Professional Documents
Culture Documents
Determination of As Built Properties of Fiber Reinforced Po - 2022 - Composites
Determination of As Built Properties of Fiber Reinforced Po - 2022 - Composites
Keywords: The fiber volume fraction (FVF) and porosity in fiber reinforced polymers (FRPs) depends strongly on the
Porosity manufacturing process. These parameters influence the mechanical properties and thus the performance of
Fiber volume fraction an FRP. For this research, an epoxy-pre-impregnated glass FRP was investigated to determine the FVF and
Void content
matrix mass fraction taking into consideration all material constituents, including sizing, stitching thread, as
SEM
well as the porosity, the area density and the fiber orientations of each lamina, the filament fiber diameter,
Micro-CT
Calcination
and the inter-laminar void size and shape. Therefore, samples from a commercially manufactured wind
Image processing turbine rotor blade were experimentally investigated using scanning electron (SEM) and high-resolution X-ray
Fiber reinforced polymer microscopy (micro-CT), as well as a standardized calcination method and geometric measurements. Post-
Pre-preg processing techniques such as thresholding and edge detection were used to analyze the images. There was
good FVF agreement between SEM and the method of calcination. Micro-void cross-sectional shapes were well
captured by SEM while meso- and macro-voids were volumetrically resolved with a reproducible void size
distribution for two sample volumes by micro-CT.
1. Introduction of the preform typically observable with the naked eye. The formation
of micro- and meso-voids is controlled by the resin flow velocity and
The mechanical performance of fiber reinforced polymers (FRPs) is capillary within and between the fiber bundles, whereas macro-voids
determined by the properties of the constituent materials, the quality form as a result of the macroscopic flow, i.e., between fabrics [12].
of the adhesion at the interface between fibers and matrix, the fabric The void shapes, also called morphologies, for uni-directionally
texture, e.g., non-crimp or woven, the stacking sequence, and the oriented pre-impregnated fabrics were reported to change with an in-
manufacturing processes, i.e., curing cycle and vacuum application. In crease in the void content, from small and spherical (volatile-induced)
particular, the as-built fiber volume fraction (FVF) affects the stiffness voids between laminae to inter-laminar (air-entrapment) voids, which
are ‘‘flattened and elongated’’ [13]. Hsu and Uhl [14] conducted a
and strength properties of the FRPs [1,2]. It also impacts the fatigue
‘‘manual’’ tomography by cutting consecutive thin slices perpendicular
failure of a non-crimp fiber reinforced epoxy, as reported in [3,4].
and parallel to the fiber direction, and performed microscopy and
Furthermore, the local FVF on the microscopic level directly affects the
image analysis. They observed that voids, which were mainly inter-
fatigue damage growth rate in FRPs [5,6].
laminar, do not have a circular cross-section, but mainly flat irregular
Secondary manufacturing effects, like the formation or trapping of cross-sections. Most of them were elongated in the adjacent fiber direc-
air voids/inclusions during the layup process, deteriorate the properties tion, and voids with large cross-sections were quite long (‘‘needle-like
and performance of the FRP [7–11]. voids’’) [7]. Gürdal et al. [15] observed that intra-laminar voids are also
Mehdikhani et al. [7] comprehensively reviewed the state-of-the-art elongated mainly in the fiber direction to create a ‘‘cylindrical morphol-
in void size and shape in FRPs. Excerpts related to this research are ogy and elliptic cross-section’’, with an average length-to-radius ratio
presented in the following paragraphs. of ≈ 40.
Within FRPs voids appear on three different scales: Micro-voids are Traditionally, the FVF, fiber mass fraction, and porosity of an FRP
formed between fibers of a fiber bundle (intra-laminar), meso-voids in have been characterized by physical removal of the fiber surround-
between the bundles (inter-laminar), and macro-voids in a larger zone ings, e.g., through a calcination (burnoff in a muffle furnace) [16]
∗ Corresponding author.
