International Journal of Material Forming

https://doi.org/10.1007/s12289-019-01477-4

ORIGINAL RESEARCH

Prediction of process-induced void formation in anisotropic

Fiber-reinforced autoclave composite parts
Bamdad Barari 1 & Pavel Simacek 2,3 & Shridhar Yarlagadda 3,4 & Roger M. Crane 3,4 & Suresh G Advani 2,3

Received: 9 October 2018 / Accepted: 18 January 2019

# Springer-Verlag France SAS, part of Springer Nature 2019

Abstract
A numerical methodology is proposed to predict void content and evolution during autoclave processing of thermoset prepregs.
Starting with the initial prepreg void content, the void evolution model implements mechanisms for void compaction under the
effect of the applied pressure, including Ideal Gas law compaction, and squeeze flow for single curvature geometries. Pressure
variability in the prepreg stack due to interactions between applied autoclave pressure and anisotropic material response are
considered and implemented. A parametric study is conducted to investigate the role of material anisotropy, initial void content,
and applied autoclave pressure on void evolution during consolidation of prepregs on a tool with single curvatures. The ability of
the model to predict pressure gradient through the thickness of the laminate and its impact on void evolution is discussed.

Keywords Void prediction . Prepreg . Anisotropic prepreg . Laminate . Autoclave process

Introduction unidirectional prepregs that can be stacked in the required

orientation sequence, with the resin system tailored for the
One of the traditional methods to fabricate composites is specific application by adding additives, such as flame
using prepreg layups. Prepregs contain reinforcing unidi- retardants, or toughening agents, in order to meet specific
rectional fibers or woven fabrics that are preimpregnated requirements. Flat laminates or structures can be reliably
with either a thermoplastic or thermoset resin. If a ther- made with this process without any anomalies such as
moset matrix is used, the resin is partially cured for ease wrinkles or other defects [3].
of handling and this b-stage prepreg must be heated in an For complex geometry parts with prepreg stacks,
oven or an autoclave for consolidation and complete cure manufacturing defects such as porosity, possible wrinkling
[1]. Composite parts can be made from such prepregs by and reinforcement disorientation which introduce gaps and
stacking them, heating the assembly, curing and consoli- overlaps within the laminate, will detrimentally affect the final
dating them under pressure. Autoclave prepreg processing mechanical properties (stiffness, strength) of the component
is one of the composite manufacturing methods typically being constructed especially for parts that have single and/or
used for high-end structural applications especially in the double curvature [4–9]. Curved cross-sectional laminate parts
aerospace industry, since it provides excellent mechanical are very diverse in design including, T-shaped, L-shaped, J-
characteristics with good reproducibility [2]. Moreover, shaped parts. The L-shaped part is one of the simplest designs
the m ater ia l be hav io r c an be tailored b y u sing of complex structures and is extensively used in the aircraft
industry. Several experimental and numerical papers have
been published regarding the compaction quality of the parts
* Bamdad Barari using the autoclave process [10–15]. The effect of the process-
bbarari@mit.edu
ing parameters, such as cure cycle, mold design, bagging con-
figuration and stacking sequence have been investigated to
1
Massachusetts Institute of Technology, Boston, MA, USA some extent [16, 17]. Several models have been developed
2
Department of Mechanical Engineering, University of Delaware, to predict composite resin flow in autoclave processing
Newark, DE, USA [18–25]. These studies were mainly focused on the consolida-
3
Center for Composite Materials, University of Delaware, tion of simple shaped laminates. Hubert et al. [19] and Li and
Newark, DE, USA Tucker [24] developed and implemented a finite element-
4
Composites Automation LLC, Newark, DE, USA based flow-compaction model for L shaped composite
 Int J Mater Form

laminates. Hubert et al. [19] used an incremental, quasi-linear incrementally Bdebulked^ under vacuum several times
elastic finite element model to simulate the multiple physical during the layup to reduce the content of trapped vola-
phenomena in autoclave processing. Li and Tucker [24] de- tiles. While this does not necessarily change the consoli-
veloped a hyper-elastic model for fiber reinforcement stress dation process, the amount of volatiles to be disposed of
where the mesh geometry and fiber orientation were updated is significantly reduced that way.
during the consolidation process. Dong [25] developed a mod- Thus, as the temperature and pressure varies during the
el for predicting the formation of the resin-rich zone in angled process, the material behavior has to be properly described
composite laminates. Dong’s model quantitatively related gap as a function of temperature and pressure to properly describe
thickness with the radius, fiber volume fraction, stacking se- consolidation and the initial porosity and void content should
quence and enclosed angle. Gutowski et al. [22] proposed a be reasonably described. These steps go beyond the scope of
3D flow and a 1D compaction model and characterized the this work and we will, optimistically, assume that they can be
composite as a deformable unidirectional fiber reinforced accomplished to full satisfaction.
structure where the load is balanced by the average resin At the beginning of the process, the stacked prepregs con-
pressure and the average effective stress in the fiber net- tain significant voids and potentially, volatiles such as mois-
work. Dave et al. [21] used a similar approach but their ture. Additional porosity may also develop during the process
model considered the flow in different directions to be because (i) the increase in temperature may evaporate the
coupled. However, the effect of compaction pressure and liquid volatiles in the system and (ii) curing reaction may
its interaction with the void content has not been addressed release by-products in gaseous state, however these effects
especially for part geometries with curvature. The evolu- are not being considered at this time.
tion of void content during processing and sensitivity to During the consolidation process, the void content with-
complex interactions between the locally applied pressure in the composite should be reduced to acceptable (~1 to 2%
and the non-uniform resin pressure experienced by the res- for aerospace components) range, which is generally ac-
in within the prepreg plies needs to be investigated. complished by subjecting the prepreg assembly to higher
In the present study, a pressure model is developed to pre- applied pressures. The resin pressure in the system during
dict through thickness resin pressure distribution and the con- the consolidation is not the sole force to carry the applied
solidation stress components on a ply-by-ply basis for an L- autoclave pressure. Consolidation requires volumetric and
shaped geometry. Ply-by-ply void content is tracked and void shape change of the prepreg stack and under this deforma-
evolution during processing is modeled considering void tion the fibrous bed within the prepreg develops the stress
compaction due to pressure (Ideal Gas law) and squeeze flow field and carries part of the applied load. This combined
of resin into voids. Constitutive equations are developed to with the resin pressure, according to the Terzaghi equation,
describe the anisotropic material behavior of the prepreg as is equal to the total applied consolidation pressure.
well as the stress-strain relation which accounts for the resin Consequently, the resin pressure will always be less than
pressure term due to the squeeze flow and Ideal Gas law the applied consolidation pressure and an understanding of
compaction. The model is fully coupled with the the mechanisms that determine resin pressure as a function
ABAQUS solver through an UMAT subroutine. This of incoming material characteristics, geometry, and process
ABAQUS integrated approach allows one to import com- conditions is critical.
plex geometries, establish process conditions and address The resultant resin pressure can reduce void content by the
local void evolution due to non-uniform resin pressure at following mechanisms
corners or in specific complex geometry zones.
& Void size reduction due to volatile gas compression by the
higher external resin pressure than internal void pressure –
Model development and mechanisms Ideal Gas law-based compression
& Void size reduction by driving resin flow into voids from
In a typical autoclave process, thermoset prepreg layers are neighboring regions
processed at elevated temperature and pressure. Vacuum is & Dissolution of void gas into resin driven by solubility
drawn on the part to remove volatiles that can be removed (Henry’s Law)
that way, temperature is raised to reduce resin viscosity and & Diffusion mechanisms enabling transport of dissolved
pressure is build up to dissolve more volatiles and fill the species to the surface of the prepreg stack, typically
empty pores with resin. Finally, temperature – and sometimes driven by concentration gradients and thus related
pressure, too - are elevated even more to cure the resin. with the dissolution
Consolidating pressure is applied through a compliant bag
against the tool surface. Additional steps may be involved Note that the pressure in voids needs to be lower than
in the process. For example, thick layup stacks may be the resin pressure for all these mechanisms to occur. As
Int J Mater Form

