to be ½02 and ½902 with their 175 lbs), and the full-field normal strain in the bottom flange of
associated error listed in Table 1 based on Eq. (11). The plies orien- the I-beams was recorded for each stage. The flanges of the I-
tation 0, 90 and 45 were considered intentionally to design a beam were the best candidates for the strain measurement
scaled-model feasible to manufacture, however 45 plies did not because they exhibit the highest level of strain compared to other
qualify for this design as they do not satisfy the laminate symmetry regions of the beam. The applied-load steps were adjusted to gen-
condition. The layup ½02 was selected for manufacturing the erate strain levels in the flanges high enough to be captured by the
scaled-model because of showing less discrepancy in prediction cameras while remaining to be in the elastic range but far enough
of the transverse deflection of the prototype for both response scal- away from loading levels that could induce cracking and subse-
ing laws Eq. (4a) and Eq. (4b), as shown in Table 1. Although layup quent beam failure. The noise floor for the normal strain ezz during
½02 shows less error in predicting the deflection compared to layup the measurement was 20 lm=m and 0:002 mm for the trans-
½902 for both deflection scaling laws (i.e. Eqs. (4a) and (4b)), error verse displacement. Because the primary purpose of the experi-
values in Table 1 suggest that Eq. (4a) is the more accurate choice ment was to investigate the accuracy of the derived scaling laws
to be used in conjunction with layup ½02 to estimate the deflection derived from a formulation valid for the linear elastic bending,
of the prototype. This observation which is based on the analytical the beams were loaded in their elastic bending region to keep both
derivations is verified experimentally in Section 4.2. governing equations and the scaling laws valid for the entire study.
The average of the three experimental strain results was com-
pared to the ‘‘scaled” analytical model and to an Abaqus finite ele-
3. Numerical model
ment model. Fig. 6 shows the normal strain values in the bottom
flange of the I-beam for the analytical model, the Abaqus simula-
For the finite element analysis Abaqus version 6.13 was used.
tion and the experimental data. Each solid line represents a pre-
The beam was modeled using the doubly-curved general purpose
scribed load in the analytical model, the associated dashed line
shell element S8R with reduced integration. The model was dis-
and data points denote the Abaqus solution and the experimental
cretized with 7500 nodes with the approximate characteristic
data, respectively. From the results shown in Fig. 6, it can be con-
length element of 0.003 m. The displacement of the bottom flange
cluded that there is good correlation for the normal strain values
elements were restrained at the support locations in the direction
among the three approaches. The degree of correlation between
of the applied load. As shown in Fig. 3, a path was defined along the
the experimental and simulation results increases as the load
bottom flange of the beam to extract the strain and displacement
increases. This increase is correlation is because the effect of the
values to be compared to the experimental and analytical values.
noise floor on the measurement decreases as the level of strain
The distance of the supports from the free ends of the beam was
increases.
25.5 mm identical to that of the experimental setup. A static load
To investigate the applicability of the derived scaling laws and
was applied in four increments from 222.4 N to 7778.5 N (i.e. 50
the model in predicting the strain distribution of the prototype,
lbs to 175 lbs) identical to the loads applied to the analytical model
analytical strain values of the model were projected to those of
and the loaded beams in the experiment.
the prototype using the derived scaling law. The load applied to
the prototype was determined from Eq. (5) to have a normal strain
4. Experimental observations
distribution in the bottom flange to be the same as the model (i.e.
kezz ¼ 1Þ. As shown in Fig. 7, the analytical strain values in the bot-
Based on the similitude analysis in Section 2, three identical I-
tom flange of the designed I-beam under maximum load (i.e.
beams with the specified geometries and ½02 layup for the flanges
P = 777 N) were projected to those of the prototype under the
were manufactured as shown in Fig. 4. The flanges were made from
adjusted load (i.e. kp ¼ 4) using the scaling law Eq. (5). An overlay
Table 1
of the strain values of the prototype and the values as predicted by
Symmetric and balanced ply schemes for a model with a total of two layers in the the scaling laws indicate excellent agreement for the normal strain
flanges that only include 0 and 90 plies for the prototype layup ½452 =046 s . scaling law Eq. (5) along with designed model in predicting the
strain distribution of the prototype. In this study, the designed
Layup e Error in predicting deflection Error in predicting deflection
(%) using Eq. (4b) (%) using Eq. (4a) (%) scaled model was loaded within the elastic region and away from
its ultimate strength limit. The analytical and simulation results
½02 11.4 4.5 1.0
½902 51.7 39.2 5.0 were in good agreement with the experimental data of the scaled
model. Therefore the analytical model and the finite element sim-
818 M.E. Asl et al. / Composite Structures 176 (2017) 812–822