E-mail address: malo.rosemeier@iwes.fraunhofer.de (M. Rosemeier).
https://doi.org/10.1016/j.jcomc.2022.100310
Received 7 June 2022; Received in revised form 11 August 2022; Accepted 22 August 2022
Available online 27 August 2022
2666-6820/© 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
2
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
Table 3 where 𝑛f denotes the total number of fabrics, 𝜚At𝑖 the area density of the
Raw material mass densities. stitching thread, and 𝜚Af𝑖 the sum of all fiber area densities including
Material 𝜚 in kg m−3 sizing of the 𝑖th fabric:
Matrix (epoxy for pre-impregnation) [34]a 1100–1220 𝑛k
Fiber (E-glass) [35] 2530–2600 1 ∑
𝜚Af𝑖 = 𝜚 , (5)
Stitching (polyethersulfone) [2] 1370 𝑛k 𝑘=1 Af𝑘
Sizing [(3-aminopropyl)triethoxysilane] [35,36] 996
where 𝑛k denotes the total number of laminae. The pure fiber mass
a
Assumption derived from [37]: ‘‘Prepregs which use lower molecular weight [. . . ]
fraction of one fabric type is calculated with
epoxy resins [. . . ] include those sold [. . . ] under the product name M9 [34] [. . . ] and ( )
[. . . ] WE91 [29,33]’’. 𝜚Ag𝑖 1 − 𝜓m𝑖
𝜓g𝑖 = , (6)
𝜚Ag𝑖 + 𝜚As𝑖 + 𝜚At𝑖
where
( )
𝜚Ag𝑖 = 𝜚Af𝑖 1 − 𝜓sf . (7)
Assuming that voids do not significantly contribute to the total
mass, the mass fraction of the matrix is calculated using
𝜓m = 1 − 𝜓g − 𝜓s − 𝜓t . (8)
The pure FVF without sizing, i.e., considering four phases, is calculated
using
𝜓g 𝜚m
𝜑4p = . (9)
𝜓g 𝜚m + 𝜓m 𝜚g
Eq. (9) can also be expressed as a function of the respective volume of
each phase 𝑉𝑗 as
𝑉g
Fig. 2. Representative volume element (RVE) of a laminate stack (a), lamina orienta- 𝜑= . (10)
tions of 𝑖th fabric (b), fiber bundles and inter-lamina voids of 𝑘th lamina (c), fiber,
𝑉g + 𝑉m
sizing, and intra-laminar voids (d). Assuming that the area of each phase 𝐴𝑗 within a cross-sectional cut
in the 𝑠𝑡- or 𝑧𝑡-plane extrudes in the respective perpendicular direction
through the RVE, Eq. (10) can be expressed as
2.2. Calculation of fiber volume fraction and porosity 𝐴g
𝜑= . (11)
𝐴g + 𝐴m
To determine the as-built constituent material fractions on three
different levels, i.e., on stack, fabric, and lamina level (Fig. 2), we Taking into consideration the density of each phase (Table 3), the total
introduce the four-phase model of a representative volume element density of all four phases without voids is then calculated using [2]
(RVE) of an FRP as suggested by Krimmer [2]. Three phases define 1
𝜚tot = 𝜓g 𝜓s 𝜓t 𝜓m
. (12)
the load carrying constituents, i.e., fiber (index g), matrix (index m), + + +
𝜚g 𝜚s 𝜚t 𝜚m
and stitching thread (index t). These three phases define the structural
properties of a laminate such as stiffness or coefficient of thermal The actual density of the FRP stack including voids can be either deter-
expansion, which are typically determined using classical laminated mined experimentally, e.g., in accordance with [40], or by geometric
plate theory, as described by Reddy [38], for example. Moreover, sizing measurements of the RVE with
(index s) is considered as the fourth phase, as well as voids (index v), 𝑚
𝜚totv = tot . (13)
which represent gas inclusions, e.g., air. The fabric (index f) consists of 𝑉tot
the three phases: fiber, sizing, and stitching thread. According to Method II in ASTM D3171 [17] (ASTM II), which consid-
The mass fraction of the pure glass fibers is calculated using ers only two phases, i.e., fiber and matrix, and neglects the presence of
𝑚g voids, the FVF for one fabric type is calculated as
𝜓g = , (1) ∑𝑛f
𝑚tot 𝜚Af𝑖
𝜑2p = 𝑖=1 , (14)
in which 𝑚g is the mass of the fibers and 𝑚tot the total mass of a stack 𝜚g 𝑡 𝑖
of fabrics within the RVE. The mass fraction of the sizing is calculated
where 𝑡𝑖 denotes the measured thickness of one fabric within the RVE.
