1

Download as pdf or txt
Download as pdf or txt
You are on page 1of 151

MICROSTRUCTURAL FINITE ELEMENT MODELING OF

HIGH-PERFORMANCE CARBON FIBER-REINFORCED COMPOSITES


TO CHARACTERIZE MICROSCALE MATERIAL PROPERTIES
AND FIBER-DIRECTION COMPRESSIVE STRENGTH

by

QUY TUNG LINH VU

DISSERTATION

Submitted in partial fulfillment of the requirements


for the degree of Doctor of Philosophy at
The University of Texas at Arlington
International cotutelle with ISAE-Supaero, MEGEP, Toulouse, France

December, 2023

Arlington, Texas

Dissertation Committee:

Andrew Makeev, Supervising Professor


Erian A Armanios
Endel Iarve
Paul Davidson
Frédéric Lachaud, Supervising Professor
Yves Gourinat, Supervising Professor
Christian Hochard, External Referee
Marco Gigliotti, External Referee
Copyright by
Quy Tung Linh Vu
2023
ABSTRACT

MICROSTRUCTURAL FINITE ELEMENT MODELING OF


HIGH-PERFORMANCE CARBON FIBER-REINFORCED COMPOSITES
TO CHARACTERIZE MICROSCALE MATERIAL PROPERTIES
AND FIBER-DIRECTION COMPRESSIVE STRENGTH

Quy Tung Linh Vu, Ph.D.

The University of Texas at Arlington, 2023

International cotutelle with ISAE-Supaero, MEGEP, France

Supervisors:

Andrew Makeev

Guillaume Seon

Frédéric Lachaud

Miguel Charlotte

Yves Gourinat

Carbon fiber-reinforced composites are increasingly used in aerospace


application thanks to their excellent specific stiffness and strength. However, incomplete
understanding of their failure behaviors leads to composite components often being
designed with a high Safety Factor, limiting their advantages. Using computational
methods, this project studies carbon fiber-reinforced composite microstructures for the
objective of understanding and improving material performances under compression
load which is one of the main considerations in primary aerospace structure design. A
Finite Element (FE) model for the evaluation of composite compression behaviors was

i
developed, allowing the identification of several key material properties affecting
composite performance.

The numerical micromodels require information of the microscale components


including the reinforcing fibers, the matrix and the fiber-matrix interface. An extensive
literature review was done on the microscale component properties including their
potential impact on composite strength. Combined with results from the FE modeling, a
strategy was decided where the assessment of important material properties was given
priority.

Several authors, numerically and experimentally, demonstrated the significant


impact of microscale residual stress on composite strengths. Yet, there was no
established method for the characterization of residual stress in matrix at microscale.
This project proposes a new method for the assessment of microscale residual stress in
composite matrix. The new method is based on the fiber push-out experiment that
creates local matrix deformation induced by the relaxed stress and evaluation by the
Finite Element Method Updating technique for the inverse characterization of residual
stress field in the corresponding specimen.

Literature review and FE modeling results identified the Interfacial Shear


Strength (IFSS) as a key material property affecting composite strength. Meanwhile, the
IFSS measurement using the standard push-out method suffers from several unwanted
effects, potentially lowering its accuracy. These effects were analyzed using FE
modeling. Notably, the results suggest that the effect of microscale residual stress on
the IFSS measurement is rather insignificant.

ii
ACKNOWLEDGEMENTS

I would like to express my gratitude to my supervisors: Dr. Guillaume Seon,


Professor Frederic Lachaud, Professor Miguel Charlotte, Professor Yves Gourinat and
Professor Andrew Makeev for their guidance, advice, and support during my PhD
project.

I would like to thank the members of my Supervisory Committee and Dissertation


Committee: Professor Erian Armanios, Professor Paul Davidson, Professor Endel Iarve,
Professor Warda Ashraf, Professor Christian Hochard, and Professor Marco Gigliotti for
their insightful comments and reviews.

I would like to thank my colleague, Dr. Sarvenaz Ghaffari, for her help and for the
cooperation with my research. Dr. Ghaffari provided significant experimental data which
helped with verification and validation of my hypotheses and numerical results.

I would like to thank my colleague, Mr. Brian Shonkwiler, for his insights into the
experimental aspects of composite materials behaviors and for providing experimental
data which helped with development and verification of my numerical models.

I would like to thank my colleague, Dr. Yuri Nikishkov for his helpful advice on the
application of various mathematical and computational techniques in my research topic.

I would like to thank my friends, Mr. Hong Anh Vu and Dr. Thanh Huy Nguyen for
the excellent discussions and advice regarding mathematical aspects and
computational techniques which were subsequently applied to my research project.

The PhD project received full funding through Research Assistantships from the
ISAE-Supaero, France and the University of Texas at Arlington, USA.

This research was partially funded by the U.S. Government under Agreement
No. W911W6-17-2-0002. Such support is gratefully acknowledged. The U.S.
Government technical monitor is Dr. Mahendra Bhagwat. The views and conclusions
contained in this document are those of the author and should not be interpreted as
representing the official policies, either expressed or implied, of the U.S. Government.

iii
DEDICATION

In addition to the academic assistance directly related to this scientific work, I


received valuable support and inspiration in life in the upcoming of and during my time
as PhD student.

I would like to thank my family who have always been beside me during my
academic adventure. My grandparents, grand aunt, and grand uncle, for teaching me
the value of education, the love for sciences and the aspiration for academic
achievements. My parents, for not only giving me a strong body and a smart brain, but
also teaching me the value of hard work, independence and integrity which all scaled so
well into my life as an engineer and a scientist. My aunts and my uncles, for the
encouragement, the helpful advice and teaching, and for lending me their wisdom, all
free of charge. My little sister and my cousins, for the fun we shared and the
discussions on different aspects of life with various levels of seriousness. Nhon The
British Longhair, for the unlimited amount of entertainment over my long honorable
years in the quest for the graduate degrees.

I would like to thank my dearest cutest girlfriend Lan Lam Le, for the love and
care, for the memories we made together, for patiently watching me slowly graduating
from a high school cowboy into a slightly wiser man today. Trying to express my
gratitude to you, words don’t come easy.

All through my education, I have been blessed with excellent teachers. I would
like to thank my professors and advisors Guillaume Seon, Frederic Lachaud, Miguel
Charlotte, Yves Gourinat and Andrew Makeev for their teaching in sciences, in
aerospace engineering, in airmanship, and for the wisdom they lent to me. I would like
to thank my physics teacher Cong Toan Nguyen and my math teacher Thi Nhuy Dam,
for not only teaching sciences to the younger me, but also offering me continuous
advice and encouragements throughout my academic voyage.

My special thanks go to my teachers, mentors, senior fellow graduates Guillaume


and Yves for showing me the way. Your wise words at the direst of times were and still
are a treasure.

iv
My deepest gratitude goes to my godfather Jean-Jacques Augier for the
guidance, encouragement and inspiration to pursue my dreams in life.

My deepest gratitude goes to my life mentor Thoa Nguyen for the words of
wisdom and the inspiration to follow through hardship and achieve what is dearest in
life.

I would like to express my special appreciation to my brothers-in-arms, Hong Anh


Vu and Thanh Huy Nguyen. Our brotherhood that was forged through thousands of
hours of combat for the Defense Of The Ancient certainly is immortal.

I would like to convey my special gratefulness to my bros-at-lab, Louis Sponton,


Maksim Larionov and Abdullah Metiner. Our brotherhood that was hardened through
science, engineering, culture and inflation shall be eternal. I would like to express my
special gratitude to my friends and colleagues Brian Shonkwiler, Yuri Nikishkov, Sarvi
Ghaffari and Bastiaan van der Vossen. The time we shared contributed to the shaping
of my PhD student life into a The-Big-Bang-Theory-like experience.

I would like to mention my wonderful friends, Pham Truong Son, Thi Thu Hien
Nguyen, Thi Thanh Huyen Nguyen and Phuong Anh Nguyen, for the timely precious
support at the hardest moment of my life. Words fail to express my gratitude towards
everything that they have done for me.

I would like to express my deepest gratitude to my new friends (no longer new at
the moment of writing of this document), Duc Tam Vu, Quang Truong Pham, Kieu Trinh
Pham and Phuong Mai Trinh, for having appeared at a critical time of my life, offering
valuable help and support such that we might soon have a Dr. Vu Quy reborn.

v
List of Abbreviations

For the sake of brevity, in this document, the term “composite” refers to the Fiber-
Reinforced Composite (FRC) which is the main object of study of this project.

For a quick reference, below is a list of common abbreviations utilized in this


document.

BC Boundary Condition

CFRP Carbon Fiber-Reinforced Polymer

CDP Concrete Damaged Plasticity

CTE Coefficient of Thermal Expansion

DIC Digital Image Correlation

DOC Degree of Cure

FE Finite Element

FEM Finite Element Method

FEMU Finite Element Method Updating

FRC Fiber-Reinforced Composite

HM High-Modulus

IFSS Interfacial Shear Strength

IM Intermediate-Modulus

PBC Periodic Boundary Conditions

RVE Representative Volume Element

SEM Scanning Electron Microscope

UD Uni-Directional

vi
Table of Contents

ABSTRACT ..................................................................................................................... i
ACKNOWLEDGEMENTS ............................................................................................... iii
DEDICATION ..................................................................................................................iv
List of Abbreviations.....................................................................................................vi
CHAPTER 1: Introduction ............................................................................................. 1
1.1. Background ........................................................................................................ 1
1.2. Objectives .......................................................................................................... 2
CHAPTER 2: Study Uni-directional (UD) Carbon Fiber-reinforced Plastic (CFRP)
compression strength in fiber direction using Finite Element Method .................... 4
2.1. Introduction ........................................................................................................ 4
2.2. Literature reviews ............................................................................................... 5
Analytical methods ................................................................................................... 5
Numerical methods .................................................................................................. 6
Hybrid composites.................................................................................................... 7
Kink-band formation as a mode of longitudinal compressive failure......................... 8
2.3. Analysis of composite fiber-direction compressive strength using FEM ........... 11
2.4. Effects of some microscale properties on UD composite longitudinal
compression strength ................................................................................................ 14
2.5. Parametric analysis: comparing axial compression strengths of HM and IM
composites ................................................................................................................ 16
2.6. Proposals for improvements and future works ................................................. 19
Validation of Micromodel: 4-point-bending notched specimen ............................... 19
Proposals for future works ..................................................................................... 20
2.7. Conclusions...................................................................................................... 23
CHAPTER 3: CFRP component materials properties at microscale: review and
suggestions for future experimental works .............................................................. 25
3.1. Introduction ...................................................................................................... 25
Color Code for physical parameters review table ................................................... 25
3.2. Fiber microscale properties .............................................................................. 26
Summary ................................................................................................................ 26
Experimental measurement methods..................................................................... 27
3.3. Matrix microscale properties ............................................................................ 31
Summary ................................................................................................................ 31
Experimental measurement methods..................................................................... 32

vii
3.4. Fiber-matrix interface properties ...................................................................... 34
Summary ................................................................................................................ 34
Experimental measurement methods..................................................................... 35
3.5. PhD works on the microscale material characterization ................................... 37
FEMU method to determine carbon fiber orthotropic elastic properties ................. 38
Fiber longitudinal compression strength: Bending Beam ....................................... 40
Berkovich indentation technique for epoxy microscale properties .......................... 42
3.6. Conclusions...................................................................................................... 45
CHAPTER 4: Study microscale residual stress in CFRP using Fiber Push-out
experiment and FEMU evaluation .............................................................................. 47
4.1. Introduction ...................................................................................................... 47
4.2. Methodology..................................................................................................... 50
4.3. Finite Element Model ....................................................................................... 52
Setup for the Fiber Push-Out experiment ............................................................... 52
Specimen geometry and mesh generation ............................................................. 53
FEM Analysis Steps and matrix Sink-in measurement........................................... 56
Material models and properties .............................................................................. 58
4.4. Results and discussion .................................................................................... 64
FEMU results and application to predict matrix sink-in [109] .................................. 64
Assessment of residual stress in composite [109].................................................. 66
4.5. Ongoing works and future improvements ......................................................... 70
Improving FE model for residual stress analysis using fiber push-out experiment . 70
Effects of post-failure friction stress on matrix sink-in deformation......................... 74
Effects of matrix plasticity and damage on post-push-out matrix deformation ....... 79
Proposed improvements and future works ............................................................. 80
4.6. Conclusions...................................................................................................... 84
CHAPTER 5: FEM evaluation of the IFSS measurement using Fiber Push-out
method and effects from microscale properties including residual stress ........... 86
5.1. Introduction ...................................................................................................... 86
5.2. Methodology..................................................................................................... 87
5.3. Finite Element Model ....................................................................................... 87
5.4. Results and discussion .................................................................................... 89
Strong interaction and coupling between different material properties and effects
during the fiber push-out experiment...................................................................... 90

viii
Sensitivity studies of effects from different parameters on the IFSS measurement
using fiber push-out method ................................................................................... 94
5.5. Conclusions.................................................................................................... 109
CHAPTER 6: Conclusions ........................................................................................ 110
References ................................................................................................................. 113
Appendices ................................................................................................................ 122
Appendix A: RVE implementation in Abaqus FEM simulation ................................. 122
Appendix B: Bending Beam method for fiber longitudinal compression strength
measurement, design and evaluations .................................................................... 125
Appendix C: Review: longitudinal properties of some carbon fibers ........................ 133
Appendix D: Mesh and size convergence studies for Residual Stress Analysis using
Fiber Push-Out experiment [109] ............................................................................. 136
Appendix E: Size convergence analysis of Residual Stress to validate the application
of far-field Boundary Conditions on micromodels [109] ........................................... 139

ix
CHAPTER 1:
Introduction

1.1. Background

The composite materials are finding increasing usage in aerospace engineering,


particularly in the vertical lift aircraft structures, thanks to their superior stiffness and
strength for a given weight. However, the optimization of composite structures still
suffers from the incomplete understanding of their damage mechanism and their often
sudden and brutal failure. As a result, composite structures are often designed with a
high safety factor, limiting their advantages in weight-saving applications.

Amongst the design constraints of the aerospace primary structures, component


compressive strength including post-damage residual compressive strength has been
one of the main considerations [1]–[7], especially in the development of aircraft with
high use of lightweight composites including the traditional fixed-wing aircraft and the
newer tiltrotor vertical lift aircraft [5], [7], [2]. Fiber-direction compression strength of
common aerospace fiber-reinforced composites is significantly lower than tension
strength, while the understanding of the mechanism of compression failure is
incomplete. The high demand for high-performance composites in aerospace
engineering, coupled with their weakness in axial compression strength and the
incomplete understanding of this important failure mode, prompted very active research
seeking better knowledge on axial compressive failure mechanism and improving axial
compressive strength in new composite materials [8]–[18].

Recent development of High-Modulus (HM) Carbon Fibers permits stiffer carbon-


fiber reinforced plastic (CFRP) composites compared to the legacy Intermediate-
Modulus (IM) Carbon Fibers, promising applications in manufacturing even lighter-
weight aerospace components. However, these HM-based composites suffer from
lower compression strength in fiber-direction than traditional IM-based materials.
Coupling with their well-known sudden and brutal rupture characteristics, this
complicates the design of aerospace components using these novel HM-based CFRPs.

1
To overcome the weaknesses of HM composites, researchers at AMSL-UTA
developed a hybrid composite with a mixture of HM and IM fibers. The result is
promising: the hybrid material possesses higher Modulus (stiffer), all the while having
only a small reduction in fiber-direction compression strength [17]. It is important to note
that hybridizing fibers in a composite is not a straightforward task: in some cases, hybrid
composite is weaker than its parent, such as the glass-carbon composite [9].

This PhD project is part of an effort at the AMSL laboratory at Department of


MAE, UTA with cooperation from the international partner DMSM-ICA laboratory, ISAE-
Supaero (Toulouse, France) for the development of diagnostics and prognosis methods
relying upon 3D imaging and performance prediction based on computational tools. The
new methods might offer significant reduction of costly and lengthy experimental
iterations in the development of new composites through the (1) – high-fidelity 3D non-
destructive measurements; (2) – experimental methods to determine key material
properties; and (3) – computational techniques relying on key structure and material
properties to predict lifetime performance.

This PhD project focuses on the study of composite at microscale using


numerical methods to evaluate effects of microscale component properties (fiber,
matrix, interface etc.) on the macroscale composite properties. As key microscale
properties affecting composite performance are gradually identified, the project seeks to
propose, evaluate, and calibrate experimental methods for the measurement of these
microscale properties using computational tools.

1.2. Objectives

Chapter 2 deals with the fiber-direction compression failure of Uni-directional


(UD) composites. An extensive literature review is provided, including existing
analytical, numerical and experimental studies, offering some clues on the potentially
important material properties and effects to be considered in the FE modeling of the
phenomenon. The FE models and numerical results are then presented along with
some conclusions and suggestions for future works on the topic.

2
Chapter 3 deals with material properties and material models at microscale. An
extensive literature review on both the relevant microscale material properties and their
measurement methods is provided. Some material properties are identified as critical to
the composite performance. The PhD works on some of the identified-as-important
material properties are presented.

Chapter 4 deals with the microscale residual stress in the composite, where a
new method for the characterization microscale residual stress is proposed. Literature
reviews suggest the very significant effects of microscale residuals stress on the
composite performance. Combined with the lack of experimental measurement method
at the time, this study was given a high priority in the PhD project. The new method for
microscale residuals stress characterization consists of the fiber push-out experiment
where matrix deformation due to stress relaxation are evaluated using the Finite
Element Method Updating (FEMU) for the reproduction of the residual stress field in the
FE models. Results, the ongoing work along with proposals for future improvements are
presented.

Chapter 5 deals with the calibration of the Interfacial Shear Strength (IFSS)
measurement using the fiber push-out method. Results from Chapter 2, supported by
other studies from the literature, indicate a strong effect of the IFSS on the composite
fiber-direction compression strength. At the same time, the IFSS measurement using
the common fiber push-out method suffers from unwanted effects affecting its accuracy.
In this Chapter 5, interfering effects from different material and specimen properties are
evaluated, providing ground for a better interpretation of experimental results as well as
suggestions for a better design of the fiber push-out experiment.

Chapter 6 summarizes the conclusions and perspectives of the project.

3
CHAPTER 2:
Study Uni-directional (UD) Carbon Fiber-reinforced Plastic (CFRP)
compression strength in fiber direction using Finite Element Method

2.1. Introduction

This Chapter presents the PhD work on the Research Topic: Study Fiber-
direction Compression Failure of Uni-directional (UD) Composites using Analytical and
Numerical methods.

Composite materials are widely considered as being used under potential due to
the complexity and variety of their damage and failure, hence the notorious difficulty of
predicting their failure properties. As a result, components using composites (including
CFRPs) are usually designed with a large Safety Factor, partially decrease their
advantage in weight-saving. Particular attention is paid to the compressive damage
𝑜
properties of composite, where Fiber-Direction Compressive Strength (𝜎𝑐11 ) is usually 2
times lower than in tensile strength [19]. There are many compression failure modes of
composite which depend on properties of component fiber, matrix, and the cohesion of
fiber-matrix interface [10]. Composite may fail via. longitudinal splitting, shear crippling
or simple compression. Fiber may fail in shear, in kinking or in bending. Ductile matrix
fails via. plastic flow while brittle matrix fails via. shear banding.

To address the limitations of the HM-based composites, researchers at AMSL-


UTA develop new hybrid composites by mixing the HM and IM reinforcing carbon fibers.
A new design, the HS40/MR70/3831 hybrid material, is significantly stiffer, with only a
small reduction in axial compression strength comparing to the legacy aerospace IM
composites [17]. This promising result motivated the breaking down of the Research
Topic into smaller questions/objectives:

- Question 1: Explain the low axial compression strength of HM composites


compared to IM composites.

- Question 2: Explain the improvement of axial compression strength of the


new hybrid composite designed at AMSL.

4
- Question 3: Propose an optimal configuration of IM/HM hybrid FRC to
maximize axial compression strength.

2.2. Literature reviews

This section reviews the analytical and numerical methods that have been
employed to study axial compression failure of FRC. This section also reviews the
studies on the hybrid composites and on the kink-band formation, likely the principal
axial compression failure mechanism in high-performance carbon FRC.

Analytical methods
Analytical models have been proposed to explain the failure mechanics and
𝑜
attempt to predict composite’s 𝜎𝑐11 . By their nature, analytical models can be classified
into two groups. The first group predicts compressive strength based on properties of
the components (fiber, matrix, interface etc.). The second group predicts compressive
strength based on other macroscopic properties of composite such as shear modulus or
shear strength (Barbero et al. [20]). Models of the second group require composite
specimens to be manufactured first (and have macroscopic properties measured),
hence are of little interest in our project which focuses on composite properties
prediction from components’ properties.

In the first group, Rosen’s 2D analytical model [8] might be the first to propose a
compressive strength value based on the buckling of fibers in two modes: Extension
mode (or Out-of-phase) where:

1
𝑉𝑓 𝐸𝑚 𝐸𝑓 2
𝜎𝑐∗ = 2𝑉𝑓 [ ]
3(1 − 𝑉𝑓 ) (1)

and Shear Mode (or In-phase) where:

𝐺𝑚
𝜎𝑐∗ = (2)
1 − 𝑉𝑓

5
In most applications, compressive strength given in (1) is much higher than (2),
and (2) is frequently cited as the Rosen model’s compressive strength. Nevertheless,
this value is widely recognized as being much higher than the normally measured
compressive strength [11], [12]. Later authors proposed various models (Steif et al. [11],
Budiansky et al. [21], Gutkin et al. [14]) with added complexity and refinement, including
various component parameters such as fiber’s Young Modulus, fiber geometry and
misalignment, matrix shear strength etc.

Due to the complex failure mechanics and heterogenous nature of the composite
materials, it might be complicated for analytical models to capture all these effects. The
difficulty increases further with the more complex geometry of hybrid composites.
Considering these difficulties and potential limitations, my research efforts focus on
numerical modelling.

Numerical methods
Numerical simulation using Finite Element Method (FEM) has been employed to
study the fiber-matrix and fiber-interface-matrix systems. A 2D FEM model was
developed by Gutkin et al. [14] to validate their analytical model with good correlation on
the kink-band formation part of the failure envelope. In addition, Gutkin’s FEM model
also created the shear-driven compressive fiber failure of the failure envelope which
was not described by their kink-band formation analytical model. Naya et al. [15] studied
a multiple fiber Representative Volume Element (RVE) with random fiber distribution,
matrix and fiber-matrix interface. However, due to high computation costs, their
parametric studies were run on a single-fiber RVE. One of the most notable results is
the observed composite strength reduction due to weaker fiber-matrix interface, which is
confirmed by Herraez et al. [22] and Ghaffari et al. [23] also by analyzing a single-fiber
RVE. The use of Periodic Boundary Conditions on the RVE by Naya and Herraez
means a realistic non-zero kink-band boundary angle cannot be reproduced.

With the increasing computation power and improved sophistication of numerical


models, computer simulation is especially useful in studying and engineering
heterogeneous composite material with complex geometry. As part of my project, FEM

6
numerical models are being developed and calibrated using experimental data from
works at AMSL and from literature.

Hybrid composites

Considering the advantages and the setbacks of IM-based and HM-based


composites, it might be a natural idea to combine their strengths in a “Hybrid
Composite” (from now shortened as “hybrid”), which offers high stress resistance of IM
with high rigidity of HM composites. Makeev et al. successfully designed an IM/HM
carbon fiber hybrid system having a longitudinal Modulus at 32% higher than reference
IM composite, while being only 5% weaker in compression strength [17]. This approach
is, however, not straightforward. Indeed, Piggott et al. [9] reported that Carbon-Glass
hybridization actually reduced Compression Strength compared to either pure Glass or
pure Carbon reinforced composite. Other studies [24], [25] confirmed this possibility of
(severe) strength reduction in hybrids.

Aveston et al. [26] proposed a simple theoretical model and failure mechanism
for hybrid composite in tension. Their model was compared with experimental results on
a Carbon-Glass-Epoxy hybrid, showing relative correspondence, differences can be
explained by the fact that not all reinforcing fibers fail at once.

Piggott et al. [9] studied experimentally the behavior in compression of a range of


different hybrids: Glass-Kevlar, Glass-Carbon, Kevlar-Carbon, Carbon-Carbon, with
varying ratio of mixing. The most important conclusion is that the “mixture rule” is not
valid in most cases for strength i.e., the observed strength deviates positively or
negatively from the “linear prediction line”. Particularly, in the case of carbon-glass
composite, compression strength is reduced. Yerramalli et al. [24] confirmed this
observation. They proposed different theoretical mechanisms of failure for hybrids with
different mixture ratios which offered an explanation for their experimental results.
These results also confirm the complexity and variety of failure mechanisms of hybrid –
which is unsurprising, given that even non-hybrid composites exhibit different
mechanisms according to their specific configuration (fiber and matrix properties, fiber
volume fraction, etc.).

7
There are few numerical studies on hybrid composites. Koppisetty et al. [16]
studied the effect of fiber distribution in Glass/Carbon hybrid using FEM on a RVE
containing different distributions of fibers. The studied configurations appear relatively
simple and idealistic. Nevertheless, the simulations confirmed the dependence of
Compressive Strength on fibers’ placement and suggested positive effects of
glass/carbon hybridization if fibers placement could be controlled at tow-level.

Kink-band formation as a mode of longitudinal compressive failure

Kink-band formation is widely agreed as the principal mode of longitudinal


compressive failure in UD composite [10], [27]. Therefore, any credible FE model of
fiber-direction compressive composite failure should be able to reproduce kink-bands
similar to the experiments. Thus, it is helpful to review the literature on the kink-band
properties including their formation mechanics in the UD composites.

Kink-band is typically created due to the structural buckling of oriented material.


As a result, kink-band can be formed at both meso/ply-scale [28] and micro/fiber-scale
[13]. At microscale, kink-band is supposed to be formed through local fiber buckling,
fracture and rotation. Kink-bands frequently appear in conjunction and might have
causality relation with other failure modes such as composite shear failure and shear
splitting, in which shear splitting can initiate kink-band formation and vice versa [29],
[30].

Wang et al. [30] studied kink-band formation in notched UD CFRP under four-
point bending, and described the two main types of kink-band geometry: 1-Shear type
and 2-V-Shaped type. Figure 1 illustrates the kink-bands of these two types. Meanwhile,
several experimental studies show the more frequent presence of Shear-type kink-
bands [29], [31], [32]. As a result, my project concentrates on the Shear-type kink-band.