Fig. 3. The Designed scaled model was loaded in a three-point bending configuration in four increments from 222.4 N to 7778.5 N and displacement and strain values were
extracted from the defined path on the bottom flange of the beam (red dotted line). (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)

Fig. 4. (Left): The three manufactured I-beams each consisting of two flanges made of unidirectional glass/epoxy materials and a sandwich shear web constructed from a
foam core wrapped in a 45 ply of glass/epoxy. (Right): The flanges and the web were bonded together using the adhesive paste with a consistent adhesive thickness
throughout the length of the beams.

Fig. 5. Three-point bend test: (Left) Three-point bending test configuration of the designed I-beam for the full-field strain measurement using DIC cameras. (Right) Strain and
displacement values for transverse deflection of the I-beam were selected from the midline of the bottom flange of the I-beam.
 M.E. Asl et al. / Composite Structures 176 (2017) 812–822 819

Fig. 6. Comparison of the normal strains in the bottom flange of the designed I-beam for different prescribed loads in a three-point bending test configuration.

Fig. 7. Comparison between the predicted strain values of the prototype (red circles) with the analytical prototype strain values (blue dots) by projecting the analytical strain
values (green asterisks) of the model using derived scaling law Eq. (5) for the maximum load P = 777 N. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)

ulation are validated and verified using the experimental data from prototype, a comparison between the displacement fields of the
testing of the scaled beam. As scaling doesn’t have any significant designed scaled-down model and the prototype was pursued.
effect on the beam stiffness [19–21] and the study is carried out in The transverse deflection for the bottom flange of the designed
the elastic range of the response, the analytical model and finite beam was measured using the same test setup using DIC cameras
element simulation are still applicable for a larger scale and are as shown in Fig. 5 and applied loads described in the previous sec-
used to accurately predict the strain and deflection of the tion. The measured deflection of the bottom flange of the beam
prototype. along its length span was compared to the analytical and finite ele-
Next, the accuracy of the derived scaling law Eq. (5) and the ment models as shown in Fig. 9. A very good correlation between
designed model were investigated for prediction of the strain dis- measured data, the finite element and analytical models can be
tribution of the prototype using the experimental data. Fig. 8 observed especially for the two highest loading stages (i.e.
shows the comparison between the predicted strain distribution P = 666 N and P = 777 N). As the load increases, the correlation
of the prototype using the experimental values and their corre- between experimental data and analytical data improves because
sponding analytical values for the two highest loads (i.e. the effect of the noise floor on the measured displacement field
P = 666 N and P = 777 N). These two loading steps were selected decreases for larger deflections.
because they had the least influence from the noise floor of the Next, the displacement field of the prototype is predicted by
measurement setup. A very good correlation between the pre- analytical and experimental displacement fields of the designed
dicted and nominal strain values for both loading stages suggests beam. The transverse deflection of the bottom flange of the model
that the designed model can replicate the strain field of the proto- along its length span is mapped to those of the full-scale prototype
type for the bottom flange. Additionally, the derived scaling law using the scaling factors Eqs. (4a) and (4b) to evaluate the accuracy
accurately predicted the normal strain of the prototype when using of the scaling laws in predicting the displacement field of the pro-
this experimental data. totype. Fig. 10 shows the predicted values for the deflection of the
bottom flange of the prototype by projecting the experimental and
4.2. Displacement field analytical values of the model using scaling law Eq. (4a). Predicted
values for the deflection of the prototype using scaling law Eq. (4b)
To investigate the ability of the derived scaling laws and the is shown in Fig. 11. According to Fig. 10 and Fig. 11, both scaling
designed scaled beam in predicting the displacement field of the laws show a very good accuracy in predicting the transverse
820 M.E. Asl et al. / Composite Structures 176 (2017) 812–822

Fig. 8. The comparison between the predicted strain values of the prototype (red circles) and their corresponding analytical values (blue dots) by projecting the experimental
strain values (pink asterisks, zoomed in) using derived scaling law Eq. (5) for the last two loading stages. (For interpretation of the references to colour in this figure legend,
the reader is referred to the web version of this article.)