using the equation from [2]
The smeared porosity of the FRP stack is calculated as in [2]
𝜓sf
𝜓s = 𝜓g , (2) 𝜚
1 − 𝜓sf 𝜉FRP = tot − 1. (15)
𝜚totv
in which 𝜓sf denotes the sizing mass fraction of the pure glass fiber
Eq. (15) can also be expressed as a function of the respective volume
plus sizing mass of the fabric. Technical data sheets (TDSs) of typical
𝑉𝑗 as
glass fiber bundles give a mass fraction of 𝜓sf = 0.55% for FRPs of type
Hybon 2001/2002 [39], for example. The mass fraction of the stitching 𝑉v
𝜉FRP = . (16)
thread is calculated as in [2] 𝑉g + 𝑉s + 𝑉t + 𝑉m
( ) 𝜓tf or in accordance with the assumptions made for Eq. (11) as a function
𝜓t = 𝜓g + 𝜓s , (3)
1 − 𝜓tf of the respective cross-sectional area 𝐴𝑗
in which 𝜓tf denotes the stitching thread mass fraction of the whole 𝐴v
stack of fabrics (Fig. 2a) using 𝜉FRP = . (17)
𝐴g + 𝐴s + 𝐴t + 𝐴m
𝑛f
1 ∑ 𝜚At𝑖 In reality though, the fibers, the sizing, and the stitching thread do
𝜓tf = , (4)
𝑛f 𝑖=1 𝜚Af𝑖 + 𝜚At𝑖 not contain voids, so the total density including voids 𝜚totv is corrected
3
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
Fig. 3. Original SEM image (a), void isolation (b), void mask applied to analyze the remaining area (c), and segmented image highlighting pure glass fibers (white), matrix (black)
and voids (gray) (d); red rectangles show partitions in the 𝑠-direction. (For interpretation of the references to color in this figure legend, the reader is referred to the web version
of this article.)
by these three phases to obtain the actual matrix density including 2.4. Scanning electron microscopy and image processing
voids [2]
𝜓m 𝜚totv SEM was performed using a high-resolution, low-voltage scanning
𝜚mv = ( ). (18) electron microscope of type Zeiss DSM 982 Gemini. For the SEM
𝜓 𝜓 𝜓
1 − 𝜚totv 𝜚 g + 𝜚 s + 𝜚 t assessment and the subsequent image analysis, the scanning parameters
g s t
selected are described in the following. The intention of the SEM
The porosity of the matrix is then calculated using [2] scan was to achieve the largest possible homogeneous image quality.
𝜚 This reduces potential drawbacks during the image stitching for the
𝜉m = m − 1. (19)
𝜚mv subsequent image analysis. To achieve this target, the largest possible
working distance was set. According to the Intercept theorem, the
This value describes the deviation of the matrix density including the
lowest magnification is attained in this geometric position. In the
void volume from the matrix density without void volume. Eq. (19) can
microscope, the maximum possible working distance is 33.9 mm. For
also be expressed as a function of the respective volumes 𝑉𝑗
the optimal magnification of 20x, an acceleration voltage of 15 kV was
𝑉v selected. The aperture was selected to be 10 μm since a larger diameter
𝜉m = (20)
𝑉m would result in a noticeable defocusing at the sample edges, which
would negatively impact the image analysis. The number of pixels is
or in accordance with the assumptions made for Eq. (11) as a function
limited both by the resolution of the smallest structures to be resolved
of the respective cross-sectional area 𝐴𝑗
and by the design of the scanning generator. In combination with the
𝐴v image analysis and the size of the relevant structures of approximately
𝜉m = . (21)
𝐴m 16 μm, a pixel resolution of 13,000 × 13,000 pixels was chosen. Further-
more, the pixel resolution in conjunction with the signal-to-noise ratio
and the size of the aperture is decisive for determining the integration
2.3. Calcination method time per pixel. Therefore, an integration time of 10 μs per pixel was
chosen, which resulted in an image acquisition time of 30 min.