8
Figure 1. Illustration of two main types of kink-band geometry: shear and V-shaped.

Figure 2 provides an illustration of the shear-type kink-band with its important


geometric features including:

- Kink-band width along fiber direction: 𝑤

- Kink-band fiber rotation angle, relative to original fiber direction: 𝛼

- Kink-band inclining angle, relative to original transversal plane: 𝛽

- Kink-band orientation: out-of-plane, in-plane

-
- Figure 2. Label system of kink-band geometric properties: width 𝑤; fiber rotation 𝛼; inclining
angle 𝛽.

For shear-type kink-band on the T800 CFRP, Wang et al. [30] reported the
following properties: kink-band width 𝑤 = 15 − 30 𝜇𝑚; kink-band inclination 𝛽 = 20𝑜 −

9
35𝑜 ; fiber rotation angle 𝛼 = 45𝑜 − 60𝑜 under load and 𝛼 = 20𝑜 − 50𝑜 when unloaded. It
seems obvious that kink-band width and inclination are crack properties so they do not
change after unloading, while fiber rotation would be reduced when the applied load is
relaxed. Hahn et al. [10] summarized kink-band geometry of several FRC types. For
carbon FRC, fiber breaks are usually observed at kink-band boundaries. Fiber kinking
without fracture is typical for the less brittle Kevlar FRC. Glass FRC even shows
extreme fiber bending without fiber fracture. Hahn et al. also reported the approximate
relation between the angles: 𝛼 ≈ 2𝛽. This relation is supported by some analytical
models on the kink-band formation. Daniel et al. [29] studied kink-band features on the
thick specimen (72 plies, 9.1𝑚𝑚 thick) of the CFRP IM6G/3501-6. Daniel et al. reported
some important behaviors of the initiation and development of the kink-bands:

- First failure mechanism: failure seems to be initiated by matrix shear


yielding/failure precipitated by local fiber misalignment, leading to kink-band
formation.

- Fiber buckling and fracture: follows matrix shear yielding/failure. Fiber fracture
increases shear stress in its vicinity, favoring matrix yielding and leads to
further propagation of fiber buckling and fracture. This cascade of events
eventually leads to the formation of kink-bands.

- Kink orientation angle: starts very small, observed to be 𝛼0 = 5.3𝑜 , but


increases with higher compressive stress. The angle 𝛼 can increase up to a
maximum value of 𝛼𝑚𝑎𝑥 = 2𝛽.

- Once the maximum rotation angle 𝛼𝑚𝑎𝑥 is reached, further stress increase
leads to kink-band multiplication and broadening. In parallel, kink-band
multiplication and broadening leads to decreased fiber rotation angle 𝛼,
suggesting a relation in which the local stress is partially relaxed by the kink-
band propagation.

- Multiple kink-bands can originate from one source region.

10
- There is no clear relation between kink angles 𝛼 and 𝛽. However, the
maximum value of local 𝛼 is reported to be: 𝛼𝑚𝑎𝑥 = 2𝛽, which is similar to the
observation of Hahn et al. [10]. 𝛼 and 𝛽 are frequently but not always equal.

- Fiber rotation angle 𝛼 ranges from very small to up to 60𝑜 .

- Kink inclination 𝛽 seems to depend on both local and global states of stress.
Its value varies between 20𝑜 − 30𝑜 .

- Kink-band width 𝑤 varies between 4 to 20 fiber diameters. For IM6 carbon


fiber, these equal to 20 − 100 𝜇𝑚.

- Within the same kink band, both width 𝑤, kink angles 𝛼 and 𝛽, and kink-band
orientation (in-plane, out-of-plane) could vary along the kink band.

- Insignificant effect of specimen thickness on compressive strength. However,


kink-band geometry is more complex in thicker specimens.

The most interesting result from the study of Daniel et al. [29] is probably the
observation that, for CFRP with high performance fiber such as the IM6G/3501-6, kink-
band formation occurs before fiber fracture at kink-band boundaries. This strongly
supports the hypothesis that fiber buckling leading to kink-band formation is the
principal initiator and mode of compressive failure in high-performance CFRP.

2.3. Analysis of composite fiber-direction compressive strength using FEM

The FE models are developed in Abaqus using integrated Python Scripting


feature, allowing the flexibility and rapidity of modifying and parameterizing the FE
models, saving time both in building the models and processing the results.

By definition, the RVE is the smallest volume over which an estimation of a


certain property of the material as a whole can be extracted with a given precision. For
elastic properties, RVE size can be small (an RVE containing five fiber is sufficient, see
Appendix A), while damage properties study requires RVE of much larger size. RVE
numerical studies by Gonzalez et al. [33] suggested that a RVE containing 30 fibers
(with 50% fiber volume fraction, the cross-section size is about 13.7 times fiber radius)

11
is sufficient to capture damage behaviors. Naya et al. [15] utilized an RVE containing 50
fibers with 60% fiber volume fraction (the cross-section size is therefore 16 times fiber
radius) for compressive strength analysis. Due to high computational cost, Naya utilized
a RVE containing 1 fiber for parametric studies, reasoning that 1-fiber RVE produces
the same compressive strength as the 50-fiber RVE. However, this equivalence was
shown on only one single configuration (fiber distribution, fiber misalignment), and has
not been verified on other configurations.

In this project, Periodic Boundary Condition (PBC) is applied on the RVE using
the algorithm and technique of “dummy nodes” as explained by Gomez et al. [34]. The
current algorithm applies PBC to any 3D rectangular cuboid meshed volume where the
meshes on opposite surfaces/edges are periodic. Details of the PBC and the RVE size
convergence study are presented in Appendix A.

Figure 3. Example of FE RVE with PBC, fiber waviness for longitudinal compression test: a) 1-fiber RVE;
b) 65-fiber RVE

12
Figure 3 presents the RVE for compression strength study. To create FE model
geometry, circular cross-sections of fibers are randomly generated as circles on a 2D
square surface, then the 3D model is created by extruding this 2D surface along the
fiber direction. For each square area, the number of fibers is appropriately calculated so
that the RVE has the same Fiber volume fraction 𝑉𝑓 = 0.591 as the reference material.
PBC is applied on the model with some modifications accounting for hypotheses of a
fiber-direction compression test. Fiber misalignment is introduced by applying sinusoidal
coordinate displacements to RVE’s nodes. Fiber’s material orientation is defined
following the finite elements’ inherent geometry, hence the local fiber material
orientation always follows the fiber geometrical orientation, including their
misalignment/waviness. Fiber and Matrix are modelled by a mixture of C3D6 and C3D8-
types 3D elements.

The FE RVE models the fibers, the matrix and the fiber-matrix interface. Fiber
material model is homogenous, orthotropic elastic. The Matrix material model is
homogenous, isotropic, with non-linear behavior modelled using the Concrete Damaged
Plasticity available in Abaqus. Interface is modeled with Surface-to-Surface cohesive,
damageable contact. The RVE presented in this study utilizes the IM7/8552 CFRP
composite as reference. Component material properties of the IM7/8552 composite are
presented in detail in Chapter 3. The current fiber material model does not describe its
damage behaviors. However, literature reviews shows that fiber material damage has
potentially important role in kink-band formation and propagation as well as geometry.
As a result, future improvements to the FE model should include fiber damage behavior.

The composite RVE represents an elastic column prone to compressive buckling.


Therefore, the Riks method available in Abaqus is implemented as an efficient strategy
to study the RVE structure buckling.

13
2.4. Effects of some microscale properties on UD composite longitudinal
compression strength

In this Section and the Section that follows, “Stress” result refers to the global
compression stress, which equals applied Load divided by the RVE cross-section Area
orthogonal to fiber direction. “Strain” refers to change of RVE length in the original fiber
direction.

Figure 4 presents a preliminary analysis of fiber misalignment effects over RVE


compression strength. The model consists of a single fiber (Figure 3.a). Misalignment is
varied by changing the amplitude of the sinusoidal waviness. The results indicate that
the Peak Load is reduced with increasing misalignment. The configuration with 0.9𝑜
misalignment (similar to the standard deviation of typical experimentally measured fiber
misalignment [35], [36]) produced a buckling load/ peak load higher than the
experimental results. The tendency indicates that lower fiber misalignment would lead to
even higher buckling loads. These results are similar to the numerical results by Naya et
al. [15].

Figure 4. Effects of fiber misalignment on compressive behavior of 1-fiber RVE: a) Stress-Strain curves of
different misalignment angles; b) Peak-Load in function of fiber misalignment angle.

Figure 5 presents the FEM results on effects of weak fiber-matrix interface on the
RVE buckling stress. FE RVEs with 65 fibers and 1 fiber (depicted in Figure 3) were
studied. Fiber misalignment angle is 0.9𝑜 . The simulation of 65-fiber models took
approximately 25 hours to finish, while the 1-fiber model took approximately 10 minutes.

14
Stress-Strain results are compared between the two cases: 1-interfacial damage is
deactivated, and 2-the interface is weak and damageable. Interfacial strength is taken
from experimental measurement of IFSS on IM7/8552 composite [23]. On the FE RVEs
of both sizes, weak and damageable interface leads to lower Compressive Strength, a
tendency observed in FEM numerical results on single-fiber RVE by Naya et al. [15],
Herraez et al. [22] and Ghaffari et al. [23].

Figure 5. Damageable interface leads to lower compressive strength in fiber-direction. Stress-Strain


curves of: a) 1-fiber RVE; b) 65-fiber RVE

Figure 6 compares compression stress-strain curves of the 65-fiber RVE and the
1-fiber RVE (shown in Figure 3), both having the same fiber volume fraction and fiber
misalignment (0.9𝑜 ). Figure 6.a compares results from models without interface
damage, and Figure 6.b compares results from models with interface damage activated.
Mesh densities on both RVEs are approximately the same. The initial, linear part of the
two curves shows no difference, due to similarities in material compositions and fiber
volume fraction. However, in both cases with interface damage and without interface
damage, 1-fiber RVE shows a significantly lower peak-load than the 65-fiber RVE: 6%
lower in case of no interface damage, and 18% lower in case with interface damage.
These results suggest that RVE containing only 1 fiber is insufficient for composite’s
𝑜
𝜎𝑐11 strength analysis.

15
Figure 6. Comparing stress-strain curves of 65-fiber and 1-fiber RVEs: a) Without interface damage; b)
With interface damage activated

2.5. Parametric analysis: comparing axial compression strengths of HM


and IM composites

Using the 1-fiber RVE at 0.9° of misalignment, this Section presents an analysis
of axial compression strengths of the MR70/3831 (IM) and the HS40/3831 (HM)
carbon/epoxy composites.

Table 1 presents the reference elastic parameters of HS40 and MR70 carbon
fibers in the model. The axial modulus values (𝐸11 ) are provided by the manufacturer.
Other properties are assumed equal to the IM7. Table 2 presents the elastic and plastic
properties of the 3M 3831 epoxy. Modulus and Tension Strength are provided by the
manufacturer. Other properties are estimated based on the 8552 epoxy. For a detailed
presentation on the micro-properties of IM7 carbon fiber and 8552 epoxy, please refer
to Chapter 3. Table 3 presents the reference IFSS of MR70/3831 and HS40/3831
composites which were obtained experimentally through fiber push-out test [18].

16
Table 1. Reference mechanical properties of HS40 and MR70 carbon fibers.
𝐸11 𝐸22 𝐸33 𝜈12 𝜈13 𝜈23 𝐺12 𝐺13 𝐺23
(GPa) (GPa) (GPa) (GPa) (GPa) (GPa)
HS40 (HM) 455 15.0 15.0 0.20 0.20 0.07 15.0 15.0 7.03

MR70 (IM) 325 15.0 15.0 0.20 0.20 0.07 15.0 15.0 7.03

Table 2. 3M 3831 reference mechanical properties


𝐸 (GPa) 𝜈𝑚
Elastic
6.16 0.39
Concrete Damaged Ψ 𝜖 𝜎𝑏0 /𝜎𝑐0 𝐾𝑐 𝜇
Plasticity 29 0.1 1.29 1 1E-5
Tensile (MPa) Compressive (MPa)
Yield Stress
125 180
Mode I (J/m2)
Fracture Energy
100

Table 3. Experimental IFSS of HS40/3831 and MR70/3831 carbon/epoxy composites.

IFSS (MPa)

HS40/3831 109

MR70/3831 154

The IM and HM carbon fibers show a very apparent difference in fiber axial
modulus. In addition, experimental data show a significant difference of IFSS between
the two types [18]. As such, sensitivity studies on the effects of these parameters on the
composite strength are performed. Using the HS40/3831 as a baseline, Figure 7
presents the effects of fiber axial modulus (Figure 7.a) and of the IFSS (Figure 7.b) on
the RVE compression Stress-Strain response. Fiber Modulus visibly affects the initial
slope, but has almost negligible effects on the Buckling Stress. On the other hand, IFSS
has significant influence on the Peak/Buckling Stress. Figure 7.c presents the effects of
IFSS on the RVE compression strength. At lower values of IFSS, the effects on
compression strength are very significant. The effects become less significant at higher
values of IFSS as stronger interfacial strengths produce the results approaching the
case of perfect interface with no damage.

Figure 7.d compares the RVE Stress-Strain responses of the HS40/3831 (HM)
and MR70/3831 (IM) composites. A 30%-lower IFSS in HM composite RVE leads to an
8% reduction of compression strength compared to the IM. This reduction is less

17
significant than experimental observations (from 20% to 30% reduction [18]). The
difference might come from the incomplete RVE (in size, fiber distribution
representation, material models etc.).

Figure 7. On 1-fiber RVE: a) Fiber Modulus effects on Stress-Strain response; b) IFSS effects on Stress-
Strain response; c) IFSS effects on Buckling Stress; d) Stress-Strain responses of HS40/3831 (HM) and
MR70/3831 (IM) composites.

These results might offer an explanation to the Question 1 regarding the low axial
compression strength of the HM composite. The analyses suggest a strong correlation
between the IFSS and the Buckling Stress of the Composites. As micro-buckling is
likely the principal mechanism of axial compression failure in high-performance carbon
FRCs, low IFSS is likely the cause of lower axial compression strength in HM
composites.

18
2.6. Proposals for improvements and future works

Validation of Micromodel: 4-point-bending notched specimen

The microscale RVE has traditionally been implemented with Periodic Boundary
Conditions [15], [23], [37], representing a small volume located deep within the
specimen of much larger scale. Literature review suggests that for damage analysis
purposes, the RVE cross-section should be of at least 50 times the fiber radius [38],
[39]. In HS40/3831, this minimum size equals 125 𝜇𝑚. Meanwhile, UD composite plies
have the typical thickness of 400 𝜇𝑚. The RVE size is therefore of almost the same
scale as the ply thickness, making the argument of “small volume located deep inside
the specimen” and the application of Periodic Boundary Conditions questionable. As a
result, a direct validation of the micromodel is difficult due to the uncertainty in
determining appropriate boundary conditions.

To alleviate this difficulty to validate the micromodels due to uncertain boundary


conditions, the AMSL/UTA Research Team proposed the experiment of 4-point-bending
notched specimen (Figure 8.a). Due to stress concentration at the notch, the location of
initial compression failure can be easily predicted (Figure 8.b, 8.c). As a result, with the
micromodel being inserted at the notch, compression failure analysis can be conducted
on the micromodel with boundary conditions being determined precisely. Thus, this
experimental setup potentially offers a strong, direct validation of the micromodel
without any assumption required on the boundary conditions. Figure 8.d presents the
FE model of the 4-point bending notched specimen with the composite micromodel
embedded at the central notch.

19
Figure 8. Four-point bending notched specimen: a) Experiment setup; b) DIC image of the notch area,
showing 𝜀𝑥𝑦 ; c) Compression fracture initiation at the notch; d) FE model of the experiment, with
micromodel embedded at the notch.

Proposals for future works

Below are some perspectives on numerical simulation model:

- Improve geometry of the RVE: fiber misalignment should be random and follow
a configurable distribution. Figure 9 compares the current RVE having uniform fiber
misalignment and the nano-CT image of a microscale composite volume where random
fiber misalignment is apparent.

- Calibrate the stress concentration field at the notch where the embedded
micromodel is located. Figure 10 presents an example of longitudinal stress
concentration in this area of interest where a token material volume with homogenized
composite properties is embedded (Figure 10.a). The low computational cost of the

20
homogenous token material volume is useful at the stage of feasibility and calibration
study. Figure 10.b presents the global Load-Displacement response registered at the
numerical Loading Heads in the 4-point-bending model, showing good correlation up to
the point of abrupt failure.

- Develop a simulation strategy to optimize calculation resources. Both the


microscale RVE and the 4-point-bending FE model are potentially very expensive in
terms of computational cost due to the complex geometry of huge number of fibers in
the micromodels.

- Improve Abaqus simulation model: add damage model to fiber material model,
refine material and interface properties. Perform mesh and size convergence analyses.

- Validate the numerical micromodel using the 4-point-bending notched specimen


setup which offers precise determination of the Boundary Conditions applied on the
embedded micromodel.

- Once the micromodel for axial compression strength has been validated,
perform parametric studies to determine the impact of different material parameters on
𝑜
composite axial compression strength 𝜎𝑐11 . Identify key parameters with the most
𝑜
significant impact on 𝜎𝑐11 .

- The method of Barbero et al. [20] provides an interpretation on effects of fiber


𝑜
misalignment distribution on composite 𝜎𝑐11 at microscale and at macroscale. This
method might offer a prediction of material strength without the need for a RVE with
realistic fiber misalignment distribution. However, an improved RVE geometry with
realistic fiber misalignment distribution, combined with direct micromodel validation
using the 4-point-bending notched specimen presented above, is believed to provide a
more robust representation of composite failure and would be more useful for the
optimization of new hybrid composite designs.

- With the RVE having realistic fiber misalignment representation and the
micromodel being validated, these FE micromodels can be applied to the engineering of
hybrid composites. Optimize hybrid compressive strength by studying the impact of

21
different design variables, including the choice of fibers, distribution and placement of
two types of fiber, etc.

Figure 9. a) Current RVE with uniform fiber misalignment; b) Nano-CT image showing realistic, random
fiber misalignment within a small composite material volume [31].

Figure 10. a) Heterogenous embedded micromodel is replaced by homogenous volume with


homogenized composite properties for efficient stress concentration calibration analysis; b) Comparing
Load-Displacement response at Loading Heads, FEM and experiment.

22
2.7. Conclusions

A literature review was performed on the analytical and numerical methods for
the analysis of UD composite axial compression failure in carbon FRC as well as their
applications in hybrid composites that combine different types of reinforcing fibers. The
numerical FE modeling is suggested as the preferred methodology for the project
thanks to its ability to represent complex material models, complex failure mechanisms,
and the complex geometry of the hybrid carbon FRCs at microscale. The literature
review also found the kink-band formation to be a principal compression failure
mechanism in high-performance CFRP composites. Common geometric features of
kink-bands in different types of FRC were reviewed and summarized, which would
serve as a helpful reference for the future studies on numerical modeling of kink-band in
FRC materials.

An RVE with uniform fiber misalignment was developed and analyzed for
instability/buckling failure under longitudinal compression. Parametric analyses
identified several microscale properties with significant impact on composite axial
compression strength: fiber misalignment and fiber-matrix IFSS. An RVE containing one
fiber is likely insufficient to represent axial compression failure in composite.

A study comparing HM and IM carbon FRC compression strength was performed


using parametric analysis. Fiber axial modulus has almost negligible effects on
composite strength, while IFSS has very pronounced effects. Lower IFSS in HM
composites is likely the reason for the lower compression strength compared to IM
composites due to the lower resistance to compression instabilities/buckling failure.

The application of Periodic Boundary Conditions on the micromodel RVE was


discussed and questioned. Direct validation of the micromodel has long been difficult
due to the inability to represent precise, realistic boundary conditions in the microscale
numerical model. To alleviate this difficulty, AMSL/UTA Research Team proposed the 4-
point bending notched specimen experimental setup. In this experiment, boundary
conditions on the micromodel embedded at the notched can be determined precisely.
An FE model of the experiment was developed, and preliminary feasibility and
calibration analyses were performed.

23
Future works on the research topic of composite axial compression failure were
proposed. Microscale RVE needs improvement to represent realistic, random fiber
misalignment. Further development of the 4-point-bending notched specimen FE model
is required for the strong validation of the micromodel with realistic boundary conditions.
Numerical models need to incorporate improved material models, including fiber
damage for realistic kink-band formation. With the improved RVE validated to represent
axial compression failure in FRC composite, works could then be done on the
optimization of hybrid composites with different types of reinforcing fibers using the
numerical micromodel.

24
CHAPTER 3:
CFRP component materials properties at microscale: review and
suggestions for future experimental works

3.1. Introduction

The FEM has the advantage of explicitly representing different phases of the
heterogenous Fiber-reinforced Composite (FRC) at microscale. However, the
understanding of the microscale component properties (fiber, matrix, interface etc.) is
incomplete. In addition, measurement methods for some microscale properties cannot
be found in the literature. On the other hand, for the specific objective of modelling UD
FRC fiber-direction compression strength, preliminary numerical results and literature
review suggest some material properties have more significant effects than others. It is
therefore unnecessary to measure precisely some properties i.e., a good estimation
within an order of magnitude is sufficient.

This Chapter provides an extensive review of FRC components microscale


properties, their available measurement methods, the significance of their effects on the
UD FRC compression strength whenever possible, and the recommendation if future
experimental measurement of that material property should be conducted. The relatively
well-studied IM7/8552 carbon/epoxy CFRP is also the reference material for this review.

Color Code for physical parameters review table

Physical properties of each component of the FRC system (fiber, matrix, fiber-
matrix interface) are summarized in a review table at the beginning of each
corresponding section. Each physical parameter is described by:

1. Value: is this parameter’s value available from literature (for IM7/8552


CFRP)?

2. Importance on the study of composite’s fiber-direction compressive strength


𝜎𝑐𝑐𝑟𝑖𝑡 : based on numerical sensitivity studies and literature review.

3. Measurement methods available.

25
4. Recommendation for experimental work: considering the parameter’s
availability, impact and importance towards the microscale modelling project
at AMSL laboratory.

To visualize the availability and the importance of parameters and corresponding


measurement methods, a system of color is applied. Table 4 defines the color code for
some scenarios.

Table 4. Color code for material properties summary tables.


Parameter 1-Value 2-Important? 3-Measurement 4-Recommendation for
Impact? method experimental works
Green: Green: Green: established
determined or Unimportant
estimated
White: Red: Orange:
unknown Important Value is estimated
or dubious method
White: White:
Unknown unknown

3.2. Fiber microscale properties

Summary

Table 5 summarizes mechanical properties of the IM7 carbon fiber.

Table 5. Mechanical properties of IM7.


Parameter 1-Value 2-Important? 3-Measurement 4-Recommendation for
Impact? method experimental works
𝐸11 276 GPa Unknown, FEMU; Supplier- FEMU available at
[40] likely important provided AMSL
𝐸22 15.0 GPa Unknown FEMU; Ultrasound FEMU available at
eigenfrequencies AMSL
etc.
𝐺12 15.0 GPa Unknown, FEMU; Torsional FEMU available at
likely important Pendulum AMSL
𝐺23 7.03 GPa Unknown FEMU FEMU available at
AMSL
𝜈12 0.2 Unknown FEMU FEMU available at
AMSL
𝜈23 0.072 Unknown FEMU FEMU available at
AMSL
Longitudinal Unknown Important Bending Beam; Recommended due to
Compressive Micro-Compression; high importance
𝑜
Strength etc. Related to 𝜀𝑐11

26
𝑜
𝜎𝑐11
Longitudinal Unknown Important Bending Beam; Recommended due to
Compressive Micro-Compression; high importance
𝑜
Critical Strain etc. Related to 𝜎𝑐11
𝑜
𝜀𝑐11
Longitudinal 5516 MPa Unknown Supplier-provided;
Tensile [40] Tension test

Strength
𝑜
𝜎𝑡11

Experimental measurement methods

Longitudinal Modulus 𝑬𝟏𝟏 and Longitudinal Tensile Strength 𝝈𝒐𝒕𝟏𝟏

Experiments determining carbon fiber microscale properties are difficult due to


𝑜
the small scale involved. The longitudinal tension properties 𝐸11 and 𝜎𝑡11 are the
exceptions thanks to the fiber physical length that allows easy tension test setup. In
addition, these properties are typically provided by the fiber manufacturers.

It is important to note that CFRP composite’s longitudinal modulus in


compression is different from the longitudinal modulus in tension. The compression
modulus is typically equal to 90% the tension modulus [40]–[42]. By rule of mixture, we
can infer the relation between carbon fiber longitudinal modulus in compression and in
tension:
𝑓𝑖𝑏𝑒𝑟 𝑐𝑜𝑚𝑝𝑜𝑠𝑖𝑡𝑒
𝐸𝑐 𝐸𝑐
𝑓𝑖𝑏𝑒𝑟
≈ 𝑐𝑜𝑚𝑝𝑜𝑠𝑖𝑡𝑒
≈ 90%
𝐸𝑡 𝐸𝑡

Transversal Modulus 𝑬𝟐𝟐

The Raman spectroscopy method was applied by Miyagawa et al. [43] to


measure 𝐸22 and compare with results from other methods: Dynamic Mechanical
Analysis, Berkovich Nano-indentation and FEM analysis. They found significant
difference amongst the values obtained by different methods. Amongst these methods,
Gross et al. [44] demonstrated the weaknesses of the Nano-indentation technique on
anisotropic materials. Mounier et al. [45] proposed a method to measure 𝐸22 using laser
resonant ultrasound spectroscopy and FEMU analysis.

27
Amongst the 𝐸22 measurement methods, the FEMU for inverse determination of
fibers’ microscale properties might be the most efficient. Measurement of composite
macroscale mechanical properties and implementation of FEMU are typically
inexpensive and comparatively easier than other proposed methods.

Longitudinal Shear Modulus 𝑮𝟏𝟐

Deteresa et al. [46] described the Torsional Pendulum experimental method for
measuring a single Fiber’s Shear Modulus and Shear Strength. Based on a radial
distribution of shear strain over the fiber cross-section, from maximum at the surface to
zero at the core, Deteresa suggested the formula to calculate Shear Modulus 𝐺23 :

8𝜋𝐿𝐼
𝐺=
𝑟4𝜏 2

With fiber length 𝐿, fiber radius 𝑟, disk angular moment of inertia 𝐼, system
oscillation period 𝜏. In 1999, Tsai and Daniel [47] proposed a similar system, with a
more sophisticated analytical model which also considers pendulum’s damping, but
their experiments showed the damping effect can be neglected, hence giving the same
formula as Deteresa.

The Shear Modulus 𝑮𝟏𝟐 can also be measured using the FEMU technique.

Transversal Shear Modulus 𝑮𝟐𝟑 , longitudinal Poisson ratio 𝝂𝟏𝟐 , transversal


Poisson ratio 𝝂𝟐𝟑

Apart of the FEMU technique for inverse characterization of properties, my


literature review found no experimental measurement methods for 𝐺23 , 𝜈12 and 𝜈23 .