thermoset prepreg is typically stored under ambient pres- the stress field that influence the deformation need to be
sure conditions, the general assumption is that the initial modeled. First, the reinforcement stress-deformation rela-
internal void pressure is in equilibrium with atmospheric tion due to the elastic, elasto-plastic, or viscoelastic aniso-
pressure prior to autoclave processing. tropic stress needs to be included depending on the mate-
The unidirectional fibers (e.g. carbon or glass) within rial type. Second, the average pressure within the liquid
each prepreg layer make the prepreg properties highly resin has to be related to the volumetric deformation of
anisotropic. The transverse compliance is relatively high the resin which is affected by the void hydrostatic pres-
for such prepreg layers with the in-plane stiffness in fiber sure developed during the compaction/consolidation pro-
direction being orders of magnitude larger. The latter cess and, consequently, the volatile concentration for the
plays an important role in curved regions as the mem- dissolution/diffusion process. To this one should add pres-
brane stresses can carry a part of the transverse compac- sure increase necessary to transfer the resin to where it is
tion pressure from the autoclave, if fibers are oriented needed (squeeze flow). Lastly, for any curved structure,
along the curve and may require significant strain in fiber the reaction between layers will influence the stress field
direction or sliding between layers to enable thickness within the reinforcement as it generates in-plane stresses
reduction during compaction. Note that this is specific to which, for curved surface, influences the transverse mo-
concave geometries, where compaction requires either ex- mentum conservation. Note that practically, the in-plane
tension or layer sliding due to increase in radius. On the stiffness in the fiber direction tends to be very significant
other hand, convex geometries undergo decrease in effec- and this effect is hardly negligible.
tive radius and consequent compression along the ply di- In this work, we address two of the above-mentioned
rection. This is the primary cause for wrinkling defects, three mechanisms. The first mechanism of stress defor-
and is not a focus of this paper. mation is described by anisotropic elasticity. The second
A resin with significant porosity is not incompressible. The mechanism relates the non-linear elastic component to
voids (bubbles) within resin contain certain amount of vola- the compression of voids and has a viscous component
tiles at some pressure (volatile pressure) which is related to the which in accordance with Darcy’s flow redistributes the
volume and the amount of captured gas. Change of volatile resin. The former utilizes ideal gas state equation for the
volume (porosity), pressure (volatile pressure) and mass is volatiles. The latter brings dependency on the mean
linked, say through ideal gas law though other constitutive spacing of the voids, resin viscosity and permeability of
relation may be used. the reinforcing bed. The third mechanism of sliding be-
The pressure in the resin is coupled with the volatile pres- tween layers is not considered as a laminate description
sure in the voids. Increase the resin pressure, the porosity was used for the cases we investigated, and shear defor-
gets compressed (and mass gets dissolved) and the volume mation is employed as the sole mechanism for the rela-
decreases. Thus, the entire system acts like a progressively tive displacement of the layers. Sliding or partial sliding
stiffening spring which reaches the incompressibility only – whether Coulombian or viscous – is not allowed in our
when the porosity vanishes. Note that the resin pressure model. This is equivalent to mechanical interlocking of
varies with location and is not identical to volatile pressure plies which would provide sufficient Bstatic friction.^
even if the surface tension in voids is neglected. If the voids This is a significant simplification but qualitatively it
are large and embedded in the fibrous bed, filling them with provides the in-plane stress field. The scope of this work
resin requires additional pressure to transfer (Bflow^) the is to investigate the effect of anisotropy and voids during
resin from locations without voids to those containing the consolidation of curved parts in an autoclave process
voids. This adds a potentially significant viscous Bdamper^ on final void content in the composite. Summary of the
to the behavior. Relative motion of resin (Bsqueeze flow^) main mechanisms that been used in the model are pre-
through the reinforcement needed to fill the void regions is sented in Table 1.
modeled using Darcy’s law, and has been shown to require
significant pressure and/or time to complete. Elastic behavior of fibrous reinforcement
The above approach focuses on calculation of the resin
pressure around voids which will be lower than the applied The first mechanism can be described through the aniso-
autoclave pressure due to the fiber bed carrying part of the tropic elasticity for the fibrous reinforcement in the pre-
applied load. Low resin pressure due to poor pressure transla- preg. For real materials, this is a rather complex relation.
tion is not desirable but it happens non-uniformly especially In the scope of this work we opted for simple but aniso-
around curved geometries. This can promote the evolution of tropic linearly elastic material, as the goal is to understand
voids within the resin due to insufficient resin pressure. the effect of the degree of anisotropy on resin pressure
To address the prediction of resin pressure within the and ultimately void content in complex geometries. The
layers of a complex geometry part, three components of fibrous bed is assumed to behave linearly and exhibits
 Int J Mater Form