The mass fraction of the fiber was determined experimentally. The SEM-sample cut-outs (Fig. 1c,d) were cast in epoxy, and the
Therefore, the two volumes of the Cube-sample (Fig. 1c), i.e., SS and surface was ground to achieve sufficient smoothness. The SS and PS
PS laminate stack, were mechanically separated from the coating and were scanned separately to increase the resolution per scan. For the
adhesive layer (Table 1). A calcination method in accordance with [16] examination of the fibers oriented in both the 𝑧- and 𝑠-directions, scans
was conducted for the two samples. Each sample was placed and dried were performed in the perpendicular 𝑠𝑡- and 𝑧𝑡-planes. This resulted in
in a ventilated heating cabinet at 105 °C. After 8 h, each sample was four sample scans: SS-𝑠𝑡, SS-𝑧𝑡, PS-𝑠𝑡, and PS-𝑧𝑡.
cooled to room temperature in a vacuum and subsequently scaled to The images were processed with an in-house script, which utilizes
determine the total mass. The two sample dimensions were measured the open-source tool OpenCV [41]. The high-resolution SEM scan, see
with a calibrated digital calliper, with a measurement precision of Fig. 3a, was processed using the script. The first step was to isolate the
±5 μm. Each sample volume 𝑉tot was calculated from the measured voids, which was accomplished through blur techniques, thresholding
dimensions. Subsequently, each sample was heated in a heating cabinet as introduced by Otsu [42], morphing, and contour find functions [43],
at 625 °C, resulting in the burn-off of the surrounding matrix, sizing and see Fig. 3b. The grid size and interactions were determined by a
stitching threads. Afterward, the remains were again scaled to derive trial-and-error process until the processed image was considered as rep-
the mass of the pure fibers 𝑚g . resentative of what the human eye interprets. To isolate the voids, large
grid sizes were necessary, whereas fibers have a much smaller cross-
The matrix mass fraction was subsequently calculated with Eq. (8).
sectional area and therefore require finer grids. The voids often caused
The FVF was calculated with Eq. (9), the smeared porosity with
a charging effect, which in turn caused shadows on the surrounding
Eq. (15), and the matrix porosity with Eq. (19). To account for the
fibers that were corrected manually in the event of severe disturbances.
uncertainty in the input values, a sensitivity analysis was carried out.
After being isolated, the voids were masked so that the subsequent
Therefore, to obtain the maximum and minimum deviation from the processing ignored those pixels, see Fig. 3c.
average, the uncertainty range of the raw material densities, as given The next step was to separate fibers from matrix. Another series
in Table 3, and the geometry measurements were combined. of morphology and thresholding was used to process the image into
Moreover, the FVF was further determined for the two fabrics which Fig. 3d, which shows the segmentation into gray voids, black matrix,
were contained in the FRP stack, i.e., XE600 and YE900. To this end and white fibers.
the matrix mass fraction (Table 2) and fiber mass fractions of each The FVF was calculated with Eq. (11) by assuming that the black
fabric type (Eq. (6)) were weighted according to the laminate plan (Ta- area of a segmented image (Fig. 3d) represents the matrix cross-
ble 1). The FVF was subsequently calculated for the four-phase-model sectional area 𝐴m . Since the images do not show any sizing area 𝐴s and
(Eq. (9)). only vaguely any stitching thread area 𝐴t , they were neglected. This
4
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
5
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
Table 4
Images and processing results of the SEM suction-side samples.
𝜚Ag𝜃
Sample image No. 𝜃 in ° 𝜉FRP in % 𝜑 in % 𝜚Ag
in %
a
Backing fibers.
A void area of 25 μm2 is encountered most frequently in the SS-𝑠𝑡- were predicted to appear most frequently in this sample
(Fig. 5(b)).
sample (Fig. 5(a)). This size was mainly allocated to inter-fiber voids.
On average, the porosity in the two 𝑧𝑡-samples is lower than in the
After statistically extrapolating the cross-sectional areas of these inter-
two 𝑠𝑡-samples (Fig. 6). When considering the porosity distributions
laminar voids to ellipsoids, void volumes between 500 to 2,000 μm3 through the thickness of the stacks, they seem to be in phase on the SS
6
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
Table 5
Images and processing results of the SEM pressure-side samples.
𝜚Ag𝜃
Sample image No. 𝜃 in ° 𝜉FRP in % 𝜑 in % 𝜚Ag
in %
a
Backing fibers.
Table 6 Table 7
Filament diameters of fibers identified in SEM images. Cube-sample; matrix mass fractions in %.