For an ideal transversely isotropic material, transversal shear modulus 𝐺23 can
be related to transversal Poisson ratio 𝜈23 and transversal tension modulus 𝐸22 via the
equation:

28
𝐸22 (3)
𝐺23 =
2(1 + 𝜈23 )
However, Gusev et al. [48] demonstrated the high error of the equation (3) in
application to UD composites. The microstructure of carbon fibers also shows
transversal anisotropy [49]. Therefore, the three elastic parameters 𝐺23 , 𝐸22 and 𝜈23 are
considered independent in my studies for both the macroscale composites and
microscale carbon fibers.

Longitudinal Compression Strength 𝝈𝒐𝒄𝟏𝟏 and Critical Strain 𝜺𝒐𝒄𝟏𝟏

In UD CFRP composites, literature review and preliminary FEM analysis suggest


𝑜
the high importance of carbon fiber longitudinal compression strength 𝜎𝑐11 on both the
compression peak load and the post-failure behaviors including kink-band formation.
Since carbon fiber exhibits linear behavior prior to its brittle fracture, it can be assumed
𝑜 𝑜
that 𝜎𝑐11 and 𝜀𝑐11 are linearly correlated, and thus the determination of the two
parameters is equivalent:
𝑜 𝑜
𝜎𝑐11 = 𝐸11 × 𝜀𝑐11
𝑜 𝑜
Several methods exist in the literature for measurement of 𝜎𝑐11 (and/or 𝜀𝑐11 ). The
𝑜
Elastica Loop method was first developed by Sinclair et al. [50] to measure 𝜎𝑐11 of glass
fiber. Furuyama et al. [51] utilized the method to study carbon fibers. The method has
the advantage of a relatively simple and inexpensive setup. However, the precise
moment of fiber damage initiation is hard to determine, complicating interpretation of
results. Deteresa et al. [46] proposed the Bending Beam method. In the setup, the
microfiber is bonded on the upper surface of an elastic beam with rectangular cross-
section. The bonded microfiber follows the beam’s upper surface strain while the beam
is bent under either cantilever or three-point bending setup. Using beam theory, bonded
fiber compression strain can be demonstrated to have a linear distribution along beam
𝑜
length. By observing the extent of fiber damage, fiber critical compression strain 𝜀𝑐11
can be deduced. This method allows complete prevention of Euler buckling in tested
𝑜
fiber which is the main difficulty in 𝜎𝑐11 measurement. The Micro-Compression method
𝑜
allows direct measurement 𝜎𝑐11 by pushing a free-standing part of a microfiber in the

29
longitudinal direction. The free-standing fiber part is short to prevent Euler buckling. The
microfiber itself is typically held in place by some kind of fixation (glue, polystyrene etc.).
Oya et al. [52], Shioya et al. [53], Ueda et al. [54], Leal et al. [55] proposed different
setups and sample preparation techniques of the Micro-Compression method. The
method requires careful monitoring of fiber damage to avoid possible unwanted effects
(matrix damage, fiber-matrix interface damage etc.). In fact, deduced fiber critical
𝑜
compressive strain 𝜀𝑐11 from micro-compression results from the literature [52] is lower
than critical compressive strain of the corresponding composite (view Table A6),
suggesting an unrealistic situation where the reinforcing fibers failed but the composite
is still intact. This comparison suggests that these fiber compressive strength results
might be underestimated due to multiple unwanted effects affecting Micro-Compression
measurements. Another disadvantage of the method is the time-consuming preparation
of samples, which was partially alleviated by the sample preparation technique of Leal
et al. [55]. In the Recoil Test method [56], the fiber is put under tensile loading with two
ends fixed on a frame. When the fiber is cut (typically using laser), the elastic recoil
fractures the fiber in compression. This method’s greatest advantage is its simplicity.
However, the interpretation of the result relating compression strength to initial tension
stress might be difficult. Newell et al. [57] provides an analytical expression of axial
stress history in microfiber which suggests strong dependence of axial stress on various
assumptions on boundary condition (BC) as well as uncertainties in measures. The
Embedded In Resin method was proposed by Hawthorne et al. [58], where a
rectangular cuboid of transparent elastic material contains one single fiber. Fiber is
monitored for damage while compression strain parallel to the fiber is applied on the
cuboid specimen. The method provides a straightforward test procedure and result
interpretation is simple. However, sample preparation in this method is extremely
laborious, with the author having to discard many specimens due to fiber orientation
issue.
𝑜
Due to the importance of reinforcing fibers 𝜎𝑐11 on the modelling of composite
𝑜
strength, it is advised that 𝜎𝑐11 measurement is performed on interested fiber. The
Bending Beam technique by Deteresa et al. [46] is proposed as the preferred method
thanks to its relatively simple and inexpensive experimental setup, as well as the

30
robustness and the simplicity of result interpretation. In addition, this method was
demonstrated on carbon fibers [59].

3.3. Matrix microscale properties

Summary

Table 6 summarizes mechanical properties of the 8552 epoxy resin. The


Concrete Damaged Plasticity (CDP) model available in Abaqus is commonly used to
describe non-linear behavior of the brittle epoxies [60], [61]. Table 7 summarizes the
properties/parameters for the CDP modelling of the 8552 epoxy.

Table 6. Mechanical properties of the 8552 epoxy.


Parameter 1-Value 2-Important? 3-Measurement method 4-Recommendation for
Impact? experimental works
Modulus 4670 𝑀𝑃𝑎 [62] Yes Tensile Test; Supplier-
𝐸 provided [62]

Poisson ratio 0.35 [60] Unknown Typical of Epoxy. Ultrasonic


𝜈 Velocity Measure [63]; Fluid
Compressibility [64]
Compressive 174 𝑀𝑃𝑎 [60] Unknown, Nano-indentation [65] Equipment and FEM
Yield Stress likely analysis available at
𝑦
𝜎𝑐 important UTA
Tensile Yield 121 𝑀𝑃𝑎 [62] Yes Provided by Supplier [62]; Recommended, using
Stress Potentially different at microfiber tensile test
𝑦
𝜎𝑡 microscale [66] [66]
Fracture 1.62 𝑀𝑃𝑎√𝑚 Not known Provided by Supplier [62]
toughness [62] yet
𝐾𝐼𝐶
Energy 679.5 𝑁/𝑚 Not known Provided by Supplier [62]
release rate [62] yet
𝐺1𝐶
Fracture 100 𝑁/𝑚 Yes Provided by Supplier [65]; Potentially important
energy [65] Or estimated based on 𝐺1𝑐
𝐺𝑓

Table 7. Concrete Damaged Plasticity model parameters for the 8552 epoxy.
Parameter 1-Value 2-Important? 3-Measurement method 4-Recommendation
Impact? for experimental works
Dilation 29𝑜 [65] Unknown Assumed equal Friction Angle Significant effects on
Angle (associated flow assumed [65]); nano-indentation
Triaxial Test [67]; Parametric Load-Displacement

31
study [68] curve.
Eccentricity 0.1 Unknown Typical value for concrete [69];
Triaxial Test [67]; Parametric
study [68]
𝑓𝑏0/𝑓𝑐0 1.29 Unknown Calculated from Internal Friction
Angle – obtained from nano-
indentation [65]
𝐾𝑐 1.0 Unknown Assumed associated flow [65];
Parametric study [68]
Viscosity 0.00001 Numerical value Recommended small or zero
Parameter for stabilization [70]

Experimental measurement methods


In the topics studied during my PhD Project (the Composite Compressive
Strength, Residual Microstress and IFSS Measurement using Push-Out method), the
following matrix properties were identified as important and have significant impact:

- Modulus 𝐸: significant effects on microscale residual stress.


𝑦
- Tension Yield 𝜎𝑡 : significant effects on microscale residual stress and IFSS
measurement using Push-Out method.
𝑦
- Tension Yield at microscale 𝜎𝑡 : various authors [66], [71]–[73] studied epoxy
tensile behavior at microscale using tensile test on epoxy microfiber samples.
They demonstrated very significant size effects: microscale epoxy is ductile
while macroscale epoxy is brittle; ultimate failure strain and stress of
microscale epoxy is much higher than macroscale epoxy. This suggests the
𝑦
importance of determining microscale epoxy matrix 𝜎𝑡 in modelling composite
at microscale.
𝑦
- Compression Yield 𝜎𝑐 : sensitivity study was not conducted on this parameter,
but matrix compression yield is likely important due to its role in keeping
reinforcing fibers orientation under compression load.

Matrix Modulus 𝑬

32
Epoxy Modulus can be obtained using tension test. Results from literature [73]
and preliminary results at AMSL, UTA suggest that at small, elastic strain, the epoxy
Modulus at microscale (microfiber testing) and macroscale (dog-bone specimen testing)
are similar.

𝒚
Tension Yield 𝝈𝒕 and size effects

Different authors [66], [71]–[73] have consistently demonstrated the different


behaviors of epoxy at micro- and macro-scales. The most common measurement
method of microscale 𝜎𝑦𝑡 is microfiber specimen tension testing. Different methods exist
for stress-strain result analysis, including direct fixture load-displacement readings [72],
[73] and in-situ Scanning Electron Microscope (SEM) observation [66].

The epoxy microfiber tension testing method is being investigated at AMSL, UTA.

𝒚
Compression Yield 𝝈𝒄

Epoxy microscale behaviors in compression can be characterized using nano-


indentation test. For example, the Berkovich indentation technique was successfully
applied by Sanchez-Carmago et al. [74] to extract elasto-plastic behavior of the 316 L
austenitic stainless steel. Unlike steel however, thermoset epoxy exhibits different
properties in tension and in compression, at least at macroscale. Naya et al. applied the
technique to extract compression properties of the Hexcel 8552 epoxy [65]. A potential
advantage of the nano-indentation technique is the ability to realize in-situ measurement
of the composite epoxy matrix at easily selected positions.

To investigate the capabilities of the nano-indentation technique to measure


matrix microscale properties, Berkovich-indentation experiments were realized at UTA
on the 8552 epoxy matrix, and a FEM model for result analysis were developed. Details
on this study are presented later in Section 3.5.

33
3.4. Fiber-matrix interface properties

Summary

Table 8 summarizes properties of the reference composite IM7/8552 fiber/matrix


interface.

Table 8. Mechanical properties of the IM7/8552 fiber/matrix interface.


Parameter 1-Value 2-Important? 3-Measurement method 4-Recommendation for
Impact? experimental works
Interfacial 64.0𝑀𝑃𝑎 No Estimated 2/3 IFSS [75]; No
Normal Strength Single Fiber Specimen in
IFNS Tension with strain [76], [77]
or debonding [78] measure
Interfacial Shear 95.8𝑀𝑃𝑎 Yes Fiber Push-Out [23] Equipment for Push-
Strength, [23] Out experiment
longitudinal IFSS available at UTA

Interfacial Shear 95.8𝑀𝑃𝑎 No Assumed equal IFSS in No


Strength, longitudinal;
transversal Single Fiber Specimen in
IFSS Tension with strain [76], [77]
or debonding [78] measure
Normal Fracture 2 𝑁/𝑚 No Assumed similar to No
Energy [75], Glass/Epoxy interface.
𝐺𝐼𝑐 [79], [80] Transverse single fiber test
for interfacial debonding [80]
Shear Fracture 50 𝑁/𝑚 No Estimated with 𝛿𝑠𝑐 = 1𝜇𝑚 [75] No
Energy, Transverse single fiber test
𝑐
transversal 𝐺𝐼𝐼𝐼 for interfacial debonding [80]
Shear Fracture 50 𝑁/𝑚 Yes assumed equals 𝐺𝐼𝐼𝑐 [75] Measurement method
Energy, Measurement Method unavailable
longitudinal 𝐺𝐼𝐼𝑐 Unavailable
Benzeggagh- 1.45 Unknown Assumed similar to ply-level
Kenane (BK) [75], [81] parameter [75]
exponent 𝜂 𝐵𝐾 Mixed-Mode Fracture Test
Unavailable at Fiber scale
Interfacial 0.4 YES Fiber direction: Measurement Under development.
Friction [60] (kink-band method under development Residual stress
𝜇 formation) Transversal direction: Mixed- combined with friction
Mode transversal Load [60] stress measurements.
Viscosity for 0.0005 Numerical Numerical parameter for
Damage value simulation stabilization,
stabilization 𝜂 should be near-zero

34
Experimental measurement methods

First of all, it is important to note that the term “interface” between fiber and
matrix in composite is not clearly defined. Cech et al. [82] supported the assessment
that the “interphase” between glass fiber and polyester matrix has non-zero thickness –
it is instead a 3D region extending into both sides of the “interface” where the local
matrix and fiber properties are modified and discernable from bulk properties.

Meanwhile, SEM images of Fiber Push-Out experiment aftermath by Ghaffari et


al. [31] show a “clean” fiber surface which was previously separated from matrix during
the Push-Out test. This suggests the simplification of “infinitesimally small interface
thickness” is acceptable for the purpose of studying fiber/matrix composite system at
the fiber diameter scale.

On the other hand, there exists the situation where the fiber (or the matrix)
surface post-test is not smooth with pieces of matrix still attached to the fiber surface,
such as the Fiber Push-Out test results by Medina M et al.’s 2015 [83]. This
phenomenon might occur due to the high Interfacial Shear Strength compared to the
matrix material strength. Precise interpretation of interface properties and its nature in
this case might be complicated.

Preliminary FEM analysis on the fiber-direction buckling stress of the UD


composite (Section 2.3) suggests a very significant effect of the fiber-matrix IFSS
(longitudinal, fiber-direction) on the composite longitudinal compression strength. This
strong effect was also confirmed by other authors [15], [22], [23]. It was therefore
concluded that the precise determination of the IFSS is essential for the accurate
modeling of the UD composite compression buckling.

The fiber-matrix IFSS can be characterized by the Push-Out or the Push-In


method [83], each method having its own advantages and disadvantages. The Push-
Out method is considered more complicated and more time-consuming due to the need
of polishing the thin membrane specimen, but allows comparatively simpler and more
robust result interpretation. In addition to the traditional membrane polishing technique,

35
Ghaffari et al. utilized a new sample preparation technique where a “cave” is carved into
a block of material [61] using femtosecond laser. The material above the cave becomes
the “membrane” for the fiber Push-Out test. The new technique allows much faster
production of specimens at much higher volume. Considering the relative robustness
and the ease of sample preparation thanks to the new technique, the Fiber Push-Out
test is the preferred method for the IFSS evaluation. My PhD works on the evaluation of
IFSS using Fiber Push-Out and FEM analysis are presented in Chapter 5.

FEM analysis suggests a very significant effect of interfacial shear fracture


energy, longitudinal 𝐺𝐼𝐼𝑐 on the accuracy of the IFSS measurement using Fiber Push-Out
method. However, my literature study did not find any conclusive method for the
𝑐
measurement of this parameter. The reference value for 𝐺𝐼𝐼𝐼 is an estimation made by
the assumption that 𝐺𝐼𝐼𝑐 = 𝐺𝐼𝐼𝐼
𝑐
, similar to the approach taken by Herraez et al. [75]. In
𝑐
turn, 𝐺𝐼𝐼𝐼 was not measured for carbon-epoxy interface and was taken to be similar to
the glass-epoxy interface [75], [80]. Because of importance of 𝐺𝐼𝐼𝑐 , of the dubious
estimations for the current value and the lack of experimental method to determine this
parameter, 𝐺𝐼𝐼𝑐 is often the subject of many parametric studies in my PhD project.

FEM analysis of the Fiber Push-Out test in the presence of residual stress shows
the very significant effect of the interfacial friction 𝜇 on test results. However, the effects
of the matrix residual stress (squeezing the fiber) and interface friction on the measured
Load are coupled, making it difficult to conclusively determine 𝜇 without knowing the
residual stress beforehand. Works are being performed on the characterization of matrix
residual stress, after which interfacial 𝜇 can be decoupled and deduced.

Regarding other interface properties: normal strength IFNS, normal fracture


energy 𝐺𝐼𝑐 , transversal shear strength IFSS, transversal shear fracture energy 𝐺𝐼𝐼𝐼
2
, FEM
analysis suggested the weak to negligible effects of these parameters on both the
composite fiber-direction compression strength and on the IFSS measurement using
Fiber Push-Out method. Therefore, it might be unnecessary to obtain precise values for
these parameters, and a reasonable estimation is sufficient for the current purposes.

36
3.5. PhD works on the microscale material characterization

PhD work has been done on the characterization of microscale material


properties. Extensive work has been done on the two topics: 1- Microscale residual
stress in composite and 2- Evaluation of IFSS measurement using Fiber Push-Out
method and effects from different factors including micro-residual stress. Therefore,
each of these topics is covered in their own respective Chapter 4 and Chapter 5.

This Section presents other works on microscale material characterization at


different stages of completeness:

3- The FEMU method for inverse characterization of fiber elastic material. The
method implements FEM analysis combined with the Newton-Raphson algorithm. This
method is relatively simple but extremely helpful in providing fiber elastic properties,
some of which have no known experimental measurement method available.

4- The nano-indentation to characterize microscale compression properties of the


matrix. Preliminary analysis demonstrated the feasibility of the method and identified
some microscale matrix properties that can be studied with this method. This technique
enables the in-situ characterization of matrix at precisely selected locations, which is
potentially useful since material properties at microscale might have significant spatial
variations, and might differ from bulk material properties at macroscale.

5- The Bending Beam to measure fiber compression strength 𝜎𝑐11 . Literature


review and analytical analysis have been done to demonstrate the feasibility of the
technique with results interpretation, as well as provide important recommendations
regarding engineering requirements for the experiment. The characterization of micro-
fiber compression failure is strongly recommended, as literature review on the kink-
bands clearly show fiber breakage at the kink-band boundaries in high-performance
carbon FRCs. As kink-band formation is a principal failure mode in high-performance
carbon composites under axial compression, accurate representation of fiber damage is

37
important in reproducing realistic kink-band formation which is necessary for a good
prediction of composite compression failure.

FEMU method to determine carbon fiber orthotropic elastic properties

The Representative Volume Element (RVE) is the smallest volume over which an
estimation of a certain material property can be extracted with a given precision. Size
convergence study suggests that a relatively small RVE containing five (5) fibers with
random cross-sectional distribution is sufficient to reproduce macroscopic elastic
properties for UD FRC materials. Details on the implementation of the RVE with
Periodic Boundary Condition (PBC) and the convergence analysis are presented in
Appendix A.

In the study concerning fiber and composite orthotropic elastic properties, the FE
RVE takes the following inputs:

- Fiber orthotropic elastic properties: 𝑃 𝑓𝑖𝑏𝑒𝑟 = 𝐸11 , 𝐸22 , 𝜈12 , 𝜈23 , 𝐺12 , 𝐺23 . It is
assumed that properties in transversal directions equal, thus: 𝐸22 = 𝐸33 and
𝜈12 = 𝜈13 .

- Matrix elastic properties: 𝐸𝑚 , 𝜈𝑚

- Fiber volume fraction: 𝑉𝑓 . Fiber cross-sectional distribution is randomly


generated using the Improved Random Search Algorithm (IRSA) that is
described in Section 4.3.

The FE RVE produces the following outputs:

- Composite orthotropic elastic properties:

𝑃𝑐𝑜𝑚𝑝𝑜𝑠𝑖𝑡𝑒 = 𝐸11 , 𝐸22 , 𝜈12 , 𝜈23 , 𝐺12 , 𝐺23

Amongst the input parameters, matrix elastic properties 𝐸𝑚 , 𝜈𝑚 and fiber volume
fraction 𝑉𝑓 are fairly easy to determine experimentally and are considered constants.
𝐸𝑚 , 𝜈𝑚 of 8552 epoxy (Table 9) are provided by the manufacturer [62]. Therefore, the
RVE FE simulation can be represented as a function:

38
𝑅𝑉𝐸 𝐹𝐸 𝑠𝑖𝑚𝑢𝑙𝑎𝑡𝑖𝑜𝑛
𝑃 𝑓𝑖𝑏𝑒𝑟 → 𝑃𝑐𝑜𝑚𝑝𝑜𝑠𝑖𝑡𝑒

Both the input 𝑃 𝑓𝑖𝑏𝑒𝑟 and output 𝑃𝑐𝑜𝑚𝑝𝑜𝑠𝑖𝑡𝑒 are vectors of 6 independent
dimensions. It is thus possible to implement the FEMU scheme to determine fiber
properties inversely from composite macroscopic properties. In this study, the inverse
algorithm is implemented using the Python scripting feature of Abaqus/CAE. This FEMU
procedure utilizes the Newton-Raphson algorithm with the Jacobian matrix computed
through the finite difference method. Figure 11 presents the flowchart of this FEMU
procedure. A consistent result on fiber properties is typically obtained within three
iterations. Table 9 presents elastic properties of IM7/8552 carbon/epoxy matrix which
were obtained through in-house experiments with macroscale specimens.

Figure 11. Flowchart of the FEMU process to determine reinforcing fiber elastic properties inversely from
composite elastic properties.

Convergence study (details presented in Appendix A) suggests that a 5-fiber


RVE is sufficient for elastic properties homogenization analysis. Using a 5-fiber RVE the
FEMU analysis, IM7 carbon fiber’s elastic properties result is presented in table 10 and
compared with some values from the literature. The FEMU results are very close to the
IM7 properties that Arteiro et al. utilized for their micromodel [84], while some results
(most notably 𝜈23 and 𝐺12 ) differ significantly from the properties utilized by Shah et al.
[85] and Pagano et al. [86]. In these articles, however, the authors did not provide
information on how those properties were obtained. FEMU results are rather close to
the properties found by Ballard et al. using a similar inverse characterization technique

39
with FE modeling [87]. Differences are possibly due to the difference of composite
laminate properties provided to the FEMU algorithms.

Table 9. Elastic properties of IM7/8552 composite 𝑃𝑐𝑜𝑚𝑝𝑜𝑠𝑖𝑡𝑒 for objective function. Elastic properties of
8552 epoxy matrix are considered constant.
𝐸11 𝐸22 𝜈12 𝜈23 𝐺12 𝐺23
IM7/8552 composite (GPa) (GPa) (GPa) (GPa)
157 10.3 0.20 0.07 4.65 3.01
𝐸𝑚 𝜈𝑚
8552 epoxy matrix (GPa)
4.67 0.39

Table 10. Elastic properties of IM7 carbon fiber. Result from FEMU analysis of IM7/8552 composite.

IM7 orthotropic elastic parameters


𝐸11 𝐸22 𝜈12 𝜈23 𝐺12 𝐺23
(GPa) (GPa) (GPa) (GPa)
FEMU Results 276 15.0 0.20 0.07 15.0 7.03

Arteiro et al. [84] 276 15.0 0.20 – 15.0 7.00

Shah et al. [85] 276 19.5 0.28 0.25 70.0 7.80

Pagano et al. [86] 276 27.6 0.30 0.80 138 –

Ballard et al. [87] 276 22.4 0.27 – 12.0 7.53

Fiber longitudinal compression strength: Bending Beam

Figure 12 provides an illustration of the Bending Beam experiment for


measurement of fiber critical compressive strain. The method measures the fiber
𝑜 𝑜
compressive critical stress 𝜀𝑐11 , from which the compressive strength 𝜎𝑐11 can be
deduced.

40
Figure 12. Concept of the Bending Beam method for fiber critical compression strain measurement. From
fiber compression damage extension and beam dimensions, fiber critical strain can be deduced.

For the detailed presentation of the analytical analysis and feasibility study,
please refer to Appendix B. Below are some important results from the analysis:

- Fiber critical compressive strain can be deduced from the dimensional


extension of fiber damage 𝑑 (Figure 12) through a linear relation:
𝑜
𝜀𝑐11 =𝐶×𝑑

where the coefficient 𝐶 depends on the parameters: Load, beam material


Modulus, beam geometry which are relatively easy to measure. This allows
easy and robust result interpretation.

- Result interpretation can be further supported by FEM analysis and DIC


analysis of the Bending Beam strain field.

- Bending Beam material should be PolyCarbonate thanks to its advantages:


high Yield Strains allowing a wide range of compressive strain measurement
for the bonded fiber; relatively low Modulus requiring lower Load from the
bending machine; being ductile thus safer and easier to handle during test.

- Clear Acrylic or Urethane spray coating can be used as adhesive to bond the
fiber to the Bending Beam. These adhesive agents were used and
demonstrated by Deteresa et al. [46] and Fidan [59] in their studies. The

41
strength of these adhesives was reviewed and demonstrated to be sufficient
to securely bond the fiber to the Beam.

- A reasonable example of Bending Beam dimensions was proposed and


demonstrated to be compatible to all engineering requirements: maximum
bonded fiber compressive strain; shear stress at fiber-bonding adhesive
interface; maximum Load available from the bending test machine.

- Several recommendations regarding the test setup and specimen preparation


to avoid some potential interference on the experimental results.

Berkovich indentation technique for epoxy microscale properties


Berkovich Tip assumes perfect geometry as described in ASTM - E2546 - 15.
The manufacturer of the Berkovich Tip at AMSL provided the half angle 𝛼 = 65.3.
Sanchez-Carmago et al. demonstrated significant effects of the imperfect tip geometry
on the nano-indentation results (Load-Displacement curve) [74]. To verify the FEM
model’s geometry, total included angle of FEM Tip geometry is measured to be 𝛼 + 𝜃 =
142.3. The SEM image of the newly purchased Berkovich tip also shows no sign of wear
or damage (Figure 13).

Figure 14 presents the Abaqus FEM simulation of the nano-indentation test with
the Berkovich tip on the 8552 epoxy.

The Berkovich tip is modelled as an analytical rigid, having infinite stiffness. The
infinite stiffness simplification is taken thanks to the huge difference in Modulus of the
diamond tip (order of 1000 𝐺𝑃𝑎) and the epoxy sample (order of 5 𝐺𝑃𝑎). The tip
dimension is chosen to make sure the tip-sample contact area is fully covered within the
pyramid surface.

The 8552 epoxy is modelled as an isotropic material. The epoxy is linear elastic
before yield. Non-linear behavior is modelled with Concrete Damaged Plasticity
properties. Reference properties and material parameters of the 8552 epoxy are
presented in Table 6 and Table 7.

42
Figure 13. The newly acquired Berkovich tip shows no wear or damage.

Figure 14. Berkovich test simulation: a) FE model, with high mesh density at the indented matrix area;
b) Significant plastic damage area (𝑃𝐸𝐸𝑄 > 0.01%) fully contained within high-density mesh volume.

The simulation utilizes the Static solver. The FEM epoxy sample is fixed at the
lower cylinder surface (Figure 14.a, left). During the test, the FEM Berkovich tip is
moved down to the same depth (compared to the initial sample surface) as in the

43
experiment, then the FEM tip is moved back to its initial position. The FEM tip assumes
no rotation.