Table 1 Summary of the actual mechanism and the mechanism we used in model

Actual autoclave process Proposed model assumption

Prepreg model Viscoelastic anisotropic response from reinforcement Linearly elastic anisotropic reinforcement behavior
Reinforcement
Prepreg model Resin pressure ▪ Resin pressure caused by hydrostatic pressure in voids ▪ Ideal gas law
▪ Additional resin pressure caused by resin redistribution ▪ 1D Darcy squeeze flow
(squeeze flow) ▪ Terzaghi equation used to combine pressure
and stress in fiber bed
Interface model ▪ Viscous Resin Interface (Viscous) and dry Friction (Plastic) ▪ Shear deformation is the dominant mechanism
▪ Failure/Initial Friction Before Relative Motion Can Occur in compaction process
▪ Shear deformation both before the initial failure and after ▪ No sliding is considered due to the laminate assumption
Dissolution model ▪ Dissolution of void gas into resin driven by solubility ▪ No dissolution of void gas into resin driven by
Diffusion model (Henry’s Law) solubility (Henry’s Law)
▪ Diffusion mechanisms enabling transport of dissolved ▪ No diffusion is considered
species to the surface of the prepreg

different material properties (i.e., elastic modulus and Modeling the average resin pressure field
Poisson’s ratio) in the transverse and fiber directions.
Therefore, nine material properties (E1, E2, E3, ν12, ν13, The resin itself is, for the loads considered herein, incompress-
ν23, G12, G13, G23) are needed to define the linear aniso- ible. It can carry unlimited and indeterminate hydrostatic pres-
tropic elastic material. sure to be added to the elastic force. However, the volatile
8 9 2 38 9 voids within the resin are compressible. Thus, for the second
>
> σ11 >> C 1111 C 1122 C 1133 0 0 0 >
> ε11 >
>
>
> σ22 >
> 6 7>> >
ε22 >
mechanism, volatiles resist compaction by developing hydro-
>
> >
> 6 C 2222 C 2233 0 0 0 7>> >
>
< = 6 <
7 ε33 = static pressure related to their compression. This pressure
σ33 C 3333 0 0 0
σ
¼6
6
7
7 > ε12 > ð1Þ
>
> 12 >
> 6 sym C 1212 0 0 7> > translates to resin pressure within the composite prepreg. If
>
> >
> 5>> >
>
> σ13 >>
4 C 1313 0 >
> ε > >
: ; : 13 >; resin is available around the voids, the resin pressure on the
σ23 C 2323 ε23
void interface can be assumed to be the pressure in the vola-
With tiles, the resin is essentially in a hydrostatic state and does not
move relative to the fiber bed. The existing models tend to add
C 1111 ¼ E 1 ð1−ν 23 ν 32 Þγ a viscous delay as the fluid shell around the void must deform,
C 2222 ¼ E 2 ð1−ν 13 ν 31 Þγ but the characteristic times of such process are so low that
C 3333 ¼ E 3 ð1−ν 12 ν 21 Þγ there is no need to consider it for the autoclave consolidation.
C 1122 ¼ E 1 ðν 21 þ ν 31 ν 23 Þγ However, for the resin to fill the voids insulated from the resin
C 1133 ¼ E 1 ðν 31 þ ν 21 ν 32 Þγ rich areas by the fiber bed, a pressure gradient must develop to
ð2Þ
C 2233 ¼ E 2 ðν 32 þ ν 12 ν 31 Þγ squeeze and drive the resin from these areas to resin starved
C 1212 ¼ G12 areas by the local relative motion of the resin through the fiber
C 1313 ¼ G13 bed. Thus, the average resin pressure will need to be larger
C 2323 ¼ G23 than the hydrostatic pressure in the voids. The squeeze flow is
γ ¼ 1=ð1−ν 23 ν 32 −ν 31 ν 13 −ν 12 ν 21 −2ν 21 ν 32 ν 13 Þ induced due to the compaction of the reinforcement layers
which creates a higher pressure and a pressure gradient be-
where E1 and E2 are the modulus in the fiber direction tween the voids which will drive the resin into the voids as
and transverse in-plane direction respectively, E3 is the governed by Darcy’s law. This increases the average resin
modulus normal to the plane of fiber bed, ν12, ν13, ν23 pressure above the Bboundary^ value in the voids. The effect
are the Poisson’s ratio and G12, G13, G23are the shear of this phenomenon is that the consolidation is a transient
moduli in the fiber and the transverse directions respec- process, despite the pressure in the voids and the stress field
tively. The other three Poisson ratios,ν21, ν31 andν32can in fiber bed being purely elastic.
be calculated as. Note that should the diffusion and dissolution be
E2 modeled (not considered in our model), Henry’s law
ν 21 ¼ ν 12 should still be applied using the resin pressure at the in-
E1
E3 terface with the void, not the averaged resin pressure. As
ν 31 ¼ ν 13 ð3Þ
E1 the squeeze flow related pressure gradient builds fast [26],
E3 this issue should be considered even on length scales as
ν 32 ¼ ν 23
E2 small as 10 fiber diameters (0.1 mm). Generally, the void
Int J Mater Form

spacing is restricted by the ply thickness, so that it likely where, η(T)is the resin viscosity as a function of the tem-
remains in the sub-millimeter range. The two mechanisms perature (and time), Dis the distance between the voids, εbulk •
for the pressure in case of stationary situation and the is the bulk strain rate (Bsqueeze^ rate) that can be in practical
moving resin are shown in Fig. 1. implementation defined as strain increment over time incre-
Through the schematic shown in Fig. 1, the average resin ment Δt and K(εbulk)is the permeability as a function of the
pressure consists of two terms as. bulk strain which can be measured and fitted, or computed
from a constitutive equation as.
¼P
P  hydrostatic þ P
 squeeze ð4Þ
 3
v 0
In order to define the hydrostatic pressure term, we can start 1− 1þεf bulk
K ðεbulk Þ ¼ K 0 ð7Þ
with the ideal gas law for the volatiles pressure, so the resin v f 02
pressure could be defined as. ð1 þ εbulk Þ2
 