Filament diameter in μm Sample size Volume TDS-weighted Measured 𝛥𝜓m
Main fiber 15.5 ± 2.3 51 SS 36.7 39.2 +2.5
Backing fiber 8.4 ± 1.0 49 PS 38.2 41.8 +3.6
and PS. In the SS sample, a peak of 25.2% is observed at approx. 95% these two values (Table 7) for the SS- and PS-volume, the measured
of the thickness on lamina 1.1. In the PS sample, a peak of 27.8% is matrix mass fraction is slightly greater than the one according to the
observed at approx. 55% of the thickness on lamina 3.1. TDS.
Table 6 summarizes the filament diameters of the main and backing The void size within two volumes of the Cube-sample (Fig. 7) varies
fibers, which were measured on a sample basis in the SEM images. between 1.6e+4 to 1.4e+10 μm3 (Fig. 8). The most frequent void volume
size counted is at approx. 0.12e6 μm3 in the SS- and PS-volume. Void
3.2. Cube-sample volumes <1.6e+4 μm3 were not captured due to the micro-CT settings
and the resulting resolution limitations used in this research.
This section presents the following metrics determined for the Cube- The void shape differs depending on its size and location. Larger
sample: matrix mass fraction, void size, void shape, and porosity. voids or void clusters are located between adjacent fabrics (inter-
First, the matrix mass fraction 𝜓m was weighted using the TDS laminar voids). They are of arbitrary shape when projected onto the
values according to the laminate plan. Second, 𝜓m was determined 𝑠𝑧-plane. If projected onto the 𝑠𝑡- or 𝑧𝑡-plane, they appear flat or elliptic.
according to Eq. (8) from the fiber mass measurement using the cal- There are also a few voids with large length-to-diameter ratios which
cination method and the area densities of the fabric types. Comparing appear cigar-like and are oriented parallel to adjacent fiber bundles
7
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
Fig. 6. Smeared porosity as function of the relative position along the 𝑡-direction in
SS (a) and PS (b) samples.
Fig. 5. Frequency of void cross-section areas (a) and extrapolated minimum and
maximum void volumes (b) of SEM-sample in SS-𝑠𝑡-sample.
and between 0.1% and 16% in the SS-volume (Fig. 9(a)). The standard
Table 8 deviations of the porosity extracted from the spatial distributions in the
Phases considered by methods. 𝑧-, 𝑠- and 𝑡-directions are 1.0%, 2.4%, and 2.9% in the SS-volume, and
Abbrev. Phases/voidsa Method 1.1%, 2.8%, and 6.1% in the PS-volume.
4pv gstmv Calcination
2pv g (s t m) v SEM 3.3. Summary
2p (g s) (t m v) ASTM II
1pv (g s t m) v micro-CT
This section presents the following metrics determined for the SEM-
a
g = fiber, s = sizing, t = stitching thread, m = matrix, v = void; brackets indicate a samples and the Cube-sample: FVF, smeared porosity, and matrix poros-
smeared phase. ity. Table 8 summarizes the material phases considered and how voids
were taken into account by each method.
The FVF on the SS is slightly smaller than on the PS when consider-
visible in ±45° lamina, e.g., lamina 5.2 or 5.3 in the PS, cf. Table 5 and ing the average values determined by calcination for the Cube-sample
Fig. 7e. A very large void is located in the mid-plane of the PS-volume and by SEM for the 𝑠𝑡-sample (Fig. 10(a)). On the SS, the FVF deter-
(Fig. 7e). Smaller and most frequent void volumes become visible after mined by SEM for the 𝑠𝑡-sample reveals almost the same average and
excluding larger voids >2.5e6 μm3 as shown in Fig. 7c,d. Most small uncertainty as the FVF determined by calcination for the Cube-sample;
voids appear spherical and are located in matrix-rich areas between a similar situation applies to the SEM for the 𝑧𝑡-sample compared to
fiber bundles (meso-voids). calcination of the PS-volume of the Cube-sample. Taking into account
The average smeared porosity in the Cube-sample is 4.4% in the the uncertainties, however, the FVF determined by SEM for the PS-𝑧𝑡-
SS-volume and 4.8% in the PS-volume. There is a significant spatial sample ranges over the FVFs revealed on the PS. In other words: All
scatter along the volume directions (Fig. 9). The very large void seen error bars overlap each other on both sides except the FVF determined
in the mid-plane of the PS-volume (Fig. 7e) is visible as a 40%-peak by calcination of the PS-volume of the Cube-sample.