The mesh and size convergence of the model is based on the Peak Load
attained. The size of the central, denser mesh area (Figure 14.a, right) is selected such
that the plastic damage volume surrounding the indented epoxy area is fully contained
within the dense-mesh volume (Figure 14.b).

Following the indentation, an indented mark is left behind (Figure 15.a). This
effect is also produced in the FE simulation (Figure 15.b). In the indentation
experiments, both the tip load-displacement curve and the indented mark geometry
provide data for the determination of sample material properties.

Figure 15. Following nano-indentation experiment: a) SEM image of indentation mark on HM63/8552
specimen; b) FEM result: vertical displacement showing indentation mark on the 8552 epoxy specimen.

Figure 16 presents results of the preliminary sensitivity studies on the effects of


different epoxy properties on the indentation Load-Displacement curve. Tension Yield
(Figure 16.a) and Fracture Energy (Figure 16.b) have almost negligible effects on the
Load-Displacement curve. Meanwhile, Compression Yield (Figure 16.c) and plastic
Dilation Angle (Figure 16.d) have very significant influence. This difference can be
explained by the fact that the indented matrix is deformed mainly in compression.

44
These preliminary results suggest that the nano-indentation technique can be
utilized to evaluate material compression properties. For the tension properties,
however, other methods must be used.

Figure 16. Effects of different epoxy properties on indentation Load-Displacement curve:


a) Tension Yield; b) Fracture Energy; c) Compression Yield; d) Plastic Dilation Angle (phi)

3.6. Conclusions

A literature review was performed on microscale material properties of composite


components, including fiber, matrix and fiber-matrix interface. Material parameters were
classified based on the availability of their values in the literature, the potential impact
on composite axial compression strength, and the availability of experimental
characterization methods. Combined with results from FEM parametric analyses, in-

45
house experimental characterization is suggested to be performed on several selected
material properties.

Amongst the selected properties, works on the micro-residual stress and the
IFSS are presented in their own respective Chapter 4 and Chapter 5. For the
characterization of micro-fiber orthotropic elastic properties, an FEMU procedure using
macroscale composite properties as input was developed and demonstrated. For the in-
situ characterization of matrix compression properties at microscale, a FE model was
developed to evaluate results from the Berkovich indentation experiment. The Bending
Beam method is recommended for the characterization of axial compressive failure in
reinforcing fibers. Analytical evaluations on the feasibility and the results analysis of the
Bending Beam method were provided.

46
CHAPTER 4:
Study microscale residual stress in CFRP using Fiber Push-out
experiment and FEMU evaluation

4.1. Introduction

Microscale residual stress in CFRP composite develops during the curing


process due to the properties mismatch between the stiff fibers and the much softer
epoxy matrix. For instance, the chemical shrinkage strain of CFRP epoxies is of the
order of several percent [88]–[90], while the chemical strain of carbon fibers is relatively
negligible [49]. During the cooldown phase, Coefficient of Thermal Expansion (CTE)
mismatch between fibers and matrix generates residual stress. The CTE of common
epoxies is around 50𝑝𝑝𝑚/𝑜 𝐶, while the CTE of common carbon fibers is typically
negative and of two orders of magnitude smaller. The manufacturing process of the
composite specimens can generate additional residual stress. The presence of
microscale residual stress can negatively affect macroscale performance of composites
[15], [37], [60], [91], [92]. Therefore, accurate representation of residual stress in
numerical micromodels might be critical for reliable computation of CFRPs properties.

Microscale residual stress is commonly introduced into the numerical model by


applying the prescribed cure process of the corresponding composite [15], [37], [60],
[91], [92]. However, key properties involving the microscale residual stress built-up are
difficult to characterize experimentally. In addition, there exists a strong coupling of
different effects including cross-linking shrinkage, thermal expansion, temperature and
degree of cure affecting epoxy mechanical properties. These coupling effects contribute
to the difficulty of a reliable representation of microscale residual stress in a
heterogenous composite [93]. The very small scale of the numerical micromodel adds to
the complexity of any experimental verification of the computed residual stress.

While there exists many experimental methods for residual stress measurement
in polymer matrix composite at ply scale, very few methods are currently available at
microscale [94]. Residual stress is commonly characterized indirectly through the
residual strain that is deformation from material’s relaxed state. Experimentally

47
observed strains can be compared with strains generated by the numerical models to
validate and deduce the residual stress field present in the specimen. At microscale,
two main challenges exist: the material removal for destructive methods in which
removed materials relaxes some residual stress leading to observable relaxing
deformation; and the measurement/observation of strain and displacement with both
destructive and non-destructive methods.

Some difficulties related to material removal might be overcome using non-


destructive methods. The Raman spectroscopy is a non-destructive method measuring
the molecule vibration modes which can be related to residual stress in a crystalline
microstructure and allows sub-micrometer resolution [95]–[98]. However, application of
this method to amorphous materials leads to poor resolution [99]–[101], making the
method unsuitable to characterize residual stress in the thermoset epoxy matrix of the
CFRPs. X-ray diffraction allows non-destructive measurement of microscale residual
stress, but a crystalline microstructure is also required [102]. Predecki et al. [103], [104]
investigated a solution for the application of X-ray diffraction method in which crystalline
microparticles are dispersed into the amorphous epoxy resin medium and X-ray
diffraction associated with the microparticles is measured. However, interaction
between different materials as well as particle geometric effects are complex and
incompletely characterized, making it difficult to interpret, quantify and validate
microscale residual strain in the epoxy medium.

The Photoelasticity method, a non-destructive method allowing visualization of


strain fields at microscale [105], [106], is worth discussing. The method was initially
limited to composites with low fiber volume fraction due to the requirement of light
transmission through the specimen. This limitation was solved by Andersson et al. [107]
by cutting and polishing a thin composite membrane specimen with plane perpendicular
to the fiber direction. The authors [107] reported good correlation between FEM-
generated strain field and the experimental optical patterns. However, when the authors
applied an extra in-plane load on the specimen, FE model predicted very high additional
strain due to this extra load, but the experimental optical patterns changed very little and
do not verify the FEM prediction. From these results, the nature of the visualized optical

48
features seems to be unclear, and any interpretation of the residual strain from
experimental optical patterns should be performed with care.

The Incremental Micro-Hole-Drilling with DIC (digital image correlation) method


[108] is potentially applicable to measure microscale residual stress in composites. The
destructive method utilizes ion beam to remove material and create a hole at
microscale. Microparticles are randomly dispersed on the specimen surface, allowing
DIC observation of relaxed residual strains, providing input for FEM analysis. The
method works with both crystalline and amorphous materials, allowing potential
application to the epoxy matrix of CFRPs. In the published literature, however, the
method has not been applied to composite materials.

This study proposes an approach to characterize residual stress in composite


matrix at the scale of reinforcing fibers by evaluating the Fiber Push-Out experiment
[109]. In this methodology, fibers surrounding a matrix-rich pocket are pushed out. After
the Push-Out of this group of fibers, local in-plane and out-of-plane matrix deformations
are visible under SEM. These deformations are attributed to the relaxation of the
residual stress stored in nearby matrix due to broken fiber-matrix interfaces. The
resulting matrix deformation is measured in the out-of-plane direction using a
nanoindenter and compared with results from the corresponding numerical model. A
FEMU procedure is proposed to evaluate and characterize the residual stress field
present in the specimen [109]. The FE geometry is created based on SEM images of
the specimen. In the FEMU algorithm, material properties related to residual stress
build-up are optimized such that the difference between FEM-predicted and
experimentally measured matrix deformation is minimized. After algorithm convergence,
the residual stress can be assessed using the resulting FE model. As some fiber-matrix
interfaces are irreversibly broken in the push-out experiment, this technique can be
classified as a destructive method.

The proposed method is applied and demonstrated on the HS40/F3G


carbon/epoxy composite that represents CFRP composites in aerospace application.
This work therefore bridges the gap between microstructure modelling to the prediction

49
of macroscale properties by providing an experimental validation of the numerically
generated residual stress at microscale.

4.2. Methodology

The technique of fiber push-out experiment with in situ SEM was originally
developed at AMSL, UTA for the measurement of interfacial shear strength (IFSS) in
CFRPs. Live monitoring of the process showed significant matrix deformations in both
in-plane and out-of-plane directions. The out-of-plane deformation increases as more
fibers are pushed out around a matrix-rich area. This is attributed to the increasing
amount of residual stress being relaxed in the matrix as the bonding with the much
stiffer fibers are broken. This phenomenon is illustrated schematically in Figure 17.
Figures 18.a and 18.b present an example of the observed matrix sink-in after fibers
push-out. The vertical position of the central matrix-rich area is measured using the
nanoindenter readings by gradual lowering the probe until the non-zero load was picked
up. The matrix sink-in was verified and measured by comparing the matrix vertical
positions before and after fibers push-out (Figure 18.c).

Figure 17. Illustration of the fiber push-out experiment. Residual stress in matrix is partially relaxed due to
the broken interface, causing matrix deformation.

50
Figure 18. Matrix sink-in measurement using the nanoindenter: a) Before fiber push-out; b) After fiber
push-out; c) Matrix sink-in is measured by comparing load-displacement data. [109]

An FEMU procedure is employed to perform an inverse characterization of the


material properties inducing the residual stress associated with the matrix sink-in
deformation [109]. Figure 19 presents the flowchart of the FEMU process. In this study,
the FEMU procedure optimizes the epoxy chemical shrinkage due to curing, to minimize
the difference between experimentally measured matrix sink-in and the corresponding
FEM-computed matrix sink-in. The choice of epoxy chemical shrinkage as the
optimizing variable is driven by several factors: chemical shrinkage is an important
contributor to the residual stress formation; experimental measurement of chemical
shrinkage is challenging and random variations might occur at microscale due to the
heterogenous nature of the composites. The FEMU procedure is written and
implemented in Abaqus/CAE using the integrated Python scripting feature.

The FEMU procedure utilizes the Gauss-Newton optimization algorithm. The


Jacobian matrix is computed through the finite difference method. By iterating the epoxy
chemical shrinkage 𝜀0𝑐ℎ𝑒𝑚 , the algorithm minimizes the difference between FEM-
computed and experimental matrix sink-in:

Δs = |𝑆𝑖𝑛𝑘𝑖𝑛𝐹𝐸𝑀 − 𝑆𝑖𝑛𝑘𝑖𝑛𝑒𝑥𝑝 |

The iteration procedure is stopped when the difference reaches the convergence
threshold of Δ𝑠 < 1 𝑛𝑚, corresponding to a relative difference of approximately 0.3% of

51
the experimental sink-in measurement (see Section 4.4). Convergence is typically
achieved within three iterations.

Figure 19. Flowchart depicting the FEMU procedure for the inverse characterization of material
properties. [109]

4.3. Finite Element Model

Setup for the Fiber Push-Out experiment

In this study, the HS40/F3G composite specimen is cut and polished into a thin
membrane perpendicular to the local fiber direction, with the thickness in range of 20 −
30 𝜇𝑚. The specimen is placed on a steel mount with an approximately 50 𝜇𝑚 wide
groove, with its outer parts taped down to the fixture. A flat-end diamond indenter with
4 𝜇𝑚 diameter is used to perform the fiber push-out and the measurement of matrix
vertical displacement. Figure 20 presents the experiment setup and the in-situ SEM
image of the experiment. FE models in this Chapter are based on fiber push-out
experiments under this setup.

52
Figure 20. a) Fiber push-out experiment setup; b) SEM image showing the free-standing push-out
configuration.

Specimen geometry and mesh generation

Microscopy images provide details on the in-plane fiber geometry and the out-of-
plane dimension of Push-Out specimens. The creation of FE geometry from specimen
geometry comprises three Steps, as depicted in Figure 22 [109].

In Step 1, the identification of fibers from the in-plane microscopy images is


accomplished through the application of the Hough Circle Transform algorithm for the
circle detection that is accessible through the OpenCV library of the Python
programming language. Fiber cross-sections are assumed to be circular. The algorithm
detects the circular fibers on an Inner Zone surrounding the Push-Out Fibers. A few
missed detections are corrected manually. To facilitate the meshing of the FE geometry,
pairs of circles that are too close or overlapping have their radii proportionally adjusted
to achieve a minimum distance equal to 0.08 times the typical HM carbon fiber radius of
𝑅𝑓 = 2.47 𝜇𝑚. Figure 22.a presents the in-plane fibers geometry obtained from Step 1.

In Step 2, the limit for an Outer Zone is manually defined according to specimen
in-plane geometry. The Outer Zone is populated with randomly generated fibers
distributed in a random pattern and having the same radius 𝑅𝑓 . All fibers, detected and
randomly generated, respect the minimum distance of 0.08 × 𝑅𝑓 . Fiber generation
utilizes an Improved Random Search Algorithm (IRSA) which is built upon the classical
Random Search Algorithm (RSA). The “jamming” problem of the original RSA limits the

53
resulting Volume Fraction to less than 50%, significantly lower than the typical fiber
Volume Fraction of 60% to 70% in aerospace composites. The IRSA solves the
jamming problem by iterating through the generated circle, pulling each circle towards
nearby existing circles by random distances, thus freeing spaces to generate new
fibers. In this Chapter, the fiber generation algorithm seeks an objective Volume fraction
of 54%.

The above-described process assumes several simplifications: specimen surface


is perfectly flat and parallel to the image plane and there is no distortion in the SEM
images. With SEM-provided scaling ratio, the projection 𝑷−𝟏
𝑰 transforms the specimen
image geometry (𝒙𝑰 , 𝒚𝑰 ) into an in-plane specimen geometry in the real coordinates
system (𝒙𝑨 , 𝒚𝑨 ):

𝒙𝑨 𝑀𝑎𝑔𝐼 0 𝒙𝑰 −𝐶𝑥
(𝒚 ) = 𝑷𝑰−𝟏 (𝒙𝑰 , 𝒚𝑰 ) = [ ] (𝒚 ) + 𝑀𝑎𝑔𝐼 ( 𝐶 )
𝑨 0 −𝑀𝑎𝑔𝐼 𝑰 𝑦

With 𝑀𝑎𝑔𝐼 being the image scale ratio. (𝐶𝑥 , 𝐶𝑦 ) is the real coordinate center that
can be arbitrarily selected on the image. Figure 21 provides an illustration of the relation
between the two coordinate systems.

Figure 21. Projection allowing transforming features between Image and Real/Abaqus coordinate
systems.

In Step 3, based on the in-plane specimen geometry obtained from previous


Steps, a 2D surface geometry is generated in Abaqus and meshed using the integrated
meshing tool (Figure 22.c). The 3D FE geometry is created through the extrusion of the
2D meshed geometry by the membrane specimen thickness that is assumed to be
uniform (Figure 22.d). The geometry shown in Figure 22 belongs to the Sample 4 Area
1 that is utilized as reference FE model in this Chapter.

54
Figure 22. Generating 3D FE geometry based on microscopy data: a) Inner Zone with blue colored
detected fibers; b) Brown colored generated fibers for the rest of the model; c) 3D FE model, top view; d)
3D FE model, oblique view. [109]

Mesh and size convergence analyses have been performed. Convergence is


assumed when significant changes in mesh or size parameters have insignificant
effects on the matrix sink-in deformation. Details on these convergence studies are
presented in Appendix D.

55
FEM Analysis Steps and matrix Sink-in measurement
Figure 23 illustrates the three processes replicated in the FE simulation: curing;
specimen grinding and polishing; and the fiber push-out experiment [109]. The
numerical study utilizes the Fully Coupled Temperature-Displacement analysis available
in Abaqus.

Figure 23. Three processes in the FEM analysis. Fiber and matrix shrinkage mismatches during curing
generate through-thickness tension stress in matrix. Significant matrix sink-in deformation on the upper
surface due to relaxed residual stress is visible after fiber push-out. [109]

Throughout the simulation, the model assumes a uniform temperature field


simplification. The curing temperature profile of the F3G epoxy is applied. To assess the
rationality of this simplification, a thermal conduction analysis was performed on a 1 𝑐𝑚-
thick ply of IM7/8552 using available exothermic curing heat measurements. Over a
distance of 40 𝜇𝑚 characteristic size of FE models in this Chapter, the resulting
maximum temperature gradient was significantly lower than 1 𝑜 𝐶. Therefore, the uniform
temperature field is a reasonable simplification.

During the curing process, the evaluated volume is assumed to be situated deep
inside a large composite material volume with negligible through-thickness deformation
due to high stiffness and insignificant fiber-direction curing expansion or shrinkage.
Accordingly, symmetry boundary condition is applied on model surfaces initially normal
to the fiber direction which corresponds to specimen upper and lower surfaces (Z-
surfaces). Supposing a lack of external constraint in the UD composite transversal
directions, model surfaces corresponding to the membrane edges (X-surfaces, Y-
surfaces) are free of constraints. To verify the validity of the application of macroscale
boundary conditions on the microscale FE model, a size convergence study of the

56
generated residual stress with models up to 16 times the reference in-plane area was
performed. The residual stress convergence in model size was successfully verified.
Through inductive inference, the convergence study demonstrated the validity of the
application of macroscale in-plane boundary conditions (BCs) to a micromodel bigger
than a certain minimum (convergence) size for the purpose of residual stress analysis.
Details on this convergence analysis are presented in Appendix E.

Through the grinding process, symmetry constraints are removed from the Z-
surfaces to model the manufacturing processes which expose the upper and lower
membrane surfaces of the studied specimen.

During the curing and the grinding processes, the fiber-matrix interface assumes
perfect bonding. To achieve this, the applied Penalty Contact and No Separation
properties prevent normal penetration and separation, and the applied Rough Friction
property prevents tangential sliding.

During the push-out process, the specimen fixture is represented by the fixed
boundary conditions on the X-surfaces and Y-surfaces. The model simplifies the push-
out process by dividing this into two separate sub-steps. First, the fiber-matrix interface
is broken, which is represented by allowing normal separation with frictionless tangential
slipping through a model change in contact properties. Subsequently, the fibers push-
out movements are represented by applying an out-of-plane displacement on their
corresponding upper surfaces.

In the FE model, a group of “probed nodes” represents the area of nanoindenter


contact for sink-in measurement, which are colored in red in Figure 24.a,b. The probed
nodes are located on a circular area with the same radius as the nanoindenter flat tip
(𝑟𝑎𝑑𝑖𝑢𝑠 = 2𝜇𝑚). The Z/vertical coordinate measurement takes the node with highest
vertical coordinate as representative of the probed area. Figure 24.c presents an
example of the Z-coordinate evolution of the probed area during the simulation. The
symmetry boundary conditions fix its Z-coordinate during curing. After grinding, the
vertical constraints are removed, relaxing some residual stress and leading to some
matrix sink-in. Fiber push-out breaks the interfaces and releases more residual strain,

57
causing more matrix sink-in corresponding to the sink-in that is experimentally
measured.

Figure 24. a) Probed nodes (red) before fiber push-out; b) After fiber push-out; c) Z-coordinates of the
probed area provides sink-in measurements. [109]

Material models and properties

The material behaviors during the curing of the epoxy matrix are described by the
Abaqus built-in material curing model. For the epoxy materials, the ratio: 𝛼 = 𝑄/𝑄𝑡𝑜𝑡 is
commonly utilized as the “Degree of Cure”, where 𝑄 is the accumulated exothermal
reaction heat and 𝑄𝑡𝑜𝑡 is the total exothermal heat from fully cured epoxy. The Kamal
curing kinetics is a popular model providing a semi-empirical description of the DOC
evolution in thermoset plastics [37], [110], [111]. This study utilizes the following form of
the Kamal equation for the epoxy curing kinetics:

Δ𝐸1 Δ𝐸2 𝑚
𝛼̇ = 𝐴1 exp (− ) (1 − α)𝑛 + 𝐴2 exp (− ) α (1 − α)𝑛 (4)
𝑅𝑇 𝑅𝑇

58
where 𝛼̇ is the rate of cure; 𝑅 is the gas constant; 𝑇 is the absolute temperature
and 𝑚, 𝑛, 𝐴1 , 𝐴2, Δ𝐸1 , Δ𝐸2 are effectively experimental fitting parameters (though several
authors assigned specific names to these parameters [37], [111]). The Abaqus built-in
curing kinetics model utilizes the Kamal equation with the expression as a sum of 𝑁
terms:

𝑁
𝐸𝑖
𝛼̇ = ∑ 𝑍𝑖 𝑒 −𝑅𝑇 (𝑏𝑖 + 𝛼 𝑚𝑖 )(𝛼𝑚𝑎𝑥 − 𝛼 )𝑛𝑖 (5)
𝑖=1

where 𝑍𝑖 , 𝐸𝑖 , 𝑚𝑖 , 𝑛𝑖 are material parameters. 𝑏𝑖 is the initial rate of cure. Setting


𝑏𝑖 = 0, we can define a linear transformation of parameters to rewrite Kamal equation
form (4) into Kamal equation form (5) with only six independent parameters.

The set of material parameters (𝑚, 𝑛, 𝐴1 , 𝐴2, Δ𝐸1 , Δ𝐸2 ) can be obtained through
the fitting with experimental DOC data. The curve-fitting algorithm seeks to minimize the
following sum:

𝑺 = ∑ 𝚫𝑖 (6)
𝑖

where 𝑀 is the total number of experimental cases provided to the algorithm. For
the experiment number 𝑖, 𝚫𝑖 measures the difference between experimental DOC
history and modeled DOC history:

2
𝚫𝒊 = ∑ (𝐷𝑂𝐶 𝑒𝑥𝑝 (𝑡𝑗 ) − 𝐷𝑂𝐶 𝑚𝑜𝑑𝑒𝑙 (𝑡𝑗 )) (7)
𝑗

where 𝐷𝑂𝐶 𝑒𝑥𝑝 (𝑡𝑗 ) and 𝐷𝑂𝐶 𝑚𝑜𝑑𝑒𝑙 (𝑡𝑗 ) are respectively experimental DOC and
modeled DOC measured at time point 𝑡𝑗 . Experimental data are processed such that:

• For each experiment, time points 𝑡𝑗 are spaced at roughly equal intervals.

• Each input experiment consists of DOC at about 40 time points. As such, each
experiment case carries roughly the same “weight” into the sum presented in
equation (6).

59
The inverse/curve fitting algorithm by minimizing the sum (6) is written in Python
and employs the function 𝑙𝑒𝑎𝑠𝑡_𝑠𝑞𝑢𝑎𝑟𝑒𝑠 from SciPy library. This optimization algorithm
is based on the Trust Region Reflective method. The algorithm stops when the change
in optimizing parameters (𝑚, 𝑛, 𝐴1 , 𝐴2, Δ𝐸1 , Δ𝐸2 ), in gradient, or the optimized cost
function becomes smaller than given tolerances, or when the maximum number of
iterations has been reached. The algorithm seeks not the best fit to a particular
experiment, but a single value for the set of material parameters (𝑚, 𝑛, 𝐴1 , 𝐴2, Δ𝐸1 , Δ𝐸2 )
which provides the best fit to all provided experimental cases.

Due to the lack of data on this new epoxy, the concerned F3G in this study
assumes the same curing kinetics as its base F4A, both having the standard curing
temperature of 121𝑜 𝐶. Schechter et al. [112] utilized the following expression for the
F4A curing kinetics:

𝐸𝑎 𝛼 𝑚 (1 − 𝛼 )𝑛
𝛼̇ = 𝐴 𝑒𝑥𝑝 (− )× (8)
𝑅𝑇 1 + 𝑒𝑥𝑝(𝐶(𝛼 − (𝛼𝐶0 + 𝛼𝐶𝑇 𝑇)))

where 𝐴, 𝐸𝑎 , 𝑚, 𝑛, 𝐶, 𝛼𝐶0 , 𝛼𝐶𝑇 are curing kinetics parameters. Schechter et al.


studied curing kinetics of F4A under four temperature configurations: four ramps:
1,2,3,5𝑜 𝐶/𝑚𝑖𝑛 and four isothermal dwells between 100𝑜 𝐶 and 130𝑜 𝐶. They fitted
equation (8) to these results and obtained the curing parameters in Table 11.

Table 11. Cure kinetic constants for F4A epoxy resin by Schechter et al. [112].
Kinetics Parameter Value Unit
𝐴 4.86 × 10−7 𝑠 −1
𝐸𝑎 70338 𝐽/𝑚𝑜𝑙
𝑚 1.34 −
𝑛 1.88 −
C 4.36 −
αC0 1.15 × 10−7 −
𝛼𝐶𝑇 1.21 × 10−3 𝐾 −1

To obtain Kamal equation’s parameters for F3G/F4A epoxy resin, an analytical


transformation from equation (8) to equation (4) might be complex. Instead, an alternate
approach is proposed that consists of 2 steps. First, using Schechter’s parameters and
kinetics equation (8), virtual experimental data is generated for the eight experimental
cases according to Schechter et al.’s description that were used for curve fitting to

60
obtain parameters presented in Table 11. Data for eight virtual experiments with the
following temperature profiles are generated: four ramps at 1, 2, 3, 5𝑜 𝐶/𝑚𝑖𝑛 and four
isotherms at 100,110,120,130𝑜 𝐶. Initial DOC is assumed to be 0.005. Next, data of these
eight virtual test cases is provided to the inverse algorithm described above to find
Kamal equation parameters.

Table 12 presents the Kamal curing kinetics parameters (Equation (4)) that were
found to fit the 8 described virtual test cases. Figure 25 shows the DOC data from the
virtual curing experiments along with their fits to the Kamal equation.

Table 12. F4A/F3G Kamal curing kinetics constants, obtained through virtual experiment curve fitting.
[109]
Kinetics Parameter Value Unit
𝑚 1.29 -
𝑛 2.70 -
𝐴1 2.32 × 1015 𝑠 −1
𝐴2 3.20 × 107 𝑠 −1
Δ𝐸1 1.50 × 105 𝐽/𝑚𝑜𝑙
Δ𝐸2 6.99 × 104 𝐽/𝑚𝑜𝑙

Figure 25. Virtual experiment DOC data and their fits to the Kamal curing kinetics: a) Cases with
temperature ramps; b) Cases with isotherms. [109]
This study applies the manufacturer-provided curing temperature profile for F3G
on the model. The curing cycle includes a ramp of approximately 1.6 𝑜 𝐶/𝑚𝑖𝑛 to 121 𝑜 𝐶,
a two hour-dwell at 121 𝑜 𝐶 , and cooling at approximately −1.6 𝑜 𝐶/𝑚𝑖𝑛 down to ambient
temperature. The curing temperature profile and the DOC evolution of F3G matrix in this
study are presented in Figure 26.