Fig. 9. Comparison of the transverse deflection of the bottom flange of the designed I-beam for different prescribed loads in a three-point bending test configuration.

deflection of the bottom flange of the prototype along its length the governing equations used in this study are only valid within
span. However, comparing the predicted values to the analytical the linear elastic range of the response of the structure. The goal
values of the prototype, suggests that scaling law Eq. (4a) under- in this study was to provide a ‘‘design” methodology for a subcom-
predicts the nominal deflection of the prototype whereas Eq. (4b) ponent fatigue test so that the scaled model experiences the same
over-predicts the nominal deflection values of the prototype. This strain filed as its full-scale counterpart does during the fatigue test-
effect is more observable for the points with higher deflection val- ing of the blade. This was achieved by similarity analysis of the
ues near the midpoint of the length span of the prototype. Compar- strain field and the fact that designed models and the full-scale
ing the predicted deflection values of the prototype at the midpoint subcomponent should show same strain levels during the high-
using experimental values (i.e. red asterisks) in Figs. 10 and 11 sug- cycle fatigue testing. For future work, a fracture mechanic analysis
gests that Eq. (4a) give a better prediction than Eq. (4b). Addition- may be used in conjunction with the similitude analysis to inves-
ally, overlay of the analytically predicted values (i.e. green circles) tigate the similarity conditions for the failure of the scaled lami-
to their corresponding nominal values (i.e. blue dots) for a larger nated structures.
span in Fig. 10 compared to Fig. 11 suggests that Eq. (4a) predict
the deflection of the porotype more accurate than Eq. (4b). This 5. Conclusion
observation is also in agreement with the values shown in Table 1
for prediction errors of [02] layup which indicates that the scaling An analytical design methodology accompanied by experimen-
law Eq. (4a) has less discrepancy in predicting the deflection com- tal validation was conducted to demonstrate the feasibility of using
pared to the scaling law Eq. (4b). The analysis pursued in this paper scaled models to facilitate the certification procedure of the new
is valid within the linear elastic range of the response of the model materials for commercial-scale wind turbine blades. Structural
and may not be applicable to extrapolate the failure process of the similarity was investigated for flap-wise bending of a laminated
model to that of the full-scale prototype. This limitation is because composite I-beam representative of the spar caps and the shear
 M.E. Asl et al. / Composite Structures 176 (2017) 812–822 821

lation was observed between the predicted strain values obtained

via scaling laws and the nominal values for the prototype. The
results demonstrated that the designed scaled-down I-beam can
replicate the same strain field as the prototype in its bottom flange
using a smaller scale. The transverse deflection of the bottom
flange of the prototype was predicted by two different scaling laws
with very good accuracy. The accuracy of each scaling law in pre-
dicting the transverse deflection of the prototype was in agreement
with the estimated prediction error in the similitude analysis.

Acknowledgments

The support of this research by National Science Foundation

through Grant Number 1230884 (Achieving a Sustainable Energy
Pathway for Wind Turbine Blade Manufacturing) is gratefully
acknowledged. Special thanks go to Stephen Nolet from TPI Com-
posites Incorporation to help define the composition of the speci-
mens and the tests conducted. We also thank Andy Schoenberg
for CERL Company for fabricating the specimens.