at 55% sample thickness (𝑡-direction). When neglecting this peak, the The FVF of the XE600 fabric determined by means of the 2p ASTM II
porosity varies between 0.3% and 15% in the PS-volume (Fig. 9(b)), model according to [17] varies significantly from the TDS value. The 2p
8
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
Fig. 7. Segmented micro-CT-sample; SS-volume (top) and PS-volume (bottom); fiber bundles (gray-scale) and voids (red) (a,d), isolated voids (b,e), and isolated void entities
<2.5e6 μm3 (c,f). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
ASTM II model lies on average below the 4pv model, in particular for
XE600, but with the exception of the PS-samples. The uncertainty in the
FVF determined by SEM is significant, in particular in the SS-𝑠𝑡-samples
for XE600 and YE900 and in the PS-sample of the SEM. Moreover,
the average FVF varies significantly in the SEM samples of the XE600.
There seems to be good agreement between the 4pv calc Cube-sample
and the SEM PS-sample for XE600. Also in good agreement are the
4pv calc of the SS-Cube-sample with the SS-st-sample and the 4pv calc
PS-Cube-sample with the PS-st-sample.
The error bars of the smeared porosity, which are quite wide in part,
overlap at approx. 5% (Fig. 10(c)). Note that the large void in the PS of
the Cube-sample is considered in the average of the micro-CT results,
but excluded from the uncertainty. Similar to the porosity determined
by micro-CT for the Cube-sample, the porosity determined by SEM for
the 𝑠𝑡-sample is larger on the PS than on the SS, while the porosity
determined by SEM for the 𝑧𝑡-sample is similar on the two sides.
Similar to the smeared porosity, an overlap of error bars is obtained
for the matrix porosity at approx. 8.7% (Fig. 10(d)). The relative level
of the average values as well as the error bars between the samples is Fig. 9. Cube-sample; smeared porosity along the three directions in SS- (a) and
comparable to those of the smeared porosity. PS-volume (b).
9
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
Fig. 10. Summary of results for SEM-samples and Cube-sample: fiber volume fraction (a), smeared porosity (b), and matrix porosity (c); error bars indicate: minimum and
maximum for calcination, standard deviation in 𝑠- or 𝑧-direction for SEM and standard deviation in 𝑠- and 𝑧-direction for micro-CT.
10
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
on the SS. On the PS, the qualitative results are similar to those of Volume-related methods such as micro-CT could improve the sta-
the SS. There is, however, a smaller difference between the porosity tistical representativeness. Therefore, the scan volume needed to be
for the Cube-sample determined by calcination and micro-CT, and a large enough, or a sufficient number of samples needed to be ana-
larger difference between the two SEM-samples, which contradicts the lyzed. Larger meso-voids located between fiber bundles and between
conclusions drawn for the PS-samples. The reasons for this can be the two fabrics could be captured well, and the method provided three-
larger uncertainty in the SEM-PS-samples compared to the SS-samples dimensional information on size, location, and shape of individual
or a greater spatial variation between the two SEM-samples and the voids. For the identification of micro-voids within fiber bundles, how-
Cube-sample on the PS compared to the two SEM-samples and the ever, the resolution must be very high, resulting in a small sample size,
Cube-sample on the SS. short sample-to-source distance, and long sample-to-detector distance.
The micro-CT analysis revealed a greater spatial scatter in porosity The advantage of higher resolution will come with the compromise
along the 𝑠- and 𝑡-directions when compared to the 𝑧-direction (Fig. 9). of smaller fields of view, i.e., smaller scan volumes. Consequently,
The local maxima in the 𝑡-direction are related to large voids entrapped more scans would be required to capture the same volume. A higher
between adjacent fabrics (inter-laminar voids), which can qualitatively resolution would also make it possible to segment fiber filaments,
be correlated with the porosity distributions determined for the SEM which is necessary to determine the FVF.
samples (Fig. 6). The main source for those inter-laminar voids is air None of the optical methods could representatively identify the
entrapment that occurs during the lay-up process of pre-impregnated stitching thread and sizing fractions.