61
In the epoxy curing model, the isotropic, linear chemical shrinkage 𝜀0𝑐ℎ𝑒𝑚
corresponds to the fully cured epoxy (𝐷𝑂𝐶 = 100%). Chemical shrinkage is proportional
to the epoxy DOC: 𝜀 𝑐ℎ𝑒𝑚 = 𝜀0𝑐ℎ𝑒𝑚 × 𝐷𝑂𝐶. In the FE model, the final DOC is 96.7%.

Figure 26. Curing Temperature and Modulus evolution of the F3G epoxy material model. [109]

F3G epoxy material model assumes an isotropic, linear elastic behavior. Table
13 presents final properties of the F3G epoxy following the curing process. The modulus
is obtained through tensile tests. The CTE is provided by the manufacturer. The
Poisson ratio is estimated as typical for epoxies. Amongst material properties, only the
modulus varies during the curing cycle. The epoxy modulus is dependent on both DOC
and temperature, and is provided to the Abaqus simulation using a data table. Precise
determination of modulus dependence on DOC and temperature during the curing
process is difficult due to the strong coupling between the temperature, rate of cure with
the resulting exothermal heat and the heat transfer as the specimen is being cured [93].
In this Chapter, the modulus development during the curing cycle is shown in Figure 26.
The initial modulus is 1% of the final modulus. The FE model does not describe
explicitly the chemical cross-linking and the bonding development at the fiber-matrix

62
interface. Instead, an equivalent approach is used where the tied contact leads to zero
relative displacement at the interface, and the force transmission at the interface and its
evolution happens directly through the matrix modulus development as a result.

In this study, the F3G material model assumes no plasticity damage. This
simplification is taken due to the lack of data on the new F3G epoxy on top of the lack of
an accurate, validated model for the description of the very complex dependence of
epoxy damage properties on many parameters including local temperature [15], [62],
degree of cure [37], humidity [15], [62], all of which are also difficult to determine
experimentally for a curing specimen. Due to this simplification, only the specimens
without visible significant matrix damage are considered for the FEMU studies.

Table 13. F3G epoxy final properties after the curing process.
𝐸(𝑀𝑃𝑎) 𝜈
Elasticity
4300 0.39
Thermal 𝐶𝑇𝐸(𝑝𝑝𝑚/𝑜 𝐶)
properties 42.7
Curing 𝜀0𝑐ℎ𝑒𝑚
properties 𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒

The HS40 carbon fiber model includes orthotropic, linear elastic properties with
isotropic thermal expansion. Its material parameters are presented in Table 14. HS40
axial modulus is provided by the datasheet [113]. Other properties are estimated.

Table 14. HS40 carbon fiber material properties.


𝐸11 (𝐺𝑃𝑎) 𝐸22 (𝐺𝑃𝑎) 𝐸33 (𝐺𝑃𝑎) 𝜈12 𝜈13 𝜈23 𝐺12 (𝐺𝑃𝑎) 𝐺13 (𝐺𝑃𝑎) 𝐺23 (𝐺𝑃𝑎)
Elasticity
455 13 13 0.3 0.3 0.46 11.3 11.3 44.5
Thermal 𝐶𝑇𝐸(𝑝𝑝𝑚/𝑜 𝐶)
properties −0.7
The fully coupled thermal-stress analysis also requires material thermal
conductivity, mass density, and specific heat. However, these parameters have no
influence on the FEM results due to the application of a uniform predefined temperature.

The fibers and matrix FE volumes comprises C3D6T and C3D8T-type 3D


elements.

63
4.4. Results and discussion

FEMU results and application to predict matrix sink-in [109]

Five different FE geometries corresponding to the five experiments are analyzed


using the FEMU algorithm. Only the push-out test areas without significant matrix
damage are selected for the FEMU studies to correspond with the simplification of a
linear elastic matrix. Figure 27 presents the SEM images and corresponding FE
geometries of these five fiber push-out areas. Table 15 presents the parametric
chemical shrinkage 𝜀0𝑐ℎ𝑒𝑚 results from FEMU analyses of these five experiments. The
inversely calculated chemical shrinkage values are relatively consistent with a
Coefficient of Variation of 17.1%.

Figure 27. SEM images and corresponding FE geometries of fiber push-out areas in the FEMU analysis.

64
Table 15. FEMU results of the five analyzed Fiber Push-Out areas. [109]
Chemical Shrinkage
Test Area Experimental Sink-in(𝑛𝑚)
(𝜀0𝑐ℎ𝑒𝑚 , 100% DOC)
Sample 3 Area 4 360 4.91%
Sample 3 Area 2 328 4.10%
Sample 3 Area 1 319 3.68%
Sample 4 Area 2 260 3.24%
Sample 4 Area 1 300 3.43%
Mean 3.87%
Standard Deviation 0.66%
Coefficient of Variation 17.1%

Applying the average chemical shrinkage 𝜀0𝑐ℎ𝑒𝑚 from the FEMU results, the
matrix sink-in is predicted for other fiber push-out areas. Table 16 compares FEM sink-
in predictions with experimental measurements. Large discrepancy between numerical
predictions and experimental results might be due to the matrix damage which was not
considered in the current FE models. For instance, for Sample 5 Area 3 and Sample 2
Area 3 that show little matrix damage, experimental matrix sink-in is close to FEM-
predicted matrix sink-in. For experiments with significant visible damage (Sample 5
Area 1, Sample 2 Area 4, Sample 2 Area 5), experimental measurements are much
higher than FEM predictions. Figure 28 shows a push-out area with significant visible
matrix damage.

Table 16. Matrix sink-in at other fiber push-out areas with application of FEMU result. [109]
Visible matrix FEM-predicted Experimental Difference
Test Area
damage? Sink-in(𝑛𝑚) Sink-in(𝑛𝑚)
Sample 5 Area 1 Yes 257 530 106%
Sample 5 Area 2 No 258 144 -44%
Sample 5 Area 3 No 178 131 -26%
Sample 2 Area 3 No 299 351 17%
Sample 2 Area 4 p1 Yes 120 390 225%
Sample 2 Area 4 p2 Yes 152 340 124%
Sample 2 Area 5 Yes 62 345 456%

65
Figure 28. SEM image shows significant matrix damage on the Sample 5 Area 1.

Additional factors were identified as potentially contributing to the significant


discrepancy between numerical predictions and experimental measurements. For
example, Ghaffari et al. experimentally demonstrated significant friction effects at the
fiber-matrix interface, causing interfacial stress in the order of 25% of the interfacial
shear strength [114]. High friction stress can introduce significant residual stress to the
nearby matrix and modify the stress distribution. This effect was not considered in the
above-presented FEM results with a free-sliding interface after fiber push-out. The
effects of the surface friction on the matrix sink-in measurement were confirmed in the
parametric study presented in Section 4.5, which suggested that higher interface friction
led to higher matrix sink-in. Local variance of material characteristics, including those
due to voids and defects, is another possible factor causing errors, but it is difficult to
verify these factors experimentally. The low matrix deformation sink-in measured on
Sample 5 Area 2 might be due to the local material variance.

Studies using improved FE models that include some of these effects are being
conducted to investigate these error-inducing factors.

Assessment of residual stress in composite [109]


In the referenced Sample 4 Area 1, after specimen grinding and before the fibers
push out, residual stress is assessed and discussed.

66
Figure 29 presents the through-thickness direction normal stress (𝑆𝑧𝑧 ) in the
matrix. Regions with high 𝑆𝑧𝑧 values are concentrated near the fibers, highlighting the
interaction with stiff fibers that generates residual stress in the matrix. In the areas deep
in the matrix-rich pockets and far from the fibers, the reduced blocking effect from the
stiff fibers lowers the 𝑆𝑧𝑧 . In the current model, generated residual normal stress is very
high. For a comparison, F3G manufacturer datasheet indicates a tension strength of
74.6 𝑀𝑃𝑎.

Figure 29. Before fiber push-out: normal stress 𝑆𝑧𝑧 in the through-thickness direction is shown for the
matrix. [109]

Figure 30 presents the fiber-direction interfacial shear stress (𝑆𝑟𝑧 ) for the push-
out fibers. 𝑆𝑟𝑧 provides an interesting comparison with the IFSS measurements using
fiber push-out method due to having the same orientation. Mesh convergence was
verified to produce a consistent location of the maximum 𝑆𝑟𝑧 in the FE model. On the
reference model, this point is located at the Fiber ID 11 interface (Figure 30.a), facing
towards a matrix-rich pocket (Figure 30.b). In Figure 30.c presents 𝑆𝑟𝑧 and the relative
location along thickness for all interfacial nodes of Fiber 11. The “path of maximum 𝑆𝑟𝑧 ”
on the Fiber 11 interface crosses the point of maximum shear 𝑆𝑟𝑧 and runs along the
membrane thickness. These results are plotted alongside the experimental IFSS of the
HS40/F3G interface in both directions (𝐼𝐹𝑆𝑆 = 95𝑀𝑃𝑎, Ghaffari et al. [61]). The through-
thickness symmetry of the FE model results in a symmetric through-thickness stress
distribution. The zero 𝑆𝑟𝑧 at the mid-thickness gradually increases its magnitude towards
the upper and lower surfaces and surpasses the IFSS at some locations. The high level

67
of interfacial shear stress developed is further emphasized in Figure 31, showing the 𝑆𝑟𝑧
higher than the IFSS in gray areas. In this FE model, mesh convergence on the
interfacial shear stress distribution was verified. FE models of other push-out areas also
produce high levels of interfacial stress.

Figure 30. Reference Sample 4 Area 1 before Push-Out: a) Fiber ID 11 where the maximum 𝑆𝑟𝑧 is
located; b) Path of maximum 𝑆𝑟𝑧 on Fiber 11 interface along thickness; c) Shear stress 𝑆𝑟𝑧 for all
interfacial nodes of Fiber ID 11, plotted against experimental IFSS of HS40/F3G CFRP. [109]

Figure 31. Interfacial shear stress 𝑆𝑟𝑧 for the push-out fibers. Gray areas indicate |𝑆𝑟𝑧 | > 𝐼𝐹𝑆𝑆. [109]

68
Figure 32 visualizes the radial axial shear stress 𝑆𝑟𝑧 in the matrix. The shear
stress is presented in a cylindrical coordinate system positioned along the central axis
Fiber ID 11. Gray color indicates areas where the matrix shear stress |𝑆𝑟𝑧 | > 𝐼𝐹𝑆𝑆. The
residual shear stress is high and concentrates near membrane surfaces, strongly
corresponding to the nearby interfacial shear stress distribution. These results illustrate
the interfacial shear stress transfer into the matrix. Further from the fiber and into the
matrix-rich area, matrix shear stress becomes lower.

Figure 32. Residual shear stress 𝑆𝑟𝑧 in matrix before fiber push-out. The cylindrical coordinates system is
positioned at Fiber ID 11 axis. Gray areas indicate |𝑆𝑟𝑧 | > 𝐼𝐹𝑆𝑆. [109]

High levels of residual stress in these FEM results suggest possible premature
interface failure and matrix damage due to the curing and specimen manufacturing
processes. However, no significant matrix or interface damage is visible in SEM images
before the fiber push-out. This discrepancy might be due to the simplification of linear
matrix material model with no plasticity/damage leading to an overly stiff model and too
high residual stress generated. This simplification of linear material model also leads to
a different stress distribution and consequently a different interfacial shear stress profile.
For instance, matrix material might exhibit plastic yielding that lowers the overall stress
level for a given strain.

69
4.5. Ongoing works and future improvements

Improving FE model for residual stress analysis using fiber push-out


experiment

The FE geometry is created based on SEM images of the push-out specimens


using the same process as previously described in Section 4.3. The improved FE model
now represents the grooved support along with correct relative position of the fiber
push-out area relative to the groove (Figure 33.a). The model now generates nine
probes corresponding with the nine push-out fibers (Figure 33.a). The probes and
supports (representing the grooved support) are modeled by analytical rigids according
to experimental geometry descriptions.

Contact between the specimen and the support is modelled using the Hard and
Frictionless surface-to-surface contact. This contact implements the General Contact
algorithm which greatly helps with convergence, due to how Abaqus contact algorithm
works. Each probe interacts only with its corresponding push-out fiber. Probe-fiber
contact is similarly modeled with the Hard, Frictionless surface-to-surface contact.

The Embedded Area surrounding the push-out fibers has higher mesh density
than the rest of the model (Unrefined Area, low mesh density) to capture properly and
efficiently the evolution of interfacial damage during the fiber push-out process (Figure
33.b, c). The Embedded Area is bonded to the Unrefined Area using the Tied Contact
formulation (Figure 33.c). Throughout the simulation, relative displacement at the tied
contact interface between Embedded Area and Unrefined Area was monitored and
verified to be insignificant, confirming the good “bonding” between the high mesh
density and low mesh density volumes in the FE model.

70
Figure 33. a) Improved FE model of the fiber push-out experiment, including the mount; b) Embedded
area surrounding push-out fibers; c) Tied contact connecting embedded area with the rest of the FE
model, push-out fiber-matrix interface modeled using Cohesive Contact with Damage.

The model size and mesh convergence analyses were performed, using the
convergence of the matrix sink-in and the residual stress generated in probed matrix
volume, similar to the convergence analysis of the legacy model (Appendix D,E).

The FEM simulation replicates the three processes: Curing, Grinding and Fiber
Push-Out. At Curing, the analysis is performed using Abaqus Fully Coupled
Temperature-Displacement solver. At Grinding and Fiber Push-Out, it is assumed that
temperature changes and their effects are negligible, so the analysis is performed using
Abaqus Static solver to save computational resources. Fiber and matrix are modeled
using a mixture of C3D6T and C3D8T-type 3D elements.

71
Figure 34. Illustration of Boundary Conditions at different processes of the simulation including Curing,
Grinding and Fiber Push-Out.

Figure 34 illustrates the Boundary Conditions (BCs) at different stages of the


simulation. At curing, symmetry BCs are applied on upper and lower membrane
surfaces (Z-surfaces). F3G epoxy curing temperature profile is applied on the whole
model. At Grinding, symmetry BCs are removed from the Z-surfaces.

To block FE model rigid body movements, at Curing and Grinding, a surface


node at the center of Push-Out Fiber is blocked in UX and UY (in-plane directions), and
a surface node at the upper edge is blocked in UX. During Grinding, a gravity load in
direction of the support is applied to “attach” the specimen on the support. These BCs
were verified such that no significant unrealistic reaction force is created in the FE
model.

72
At Push-Out, the two sides of the membrane (X-surfaces) are fixed, to simulate
the specimen taping and fixture on the support. The analytical rigid probes are moved
down and up according to a pre-defined order by applying a vertical displacement.

Fiber and matrix material models of the HS40/F3G composite are similar to the
legacy model. Epoxy matrix curing model applies the chemical shrinkage of 𝜀0𝑐ℎ𝑒𝑚 =
3.43% that was found for the reference Sample 4 Area 1 by the FEMU study.

The fiber-matrix interface is now modelled by the surface-to-surface cohesive


contact with damage properties and post-damage friction behavior. Table 17 presents
the reference fiber-matrix interface parameters of cohesive contact with damage model
and post-failure friction. The Longitudinal Shear Strength IFSS was experimentally
measured by Ghaffari et al. [61]. Other properties are estimated based on the literature
reviews from Section 3. Regarding the practical implementation of the model, it is
important to note that after a fiber is pushed out, complete interfacial failure is assumed
(as verified by SEM observation) so the cohesive behaviors are removed, and the
interface model changes into Finite Sliding. Consequently, the removal/moving up of the
probes is modeled as a separate Step following the Fiber Push-Out where the probes
are moved down. The Finite Sliding formulation allows realistic fiber-matrix contact
tracking, but is incompatible with cohesive behaviors which allows only Small Sliding
where the contact areas are not updated when relative movement at interface becomes
significant. Figure 35 shows the FEM result after total interface failure with the COPEN
result showing the contact area was properly updated. Fiber-matrix interface after total
failure includes only the Hard Contact property that prevents penetration and the
Friction property to model effects of friction stresses.

Table 17. HS40/F3G fiber-matrix interface reference properties.


Interface strength Normal 𝑇𝑟𝑎𝑛𝑠𝑣𝑒𝑟𝑠𝑎𝑙 𝑆ℎ𝑒𝑎𝑟 𝐿𝑜𝑛𝑔𝑖𝑡𝑢𝑑𝑖𝑛𝑎𝑙 𝑆ℎ𝑒𝑎𝑟
(MPa) 75 95 95
Interface fracture Normal 𝑇𝑟𝑎𝑛𝑠𝑣𝑒𝑟𝑠𝑎𝑙 𝑆ℎ𝑒𝑎𝑟 𝐿𝑜𝑛𝑔𝑖𝑡𝑢𝑑𝑖𝑛𝑎𝑙 𝑆ℎ𝑒𝑎𝑟
energy (𝐽/𝑚2 ) 2 50 50
𝜇
Interface friction
0.1

73
Figure 35. After fiber push-out and total interface failure: a) Visualization of significant relative
displacement at interface; b) Finite Sliding formulation allows updating contact areas at interface.

Effects of post-failure friction stress on matrix sink-in deformation

The fiber-matrix friction stress was suspected to have significant effects on the
matrix sink-in deformation due to the influences on the local residual stress distribution.
To better understand this phenomenon, matrix sink-in on both the upper membrane
surface (measured by the probed in the current fiber push-out setup) and the lower
membrane surface (currently not measurable) are investigated in the FE model. Similar
to the legacy model, upper surface Probed Nodes correspond to the matrix area where
the real probe contacts the specimen. Lower surface Probed Nodes correspond to real
probe diameter, but on the opposite, lower membrane surface. Figure 36 presents the
positions of the Probes Nodes on upper and lower surfaces of the FE membrane model.
The vertical coordinate of upper surface Probed Area is considered as that of the node
with the maximum vertical (Z-) coordinate. Symmetrically, vertical coordinate of lower
surface Probed Area is considered as that of the node with the minimum vertical (Z-)
coordinate.

For the first part, the fiber push-out process utilizes the simplification that all
fibers are pushed out at once (Synchronous fiber push-out).

74
Figure 36. Probed Area with Probed Nodes on the upper and lower surfaces of the FE specimen model.

To evaluate effects of fiber-matrix interface friction on matrix sink-in, three (3)


simulations with residual stress are performed with three values of interface friction
coefficient:

• 𝜇 = 0.0001 : The interface is almost frictionless. The small value allows some
simulation stability by preventing fiber rigid body movement after interface failure.
Tangential interface stress was verified to be insignificant as intended.

• 𝜇 = 0.1 : The reference, estimated value for HS40/F3G.

• 𝜇 = 0.4 : The value estimated for IM7/8552 carbon/epoxy interface from


Literature [60].

FE simulation results suggest that the sink-in deformation at upper and lower
surfaces is asymmetric, which is clearly visible on the result presented in Figure 37.a.
Parametric study results presented in Figure 37.b suggest several patterns:

• Higher interface friction leads to higher matrix sink-in at the upper surface.

• Higher interface friction leads to lower matrix sink-in at the lower surface.

75
• Consequently, higher interface friction leads to more pronounced upper-lower
asymmetry of the sink-in deformation.

Figure 37. Matrix sink-in deformation after fiber push out experiment: a) Case of 𝜇 = 0.4, the asymmetric
matrix sink-in is clearly visible; b) Matrix sink-in, upper and lower membrane surfaces for different values
of interface coefficient of friction.

Figure 38 presents FE simulation results on the combined effects of the epoxy


matrix chemical shrinkage and the interface friction on the matrix sink-in deformation of
the upper surface (corresponding to the matrix sink-in measured experimentally by the
nanoindenter). A series of simulations with an alternate epoxy chemical shrinkage value
of 𝜀0𝑐ℎ𝑒𝑚 = 1.7% were performed. These results suggest several patterns:

76
• For the same interface friction coefficient 𝜇, higher chemical shrinkage 𝜀0𝑐ℎ𝑒𝑚
leads to higher matrix sink-in.

• For the same chemical shrinkage 𝜀0𝑐ℎ𝑒𝑚 , higher friction leads to higher matrix
sink-in 𝜇.

• As a result of the combined effects, different values of the couple (𝜇, 𝜀0𝑐ℎ𝑒𝑚 ) can
produce the same matrix sink-in. For example, in the results presented in Figure 38,
both the values (𝜇, 𝜀0𝑐ℎ𝑒𝑚 ) = (0, 3.4%) and (𝜇, 𝜀0𝑐ℎ𝑒𝑚 ) = (0.4, 1.7%) produce matrix
sink-in similar to the experimental value.

These results suggest that due to the coupling effect, the matrix sink-in
measurement alone is not sufficient to determine both the residual stress level and
interface friction properties.

Figure 38. Combined effects of epoxy curing shrinkage and fiber-matrix interface friction on the matrix
sink-in deformation.

Since the matrix sink-in deformation is created following the irreversible


destruction of the push-out fibers interface, history effects might have strong effects on
the precise matrix sink-in value. In this second part, the simplification of Synchronous
fiber push-out is removed. Instead, the fibers are pushed out one-by-one, following the
same order as in the corresponding experiment (Ordered fiber push-out). Figure 39.a

77
presents the fiber push-out order that was done experimentally and was replicated in
the FE simulation.

Figure 39. a) Order of the fibers to be pushed out one-by-one in the FE simulation, replicating the
corresponding experiment; b) Top surface: sink-in values from simulations with Ordered fiber push-out,
compared with the simplified Synchronous fiber push-out; c) Ordered push-out: matrix sink-in, top and
bottom membrane surfaces for different values of interface coefficient of friction.

Figure 39.b presents and compares matrix sink-in deformation between


Synchronous fiber push-out simulations and Ordered fiber push-out simulations at
different interface friction coefficients. Figure 39.c presents matrix sink-in at top and
bottom surfaces in the Ordered push-out simulation. Some important remarks from
these results are:

• In the Order fiber push-out models, the same tendencies as in the simplified
Synchronous fiber push-out models are observed: higher friction leads to higher
matrix sink-in. However these effects are significantly weaker. This can be explained

78
by the fact that interface static friction stress at an already pushed-out fiber interface
is modified when nearby fibers are pushed-out which induces local matrix
deformation.

• Notably, in the case of frictionless post-failure interface, the two models of fiber
push-out order produce the same sink-in result. This might be due to the fully linear
material models and zero interface friction leading to zero history effect.
Consequently, history effect even without interface friction might still be significant if
matrix plasticity properties are included in the model.

• Overall, history effect is very significant in the analysis of fiber push-out


experiment. The history effect might be even more pronounced when coupling with
other irreversible effects such as matrix damage.

• Effects from friction in the Ordered push-out simulation is lower than in the
Synchronous push-out simulation, probably due to the stress relaxation due to
membrane being bent and relaxed each time a fiber is pushed out. However, the
general tendency, including the asymmetric top-bottom deformation, is still
significant (Figure 39.c).

The above remarks imply that proper modeling of history effects is very important
for the accurate analysis of the fiber push-out experiment, including the analysis for the
residual stress evaluation.

Effects of matrix plasticity and damage on post-push-out matrix


deformation

Numerical models without significant interface friction and matrix plasticity effects
predict matrix sink-in at both the top and bottom surfaces of the membrane. However,
permanent deformation of the matrix that develops during the push-out experiment has
potentially significant effects on the matrix deformation in the push-out area. To verify
this, the Research Team at AMSL conducted the fiber push-out experiment on the
IM7/8552 carbon/epoxy specimen and measured the matrix in-plane deformation at
both the top and bottom surfaces of the central matrix pocket. The bottom surface was

79
found to protrude instead of sink-in (Figure 40). Due to the strong fiber-matrix interface
of the IM7/8552 composite, very extensive matrix damage can be observed (Figure 40).
These results suggest that effects from matrix plasticity and damage are significant,
causing the experimental deformation to deviate significantly from the simplified model
prediction.

Figure 40. Fiber push-out experiment on IM7/8552 by the Research Team at AMSL/UTA. Out-of-plane
deformations are measured on both the top and bottom/back surfaces.

Proposed improvements and future works

New epoxy material model with rate-form constitutive stress-strain relation

The FEM analysis presented in section 4.3 utilizes the Abaqus built-in linear
elastic model for the epoxy matrix material. The stress-strain relation depends solely on
the instantaneous matrix of stiffness:

̅ = ̿̿̿̿̿̿̿̿̿̿
𝝈 𝑪(𝜶, 𝑻): 𝜺̅

Consequently, in the legacy model, precise evolution profile of the epoxy


Modulus has insignificant effect on the resulting residual stress due to the absence of
history effect. In addition, for a given value of final accumulated strain, stress becomes
artificially too high.

80
For the next stage of model improvement, it is suggested that material
constitutive model utilizes a rate-based formulation:

̅ = ̿̿̿̿̿̿̿̿̿̿
𝒅𝝈 𝑪(𝜶, 𝑻): 𝒅𝜺̅
𝑡
̅(𝒕) = ∫
𝝈 ̿̿̿̿̿̿̿̿̿̿
𝑪(𝜶, 𝑻): 𝒅𝜺̅
𝑔𝑒𝑙 𝑝𝑜𝑖𝑛𝑡

The “hypoelastic” formulation should be able to better describe the residual


stress accumulation during the epoxy curing process. The same approach was taken by
Goncalves et al. [115]. In the Abaqus 2023 version, built-in material models of
hypoelasticity and curing are incompatible. The usage of user subroutine for customized
material models is therefore necessary.

Cave Configuration for improved Boundary Conditions determination in


push-out experiment

The Boundary Conditions representing specimen fixture in the fiber push-out


setup are simplified, assuming no extra in-plane stress due to the taping on the two
sides of the membrane. This extra in-plane stress, if indeed present, effectively acts as
additional residual stress, for which there is currently no method to determine precisely.
In addition, the over-compliance of the membrane due to the imperfect and difficult-to-
characterize contact between the specimen and the mount might introduce other issues
to a model with strong history effects. This over-compliance effect would be presented
with more details in Chapter 5.

To improve the Boundary Conditions determination, the “Cave Configuration” that


was employed by Ghaffari et al. [61] provides a strong improvement. Figure 41 provides
an illustration of the specimen in Cave Configuration where the unremoved material
above the “cave” becomes the membrane for the fiber push-out experiment. Under this
configuration, the above-mentioned problems regarding boundary conditions can be
effectively eliminated.

81
Figure 41. Cave Configuration for Fiber Push-Out Experiment.

Application of DIC for experimental observation of post fiber push-out


matrix deformation

The method of Digital Image Correlation (DIC) was applied by Winiarski et al.
[108] in conjunction with their Microdrill method to characterize residual stress in
homogenous materials. In this method, a microscale hole is “drilled” into the specimen
using ion beams, leading to surface deformation around the hole due to relaxed residual
stress in its surrounding. Before the material removal, micro particles have been
randomly dispersed and coated on the specimen surface, providing features for the DIC
processing which shows the local surface strain field due to the relaxed residual stress.