 hydrostat ¼ P0 ϕ0
P
T
ð1−d v Þ ð5Þ Equation (7) is Kozeny-Carman empirical equation in which
ϕ0 þ εbulk T ref the fiber volume fraction is evaluated from bulk strain and the
Kozeny-Carman constant is converted to permeabilityK0which
where, P0 is volatile initial pressure, ϕ0 is the initial void
is the value of the permeability at the initial fiber volume
fraction, T and Trefare the current temperature and the ref-
fractionvf0. Substituting Eq. (7) into Eq. (6) and using the def-
erence temperature, εbulk is the bulk strain (usually nega-
inition for hydrostatic and squeeze flow terms for the average
tive in consolidation) and dv is the fraction of original vol-
resin pressure, Eq. (4) could be re-written as.
atiles (in mass or molar terms) that is removed from voids
(usually dissolved) during the consolidation process. In  
¼ P 0 ϕ0 T v f 0 2 D2 1 þ εbulk •
non-trivial cases, both the temperature development and P ð1−d v Þ þ ηð T Þ  3 εbulk ð8Þ
ϕ0 þ εbulk T ref 12K 0 1 þ εbulk −v f 0
volatile dissolution and diffusion need to be tracked inde-
pendently. This will also bring a transient character to an As both the void spacing D and permeability K0 depend on
otherwise purely elastic equation, even if squeeze flow is orientation and precise location, these factors (respectively the
not considered. The squeeze flow could be added as a sec- non-dimensional number D2/K0) are the Bequivalent^ void
ond term using the pressure development term, spacing and permeability. In actual model, the coefficient
must be taken as material parameter to determine to achieve
 squeeze ¼ 1 ηðT Þ D2 εbulk
P
•
ð6Þ best match between model and reality. More accurate model-
12 K ðεbulk Þ
ing is possible, but it would require very detailed information
on void distribution and size, not just porosity and spacing or
typical size on length scale below the actual consolidation
model size. Also note that the squeeze flow model is purely
viscous and given sufficient time this component will become
insignificant. Thus the characteristic time is important, be-
cause if the characteristic value is small, the final state will
not be influenced much by this term. On the other hand,
should the characteristic time be too long, the resin will gel
increasing the viscosity exponentially before the steady state
is reached. This may be important in practical situation if one
wishes to use the simpler steady state prediction to estimate
the consolidation levels.

Constitutive model

The linear anisotropic constitutive model, Eq. (1) is combined

with the average resin pressure in Eq. (8) and used to define
the stress-strain relation. In the proposed model, stress is re-
lated to strain and also the strain rate through the addition of
the pressure term to the constitutive equation as.
Fig. 1 A schematic showing pressure in the resin for (a) the hydrostatic

σij ¼ C ijmn :εmn −δij P ð9Þ
and (b) hydrostatic and the squeeze flow model
 Int J Mater Form

where Cijmnis a forth order tensor defined by Eq. (1) for an strain and pressure data with Ideal gas law and stored as a
anisotropic material and δij is the tensor index notation of state variable obtained as follow
Kronecker delta. Substituting Eq. (8) into Eq. (9), stress term
φ ¼ ðφ0 þ ɛbulk Þ ð13Þ
can be written as follow.
  note that, ɛbulk is the negative term which demonstrates the
P0 ϕ0 T
σij ¼ C ijmn :εmn −δij ð1−d v Þ ð10Þ change in thickness due to the laminate compaction during
ϕ0 þ εkk T ref
the process.
v f 0 2 D2 1 þ εkk •
−δij ηðT Þ  3 εkk
12K 0 1 þ εkk −v f 0

Note that while D2/K0 is Beffective^ value representing Pressure development case study for a flat
material and process parameter, the characteristic value obtain plate under the normal applied pressure
by averaged spacings/principal permeabilities should provide
a decent estimate for this value. Before implementing the formulated pressure model for
In the finite element model, the UMAT (User-defined consolidation of prepregs on a L-shaped tool, the one di-
Material model in ABAQUS) is developed to calculate mensional through the thickness model is verified for con-
the anisotropic modulus and the Jacobian matrix (incre- solidation of prepregs on a flat plate tool subjected to nor-
mental stiffness tensor) corresponding to the stress state mal applied pressure. Based on the geometry provided in
at each iteration. The UMAT allows the user to imple- Fig. 3, the applied pressure would be carried through the
ment a specific constitutive model other than the default thickness by the fiber bed reinforcement and the hydrostat-
model used by ABAQUS. The ABAQUS standard solv- ic and squeeze flow pressure terms developed in the resin
er uses the iterative method with the corresponding and the voids during the consolidation process (see
UMAT to solve the constitutive equations. The iterations Section 2). As a result, through thickness, one dimensional
continue on the basis of the previous solutions, until a pressure balance equation could be defined as.
reasonable convergence is reached. The Jacobian matrix Papp ¼ E:ε þ Phydrostatic þ Psqueeze ð14Þ
(C) is defined as.
∂Δσ
C¼ ð11Þ Using Eqns. (5–7), Eq. (14) could be re-written in a form of
∂Δε a non-linear, ordinary differential equation while applied pres-
where C is the Jacobian matrix (incremental stiffness ma- sure and the strain term are defined as a function of time. For
trix); ∂Δσ is the increment in stress; and∂Δε is the increment simplicity, as we are using this case only to verify the working
in strain. Regarding the new definition for stress term present- of UMAT formulation and Abaqus, we assumed constant tem-
ed by Eq. (10), stress increment as a function of strain incre- perature condition and no solubility for the voids during the
ment is defined as. process. Also, modulus of elasticity was defined only through
  the thickness (E3) while in-plane elastic modulus (E1 and E2)
P 0 ϕ0 T
dσij ¼ C ijmn dεmn þ ð1−d v Þ δij dεkk and Poisson ratios were assumed to be inconsequential
ðϕ0 þ εkk Þ2 T ref
! P 0 φ0 ηD2 υ f 2 1 þ ɛðt Þ :
v f 2 D2
þ 0 ηðT Þ 
v f 0 þ 2ð1 þ εkk Þ •
4 εkk δij dεkk ð12Þ
Papp ðtÞ ¼ E 3 :ɛðtÞ þ þ  3 ε ð15Þ
φ0 þ ɛðtÞ 12K 0 1 þ ɛðt Þ−υ f
12K 0 1 þ εkk −v f 0
v f 0 2 D2 1 þ εkk Equation (15) can be solved using several numerical tech-
− ηð T Þ  3 δij dεkk
12K 0 1 þ εkk −v f 0 :Δt niques using discretization of the strain rate over time. We
used first order Euler and fourth order Runge Kutta
The flowchart of the UMAT developed based on consti- discretization methods for this study as an example. The so-
tutive model is shown in Fig. 2. Within the UMAT, initial lutions were compared with the ABAQUS-UMAT simulation
volatile pressure and void distribution, viscosity and mate- for the flat plate geometry in which identical initial and bound-
rial parameters are provided as input and the Jacobian ma- ary conditions were applied as E3 = 0.1 GPa, P0 = 100 KPa,
trix (incremental stiffness matrix) is calculated, then the Papp = 1 MPa, φ0= 0.05, D = 0.5 mm, K0 = 1E-15 and υf=0.6.
total stress state is updated using the stress derivative de- The comparison between the two methods for flat, 1D simu-
fined by Eq. (12) and the pressure field which is stored as lation and the results attained by the UMAT implemented in
the state variable in the UMAT. After that, the stress state is ABAQUS solver are shown in Fig. 4. The good agreement
returned to the ABAQUS solver for pressure and void evo- verifies the accuracy of the implementation of the proposed
lution. Finally, voids content is updated based on updated model in ABAQUS. The negative results for strain obtained
Int J Mater Form