Fig. 10. Comparison between analytical and predicted values for the deflection of References
the bottom flange of the prototype by projecting the experimental and analytical
values of the model using scaling law Eq. (4a). [1] Mandell JF, Combs DE, Samborsky DD. Fatigue of fiberglass beam
substructures. Wind Energy 1995;16:99.
[2] Cairns DS, Skramstad JD, Mandell JF. Evaluation of hand lay-up and resin
transfer molding in composite wind turbine blade structures. International
SAMPE Symposium and Exhibition. SAMPE; 1999. p. 967–80.
[3] Mandell JF, Creed Jr RJ, Pan Q, Combs DW, Shrinivas M. Fatigue of fiberglass
generic materials and substructures. Wind Energy 1994;207.
[4] Laustsen S, Lund E, Kühlmeier L, Thomsen OT. Development of a high-fidelity
experimental substructure test rig for grid-scored sandwich panels in wind
turbine blades. Strain 2014;50:111–31.
[5] Sayer F, Post N, Van Wingerde A, Busmann HG, Kleiner F, Fleischmann W, et al.
Testing of adhesive joints in the wind industry. European wind energy
conference and exhibition 2009, EWEC 2009. p. 288–315.
[6] Sayer F, Antoniou A, van Wingerde A. Investigation of structural bond lines in
wind turbine blades by sub-component tests. Int J Adhes Adhes
2012;37:129–35.
[7] Kleiner F, Sayer F, Antoniou A, van Wingerde A, Trusheim M, Bernst P, et al. A
novel sub-component test method to qualify structural adhesives in wind
turbine blades. In: 34th Annual Meeting of the Adhesion Society; 2011. p. 13–
6.
[8] Zarouchas DS, Makris AA, Sayer F, Van Hemelrijck D, Van Wingerde AM.
Investigations on the mechanical behavior of a wind rotor blade
subcomponent. Compos B Eng 2012;43:647–54.
[9] Zarouchas D, Antoniou A, Sayer F, Van Hemelrijck D, van Wingerde A.
structural integrity assessment of blade’s subcomponents using acoustic
emission monitoring. Experimental and applied mechanics, vol. 6. Springer;
2011. p. 511–8.
[10] Jensen FM, Falzon BG, Ankersen J, Stang H. Structural testing and numerical
simulation of a 34 m composite wind turbine blade. Fifteenth International
Conference on Composite Materials ICCM-15 Fifteenth International
Conference on Composite Materials. 2006;76:52–61.
Fig. 11. Comparison between analytical and predicted values for the deflection of [11] Tippmann JD, Manohar A, di Scalea FL. Wind turbine inspection tests at UCSD.
the bottom flange of the prototype by projecting the experimental and analytical SPIE Smart Structures and Materials+ Nondestructive Evaluation and Health
values of the model using scaling law Eq. (4b). Monitoring: International Society for Optics and Photonics; 2012. p. 83451Q-
Q-7.
[12] White D, Musial W, Engberg S. Evaluation of the B-REX fatigue testing system
for multi-megawatt wind turbine blades. Collection of the 2005 ASME Wind
web of a commercial-scale wind turbine rotor blade. The scaling
Energy Symposium Technical Papers at the 43rd AIAA Aerospace Sciences
factors were derived from the governing equations based on the Meeting and Exhibit. p. 52–65.
standard similitude procedure. The scaled models were designed [13] Simitses GJ, Rezaeepazhand J. Structural similitude for laminated structures.
using the derived scaling laws and their accuracy in satisfying Compos Eng 1993;3:751–65.
[14] Rezaeepazhand J, Simitses GJ. Use of scaled-down models for predicting
the similarity conditions. The layup for the flanges of the scaled vibration response of laminated plates. Compos Struct 1995;30:419–26.
model was selected using the distorted layup scaling approach. [15] Rezaeepazhand J, Simitses GJ, Starnes Jr JH. Design of scaled down models for
The designed scaled-down I-beam was manufactured and tested stability of laminated plates. AIAA J 1995;33:515–9.
[16] Simitses GJ, Rezaeepazhand J. Structural similitude and scaling laws for
in a three-point bending test configuration under incremental buckling of cross-ply laminated plates. J Thermoplast Compos Mater
loads while DIC cameras measured the displacement field and cal- 1995;8:240–51.
culated the strain distribution of the bottom flange of the I-beam. [17] Rezaeepazhand J, Simitses GJ, Starnes Jr JH. Design of scaled down models for
predicting shell vibration response. J Sound Vib 1996;195:301–11.
The measured transverse deflection values were compared to the [18] Rezaeepazhand J, Simitses GJ. Structural similitude for vibration response of
analytical model and finite element simulation for model valida- laminated cylindrical shells with double curvature. Compos B Eng
tion. The derived scaling laws were subsequently applied to the 1997;28:195–200.
[19] Jackson KE. Scaling effects in the static and dynamic response of graphite-
analytical and experimental strain and displacement values of epoxy beam-columns. Virginia Tech; 1990.
the tested I-beam were used to estimate the strain and displace- [20] Camponeschi Jr ET. The effects of specimen scale on the compression strength
ment fields of the prototype for the adjusted loads. Excellent corre- of composite materials. Scal Eff Compos Mater Struct 1994.
822 M.E. Asl et al. / Composite Structures 176 (2017) 812–822