At the current stage of our experimental technique, it is difficult to apply the DIC
technique to analyze surface deformation following the fiber push-out due to the lack of
surface features. If microscale particles of appropriate size can be dispersed on
specimen surface at appropriate density and patterns, the same DIC processing
technique and algorithm can be applied to analyze in-plane strains resulting from
relaxed residual stress due to fiber push-out. The ability to characterize relaxed stress-
induced in-plane strain would provide more data for material characterization in addition
to the matrix sink-in measurement. Figure 42 presents the displacement vectors on the
matrix upper surface nodes before (Figure 42.b) and after (Figure 42.c) the push-out of
one single fiber in the FEM simulation, the difference being the relative surface
displacements between the two states. An in-plane strain field can be deduced from

82
these FEM-computed displacement results and compared with experimental DIC strain
measurements for the purpose of inverse characterization of material properties.

Figure 42. a) The push-out fiber at the center of the embedded mesh area, highlighted in red;
displacement vectors at the upper matrix surface: b) Before the fiber pushout; c) After the fiber push-out.

In addition to the obvious benefit of providing more data, the application of DIC
for the analysis of in-plane matrix strain in the Push-Out method for residual stress also
offers several advantages:

• Strains in a Cartesian coordinates system are displacement-independent. Thus,


strains can be analyzed independently from the absolute position of the specimen.
As a result, any problem with the fixture becomes much less of an issue. For
comparison, in the sink-in measurement, specimen position must be precisely fixed
during the test. The DIC method therefore allows a much more robust measurement.

• Just one fiber needs to be pushed out for the relaxed strain to be observed, as
shown in Figure 42. As a result, history effect is much less pronounced than the

83
sink-in measurement. This includes effects from existing damage from nearby,
previously pushed out fibers.

• Friction stress development can also be much better described thanks to the lack
of multiple deformation cycles from other nearby fiber push-outs which were
demonstrated to relax and modify the existing static friction stress.

• This analysis technique potentially allows a much higher number of observations


per each membrane specimen thanks to the fact that just one fiber needs to be
pushed out. For comparison, the existing sink-in evaluation technique requires that
fibers surrounding one resin-rich pocket to be pushed out, and suitable matrix-rich
areas can be difficult to find and few in number.

Meanwhile, the proposed technique also possesses several disadvantages:

• Potential difficulty in dispersing micro particles at random patterns and


appropriate density (for useful spatial resolution) on the microscale specimen
surface [116].

• Potential error and difficulty due to the distortion effects characteristic of SEM
systems [117]. The application of any macroscale DIC technique (for example, the
successful applications by Seon et al. [118], [119]) to microscale strain
measurements must take into account these interfering factors.

4.6. Conclusions

A new method was proposed to evaluate the microscale residual stress in


carbon/epoxy composites. The method is based on the fiber push-out experiment and
analysis of the post-push-out matrix deformation using Finite Element Modeling. Initial
models predict very high levels of residual stress, which are even higher than material
strengths, indicating a very significant overestimation. Several potential sources of
errors have been identified that significantly affect matrix deformation at the push-out
area, including effects from matrix plasticity and damage with experiments by the

84
AMSL/UTA Research Team, and post-failure friction at the interface by numerical
analysis, which had not been considered in the initial numerical models.

From experimental data and numerical analyses, it can be deduced that matrix
deformation at the push-out area is likely a combination of many effects that include
residual stress relaxation, matrix plasticity and damage, post-failure interface friction. As
a result, residual stress in the real specimen should be significantly lower than the initial
model prediction.

To improve the reliability of the method, several improvement strategies have


been suggested. The use of user subroutine for custom material model is
recommended, so that complex material behaviors of the epoxy matrix including the
Modulus development during curing, and plasticity and damage behaviors during push-
out, can be captured properly. The Cave Configuration proposed by the AMSL/UTA
Research Team is suggested as the preferred configuration for the residual stress
analysis with fiber push-out method, as it allows much more precise determination of
the Boundary Conditions which might have strong effects on the residual stress in the
test area. The application of the DIC technique is recommended for observation of post
push-out in-plane matrix deformation, as it provides extra data for FE analysis,
potentially more robust measurements, and potentially allowing higher number of
experimental observations per specimen.

85
CHAPTER 5:
FEM evaluation of the IFSS measurement using Fiber Push-out
method and effects from microscale properties including residual
stress

5.1. Introduction

Preliminary FEM analysis on the UD composites suggested the strong impact of


fiber-matrix interfacial strength on the fiber-direction compression strength (Section 2.3).
This relation is also reported by other authors [15], [22], [23]. Therefore, precise
measurement of the IFSS is an essential input for composite strength modeling. Fiber
Push-Out method is a popular technique for IFSS measurement. In a fiber push-out
experiment, a nanoindenter pushes the fiber out of a thin membrane specimen in the
through-thickness/fiber direction. Assuming the stress is uniformly distributed over the
fiber-matrix interface, the IFSS is commonly calculated using the formula:

𝑃𝑚𝑎𝑥
𝐼𝐹𝑆𝑆𝑎𝑝𝑝𝑎𝑟𝑒𝑛𝑡 = (9)
2𝜋 × 𝑟 × 𝑡

where 𝑃𝑚𝑎𝑥 is the maximum Load registered at the nanoindenter, 𝑟 is the fiber
radius, 𝑡 is the local membrane thickness. In this Chapter, 𝐼𝐹𝑆𝑆𝑎𝑝𝑝𝑎𝑟𝑒𝑛𝑡 refers
specifically to the IFSS value obtained through the ideal equation above. Figure 43
provides an illustration of the Fiber Push-Out experiment with important parameters.

Figure 43. Schematic illustration of Fiber Push-out experiment for IFSS measurement.

86
Meanwhile, there exist many undesired effects interfering with the IFSS
measurement using the ideal formula. Several authors pointed out the existence of non-
linear plastic matrix deformation [61], [120], non-uniform interfacial stress distribution
[121], elastic membrane bending [61], [83], all of which have the potential to introduce
significant deviations to the ideal equation for IFSS calculation (9). In addition, analysis
of the microscale residual stress in matrix demonstrated a high level of residual stress
of the same order of magnitude as the experimentally measured apparent IFSS.

Due to the importance of IFSS in composite strength, the complexity and


multitude of potentially interfering effects, it is concluded that the IFSS measurement
using fiber push-out should be investigated and verified using FE modeling.

5.2. Methodology

The present study utilizes the FEM analysis to replicate and study the fiber push-
out experiment for IFSS measurement. The FE geometry is reproduced as close as
possible to the real corresponding specimen. Sensitivity analyses on different material
parameters including interface properties, matrix properties and residual stress are
performed to study their influence on the IFSS measurement. FEM-computed Load-
Displacement curves at the nanoindenter are compared with corresponding
experimental to identify drawbacks of the FE model, and to propose improvements to
both the FE modeling and the experimental setup.

5.3. Finite Element Model

The FEM simulation models the push-out of the first fiber at the Sample 4 Area 1.
The push-out of only the first fiber is modeled to avoid the complex interferences from
history effects including matrix and interface damage due to the previous push-out of
nearby fibers. Figure 44 presents the Embedded Area surrounding the push-out fiber
and the FE contact surfaces. The features of the FE model for the IFSS study are
overall similar to the FE model of the fiber push-out experiment for residual stress

87
analysis. However, there are several differences to optimize the analysis of a single
fiber push-out for IFSS measurement.

Figure 44. a) Embedded Area surrounding the Push-Out Fiber; b) Tied Contact connecting Embedded
Area with the unrefined part.

Mesh density and model size are different from FE model for residual stress
analysis due to the emphasis on the IFSS analysis and the importance of modeling the
gradual interface failure. In this study, mesh and size convergences were verified such
that:

• A significant increase in mesh density leads to insignificant change of the Load-


Displacement curve measured at the probe.

• At reference configuration, damage variable CSDMG at fiber-matrix interface is


verified such that the transition from 0.01 (almost pristine) to 0.99 (almost fully
damaged) occurs over at least three surface elements along thickness. In most
cases, this transition occurs over much more than three surface elements. In
extreme cases of parametric study with low interface fracture energy, this condition
was also verified.

The reference material models are the same as the FE model for residual stress
analysis. However, this IFSS analysis also study effects from matrix non-linear plastic
behavior. The matrix plasticity is modeled using the Abaqus built-in Concrete Damage
Plasticity. For these sensitivity studies, the reference matrix plasticity properties are

88
presented in Table 18. Concrete Damaged Plasticity properties (Ψ, 𝜖, 𝜎𝑏0 /𝜎𝑐0 , 𝐾𝑐 , μ ) are
assumed to be the same as the 8552 epoxy. Tensile yield is provided in F3G datasheet.
Compression yield is estimated assuming the same ratio tension/compression yield
stress as 8552. Fracture Energy is typical for thermoset epoxies.

The fiber-matrix interface is modeled using the Abaqus built-in Surface-to-


Surface Contact formulation with Cohesive Contact, Damage Behavior. Reference
HS40/F3G interface damage properties implemented in the analyses are presented in
Table 19. The longitudinal IFSS is provided by the reference [61]. Other values are
estimated.

Table 18. F3G reference plasticity properties


Concrete Damaged Ψ 𝜖 𝜎𝑏0 /𝜎𝑐0 𝐾𝑐 𝜇
Plasticity 29 0.1 1.29 1 1E-5
Tensile (MPa) Compressive (MPa)
Yield Stress
74.6 108.5
Mode I (J/m2)
Fracture Energy
100

Table 19. HS40/F3G interface properties


IFNS IFSS, transversal IFSS, longitudinal
Damage Initiation (MPa) (MPa) (MPa)
75 95 95
𝐺𝐼 𝐺𝐼𝐼 𝐺𝐼𝐼𝐼
Damage Evolution (𝑁/𝑚) (𝑁/𝑚) (𝑁/𝑚)
2 50 50
Benzeggagh- 𝜂 𝐵𝐾
Kenane parameter 1.45

5.4. Results and discussion

The FEM simulations replicate the IFSS measurement using fiber push-out free-
standing setup. In the results presented in this Section, “Ideal max Load” refers to the
value:

𝐹𝑚𝑎𝑥 = 𝐼𝐹𝑆𝑆 𝑒𝑥𝑝 × 2𝜋 × 𝑟 × 𝑡


(10)

where 𝐼𝐹𝑆𝑆 𝑒𝑥𝑝 is the experimental IFSS, and used as the interfacial strength
input for the FE model. 𝑟 is the fiber radius. 𝑡 is the membrane thickness. Thus, 𝐹𝑚𝑎𝑥

89
refers to the ideal maximum Load measured by the indenter assuming uniform
interfacial shear stress at the moment of failure.

Similarly, “apparent IFSS” refers to:

𝑃𝑚𝑎𝑥
𝐼𝐹𝑆𝑆𝑎𝑝𝑝𝑎𝑟𝑒𝑛𝑡 = (11)
2𝜋 × 𝑟 × 𝑡

where 𝑃𝑚𝑎𝑥 is the maximum Load registered by the FEM nanoindenter. Thus,
𝐼𝐹𝑆𝑆𝑎𝑝𝑝𝑎𝑟𝑒𝑛𝑡 refers to the IFSS measured in the virtual fiber push-out experiment.

Strong interaction and coupling between different material properties and


effects during the fiber push-out experiment

As many parameters are estimated due to the lack of data (and in some cases,
total lack of experimental method for characterizing), a preliminary series of sensitivity
study is performed, aiming to give a qualitative assessment of the complex interaction
between different phenomena (such as matrix yield, interface damage, post-damage
friction) and their effects on the Load-Displacement Curve.

There is no Curing simulation in these studies. Thus, there is no residual stress


before Push-Out. In these studies, reference F3G matrix model includes Concrete
Damaged Plasticity behaviors. Reference interface model utilizes the properties
presented in table 19, where input IFSS is equal to the experimental IFSS.

90
Figure 45. Effects of GII: a) Load-Displacement curve at probe; b) Apparent IFSS.

Figure 45 presents effects of interface Longitudinal Shear Fracture Energy 𝐺𝐼𝐼 .


Several important remarks can be made:

• 𝐺𝐼𝐼 has a very strong effect on the Load-Displacement curve.

• 𝐺𝐼𝐼 has significant effects on Peak Load, especially at lower values of 𝐺𝐼𝐼 . At high
𝐺𝐼𝐼 , effects on Peak Load seems to “converge”.

• Literature review found no experimental method measuring longitudinal 𝐺𝐼𝐼 as of


2020.

91
Figure 46. Effects of interface coefficient of friction 𝜇: a) Load-Displacement curve at probe; b) Apparent
IFSS.

Figure 46 presents effects of interface Friction Coefficient 𝜇. Some important


remarks can be made:

• 𝜇 has strong effects on Load-Displacement curve and Peak Load.

• Even without residual stress from curing, post-damage Load is non-zero due to
the interfacial friction stress. The existence of friction stress is due to the permanent
plastic deformation of the surrounding matrix which applies non-zero normal stress
at the interface.

• At 𝜇 = 0.5, load due to friction stress is unrealistically higher than Ideal max
Load. For comparison, friction coefficient at the IM7/8552 carbon/epoxy interface is
estimated at 0.4. This demonstrates the limitations of the current FE model.

92
Figure 47. An example showing matrix plasticity damage at Peak Load and after the total interface failure.
Matrix areas with 𝑃𝐸𝐸𝑄 > 1% are colored in gray.

FE modeling has the useful advantage of allowing detailed observation of the


damage evolution of the matrix throughout the virtual experiment. Figure 47 presents an
example with two snapshots of the progress of matrix plastic deformation, at Peak Load
and at the Aftermath where fiber-matrix interface has been completely damaged. Areas
of significant plasticity strain with output 𝑃𝐸𝐸𝑄 > 1% are colored in gray. It can be
observed that at Peak Load, matrix plasticity damage can be extensive. The extent of
high-plasticity-strain area throughout the fiber push-out experiment is strongly
dependent on many factors including the matrix yield properties, interfacial friction,
interfacial strength and fracture energy.

In conclusion, preliminary analyses in this Section demonstrated qualitatively the


very complex coupling interaction between material parameters and their effects on the
IFSS measurement. Below are some of the identified interactions:

• Matrix plasticity damage evolution depends on different parameters in addition to


matrix plasticity properties themselves: interface strength IFSS, interface fracture
energy 𝐺𝐼𝐼 , interface friction properties 𝜇.

93
• Strong coupling interaction between different phenomena: matrix plasticity
damage evolution, interface damage evolution, interface friction post-failure.

• Permanent plastic matrix deformation induces non-zero interface friction stress


even without the curing/manufacturing residual stress due to the creation of non-
zero normal stress at the interface.

Sensitivity studies of effects from different parameters on the IFSS


measurement using fiber push-out method
Due to the combination of the complex parameter interaction and the lack of
precise characterization of microscopic properties, for the main sensitivity studies,
reference model utilizes a linear elastic without plasticity material model for the matrix.
Residual stress is also not included in the reference FE model – and is only modeled in
the specific studies concerning effects of the residual stress. This simplification of the
reference allows a better decoupling of effects from different phenomena for easier
analysis.

In the results presenting Load-Displacement curves, the corresponding


experimental Load-Displacement curve is also plotted for quick reference and
comparison. Reference interface model utilizes the properties presented in table 19,
where input IFSS is equal to the experimental IFSS. This reference input/experimental
IFSS is also plotted next to the charts of apparent IFSS for quick reference and
comparison.

Effects from input IFSS

Ideally, the IFSS derived from the Peak Load measurement in a fiber push-out
simulation (𝑎𝑝𝑝𝑎𝑟𝑒𝑛𝑡 𝐼𝐹𝑆𝑆) should be equal to the 𝑖𝑛𝑝𝑢𝑡 𝐼𝐹𝑆𝑆. In practice, different
unwanted phenomena affect this measurement, causing the inequality between these
two parameters. Figure 48 presents the effects of the input IFSS. Some important
remarks can be made:

• The input IFSS, as expected, has very significant effects on the nanoindenter
Load-Displacement curve.

94
• The apparent IFSS and input IFSS have a roughly linear correlation. However, in
some cases, the two values are not precisely equal (i.e., 𝑎𝑝𝑝𝑎𝑟𝑒𝑛𝑡 𝐼𝐹𝑆𝑆 ≠
𝑖𝑛𝑝𝑢𝑡 𝐼𝐹𝑆𝑆), probably due to other effects including the interface fracture energy.

Figure 48. Effects of input IFSS: a) Load-Displacement curve; b) Apparent IFSS.

To further evaluate the interference of interface (shear) fracture energy 𝐺𝐼𝐼 on


apparent IFSS, another series of IFSS input sensitivity is performed with fracture energy
much lower than the estimated reference value (100 𝐽/𝑚2 ): 𝐺𝐼𝐼 = 3.125 𝐽 /𝑚2 =
𝑟𝑒𝑓
𝐺𝐼𝐼 /16. Figure 49 presents the results of this study. Some important remarks can be
made:

• The input IFSS still has significant effects on the apparent IFSS. Higher input
IFSS leads to higher apparent IFSS.

• However, at higher input IFSS values, the difference between apparent IFSS and
input IFSS becomes much more significant. Effectively, the simple deduction of IFSS

95
from nanoindenter Peak Load becomes much less accurate. This is due to the
effects of fracture energy 𝐺𝐼𝐼 becoming much more dominating in the process of
interface failure.

Figure 49. Effects of input IFSS with very low 𝐺𝐼𝐼 . a) Load-Displacement curve; b) Apparent IFSS.

Effects from interface shear fracture energy 𝑮𝑰𝑰

Figure 50 presents effects of 𝐺𝐼𝐼 . Some important remarks can be made:

• The 𝐺𝐼𝐼 has very significant effects on the nanoindenter Load-Displacement


curve. However, the precise effects are different between the (comparatively) low
and high regimes of 𝐺𝐼𝐼 . At the low regime, the 𝐺𝐼𝐼 has very significant influence on
the Peak Load which is followed by a sudden “drop”, suggesting the fracture energy
being the dominant factor in the interface failure progress. At the high regime, the 𝐺𝐼𝐼
has insignificant effect on the Peak Load, but significant influence on the gradual
Load decrease after the Peak Load.

96
• At high 𝐺𝐼𝐼 , apparent IFSS converges towards the input IFSS. The virtual IFSS
measurement using fiber push-out is accurate and is independent of the precise
value of 𝐺𝐼𝐼 .

• The corresponding experimental curve exhibits a “sudden drop” following Peak


Load, suggesting that interface fracture in this case might be energy-dominated.

Figure 50. Effects of 𝐺𝐼𝐼 . a) Load-Displacement curve; b) Apparent IFSS.

Effects from interface coefficient of friction 𝝁 in absence of residual stress

In the absence of residual stress in the FE model, Figure 51 presents effects of


interface friction. Some important remarks can be made:

• The interface friction coefficient 𝜇 has some observable, but very small effects on
the Load-Displacement curve, including the Peak Load and post-Peak behaviors.

97
For reference, the coefficient of friction at the IM7/8552 carbon/epoxy interface is
estimated at 𝜇 = 0.4 [60].

• Following the complete interface failure, interface friction stress is almost zero
due to the absence of interface normal stress at this stage in a model with fully
elastic materials.

Figure 51. Effects of interface coefficient of friction 𝜇. a) Load-Displacement curve; b) Apparent IFSS.

𝒚 𝒚
Effects from matrix Tension Yield 𝝈𝒕 and Compression Yield 𝝈𝒄

In this study, the reference yield properties for the F3G epoxy matrix are provided
in Table 18.
𝑦
Figure 52 presents effects of matrix tension yield 𝜎𝑡 . Some important remarks
can be made:

98
𝑦
• 𝜎𝑡 has significant effects on the Load-Displacement curve. Higher matrix tension
yield leads to higher Peak Load, gradually approaching the Load-Displacement
curve of the model with linear elastic matrix where the Peak Load is equal to the
Ideal max Load.
𝑦
• 𝜎𝑡 also has significant effects on the post-failure load through at least two
effects. First, the gradual yielding of the matrix surrounding the push-out fiber, some
of which is still “attached” to the fiber due to incomplete interface failure, leads to a
gradual decrease of the indenter Load following the Peak Load. Second, permanent
plastic deformation creates some residual interface normal stress following total
interface failure, causing friction stress. The friction stress itself is dependent on the
plasticity deformation profile and the matrix yield properties.

𝑦
Figure 52. Effects of matrix tension yield 𝜎𝑡 . a) Load-Displacement curve; b) Apparent IFSS.

99
𝑦
Figure 53 presents effects of matrix compression yield 𝜎𝑐 . Overall, parametric
effects of the compression yield are similar to those of tension yield, including effects on
Peak Load and apparent IFSS, convergence towards the reference input/experimental
IFSS, and effects on the Load-Displacement evolution after the Peak Load.

Intuitively, this lowering of Peak Load can be seen as the “apparent effect” of the
Load measured by the nanoindenter. Due to the matrix yielding in the volume
neighboring the push-out fiber, this creates a “softening effect” on the reaction force
acting on the fiber and consequently captured by the FEM nanoindenter.

𝑦
Figure 53. Effects of 𝜎𝑐 . a) Load-Displacement curve; b) Apparent IFSS.

Combined effects of membrane thickness and matrix plastic damage

Another effect of the matrix plastic yielding is the occurrence of a “divergence” of


Load-Displacement curve from the initial linear slope, where the curve becomes less
100
steep which eventually leads to a lower Peak Load. In the article of Ghaffari et al. [61],
the authors reported this non-linear divergence on the experimental Load-Displacement
curves and theorized that the matrix plastic deformation being the cause of this
divergence. To verify this hypothesis, FE models with linear and non-linear matrix
material models are evaluated and compared. FE models of two different membrane
thickness (but otherwise identical geometry) are evaluated to investigate possible
coupling effects of the specimen thickness on the matrix plastic deformation.

Figure 54 presents effects of plastic damage on the Load-Displacement curves.


The non-linear divergence is observed on both models with the two thickness values.
Figure 55 presents the development of matrix plastic damage at several important
events during the fiber push-out simulation: at the onset of non-linear divergence; some
time after when the divergence becomes more pronounced; and at the Peak Load. It is
confirmed that the “divergence” corresponds with the initiation of matrix plastic damage,
and the continued softening corresponds with the development of matrix damage along
the push-out experiment.

In addition, membrane thickness also has some effects on the Load-


Displacement curves in both cases of linear elastic and non-linear matrix materials. In
the case of linear elastic matrix, the model of reference thickness provides a very good
measurement of apparent IFSS at 101% input IFSS, while the model of 2/3 reference
thickness provides a slight underestimation at 97% input IFSS. In the case of non-linear
matrix with plasticity damage, the model of reference thickness underestimates the
input IFSS by 14%, while the model of 2/3 reference thickness underestimates the input
IFSS by 19%. This difference is due to the thinner membrane being more compliant
under load, more susceptible to matrix plasticity damage, less uniform interface stress
distribution – the final combined effects being a more non-linear Load-Displacement
curve and more IFSS underestimation compared to the thicker membrane specimen.

101
Figure 54. Effects of matrix plasticity damage causing the “divergence” from the initial linear slope.

Figure 55. Areas where 𝑃𝐸𝐸𝑄 > 1% are colored in gray. Matrix PEEQ developments at several important
time points are presented, corresponding with effects of matrix plastic damage on Load-Displacement
curve.

Combined effects of residual stress and interface friction

Since the curing process involves the significant shrinkage of epoxy compared to
the reinforcing fiber, it is expected and verified in the FE models that residual stress

102
from this process would have a “squeezing” effect, or positive normal pressure at the
fiber-matrix interface. Figure 56 presents effects of a curing shrinkage of 3.45%
(estimated from the FEMU study) on the nanoindenter Load-Displacement curve,
compared to the models without residual stress. Two values of interface friction
coefficient 𝜇 = 0.1 and 𝜇 = 0.4 are evaluated.

• With no residual stress, interface friction coefficient 𝜇 has a slight effect on the
Peak Load. Higher 𝜇 leads to slightly higher Peak Load. After interface total failure,
negligible interfacial friction stress is observed.

• With residual stress present, interface friction coefficient 𝜇 has much more
significant effects on the Load-Displacement curves. With low 𝜇 = 0.1, a slight
decrease of Peak Load is observed, which can be explained by the premature
interfacial damage due to the presence of residual stress. In contrast, a high 𝜇 = 0.4
leads to a higher Peak Load due to the delaying effects from the high interface
friction stress. After interface total failure, significant interface friction stress is
observed which is strongly dependent on the coefficient of friction.

Figure 56. Combined effects of residual stress and interface friction.

Combined effects of residual stress and matrix plasticity damage

In this study, 4 FE model configurations are evaluated:

103
1- no Residual stress, Elastic matrix

2- Residual stress, Elastic matrix

3- no Residual stress, Plastic matrix

4- Residual stress, Plastic matrix

The study takes the simplification that matrix plasticity properties are unchanged
during the curing process. Figure 57 presents the combined effects of residual stress
and matrix plasticity. Some important remarks can be made:

• In the model with Plasticity matrix model, the divergence from the initial linear
slope happens earlier due to the residual stress-induced damage to the matrix.
Figure 58 shows the output variable PEEQ after Grinding, suggesting significant
matrix damage due to the curing and manufacturing process.

• Due to plasticity matrix damage, the model with residual stress produces a
slightly lower Peak Load. However, the Peak Load reduction due to the residual
stress is much smaller than the difference between the models with and without
matrix plasticity damage.

Figure 57. Combined effects of residual stress and matrix plasticity damage.

104
Figure 58. Plasticity damage in matrix after Grinding, shown at two different scales.

The yielding and nonlinear deformation of matrix also leads to modification of the
stress field in the model, and consequently the difference of the stress at the interface.
Figure 59 presents the interface damage variable CSDMG after Grinding. In the model
with Plasticity matrix, interface damage is less extensive due to the yielding and plastic
deformation of the surrounding matrix leading to lower stress.

Figure 59. Interface damage after Grinding, models with linear elastic matrix and with plasticity matrix
behavior. 𝐶𝑆𝐷𝑀𝐺 > 0.99 (near total failure): colored gray; 𝐶𝑆𝐷𝑀𝐺 < 0.01 (almost intact interface): colored
black.

Figure 60 presents the interface damage variable CSDMG at Peak Load for
different model configurations. It is shown that both residual stress and plastic matrix
damage have very strong effects on the CSDMG distribution at Peak Load. For
example, in the case of residual stress with plastic matrix damage, some interface areas

105
are fully damaged while other areas (at the mid-thickness, facing towards a matrix-rich
area) are still intact.

Figure 60. Interface damage at Peak Load. The combined effects of residual stress and plastic matrix
damage. 𝐶𝑆𝐷𝑀𝐺 > 0.99 (near total failure): colored gray; 𝐶𝑆𝐷𝑀𝐺 < 0.01 (almost intact interface): colored
black.