Fig. 2 Flowchart of the UMAT

integrated with ABAQUS solver

from the numerical approach at the very beginning of the is shown in Fig. 5. The geometry is essentially extruded 2D,
process (t < 5000 s) is because the applied pressure was still but the applied solution is fully three-dimensional.
lower than the assigned initial hydrostatic pressure in the vol- Based on the flowchart shown in Fig. 2, part geometry,
atiles which led to the expansion rather than the compaction. material properties and autoclave process are defined in
ABAQUS CAE. Two examples containing an assembly of
15 and 28 unidirectional plies are presented. The composite
prepreg is modeled as a linear transversely isotropic
Pressure development model parameters orthotropic material. Anisotropy ratio with elastic modulus
for composite prepregs on curved tool surface for the fiber direction compared to the transverse direction
(EE12 =1000 and 10,000) is used in the simulation so that one
The prepreg layers consisting of viscous resin and fiber rein-
forcement, are processed at elevated temperature and pressure
against the tool surface in a typical autoclave process. During
this step, the prepreg layers tend to behave like solid structures
and the porosity within the composite will depend on the
pressure developed within the resin during the consolidation
process as the resin solidifies. The schematic of the L-shaped
example modeled with finite element simulation in ABAQUS

Fig. 4 Predicted Strains from the Consolidation Model with Squeeze

flow implemented through the thickness and its comparison with the
same model implemented in ABAQUS for a flat plate with 5 mm
Fig. 3 Pressure balance schematic through the thickness thickness
 Int J Mater Form

Fig. 5 Schematic showing L

shaped geometry and applied
pressure for an L-shaped part

Tool Fiber direction

can address the stacking of unidirectional prepregs. The tem- 30,000 s as the time period used for model simulation which is in
perature is assumed to be constant throughout the process and line with 8 to 9 h of autoclave cycle time.
volatile dissolution/diffusion is neglected. Hexahedral, linear
standard elements were selected to model the L-shaped geom-
etry. The physical properties and dimensions for the laminates Results and discussion
are shown in Table 2.
We will track the variation of pressure with respect to the Resin pressure dynamics
coordinate system through the thickness from the top surface to
the bottom surface for the L-shaped laminate. Thus the coordi- During the prepreg consolidation process, two mechanisms
nate value in the thickness direction is 0 at the top of the top layer are dominant in resisting the applied autoclave pressure during
and 2.67 mm for the 15 plies and 5 mm for 28 plies laminate at the compaction process: (1) the linear elastic, anisotropic fiber
the bottom of the bottom layer in contact with the tool surface. reinforcement and (ii) the resin pressure as a result of the
For the autoclave process, the applied pressure increases lin- squeeze flow during the process which translates the applied
early with time and reaches 1 MPa at the end of the simulation. pressure through the plies. However, the reinforcement is very
The boundary conditions are zero displacement in all direction at compliant in the transverse direction and stiff in the fiber di-
the interface of the bottom ply and the tool surface (Thickness = rection, changing the behavior depending on the surface cur-
2.67 mm for 15 plies and 5 mm for 28 plies laminate). The initial vature. Thus, in an L-shaped composite prepreg laminate, two
volatile pressure is assumed to be the same as the atmospheric regions must be considered as shown in Fig. 6.
pressure of 100 KPa. Also, initial volatile fraction is varied from
5% to 20% for the parametric study, anisotropy ratio is assumed
to be 1000 and 10,000 for fiber direction vs in-plane transverse
direction and the curvature effect, ri/h is varied from 3 to 5,
where ri is the radius of the top ply curvature and h is the thick-
ness of the laminate. All the results are obtained at time equals to

Table 2 Laminates dimensions and physical properties

L-shaped Number Thickness Horizontal Vertical Resin

laminates of plies (mm) length length viscosity
(mm) (mm) (Poise)

Case 1 15 2.67 175 225 1000

Case 2 28 5 175 225 1000 Fig. 6 Schematic of how pressure translates between plies in flat and
curved regions
Int J Mater Form

In flat parts (zone 1) the transverse stiffness and the resin dry friction elastic force corresponding to the frictional resis-
pressure has to carry all the transverse load as any reinforce- tance will remain. We assumed no sliding in this example. The
ment contribution is orthogonal to the normal direction. Once viscous forces make the situation transient and development
the surface is curved (zone 2), the in-plane tension in the must be described as a time dependent function.
reinforcement (membrane behavior) carries the applied load The resin pressure contours and normalized resin pressure
as well and the other components decrease in accordance with change with respect to the applied autoclave pressure at the
the Terzaghi equation. If the ply-to-ply sliding is prevented, curvature through the thickness are shown in Figs. 7a and 8a at
the in-plane tension is Bpermanent^ and depends on the shear- time equals to 30,000 s for 5% initial void content in the L-
ing behavior (modulus) of the laminate. If the plies can slide, shaped composite laminate with 28 and 15 plies respectively.
its character may change to transient dependent on sliding The final applied pressure is set to 1 MPa at the end of the
mechanism (viscous, friction, combination…), though for simulation and the distance between voids, D is set to 500 μm.