[21] Johnson D, Morton J, Kellas S, Jackson K. Scaling Effects in the Tensile and [34] Haselbach PU, Bitsche R, Branner K. The effect of delaminations on local
Flexure Response of Laminated Composite Coupons. NASA CONFERENCE buckling in wind turbine blades. Renew Energy 2016;85:295–305.
PUBLICATION: NASA; 1994. p. 265-. [35] Branner K, Berring P. Compressive strength of thick composite panels. In:
[22] Grimes GC. Experimental observations of scale effects on bonded and bolted Symposium on materials science proceedings 2011. p. 221–8.
joints in composite structures. Scal Eff Compos Mater Struct 1994. [36] Chen X, Zhao W, Zhao XL, Xu JZ. Preliminary failure investigation of a 52.3 m
[23] Obrien TK. Damage and strength of composite materials: trends, predictions, glass/epoxy composite wind turbine blade. Eng Fail Anal 2014;44:345–50.
and challenges. Scal Eff Compos Mater Struct 1994. [37] Griffith DT, Ashwill TD. The Sandia 100-meter all-glass baseline wind turbine
[24] Meruane V, De Rosa S, Franco F. Numerical and experimental results for the blade: SNL100-00. Sandia National Laboratories, Albuquerque, Report
frequency response of plates in similitude. Proc Inst Mech Eng Part C 2015. NoSAND2011-3779; 2011.
0954406215610148. [38] Lee J. Flexural analysis of thin-walled composite beams using shear-
[25] Luo Z, Wang Y, Zhu Y, Wang D. The dynamic similitude design method of thin deformable beam theory. Compos Struct 2005;70:212–22.
walled structures and experimental validation. Shock Vibr 2015;2016. [39] Reddy JN. Mechanics of laminated composite plates and shells: theory and
[26] Torkamani S, Navazi H, Jafari A, Bagheri M. Structural similitude in free analysis. CRC Press; 2004.
vibration of orthogonally stiffened cylindrical shells. Thin-Walled Struct [40] Tesauro A, Eder MA, Nielsen M. Measurement of local relative displacements
2009;47:1316–30. in large structures. J Strain Anal Eng Des 2014;49:301–14.
[27] Asl ME, Niezrecki C, Sherwood J, Avitabile P. Predicting the vibration response [41] Ilbeigi S, Chelidze D. Model order reduction of nonlinear Euler-Bernoulli
in subcomponent testing of wind turbine blades. In: Special topics in beam.. Nonlinear dynamics, vol. 1. Springer International Publishing; 2016. p.
structural dynamics, vol. 6. Springer International Publishing; 2015. p. 115–23. 377–85.
[28] Asl ME, Niezrecki C, Sherwood J, Avitabile P. Application of structural [42] Ilbeigi S, Chelidze D. Reduced order models for systems with disparate spatial
similitude theory in subcomponent testing of wind turbine blades. Proc Am and temporal scales. Rotating machinery, hybrid test methods, v-acoustics &
Soc Compos 2014:8–10. laser vibrometry, vol. 8. Springer International Publishing; 2016. p. 447–55.
[29] Asl ME, Niezrecki C, Sherwood J, Avitabile P. Similitude analysis of composite I- [43] Asl M, Niezrecki C, Sherwood J, Avitabile P. Similitude analysis of the strain
beams with application to subcomponent testing of wind turbine blades. In: field for loaded composite I-beams emulating wind turbine blades. In:
Experimental and applied mechanics, vol. 4. Springer International Publishing; Proceedings of the American Society for Composites: Thirty-First Technical
2016. p. 115–26. Conference 2016.
[30] Kensche CW. Fatigue of composites for wind turbines. Int J Fatigue [44] Asl ME, Niezrecki C, Sherwood J, Avitabile P. Similitude analysis of thin-walled
2006;28:1363–74. composite I-beams for subcomponent testing of wind turbine blades. Wind
[31] Raman V, Drissi-Habti M, Guillaumat L, Khadhour A. Numerical simulation Eng 2017. http://dx.doi.org/10.1177/0309524X17709924.
analysis as a tool to identify areas of weakness in a turbine wind-blade and [45] Asl ME, Niezrecki C, Sherwood J, Avitabile P. Design of scaled-down composite
solutions for their reinforcement. Compos B Eng 2016;103:23–39. I-beams for dynamic characterization in subcomponent testing of a wind
[32] Haselbach PU, Branner K. Initiation of trailing edge failure in full-scale wind turbine blade. Shock & vibration, aircraft/aerospace, energy harvesting,
turbine blade test. Eng Fract Mech 2016;162:136–54. acoustics & optics, vol. 9. Springer International Publishing; 2016. p. 197–209.
[33] Branner K, Berring P, Haselbach PU. Subcomponent testing of trailing edge [46] Asl ME, Niezrecki C, Sherwood J, Avitabile P. Vibration prediction of thin-
panels in wind turbine blades. In: Proceedings of 17th European Conference on walled composite I-beams using scaled models. Thin-Walled Struct
Composite Materials 2016. 2017;113:151–61.