FRPs [9]. The scatter in the 𝑠-direction is caused by longitudinal meso-
voids (intra-laminar voids) in matrix-rich regions between 0°-oriented 5. Conclusions
(𝑧-direction) fiber bundles. The main source for those intra-laminar
voids is air entrapment that occurs during the impregnation process As-built properties of a cut-out from a commercially manufactured
of dry fabrics [47]. blade were analyzed by a procedure that combines a four-phase con-
stituent material model, which imported measurements from a stan-
4.3. Void size and shape dardized calcination method, with image analyses from SEM and micro-
CT recordings.
The shape of voids tends to be spherical if there is enough space There was good FVF agreement on the FRP stack level between
available, i.e., micro- and meso-voids in matrix-rich areas without calcination and SEM, and good agreement in part on the fabric level,
constraints. The available space is controlled by the pressure and in particular for the YE900 pre-impregnated fabric type.
capillary during the manufacture of pre-impregnated FRPs via resin Micro-void cross-sectional shapes were well captured by SEM, while
transfer molding [12]. If fiber bundles or single fibers act as constraints, meso- and macro-voids were volumetrically resolved with a repro-
voids become longitudinal and stretched. If adjacent fabric layers act ducible void size distribution for two sample volumes by micro-CT.
as constraints, voids become flat, but can have a more arbitrary shape
in the 𝑧𝑠-plane. 6. Supplemental material
If more space is available, voids tend to become larger, as multiple
voids can aggregate to large voids. The data presented in the figures are available in a data repository
These meso-voids appear almost circular in the SEM images (Ta- at [48]. Moreover, the repository contains the original as well as the
bles 4 and 5) on the 𝑠𝑡-plane in the 0° laminae, as they are cut segmented SEM images in a high resolution.
transversely. Moreover, longitudinal meso-voids appear in the matrix-
rich areas between fiber bundles (Section 3.2), which are also oriented CRediT authorship contribution statement
parallel to the fiber bundles.
The most frequent size of micro-voids extrapolated from the SEM Malo Rosemeier: Initiated and coordinated the research, Writing –
images ranges between 500 to 2,000 μm3 , while the most frequent size review & editing. Catherine Lester: Conducted the image processing
of meso-voids was counted to be approx. 120,000 μm3 . of the SEM images, Writing – review & editing. Alexandros Antoniou:
Initiated and coordinated the research, Writing – review & editing.
4.4. Applicability of methods Christoph Fahrenson: Created the SEM images, Writing – review
& editing. Nikolas Manousides: Created the micro-CT reconstruc-
Calcination is a standardized method to determine the FVF and tions and conducted the post-processing, Writing – review & edit-
porosity of an FRP. The FVF could be determined with a relatively low ing. Claudio Balzani: Conducted the post-processing of the micro-CT
uncertainty but the porosity was subject to a high degree of uncertainty. reconstructions, Writing – review & editing.
Any information about voids, i.e., frequency of size, location, or shape,
cannot be measured. The ASTM Method II overestimated the FVF Declaration of competing interest
because the FVF is determined via geometric measurements and does
not take account of for porosity. The authors declare that they have no known competing finan-
Surface-related methods such as SEM image analysis were useful to cial interests or personal relationships that could have appeared to
measure properties that are evenly distributed in the direction perpen- influence the work reported in this paper.
dicular to the image plane, e.g., the FVF could be determined accurately
in lamina cut by an 𝑠𝑡-plane and whose 0° orientation corresponded to Acknowledgments
the 𝑧-direction. The probability of cross-sectionally cutting a longitudi-
nal micro-void in a fiber bundle is expected to be high if the plane under We thank our colleagues Lisa Schudack, Irene Preuß, Konstantin
investigation is perpendicular to the fiber bundle (𝑠𝑡-plane). In the 𝑧- Kinsvater and Henning Schnellen, who prepared the samples and con-
direction, we observed the lowest variation in porosity taking meso ducted the experiments, and Basem Rajjoub, who carried out the man-
and macro voids into account (Fig. 9). Therefore, we can expect that ual segmentation of the two-dimensional images exported from the
a cut though the 𝑠𝑡-plane also represents meso-voids located between micro-CT reconstructions. Moreover, we thank our colleague Florian
fiber bundles and/or between two fabrics. Thus, we recommend that Sayer, who acquired the funding for this research. Finally, we would
a representative cutting surface, which transversely cuts mainly the 0° like to thank SSP Technology A/S for providing the samples for this
oriented laminae, be analyzed. research.
11
M. Rosemeier et al. Composites Part C: Open Access 9 (2022) 100310
12