Fig. 7 28 plies laminate with 5%

initial void content (a) normalized
resin pressure distribution with
respect to the applied pressure
through the laminate thickness at
corner (b) resin pressure (absolute
value) contour for the entire
laminate, Papp = 1 MPa, D =
500 μm at the end of the
simulation, t = 30,000 s
 Int J Mater Form

Fig. 8 15 plies laminate with 5%

initial void content (a) normalized
resin pressure distribution with
respect to the applied pressure
through the laminate thickness at
corner (b) resin pressure contour
for the entire laminate,, Papp =
1 MPa, D = 500 μm at the end of
the simulation, t = 30,000 s

The contour plots in Figs. 7b and 8b clearly show the pres- resin pressure drop in the curvature region. It can be observed
sure is constant in the thickness direction in the flat section. from the graphs and contours, the resin pressure was higher
This pressure is, however, still significantly lower than the for the 15 plies laminate compared to the 28 plies laminate at
applied pressure because the fibrous bed carries part of the flat and curved regions with respect to the same applied auto-
load. For the curved regions, the pressure further decreases clave pressure of 1 MPa.
through the thickness for both laminate thicknesses as the in-
plane stress carries additional load there. The reduction in Parametric study
volatile pressure is the main reason for void evolution in the
curved region. Two different case studies with different num- In order to illustrate the fiber bed resistance and the hydrostatic
ber of plies were used to investigate the effect of thickness on pressure vs squeeze flow effects during the compaction process,
Int J Mater Form

Fig. 9 Resin pressure parametric

study for different void
percentages and anisotropy ratios
at corner for (a) 28 plies and (b)
15 plies, final Papp = 1 MPa, D =
500 μm and t = 30,000 s

a parametric study has been performed based on three different was 30,000 s. The results for pressure development based on
initial void fractions and two anisotropy ratios for a layup of 15 hydrostatic and squeeze flow through the thickness for cases
and 28 plies. Three volatile fractions (ϕ0) of 5, 10 and 20% are with various void percentage and anisotropy ratio for 28 and
considered in the pressure model (Eqns. 8–10) and for the an- 15 plies are shown in Fig. 9a, b respectively.
isotropy ratio, the extreme cases with large difference between As can be observed from the results, highest resin pressure
the elasticity in fiber direction versus the transverse direction are drop across the thickness occurs at lower void content as the 5%
used (EE12 =1000 and 10,000), as the transverse modulus was kept void fraction shows highest pressure decrease. Overall, howev-
unchanged where the fiber direction modulus was increased er, the pressure is lower for high porosity as the elastic reinforce-
accordingly in order to reach the 1000 and 10,000 times anisot- ment will carry more compaction load than the resin which for
ropy ratio for the simulations. For all parametric studies, final higher porosity is more compliant. The effect of the anisotropy
applied pressure was 1 MPa, D, the distance between the voids, ratio is also prominent with lower initial void content percent-
was maintained at 500 μm and the time period for the simulation ages becoming less sensitive when the initial void content is
 Int J Mater Form

Fig. 10 Updated void content

distribution contours for 28 plies
laminate, initial void = 5%,
Papp = 1 MPa, D = 500 μm and
t = 30,000 s

higher. In a realistic prepreg laminate, initial porosity is closer to Applied pressure and geometry effects
5% than 20% where the effects of through-thickness pressure
drop are much more significant which will promote resin pres- The autoclave applied pressure is usually assumed as the value
sure drop in corner regions closer to the tool surface. This effect that determines the consolidation. In reality, there are only two
is more prominently shown in Fig. 11 where by reducing the extreme cases in which the autoclave applied pressure, after
initial void content from 5% to 3% the void pressure increases the squeeze flow transient behavior subsides, will equal the
up to 75% of the applied load and drops much more quickly. As pressure within the voids.
a result, the pressure model is able to show that for curved First extreme is if the applied pressure is equal to the initial
regions there is a substantial pressure drop along the thickness pressure in the voids. In that case, resin pressure will carry all
and this drop is a strong function of initial void content, anisot- the applied load and the reinforcement will be stress-free and
ropy ratio and the laminate thickness. The void formation and will not deform. This is, obviously, not very useful as no
growth around such corners will show larger voids closer to consolidation will take place Fig. 11.
the tool surface than the top surface as seen in Fig. 10. Second extreme is when the applied pressure is very
The updated void content was calculated using Eq. (13) large, the material stiffness of reinforcement becomes neg-
where the bulk strain was obtained from the constitutive ligible and the voids are compressed as the resin roughly
model in UMAT implemented in ABAQUS.