Effects from model size on the initial slope

As observed in the previous parametric studies, there is always a large


discrepancy between the initial slope of the FEM-predicted and the experimental Load-
Displacement curves. From the sensitivity analyses, it seems that this significant
difference cannot be explained simply through the uncertainties of the material
properties.

A study that focuses on the effect of membrane size on the initial slope of Load-
Displacement curve is performed. Figure 61 presents the Push-Out specimen under
SEM and two of the FE geometries in this study, the images being approximately to
scale to allow a quick comparison and reference. Since the experimental initial slope is
less steep than the FEM initial slope, the size study focuses on the width dimension of
the FE model which has the effect of lowering the apparent stiffness as measured by
the indenter at the push-out fiber.

Figure 62.a presents the effects of the FE model width dimension on the Load-
Displacement response of the Fiber 1. Effects of a significant width increase above the
reference are very insignificant, and cannot reproduce the very gradual initial slope in
the experiment. Figure 62.b presents the Load-Displacement data for all push-out fibers
in the same experiment, showing a very consistent overcompliance effect over different
fibers. Therefore, the gradual initial experimental slope is most likely the result of the

106
imperfect specimen-support contact before and during the push-out experiment, as was
pointed out by Ghaffari et al. [61]. Figure 63 shows the corresponding experimental
reference specimen under SEM, where a gap is visible between the specimen and the
mount. To solve this problem of overcompliance, the AMSL/UTA Research Team
proposed the Cave Configuration for fiber push-out experiment where boundary
conditions determination is much more precise as there is no “gap” between the push-
out membrane and the mount. With experimental and numerical results, The
AMSL/UTA Team demonstrated the elimination of the overcompliance effects using the
new Cave Configuration [61].

Figure 61. a) Fiber push-out specimen under SEM; b) Reference, size-converged FE geometry; c) A FE
geometry with extended width for size effects analysis. Images are approximately to scale.

107
Figure 62. a) Effects of model width on Load-Displacement curve; b) Load-Displacement data plotted for
all push-out fibers in the experiment, showing consistent initial overcompliance.

Figure 63. The reference fiber push-out sample under SEM. A gap is visible between the specimen and
the support, indicating imperfect contact.

108
5.5. Conclusions

In the FE models of the fiber push-out experiment, the addition of residual stress
into the model has some effect on the Peak Load. However, this effect is small
compared to the errors induced by the uncertainties in other material properties such as
matrix plasticity and interface fracture energy. In addition, the current residual stress is
believed to be overestimated, suggesting that the residual stress and its effects in reality
might be even weaker than those presented in this analysis.

As a result, it can be concluded that even though the residual stress has some
observable effects on the precise process of matrix and interface failure during the fiber
push-out, effects of the residual stress on the IFSS measurement itself is insignificant.

Elsewhere, from the sensitivity analyses, several factors were identified as


having significant negative impact on the IFSS measurement accuracy using the fiber
push-out method: (comparatively) low interface shear fracture energy; low matrix yield;
too small membrane thickness. Conversely, the IFSS measurement accuracy is better
when the interface shear fracture is high; matrix has high yield, and the membrane
specimen is sufficiently thick.

The free-standing setup for the fiber push-out experiment has the important
weakness of the imperfect specimen-mount contact, leading to the overcompliance
problem that is observed on the initial slope of the nanoindenter Load-Displacement
response. In turn, this overcompliance problem makes it difficult to calibrate the results
of the IFSS measurement due to the introduction of another unknown.

Ghaffari et al. demonstrated the absence of the overcompliance problem on the


results obtained from the Cave Configuration fiber push-out tests [61]. Therefore, if the
equipment for specimen preparation is available, it is suggested that the Cave
Configuration should be preferred over the free-standing configuration. The Cave
Configuration effectively eliminates one unknown variable (on the boundary conditions
at the limits of the membrane), allowing a much better FE modeling of the experiment
and much more robust result interpretation.

109
CHAPTER 6:
Conclusions

A FE RVE micromodel simulating the fiber-direction compressive failure of


composite has been developed. Literature review on the high-performance CFRP
suggested that fiber-direction compressive failure is initiated by the matrix shear
yielding/failure precipitated by local fiber misalignment, subsequently causing kink-band
formation and possible fiber failure due to extreme rotation at the kink-band boundaries.
As a result, it seems that the current FE model not including fiber damage should be
sufficient for the prediction of the fiber-direction compression strength. However, since
the experimental kink-bands often show fiber fractures, the FE models for the purpose
of reproducing realistic kink-bands should include fiber damage/failure model. In
addition, as the kink-band boundaries are typically inclined, the Periodic Boundary
Condition is unsuitable for the simulation of kink-band formations. The study of the FE
RVE identified several parameters with strong effects on composite strength which are
confirmed by previous studies in the literature. Parametric analysis and comparison
using the RVEs suggest that the low strength of HM composite is due to its low IFSS,
increasing its susceptibility to micro-instabilities and buckling under axial compression
load. Future works were proposed on the topic. High priority is assigned to the
validation of micromodel using the 4-point-bending notched specimen setup, and the
improvement of the RVE with realistic random fiber misalignment, which are the
essential prerequisites for the design of hybrid composites using numerical tools.

Due to the numerous parameters required for FE modeling input, an extensive


literature review on relevant microscale material properties were performed, providing
an overview on the availability of relevant parameters and the state of the art in their
measurement methods. A complete presentation and application results were provided
for the measurement of reinforcing fibers orthotropic elastic properties using the FEMU
technique. An FEM analysis was provided to demonstrate the capacities and limitations
of the nanoindentation technique for the measurement of microscale properties of
epoxy. A feasibility study including analytical analysis was provided for the
measurement of reinforcing fibers axial compression failure strain using the Bending

110
Beam method. It is strongly recommended that experimental work is conducted to
characterize reinforcing fiber compression failure due to the necessity of representing
fiber damage in the model, the relative simplicity of Bending Beam experimental result
analysis, and the availability of equipment required for this experimental method.

As several authors in their numerical and experimental works demonstrated the


significant impact of microscale residual stress on the macroscale composite strengths,
reliable reproduction of residual stress in micromodels is necessary for the accurate
modeling of composite failure. Meanwhile, at the time there was a lack of established
experimental method for the characterization of microscale residual stress in composite
matrix. This project proposed a new method for the assessment of microscale residual
stress in matrix. In fiber push-out experiments with in-situ SEM monitoring, significant
matrix sink-in deformation was observed after the fibers surrounding a matrix-rich
pocket have been pushed out. This sink-in, confirmed by the nanoindenter position
recordings, is associated with the relaxed residual stress in matrix after its interfaces
with the surrounding fibers have been broken. The matrix sink-in deformation is
evaluated using the FEMU technique for the inverse characterization of residual stress
field. Preliminary results suggested a very significant level of residual stress in
composite matrix that is of the same order of magnitude as some material strengths.
However, the residual stress values were apparently overestimated. Continuing works
were performed where several potential error-inducing factors were identified and
improvements were proposed, implemented, and analyzed.

The FE model suggested the strong influence of the fiber-matrix Interfacial Shear
Strength on the composite fiber-direction compression strength, which is also confirmed
by earlier authors in the literature. The IFSS is therefore of high priority in the efforts to
improve composite material performance, and the accurate measurement of the IFSS is
of critical importance. An FE model was developed based on the corresponding
experimental specimen for the analysis of factors interfering with the accuracy of the
IFSS measurement using fiber push-out method. Low interface shear fracture energy,
low matrix yield and too thin membrane were identified as factors with strong negative
impact on the IFSS measurement accuracy. Meanwhile, residual stress, despite its
effects of modifying the precise damage process during the fiber push-out, was found to
111
have a rather insignificant effect on the IFSS measurement. It was also proposed that
the Cave Configuration for fiber push-out experiment is preferable as the boundary
conditions can be determined and modeled much more accurately.

112
References
[1] D. J. Farrar, “The Design of Compression Structures for Minimum Weight,” J. R. Aeronaut. Soc.,
vol. 53, no. 467, pp. 1041–1052, Nov. 1949, doi: 10.1017/S0368393100120747.
[2] B. Smith, R. Grove, and T. Munns, Failure Analysis of Composite Structure Materials. Materials
Laboratory, Air Force Wright Aeronautical Laboratories, 1986.
[3] C. R. Schultheisz and A. M. Waas, “Compressive failure of composites, part I: Testing and
micromechanical theories,” Progress in Aerospace Sciences, vol. 32, no. 1, pp. 1–42, Jan. 1996,
doi: 10.1016/0376-0421(94)00002-3.
[4] D. R. Tenney, J. G. Davis Jr, N. J. Johnston, R. B. Pipes, and J. F. McGuire, “Structural Framework
for flight: NASA’s role in development of advanced composite materials for aircraft and space
structures,” 2011.
[5] R. Butler, A. T. Rhead, W. Liu, and N. Kontis, “Compressive strength of delaminated aerospace
composites,” Phil. Trans. R. Soc. A., vol. 370, no. 1965, pp. 1759–1779, Apr. 2012, doi:
10.1098/rsta.2011.0339.
[6] R. Sturm, Y. Klett, Ch. Kindervater, and H. Voggenreiter, “Failure of CFRP airframe sandwich
panels under crash-relevant loading conditions,” Composite Structures, vol. 112, pp. 11–21, Jun.
2014, doi: 10.1016/j.compstruct.2014.02.001.
[7] N. Zimmermann and P. H. Wang, “A review of failure modes and fracture analysis of aircraft
composite materials,” Engineering Failure Analysis, vol. 115, p. 104692, Sep. 2020, doi:
10.1016/j.engfailanal.2020.104692.
[8] N. F. Dow, B. W. Rosen, L. S. Shu, and C. H. Zweben, “Design criteria and concepts for fibrous
composite structures Final report,” 1967. [Online]. Available:
https://ntrs.nasa.gov/api/citations/19680004493/downloads/19680004493.pdf
[9] M. R. Piggott and B. Harris, “Compression strength of hybrid fibre-reinforced plastics,” J Mater Sci,
vol. 16, no. 3, pp. 687–693, Mar. 1981, doi: 10.1007/BF02402786.
[10] H. T. Hahn and J. G. Williams, “Compression Failure Mechanisms in Unidirectional Composites,”
Philadelphia, 1986.
[11] P. S. Steif, “A model for kinking in fiber composites—I. Fiber breakage via micro-buckling,”
International Journal of Solids and Structures, vol. 26, no. 5–6, pp. 549–561, 1990, doi:
10.1016/0020-7683(90)90028-T.
[12] M. Wisnom, “On the high compressive strains achieved in bending tests on unidirectional carbon-
fibre/epoxy,” Composites Science and Technology, vol. 43, no. 3, pp. 229–235, 1992, doi:
10.1016/0266-3538(92)90093-I.
[13] R. A. Schapery, “Prediction of compressive strength and kink bands in composites using a work
potential,” International Journal of Solids and Structures, vol. 32, no. 6–7, pp. 739–765, Mar. 1995,
doi: 10.1016/0020-7683(94)00158-S.
[14] R. Gutkin, S. T. Pinho, P. Robinson, and P. T. Curtis, “A finite fracture mechanics formulation to
predict fibre kinking and splitting in CFRP under combined longitudinal compression and in-plane
shear,” Mechanics of Materials, vol. 43, no. 11, pp. 730–739, Nov. 2011, doi:
10.1016/j.mechmat.2011.08.002.
[15] F. Naya, M. Herraez, C. S. Lopes, C. Gonzalez, S. Van der Veen, and F. Pons, “Computational
micromechanics of fiber kinking in unidirectional FRP under different environmental conditions,”
Composites Science and Technology, vol. 144, pp. 26–35, 2017, doi:
https://doi.org/10.1016/j.compscitech.2017.03.014.

113
[16] S. M. Koppisetty, S. B. Cheryala, and C. S. Yerramalli, “The effect of fiber distribution on the
compressive strength of hybrid polymer composites,” Journal of Reinforced Plastics and
Composites, vol. 38, no. 2, pp. 74–87, Jan. 2019, doi: 10.1177/0731684418804346.
[17] A. Makeev, S. Ghaffari, and G. Seon, “Improving compressive strength of high modulus carbon-fiber
reinforced polymeric composites through fiber hybridization,” International Journal of Engineering
Science, vol. 142, pp. 145–157, Sep. 2019, doi: 10.1016/j.ijengsci.2019.06.004.
[18] S. Ghaffari, A. Makeev, G. Seon, D. P. Cole, D. J. Magagnosc, and S. Bhowmick, “Understanding
compressive strength improvement of high modulus carbon-fiber reinforced polymeric composites
through fiber-matrix interface characterization,” Materials & Design, vol. 193, p. 108798, Aug. 2020,
doi: 10.1016/j.matdes.2020.108798.
[19] HexTow, “HexTow Carbon Fiber Selector Guide.” [Online]. Available:
https://www.hexcel.com/user_area/content_media/raw/HexTowSelectorGuide.pdf
[20] E. J. Barbero and J. Tomblin, “A damage mechanics model for compression strength of
composites,” International Journal of Solids and Structures, vol. 33, no. 29, pp. 4379–4393, Dec.
1996, doi: 10.1016/0020-7683(95)00236-7.
[21] B. Budiansky and N. A. Fleck, “Compressive failure of fibre composites,” Journal of the Mechanics
and Physics of Solids, vol. 41, no. 1, pp. 183–211, Jan. 1993, doi: 10.1016/0022-5096(93)90068-Q.
[22] M. Herraez, A. C. Bergan, C. Gonzalez, and C. S. Lopes, “Modeling fiber kinking at the microscale
and mesoscale,” 2018.
[23] S. Ghaffari, A. Makeev, D. Kuksenko, and G. Seon, “Understanding High-Modulus CFRP
Compressive Strength Improvement,” presented at the American Society for Composites, 34th
Technical Conference, 2019. doi: 10.12783/asc34/31352.
[24] C. S. Yerramalli and A. M. W. Waas, “Compressive Behavior of Hybrid Composites,” Norfolk,
Virginia, 2003.
[25] Sudarisman, I. J. Davies, and H. Hamada, “Compressive failure of unidirectional hybrid fibre-
reinforced epoxy composites containing carbon and silicon carbide fibres,” Composites Part A:
Applied Science and Manufacturing, vol. 38, no. 3, pp. 1070–1074, Mar. 2007, doi:
10.1016/j.compositesa.2006.10.004.
[26] J. Aveston and J. M. Sillwood, “Synergistic fibre strengthening in hybrid composites,” J Mater Sci,
vol. 11, no. 10, pp. 1877–1883, Oct. 1976, doi: 10.1007/BF00708266.
[27] R. Gutkin, S. T. Pinho, P. Robinson, and P. T. Curtis, “Micro-mechanical modelling of shear-driven
fibre compressive failure and of fibre kinking for failure envelope generation in CFRP laminates,”
Composites Science and Technology, vol. 70, no. 8, pp. 1214–1222, Aug. 2010, doi:
10.1016/j.compscitech.2010.03.009.
[28] J. Patel and P. Peralta, “Mechanisms for Kink Band Evolution in Polymer Matrix Composites: A
Digital Image Correlation and Finite Element Study,” in Volume 9: Mechanics of Solids, Structures
and Fluids; NDE, Diagnosis, and Prognosis, Phoenix, Arizona, USA: American Society of
Mechanical Engineers, Nov. 2016, p. V009T12A055. doi: 10.1115/IMECE2016-67482.
[29] I. M. Daniel, H.-M. Hsiao, and S.-C. Wooh, “Failure mechanisms in thick composites under
compressive loading,” Composites Part B: Engineering, vol. 27, no. 6, pp. 543–552, Jan. 1996, doi:
10.1016/1359-8368(95)00010-0.
[30] Y. Wang, Y. Chai, C. Soutis, and P. J. Withers, “Evolution of kink bands in a notched unidirectional
carbon fibre-epoxy composite under four-point bending,” Composites Science and Technology, vol.
172, pp. 143–152, Mar. 2019, doi: 10.1016/j.compscitech.2019.01.014.
[31] S. Ghaffari, G. Seon, A. Makeev, E. Iarve, and D. Mollenhauer, “Microstructural Methods for
Developing High-Performance Composite Materials,” Orlando, FL, 2020. doi:
https://doi.org/10.2514/6.2020-2109.

114
[32] Y. Wang, C. Soutis, and P. Withers, “X-ray microtomographic imaging of kink bands in carbon fibre-
epoxy composites,” Jun. 2014.
[33] C. González and J. LLorca, “Mechanical behavior of unidirectional fiber-reinforced polymers under
transverse compression: Microscopic mechanisms and modeling,” Composites Science and
Technology, vol. 67, no. 13, pp. 2795–2806, Oct. 2007, doi: 10.1016/j.compscitech.2007.02.001.
[34] D. Garoz Gomez, F. Gilabert, R. Sevenois, S. Spronk, A. Rezaei, and W. Van Paepegem,
“DEFINITION OF PERIODIC BOUNDARY CONDITIONS IN EXPLICIT DYNAMIC SIMULATIONS
OF MICRO-OR MESO-SCALE UNIT CELLS WITH CONFORMAL AND NON-CONFORMAL
MESHES,” Jun. 2016.
[35] S. W. Yurgartis, “Measurement of small angle fiber misalignments in continuous fiber composites,”
Composites Science and Technology, vol. 30, no. 4, pp. 279–293, Jan. 1987, doi: 10.1016/0266-
3538(87)90016-9.
[36] M. P. F. Sutcliffe, S. L. Lemanski, and A. E. Scott, “Measurement of fibre waviness in industrial
composite components,” Composites Science and Technology, vol. 72, no. 16, pp. 2016–2023,
Nov. 2012, doi: 10.1016/j.compscitech.2012.09.001.
[37] M. Maiaru, R. J. D’Mello, and A. M. Waas, “Characterization of intralaminar strengths of virtually
cured polymer matrix composites,” Composites Part B, vol. 149, pp. 285–295, 2018, doi:
https://doi.org/10.1016/j.compositesb.2018.02.018.
[38] Z. Shan and A. M. Gokhale, “Representative volume element for non-uniform micro-structure,”
Computational Materials Science, vol. 24, no. 3, pp. 361–379, Jun. 2002, doi: 10.1016/S0927-
0256(01)00257-9.
[39] D. Trias, J. Costa, A. Turon, and J. Hurtado, “Determination of the critical size of a statistical
representative volume element (SRVE) for carbon reinforced polymers☆,” Acta Materialia, vol. 54,
no. 13, pp. 3471–3484, Aug. 2006, doi: 10.1016/j.actamat.2006.03.042.
[40] “IM7 Datasheet.” Accessed: Jun. 28, 2023. [Online]. Available:
https://www.hexcel.com/user_area/content_media/raw/IM7_HexTow_DataSheet.pdf
[41] “T300 Datasheet.” Accessed: Jun. 28, 2023. [Online]. Available:
https://www.fibermaxcomposites.com/shop/datasheets/T300.pdf
[42] “T800H Datasheet.” Accessed: Jun. 28, 2023. [Online]. Available:
https://www.fibermaxcomposites.com/shop/datasheets/CarbonT800MDS.pdf
[43] H. Miyagawa, C. Sato, T. Mase, E. Drown, L. T. Drzal, and K. Ikegami, “Transverse elastic modulus
of carbon fibers measured by Raman spectroscopy,” Materials Science and Engineering: A, vol.
412, no. 1–2, pp. 88–92, Dec. 2005, doi: 10.1016/j.msea.2005.08.037.
[44] T. S. Gross, N. Timoshchuk, I. I. Tsukrov, R. Piat, and B. Reznik, “On the ability of nanoindentation
to measure anisotropic elastic constants of pyrolytic carbon,” Z. angew. Math. Mech., vol. 93, no. 5,
pp. 301–312, May 2013, doi: 10.1002/zamm.201100128.
[45] D. Mounier, C. Poilâne, C. Bûcher, and P. Picart, “Evaluation of transverse elastic properties of
fibers used in composite materials by laser resonant ultrasound spectroscopy,” 2012. [Online].
Available: https://api.semanticscholar.org/CorpusID:136502997
[46] S. J. Deteresa, S. R. Allen, R. J. Farris, and R. S. Porter, “Compressive and torsional behaviour of
Kevlar 49 fibre,” J Mater Sci, vol. 19, no. 1, pp. 57–72, Jan. 1984, doi: 10.1007/BF02403111.
[47] C.-L. Tsai and I. M. Daniel, “Determination of shear modulus of single fibers,” Exp Mech, vol. 39, no.
4, pp. 284–286, Dec. 1999, doi: 10.1007/BF02329806.
[48] A. A. Gusev, P. J. Hine, and I. M. Ward, “Fiber packing and elastic properties of a transversely
random unidirectional glass/epoxy composite,” Composites Science and Technology, vol. 60, no. 4,
pp. 535–541, Mar. 2000, doi: 10.1016/S0266-3538(99)00152-9.

115
[49] X. Huang, “Fabrication and Properties of Carbon Fibers,” Materials, vol. 2, no. 4, pp. 2369–2403,
Dec. 2009, doi: 10.3390/ma2042369.
[50] D. Sinclair, “A Bending Method for Measurement of the Tensile Strength and Young’s Modulus of
Glass Fibers,” Journal of Applied Physics, vol. 21, no. 5, pp. 380–386, May 1950, doi:
10.1063/1.1699670.
[51] M. Furuyama, M. Higuchi, K. Kubomura, H. Sunago, H. Jiang, and S. Kumar, “Compressive
properties of single-filament carbon fibres,” Journal of Materials Science, vol. 28, no. 6, pp. 1611–
1616, Mar. 1993, doi: 10.1007/BF00363356.
[52] N. Oya and D. J. Johnson, “Longitudinal compressive behaviour and microstructure of PAN-based
carbon fibres,” Carbon, vol. 39, no. 5, pp. 635–645, Apr. 2001, doi: 10.1016/S0008-6223(00)00147-
0.
[53] M. Shioya and M. Nakatani, “Compressive strengths of single carbon fibres and composite strands,”
Composites Science and Technology, vol. 60, no. 2, pp. 219–229, Feb. 2000, doi: 10.1016/S0266-
3538(99)00123-2.
[54] M. Ueda, W. Saito, R. Imahori, D. Kanazawa, and T.-K. Jeong, “Longitudinal direct compression
test of a single carbon fiber in a scanning electron microscope,” Composites Part A: Applied
Science and Manufacturing, vol. 67, pp. 96–101, Dec. 2014, doi:
10.1016/j.compositesa.2014.08.021.
[55] A. Andres Leal, J. M. Deitzel, and J. W. Gillespie, “Assessment of compressive properties of high
performance organic fibers,” Composites Science and Technology, vol. 67, no. 13, pp. 2786–2794,
Oct. 2007, doi: 10.1016/j.compscitech.2007.02.003.
[56] X. Ji et al., “Mechanical and Interfacial Properties Characterisation of Single Carbon Fibres for
Composite Applications,” Exp Mech, vol. 55, no. 6, pp. 1057–1065, Jul. 2015, doi: 10.1007/s11340-
015-0007-3.
[57] J. A. Newell and J. M. Gustafson, “An Improved Interpretation of Recoil Compressive Failure Data
for High-Performance Polymers,” High Performance Polymers, vol. 13, no. 4, pp. 251–257, Dec.
2001, doi: 10.1088/0954-0083/13/4/303.
[58] H. M. Hawthorne and E. Teghtsoonian, “Axial compression fracture in carbon fibres,” J Mater Sci,
vol. 10, no. 1, pp. 41–51, Jan. 1975, doi: 10.1007/BF00541030.
[59] S. Fidan, “Experimentation and Analysis of Compression Test methods for Single Filament High
Performance Fibres,” Air Force Institute of Technology, Ohio, 1989.
[60] F. Naya, C. González, C. S. Lopes, S. Van Der Veen, and F. Pons, “Computational micromechanics
of the transverse and shear behavior of unidirectional fiber reinforced polymers including
environmental effects,” Composites Part A: Applied Science and Manufacturing, vol. 92, pp. 146–
157, Jan. 2017, doi: 10.1016/j.compositesa.2016.06.018.
[61] S. Ghaffari, G. Seon, and A. Makeev, “In-situ SEM based method for assessing fiber-matrix
interface shear strength in CFRPs,” Materials and Design, vol. 197, 2021, doi:
https://doi.org/10.1016/j.matdes.2020.109242.
[62] HEXCEL, “HexPly 8552 datasheet.” Accessed: Mar. 03, 2023. [Online]. Available:
https://www.hexcel.com/user_area/content_media/raw/HexPly_8552_us_DataSheet.pdf
[63] I. Oral, H. Guzel, and G. Ahmetli, “Determining the mechanical properties of epoxy resin (DGEBA)
composites by ultrasonic velocity measurement,” J. Appl. Polym. Sci., vol. 127, no. 3, pp. 1667–
1675, Feb. 2013, doi: 10.1002/app.37534.
[64] A. Smith, S. J. Wilkinson, and W. N. Reynolds, “The elastic constants of some epoxy resins,” J
Mater Sci, vol. 9, no. 4, pp. 547–550, Apr. 1974, doi: 10.1007/BF02387527.
[65] F. Naya, “Prediction of mechanical properties of unidirectional FRP plies at different environmental
conditions by means of computational micromechanics,” Universidad Politénica de Madrid, Madrid,
2017.