Fig. 12 Non-dimensional study exhibiting void pressure variation

Fig. 11 Pressure reduction through the thickness at corner for 28 plies through the thickness in the corner as the applied pressure is increased
laminate with 3 and 5% void fraction, for final Papp = 1 MPa, D = 500 μm to compress the voids for 28 plies laminate with 5% initial void content
and t = 30,000 s with no squeeze flow effects, Papp = 1 MPa, D = 0 μm and t = 30,000 s
Int J Mater Form

carries the entire applied pressure. In this case the voids (Papp = 3000 KPa) in Fig. 12. This clearly shows that the vol-
control the deformation magnitude and the corresponding atile pressure close to the tool surface is just a limited fraction
fiber bed deformation. In that case, the elastic stress in (30–40%) of the applied pressure for the expected consolida-
fiber bed is equal to the elastic modulus times the porosity tion load cases.
(~bulk strain which is through the thickness). In between Figure 13 shows this behavior for volatile pressure with
these two extremes, however, the steady state pressure in respect to the initial void pressure and the applied pressure for
volatiles builds up only to a fraction of the applied pressure the case when we consider squeeze flow effects. In this case the
as the elastic stress in fiber bed carries part of the compac- steady state is not reached in 30,000 s which leads to additional
tion load. This depends on two factors: reduction of pressure within voids. The additional load is car-
ried by the pressure buildup as the resin is squeezed out of the
& The original pressure in the voids. fiber bed. The fraction of applied pressure carried in steady
& The elastic transverse stiffness of the reinforcement. state by the resin decreases until the applied pressure is equal
to 10 times the original void pressure, then starts to increase
These behaviors are shown for the case when we assume again. Initially the resin pressure is decreasing because the
no squeeze flow for 28 plies laminate in a range of low applied pressure in the voids essentially represents a stiffening non-
pressure (Papp = 200 KPa) to extremely large applied pressure linear spring, the initial load taken by the voids is much smaller
and most of the load is taken by the fibers but as the load
increases, voids get more compressed and carry more load
relative to the fiber bed (as load carried by fiber bed is limited).
The squeeze flow that must fill the voids builds additional resin
pressure, effectively reducing the void pressure and stress in
fiber bed. If omitted, the relative void pressure in Fig. 13 would
asymptotically go to 1 for large compression loads. These re-
sults were compatible with the assumption of the laminate and
the compaction mechanism without sliding, as the layers are
compacted through the thickness. As the plies are extended in
fiber direction due to the compaction through thickness, they
stretch and build the in-plane stress field which is arrested only
in the flat parts by shear stress related to the Bsliding^ mecha-
nism provided by the laminate shear deformation. Since the ply
modulus in the fiber direction is so much higher than the trans-
verse shear modulus, strain in the fiber direction is actually very
limited (calculated in a range of 0.1%).

Fig. 13 Non-dimensional study exhibiting void pressure variation Fig. 14 Sharp curvature (ri/h = 3) and lower initial void content (3–5%)
through the thickness in the corner as the applied pressure is increased results in higher pressure drops which will promote more void evolution
to compress the voids for 28 plies and 15 plies laminate with 5% initial and movement in the thickness direction at corner, final Papp = 1 MPa,
void content, Papp = 1 MPa, D = 500 μm and t = 30,000 s D = 500 μm and t = 30,000 s
 Int J Mater Form

Fig. 15 a Stress, S33 and b Linear strain results through the thickness for laminate with 5% void content and 28 plies, Papp = 1 MPa, D = 500 μ and t =
30,000 s

The effect of the curvature on the void pressure is shown in shown in Fig. 15. Based on the result presented for the stress,
Fig. 14 for the thicker laminate of 28 plies. Two curvatures’ uniform stress in the normal direction is obtained for the flat
radius over thickness are considered in order to study the surfaces which equals the applied autoclave pressure of
effects of sharp curvature on volatile pressure development. 1 MPa. This stress decreases in the region with curvature since
As expected, for the 3% and 5% initial void contents, higher the developed pressure is no longer equal to the applied
pressure drop is experienced for sharper curvatures while the pressure. This is confirmed by the strain results which is
geometry effect is not important for the laminate containing around 3.82% as seen in all top surface and the flat areas
prepregs with initial higher void content of 10%. while the strain decreases to 2.63% in the curvature area
which may also give rise to non-uniform thickness around
Stress and strain terms through the thickness the curve. The results presented in Fig. 16 for the 15 plies
laminate with the initial void content of 5% show the
The contours for the stress and strain terms through the thick- same pattern for stress and strain contours as what was
ness for the laminate with 28 plies and 5% void fraction are experienced by the 28 plies laminate.

Fig. 16 a Stress, S33 and b Linear strain results through the thickness for laminate with 5% void content and 15 plies, Papp = 1 MPa, D = 500 μm and t =
30,000 s
Int J Mater Form

volatile Bhydrostatic^ pressure is based on ideal gas law and

the resin pressure due to squeeze flow is described by
substituting Beffective^ void spacing and fiber bed permeabil-
ity into one-dimensional model. The initial void content is
assumed to be uniform. The permeability variation is based
on Karman-Kozeny eq. A parametric study is conducted in
which the role of the anisotropic fiber behavior in addition to
the initial void hydrostatic pressure and the induced resin
squeeze flow due to compaction are considered during the
consolidation process in the autoclave. The resin pressure dur-
ing the consolidation process has been predicted through the
thickness for an L-shaped laminate.
The results show that the applied pressure is never reached
in the resin as the elastic resistance of fiber bed carries some
load. More importantly, there is an additional – and very sig-
nificant - reduction in the resin pressure at the curvature region
Fig. 17 How the various pressure terms vary across the thickness for 28 near the corner of the L shape whereas the rest of the flat area
plies laminate with 5% initial void content, Papp = 1 MPa, D = 500 μm
and t = 30,000 s
exhibits a uniform pressure distribution. The reduction in resin
pressure through the thickness will promote the void growth
Effective mechanisms comparison as the resin pressure drops from the top surface to the tool
surface non-uniformly. The pressure results were confirmed
As we discussed in section 2 regarding the effects of all by the stress and strain terms for thin and thick laminates. A
dominant mechanisms on volatile pressure development, non-uniform pressure state in the corner extends to the elastic
presented in Fig. 17 is each term’s effect in qualitative stress and strains and this indicates that the laminate may
format for the 28 plies laminate under 1 MPa applied pres- undergo thickness change in the curved region. Lower initial
sure at 30000 s for the consolidation process. Based on the void content, thicker laminates, higher anisotropy and sharper
results shown, compression stress in the fibrous reinforce- curvatures seems to show the highest pressure drop in the
ment is the most significant term in transferring the applied resin and hence higher probability of void movement and
pressure. For real materials, this value may vary signifi- growth in addition to non-uniform thickness in the curved
cantly depending on the initial fiber volume fraction and region. The parametric study demonstrates the ability of the
reinforcement architecture. The squeeze flow effect is the model in simulating various case studies to model the void
least significant term at this time and should drop to zero if evolution during the consolidation process.
the load remains constant and the time is long enough.
Acknowledgements This work is supported by Composites Automation,
Note that during the pressure development at earlier times LLC and the United States Naval Air and Warfare Center under Prime
this is not necessarily true, in case shorter cycles are ex- Contract No. N68335-17-C-0093 administrated by Dr. Suresh G. Advani.
plored. The summation of all terms at the top ply (thick- The views, opinions and/or findings contained in this report are those of
ness = 0) is practically equal to 1 MPa or the applied auto- the author(s) and should not be construed as an official US Naval Air
Warfare Center position, policy or decision unless so designated by other
clave pressure while the effect of each term shrinks when documentation. The authors also declare that they have no conflict of
we reach the bottom ply close to the tool surface. The interest.
membrane in-plane stress is responsible for this reduction
as it reacts more and more of the applied load. Publisher’s note Springer Nature remains neutral with regard to jurisdic-
tional claims in published maps and institutional affiliations.

Summary and conclusion References

In this study, a new constitutive model has been developed as 1. Chandrakala K, Vanaja A, Rao R (2009) Storage life studies on RT
a UMAT subroutine for the finite element package ABAQUS/ cure glass—epoxy prepregs. J Reinf Plast Compos 28(16):1987–1997
Standard to study the resin pressure dynamics during the con- 2. Herrmann A, Zahlen P, Zuardy I (2005) Sandwich structures tech-
solidation process for prepregs with varying initial void con- nology in commercial aviation. In: Thomsen OT, Bozhevolnaya E,
Lyckegaard A (eds) Sandwich structures 7: advancing with sand-
tent. Most significant simplification is that of the constant wich structures and materials. Springer Netherlands, pp 13–26
temperature. The fiber bed model is defined based on the 3. Kardos J. L., Duduković M. P., Dave R. (2005) Void growth and
linear anisotropic behavior of the composite laminate. The resin transport during processing of thermosetting — Matrix
 Int J Mater Form

composites. Epoxy resins and composites IV. Advances in polymer manufacturing defects in autoclave molding technology. J Reinf
science, vol 80. Springer Plast Compos 28(22):2791–2803
4. Lightfoot SC, Wisnom MR, Potter K (2013) A new mechanism for 15. Li M (2001) Optimal curing of thermoset composites: thermochem-
the formation of ply wrinkles due to sheer between plies. ical and consolidation considerations. PhD thesis, University of
Composites Part A 49:139–147 Illinois at Urbana-Champaign
5. Potter K, Khan B, Wisnom MR, Bell T, Stevens J (2008) 16. Mélanie B (2010) Out-of-autoclave manufacturing of complex shape
Variability, fiber waviness and misalignment in the determination composite laminates. McGill University, Montreal, Manufacturing
of the properties of composite materials and structures. Composites 17. Xin CB, Gu YZ, Li M, Luo J, Li YX, Zhang ZG (2011)
Part A 39(9):1343–1354 Experimental and numerical study on the effect of rubber mold
6. Bloom LD, Wang J, Potter KD (2013) Damage progression and configuration on the compaction of composite angle laminates dur-
defect sensitivity: an experimental study of representative wrinkles ing autoclave processing. Composites Part A 42(10):1353–1360
in tension. Composites Part B 45(1):449–458 18. Johnston A (1997) An integrated model of the development of
7. Hsiao HM, Daniel IM (1996) Effect of fiber waviness on stiffness process-induced deformation in autoclave processing of composite
and strength reduction of unidirectional composites under compres- structures. Ph.D. thesis, The University of British Columbia
sive loading. Compos Sci Technol 56(5):581–593 19. Hubert P, Vaziri R, Poursartip A (1999) A two-dimensional flow
8. Garnich MR, Karami G (2004) Finite element for stiffness and model for the process simulation of complex shape composite lam-
strength of wavy fiber composites. J Compos Mater 38(4):273–292 inates. Int J Numer Meth Eng 44(1):1–26 86
9. Karami G, Garnich MR (2005) Effective moduli and failure con- 20. Min L, Yanxia L, Yizhuo G (2008) Numerical simulation flow and
sideration for composites with periodic fiber waviness. Compos compaction during the consolidation of laminated composites.
Struct 67(4):461–475 Wiley Intersci Soc Plast Eng 29(5):560–568 87
10. Hubert P, Poursartip A (1998) A review of flow and compaction 21. Dave R, Kardos JL, Dudukovic MP (1987) A model for resin flow
modelling relevant to thermoset matrix laminate processing. J Reinf during composite processing: part 1-general mathematical develop-
Plast Compos 17(4):286–318 ment. Polym Compos 8(1):29–38 88
11. Fernlund G, Griffith J, Courdji R, Poursartip A (2002) 22. Gutowski TG, Cai Z, Bauer S, Boucher D (1987) Consolidation
Experimental and numerical study of the effect of caul-sheets on experiments for laminate composites. J Compos Mater 21(7):
corner thinning of composite laminates. Composites Part A 33(3): 650–669 89
411–426 23. Li M, Li Y, Zhang Z, Gu Y (2008) Numerical simulation flow and
12. Naji MI, Hoa SV (2000) Curing of thick angle-bend thermoset compaction during the consolidation of laminated composites.
composite part: curing process modification for uniform thickness Polym Compos 29(5):560–568 90
and uniform fiber volume fraction distribution. J Compos Mater 24. Li M, Charles L, Tucker III (2002) Modelling and simulation of
34(20):1710–1755 two-dimensional consolidation for thermoset matrix composites.
13. Li Y, Li M, Gu Y, Zhang Z (2009) Numerical and experimental Compos A 33(6):877–892 91
study on the effect of lay-up type and structural elements on thick- 25. Dong C (2011) Model development for the formation of resin-rich
ness uniformity of L-shaped laminates. Appl Compos Mater 16(2): zones in composites processing. Compos A 42(4):419–424
101–115 26. Simacek P, Advani SG, Gruber M, Jensen B (2013) A non-local
14. Wang X, Zhang Z, Xie F, Li M, Dai D, Wang F (2009) Correlated void filling model to describe its dynamics during processing ther-
rules between complex structure of composite components and moplastic composites. Composites Part A 46(1):154–165