116
[66] O. Verschatse, L. Daelemans, W. Van Paepegem, and K. De Clerck, “In-Situ Observations of
Microscale Ductility in a Quasi-Brittle Bulk Scale Epoxy,” Polymers, vol. 12, no. 11, p. 2581, Nov.
2020, doi: 10.3390/polym12112581.
[67] T. Jankowiak and T. Lodygowski, “Identification of Parameters of Concrete Damage Plasticity
Constitutive Model,” Foundations of Civil and Environmental Engineering, 2005.
[68] M. Labibzadeh, M. Zakeri, and A. Adel Shoaib, “A new method for CDP input parameter
identification of the ABAQUS software guaranteeing uniqueness and precision,” IJSI, vol. 8, no. 2,
pp. 264–284, Apr. 2017, doi: 10.1108/IJSI-03-2016-0010.
[69] “Abaqus Documentation, Concrete Damaged Plasticity.” [Online]. Available:
https://classes.engineering.wustl.edu/2009/spring/mase5513/abaqus/docs/v6.6/books/usb/default.ht
m?startat=pt05ch18s05abm36.html
[70] M. Szczecina and A. Winnicki, “Calibration of the CDP model parameters in Abaqus,” 2015.
[Online]. Available: https://api.semanticscholar.org/CorpusID:138627326
[71] T. Hobbiebrunken, B. Fiedler, M. Hojo, and M. Tanaka, “Experimental determination of the true
epoxy resin strength using micro-scaled specimens,” Composites Part A: Applied Science and
Manufacturing, vol. 38, no. 3, pp. 814–818, Mar. 2007, doi: 10.1016/j.compositesa.2006.08.006.
[72] E. M. Odom and D. F. Adams, “Specimen size effect during tensile testing of an unreinforced
polymer,” J Mater Sci, vol. 27, no. 7, pp. 1767–1771, 1992, doi: 10.1007/BF01107202.
[73] J. Misumi, R. Ganesh, S. Sockalingam, and J. W. Gillespie, “Experimental characterization of
tensile properties of epoxy resin by using micro-fiber specimens,” Journal of Reinforced Plastics and
Composites, vol. 35, no. 24, pp. 1792–1801, Dec. 2016, doi: 10.1177/0731684416669248.
[74] C.-M. Sanchez-Camargo, A. Hor, and C. Mabru, “A robust inverse analysis method for elastoplastic
behavior identification using the true geometry modeling of Berkovich indenter,” International
Journal of Mechanical Sciences, vol. 171, p. 105370, Apr. 2020, doi:
10.1016/j.ijmecsci.2019.105370.
[75] M. Herraez, “Computational Micromechanics Models for Damage and Fracture of Fiber-Reinforced
Polymers - Tesis Doctoral,” UNIVERSIDAD POLITECNICA DE MADRID, 2018.
[76] G. P. Tandon, R. Y. Kim, and V. T. Bechel, “Interfacial Strength and Toughness Characterization
Using a Novel Test Specimen,” in Recent Advances in Experimental Mechanics, E. E. Gdoutos,
Ed., Dordrecht: Kluwer Academic Publishers, 2004, pp. 163–174. doi: 10.1007/0-306-48410-2_16.
[77] S. Ogihara and J. Koyanagi, “Investigation of combined stress state failure criterion for glass
fiber/epoxy interface by the cruciform specimen method,” Composites Science and Technology, vol.
70, no. 1, pp. 143–150, Jan. 2010, doi: 10.1016/j.compscitech.2009.10.002.
[78] J. Koyanagi, P. D. Shah, S. Kimura, S. K. Ha, and H. Kawada, “Mixed-Mode Interfacial Debonding
Simulation in Single-Fiber Composite under a Transverse Load,” JMMP, vol. 3, no. 5, pp. 796–806,
2009, doi: 10.1299/jmmp.3.796.
[79] T. J. Vaughan, “Micromechanical modelling of damage and failure in fibre reinforced composites
under loading in the transverse plane,” 2011. [Online]. Available:
https://api.semanticscholar.org/CorpusID:138225952
[80] J. Varna, L. A. Berglund, and M. L. Ericson, “Transverse single-fibre test for interfacial debonding in
composites: 2. Modelling,” Composites Part A: Applied Science and Manufacturing, vol. 28, no. 4,
pp. 317–326, Jan. 1997, doi: 10.1016/S1359-835X(96)00125-X.
[81] C. S. Lopes, P. P. Camanho, Z. Gürdal, P. Maimí, and E. V. González, “Low-velocity impact
damage on dispersed stacking sequence laminates. Part II: Numerical simulations,” Composites
Science and Technology, vol. 69, no. 7–8, pp. 937–947, Jun. 2009, doi:
10.1016/j.compscitech.2009.02.015.

117
[82] V. Cech, E. Palesch, and J. Lukes, “The glass fiber–polymer matrix interface/interphase
characterized by nanoscale imaging techniques,” Composites Science and Technology, vol. 83, pp.
22–26, Jun. 2013, doi: 10.1016/j.compscitech.2013.04.014.
[83] C. Medina M, J. M. Molina-Aldareguía, C. González, M. F. Melendrez, P. Flores, and J. LLorca,
“Comparison of push-in and push-out tests for measuring interfacial shear strength in nano-
reinforced composite materials,” Journal of Composite Materials, vol. 50, no. 12, pp. 1651–1659,
May 2016, doi: 10.1177/0021998315595115.
[84] A. Arteiro, G. Catalanotti, A. R. Melro, P. Linde, and P. P. Camanho, “Micro-mechanical analysis of
the in situ effect in polymer composite laminates,” Composite Structures, vol. 116, pp. 827–840,
Sep. 2014, doi: 10.1016/j.compstruct.2014.06.014.
[85] S. P. Shah and M. Maiarù, “Effect of Manufacturing on the Transverse Response of Polymer Matrix
Composites,” Polymers, vol. 13, no. 15, p. 2491, Jul. 2021, doi: 10.3390/polym13152491.
[86] N. J. Pagano, G. A. Schoeppner, R. Kim, and F. L. Abrams, “Steady-state cracking and edge effects
in thermo-mechanical transverse cracking of cross-ply laminates,” Composites Science and
Technology, vol. 58, no. 11, pp. 1811–1825, Nov. 1998, doi: 10.1016/S0266-3538(98)00047-5.
[87] M. K. Ballard, W. R. McLendon, and J. D. Whitcomb, “The influence of microstructure randomness
on prediction of fiber properties in composites,” Journal of Composite Materials, vol. 48, no. 29, pp.
3605–3620, Dec. 2014, doi: 10.1177/0021998313511654.
[88] D. U. Shah and P. J. Schubel, “Evaluation of cure shrinkage measurement techniques for
thermosetting resins,” Polymer Testing, vol. 29, pp. 629–639, 2010, doi:
https://doi.org/10.1016/j.polymertesting.2010.05.001.
[89] L. Khoun and P. Hubert, “Cure Shrinkage Characterization of an Epoxy Resin System by Two in
Situ Measurement Methods,” Polymer Composites, vol. 31, pp. 1603–1610, 2010, doi:
https://doi.org/10.1002/pc.20949.
[90] Y. Nawab, S. Shahid, N. Boyard, and F. Jacquemin, “Review: chemical shrinkage characterization
techniques for thermoset resins and associated composites,” Journal of Materials Science, vol. 48,
pp. 5387–5409, 2013, doi: https://doi.org/10.1007/s10853-013-7333-6.
[91] L. G. Zhao, N. A. Warrior, and A. C. Long, “A micromechanical study of residual stress and its effect
on transverse failure in polymer–matrix composites,” International Journal of Solids and Structures,
vol. 43, pp. 5449–5467, 2006, doi: https://doi.org/10.1016/j.ijsolstr.2005.08.012.
[92] S. L. Agius, M. Joosten, B. Trippit, C. H. Wang, and T. Hilditch, “Rapidly cured epoxy/anhydride
composites: Effect of residual stress on laminate shear strength,” Composites: Part A, vol. 90, pp.
125–136, 2016, doi: https://doi.org/10.1016/j.compositesa.2016.06.013.
[93] N. Rabearison, Ch. Jochum, and J. C. Grandidier, “A FEM coupling model for properties prediction
during the curing of an epoxy matrix,” Computational Materials Science, vol. 45, no. 3, pp. 715–724,
2009, doi: https://doi.org/10.1016/j.commatsci.2008.11.007.
[94] B. Seers, R. Tomlinson, and P. Fairclough, “Residual stress in fiber reinforced thermosetting
composites: A review of measurement techniques,” Polymer Composites, vol. 42, pp. 1631–1647,
2021, doi: https://doi.org/10.1002/pc.25934.
[95] C. Kao and R. J. Young, “Assessment of interface damage during the deformation of carbon
nanotube composites,” J Mater Sci, vol. 45, no. 6, pp. 1425–1431, Mar. 2010, doi: 10.1007/s10853-
009-3947-0.
[96] P. Jannotti, G. Subhash, J. Zheng, and V. Halls, “Measurement of microscale residual stresses in
multi-phase ceramic composites using Raman spectroscopy,” Acta Materialia, vol. 129, pp. 482–
491, May 2017, doi: 10.1016/j.actamat.2017.03.015.
[97] K. Kollins, C. Przybyla, and M. S. Amer, “Residual stress measurements in melt infiltrated SiC/SiC
ceramic matrix composites using Raman spectroscopy,” Journal of the European Ceramic Society,
vol. 38, no. 7, pp. 2784–2791, Jul. 2018, doi: 10.1016/j.jeurceramsoc.2018.02.013.

118
[98] J. Nance et al., “Measurement of Residual Stress in Silicon Carbide Fibers of Tubular Composites
Using Raman Spectroscopy,” Acta Materialia, vol. 217, p. 117164, Sep. 2021, doi:
10.1016/j.actamat.2021.117164.
[99] D. R. Tallant, T. A. Friedmann, N. A. Missert, M. P. Siegal, and J. P. Sullivan, “Raman Spectroscopy
of Amorphous Carbon,” 1997, pp. 37–48.
[100] A. Boukenter, E. Duval, and H. M. Rosenberg, “Raman scattering in amorphous and crystalline
materials: a study of epoxy resin and DGEBA,” Journal of Physics C: Solid State Physics, vol. 21,
pp. 541–547, 1998, doi: 10.1088/0022-3719/21/15/011.
[101] D. Tuschel, “Why are the Raman spectra of crystalline and amorphous solids different?,”
Spectroscopy, vol. 32, pp. 26–33, 2017.
[102] M. Pick, R. Lovell, and A. H. Windle, “Detection of elastic strain in an amorphous polymer by X-ray
scattering,” Nature, vol. 281, pp. 658–659, 1979, doi: https://doi.org/10.1038/281658a0.
[103] P. Predecki and CS. Barrett, “Stress Measurement in Graphite/Epoxy Composites By X-Ray
Diffraction from Fillers,” Journal of Composite Materials, vol. 13(1), pp. 61–71, 1979, doi:
https://doi.org/10.1177/002199837901300105.
[104] P. Predecki and CS. Barrett, “Detection of Moisture in Graphite/Epoxy Laminates by X-Ray
Diffraction,” Journal of Composite Materials, vol. 16(4), pp. 260–267, 1982, doi:
https://doi.org/10.1177/002199838201600401.
[105] J. A. Nairn and P. Zoller, “Matrix solidification and the resulting residual thermal stresses in
composites,” Journal of Materials Science, vol. 20, pp. 355–367, 1985, doi:
https://doi.org/10.1007/BF00555929.
[106] P. P. Parlevliet, H. E. N. Bersee, and A. Beukers, “Residual stresses in thermoplastic composites—
A study of the literature—Part II: Experimental techniques,” Composites Part A: Applied Science
and Manufacturing, vol. 38, pp. 651–665, 2007, doi:
https://doi.org/10.1016/j.compositesa.2006.07.002.
[107] B. Andersson, A. Sjogren, and L. Berglund, “Micro- and meso-level residual stresses in glass-
fiber/vinyl-ester composites,” Composites Science and Technology, vol. 60, pp. 2011–2028, 2000,
doi: https://doi.org/10.1016/S0266-3538(00)00099-3.
[108] B. Winiarski and P. J. Withers, “Micron-Scale Residual Stress Measurement by Micro-Hole Drilling
and Digital Image Correlation,” Exp Mech, vol. 52, no. 4, pp. 417–428, Apr. 2012, doi:
10.1007/s11340-011-9502-3.
[109] Q. T. L. Vu et al., “Evaluating Residual Stress in Carbon Fiber-Reinforced Polymer (CFRP) at
Microscale Using Fiber Push-Out Experiment and Finite Element Modeling,” Polymers, vol. 15, no.
12, p. 2596, Jun. 2023, doi: 10.3390/polym15122596.
[110] M. R. Kamal and S. Sourour, “Kinetics and Thermal Characterization of Thermoset Cure,” Polymer
Engineering and Science, vol. 13, pp. 59–64, 1973, doi: https://doi.org/10.1002/pen.760130110.
[111] Q. Li, X. Li, and Y. Meng, “Curing of DGEBA epoxy using a phenol-terminated hyperbranched
curing agent: Cure kinetics, gelation, and the TTT cure diagram,” Thermochimica Acta, vol. 549, pp.
69–80, 2012, doi: https://doi.org/10.1016/j.tca.2012.09.012.
[112] S. G. K. Schechter, L. K. Grunenfelder, and S. R. Nutt, “Air evacuation and resin impregnation in
semipregs: effects of feature dimensions,” Advanced Manufacturing: Polymer & Composites
Science, pp. 101–114, 2020, doi: https://doi.org/10.1080/20550340.2020.1768348.
[113] Grafil Inc., “PYROFIL TM HS40 12K.” Accessed: Mar. 03, 2023. [Online]. Available:
https://www.rockwestcomposites.com/downloads/HS40-12K_(07-2008).pdf
[114] S. Ghaffari, G. Seon, and A. Makeev, “Effect of Fiber–Matrix Interface Friction on Compressive
Strength of High-Modulus Carbon Composites,” Molecules, vol. 28,2049, 2023, doi:
https://doi.org/10.3390/molecules28052049.

119
[115] P. T. Goncalves, A. Arteiro, and N. Rocha, “Micro-mechanical analysis of the effect of ply thickness
on curing micro-residual stresses in a carbon/epoxy composite laminate,” Composite Structures,
vol. 319, p. 117158, Sep. 2023, doi: 10.1016/j.compstruct.2023.117158.
[116] L. P. Canal, C. González, J. M. Molina-Aldareguía, J. Segurado, and J. LLorca, “Application of
digital image correlation at the microscale in fiber-reinforced composites,” Composites Part A:
Applied Science and Manufacturing, vol. 43, no. 10, pp. 1630–1638, Oct. 2012, doi:
10.1016/j.compositesa.2011.07.014.
[117] A. D. Kammers and S. Daly, “Digital Image Correlation under Scanning Electron Microscopy:
Methodology and Validation,” Exp Mech, vol. 53, no. 9, pp. 1743–1761, Nov. 2013, doi:
10.1007/s11340-013-9782-x.
[118] G. Seon, A. Makeev, J. Cline, and B. Shonkwiler, “Assessing 3D shear stress–strain properties of
composites using Digital Image Correlation and finite element analysis based optimization,”
Composites Science and Technology, vol. 117, pp. 371–378, Sep. 2015, doi:
10.1016/j.compscitech.2015.07.011.
[119] G. Seon, A. Makeev, J. D. Schaefer, and B. Justusson, “Measurement of Interlaminar Tensile
Strength and Elastic Properties of Composites Using Open-Hole Compression Testing and Digital
Image Correlation,” Applied Sciences, vol. 9, no. 13, p. 2647, Jun. 2019, doi: 10.3390/app9132647.
[120] J. Jäger, M. G. R. Sause, F. Burkert, J. Moosburger-Will, M. Greisel, and S. Horn, “Influence of
plastic deformation on single-fiber push-out tests of carbon fiber reinforced epoxy resin,”
Composites: Part A, vol. 71, pp. 157–167, 2015, doi:
https://doi.org/10.1016/j.compositesa.2015.01.011.
[121] B. Rohrmüller, P. Gumbsch, and J. Hohe, “Calibrating a fiber–matrix interface failure model to single
fiber push-out tests and numerical simulations,” Composites Part A, vol. 150, 2021, doi:
https://doi.org/10.1016/j.compositesa.2021.106607.
[122] Instron, “2810 Series Mini Flexture Fixture”, [Online]. Available: https://www.instron.com/-
/media/literature-library/products/2018/12/2810-series-mini-flexure-fixtures.pdf
[123] MatWeb, “Steels.” [Online]. Available:
https://www.matweb.com/search/datasheet.aspx?bassnum=MS0001&ckck=1
[124] Engineering Toolbox, “Young Modulus.” [Online]. Available:
https://www.engineeringtoolbox.com/young-modulus-d_417.html
[125] MatWeb, “Copper.” [Online]. Available:
https://www.matweb.com/search/datasheet_print.aspx?matguid=9aebe83845c04c1db5126fada6f76
f7e
[126] MatWeb, “PolyCarbonate.” [Online]. Available:
https://www.matweb.com/search/DataSheet.aspx?MatGUID=501acbb63cbc4f748faa7490884cdbca
[127] MatWeb, “Polyethylene Terephthalate (PET).” [Online]. Available:
http://www.matweb.com/search/datasheet.aspx?matguid=a696bdcdff6f41dd98f8eec3599eaa 20
[128] MatWeb, “Polystyrene, Extrusion Grade.” [Online]. Available:
https://www.matweb.com/search/DataSheet.aspx?MatGUID=1c41e50c2e324e00b0c4e419ca78030
4&ckck=1
[129] MatWeb, “Acrylic, Cast.” [Online]. Available:
http://www.matweb.com/search/datasheet.aspx?bassnum=O1303&ckck=1
[130] J. Heilman, “TBP Converting, Inc. 3M Acrylic Adhesives PDS.” [Online]. Available:
https://tbpconverting.com/wp-content/uploads/2022/08/3M-Acrylic-Adhesives-DP8405NS-8410NS-
DP8425NS.pdf
[131] A. M. G. Pinto, A. G. Magalhães, R. D. S. G. Campilho, L. F. M. Silva, J. A. G. Chousal, and A. P.
M. Baptista, “Shear Modulus and Strength of an Acrylic Adhesive by the Notched Plate Shear

120
Method (Arcan) and the Thick Adherend Shear Test (TAST),” MSF, vol. 636–637, pp. 787–792,
Jan. 2010, doi: 10.4028/www.scientific.net/MSF.636-637.787.
[132] B. Nečasová, P. Liška, J. Kelar, and J. Šlanhof, “Comparison of Adhesive Properties of
Polyurethane Adhesive System and Wood-plastic Composites with Different Polymers after
Mechanical, Chemical and Physical Surface Treatment,” Polymers, vol. 11, no. 3, p. 397, Mar.
2019, doi: 10.3390/polym11030397.
[133] Songhan, “Cytec Thornel T-50 6K datasheet.” [Online]. Available:
http://www.lookpolymers.com/pdf/Cytec-Thornel-T-50-6K-Carbon-Fiber-Polyacrylonitrile-PAN-
Precursor.pdf
[134] Toray, “T700S Datasheet.” [Online]. Available:
https://www.rockwestcomposites.com/media/wysiwyg/T700SDataSheet.pdf
[135] Toray, “T1000G data sheet.” [Online]. Available:
https://fileserver.mates.it/Prodotti/1_Rinforzi/TDS/Carbonio/Fibre/T1000G_DS.pdf
[136] Toray, “M40J datasheet.” [Online]. Available: https://www.toraycma.com/wp-content/uploads/M40J-
Technical-Data-Sheet-1.pdf.pdf
[137] Toray, “M50J datasheet.” [Online]. Available: https://toray-cfe.com/wp-
content/uploads/2020/12/toray-torayca-m50j-haut-module.pdf
[138] Toray, “M60J datasheet.” [Online]. Available: https://www.toraycma.com/wp-content/uploads/m60j-
datasheet-3b0af0e171322e8c.pdf
[139] HEXCEL, “HM63 datasheet.” [Online]. Available:
https://www.hexcel.com/user_area/content_media/raw/HM63_Aerospace_HexTow_DataSheet.pdf
[140] Mitsubishi Rayon Co., “Pyrofil MR 70 12P datasheet.” [Online]. Available:
https://northerncomposites.com/wp-content/uploads/2020/07/MR-70-12P20140204.pdf

121
Appendices

Appendix A: RVE implementation in Abaqus FEM simulation

The RVE can be used to determine homogenized composite properties from its
components. In this appendix, the RVE is utilized to study elastic properties.

Inputs for the RVE include:

• Composite’s Fiber Volume Fraction

• Fiber and Matrix elastic properties

Expected output:

• Composite system’s Elastic properties

The procedure of FEM RVE analysis includes important steps:

1- Fiber positions are randomly generated on a surface, which is then projected


into a 3D model of UD composite.

2- Input elastic properties of Fiber and Matrix. Interface between Fiber and Matrix
is considered “perfect” and undamaged throughout the elastic simulation.

Table A1 presents the input elastic properties of IM7 carbon fiber and 8552
epoxy matrix considered in this study.

Table A 1. Elastic properties of IM7 carbon fiber and 8552 epoxy matrix.
𝐸11 𝐸22 𝐸33 𝜈12 𝜈13 𝜈23 𝐺12 𝐺13 𝐺23
IM7 carbon fiber (GPa) (GPa) (GPa) (GPa) (GPa) (GPa)
276 15.0 15.0 0.20 0.20 0.07 15.0 15.0 7.03
𝐸𝑚 𝜈𝑚
8552 epoxy matrix (GPa)
4.67 0.39

3- Periodic Boundary Conditions is applied on the RVE, using “dummy nodes”


technique presented by Gomez [34]. On the original, undeformed model, mesh and
nodes on opposite faces of the RVE are identical (i.e., “conformal meshes”).

In Abaqus, keyword *EQUATION is applied to constrain movements of opposite


nodes according to dummy nodes displacements. Thus, the homogenized strain is

122
applied over the opposite faces of the RVE and over the RVE itself (Figure A1). For a
2D RVE, movement equations are:

𝑢𝑗𝐵𝐶 − 𝑢𝑗𝐴𝐷 = 𝜀̅𝑥𝑗 Δ𝑥 = 𝑢𝑗𝑅𝑃𝑋


𝑢𝑗𝐷𝐶 − 𝑢𝑗𝐴𝐵 = 𝜀̅𝑦𝑗 Δ𝑦 = 𝑢𝑗𝑅𝑃𝑌

As a result, the strains 𝜀𝑖𝑗 are correctly imposed on the RVE by definition.
Application of the technique to a 3D RVE is similar.

Figure A 1. Concept of “Dummy Nodes” to apply PBC on the RVE.

4- Strains of 𝜀11 , 𝜀22 , 𝜀33 , 𝜀21 , 𝜀31 , 𝜀23 are imposed on the RVE. Stresses 𝜎𝑖𝑗 are
obtained via. Reaction Force on the dummy nodes divided by corresponding surface
areas of the RVE. Strain values are very small (≪ 1%) so there is little difference
between Engineering Stress and True Stress.

A system of equations is then created thanks to various Strain and reaction


̿:
Stress values, which permits determining the Stiffness Matrix 𝑪

̿ . 𝜺̅
̅= 𝑪
𝝈

̿, we obtain the Compliance Matrix 𝑺


Inverting Stiffness Matrix 𝑪 ̿:

̿. 𝝈
𝜺̅ = 𝑺 ̅

which can be written in terms of Global/Equivalent Elastic parameters:

123
Then we can calculate directly Global elastic properties
𝐸11 , 𝐸22 , 𝐸33 , 𝐺23 , 𝐺13 , 𝐺12 , 𝜈12 , 𝜈13 , 𝜈23 .

5- RVE size and mesh convergence study on Elastic Properties. For each RVE
size, ten different random fiber distributions are generated and studied. Fiber volume
fraction is 60%. For the IM7/8552 CFRP, results in Table A2 show that Convergence of
Elastic Properties is obtained with a relatively small RVE of 5-fiber-radius cross-section,
containing 5 fibers. For Damage assessment however, a RVE with size of ~50 times
Fiber Radius should be used [38], [39].

Table A 2. RVE size convergence study. COV: Coefficient of variation over 10 random fiber distributions.
Fibers RVE cross
𝐸11 𝐸22 𝐸33 𝐺12 𝐺13 𝐺23
in section 𝜈12 𝜈13 𝜈23
(GPa) (GPa) (GPa) (GPa) (GPa) (ksi)
RVE (in 𝑅𝑓 )
mean 167 9.72 9.72 0.276 0.276 0.377 5.05 5.03 3.67
5 5
COV 0.036% 0.33% 0.34% 0.66% 0.65% 0.53% 2.5% 2.6% 0.4%
mean 167 9.86 9.86 0.276 0.276 0.370 5.03 5.02 3.62
43 15
COV 0.003% 0.57% 0.56% 0.40% 0.40% 0.96% 1.2% 1.1% 0.8%
mean 165 9.86 9.86 0.277 0.277 0.368 5.02 4.98 3.59
150 28
COV 0.001% 0.44% 0.44% 0.28% 0.28% 0.77% 0.9% 0.9% 0.6%

124
Appendix B: Bending Beam method for fiber longitudinal compression
strength measurement, design and evaluations

The Bending Beam is proposed as the preferred experimental method for


𝑜
measurement of carbon fiber longitudinal compressive critical strain 𝜀𝑐11 and strength
𝑜
𝜎𝑐11 due to its relative simplicity and robustness. This Appendix presents a detailed
analytical analysis which demonstrates the method feasibility and provides
recommendations for the realization of the Bending Beam method.

Analytical analysis for Experiment Design Requirements

In this analysis, beam’s deformation is supposed small, elastic and linear. Figure
A2 provides an illustration of the analyzed variables.

Figure A 2. Illustration of terms in the Bending Beam analysis.

Given the beam deflection 𝑤 = 𝑓(𝑥), radius of curvature 𝜌 at a section of the


beam:

1 𝑤 ′′
= 3
𝜌 [1
+ (𝑤 ′ )2 ]2
1
If 𝑦 ′ and 𝑦 is small, then: ≈ 𝑤 ′′
𝜌

125
Given the beam’s initial, undeformed length 𝐿0 between the fixtures A and C
(Figure A2). We consider a short beam section of initial, undeformed beam length 𝑑𝐿0 .
Suppose that the beam is under pure bending which can be achieved by using low-
friction fixtures, we have the constant relation at the neutral axis: 𝜌 × 𝑑𝜃 = 𝑑𝐿0 . For an
axis at distance 𝑦 from the neutral axis, then radius of curvature at 𝑦 is (𝜌 − 𝑦) i.e.,
corresponding axis length at 𝑦 is given by:

𝑑𝐿 = (𝜌 − 𝑦)𝑑𝜃

So, the local longitudinal strain is:

𝑑𝐿 − 𝑑𝐿0 (𝜌 − 𝑦)𝑑𝜃 − 𝜌𝑑𝜃 𝑦


𝜀𝑥 = = =− (12)
𝑑𝐿0 𝜌𝑑𝜃 𝜌

If there is no lateral stress/load acting on the elastic beam, then local


longitudinal stress is simply:

𝐸𝑦
𝜎𝑥 = 𝐸𝜀𝑥 = − (13)
𝜌

Bending Moment equation:

𝐸𝐼𝑧𝑧
𝑀𝑧 = (14)
𝜌

If Bending is small and curvature 𝑤 ′ ≪ 1 then:

𝑀𝑧 = 𝐸𝐼𝑧𝑧 𝑤 ′′

Combining (13) and (14), then we have:

𝑀𝑧 𝑦
𝜎𝑥 = −
𝐼𝑧𝑧

𝜎𝑥 𝑀𝑧 𝑦
𝜀𝑥 = =−
𝐸 𝐸𝐼𝑧𝑧

We apply these equations on a 3-point bending schema, we consider only half of


the beam thanks to the symmetry, where the deflection moment is:

126
𝑃𝑥 𝐿0
𝑀𝑧 (𝑥) = 𝑓𝑜𝑟 0 ≤ 𝑥 ≤
2 2

The maximum Moment exerted on the beam is therefore:

𝐿 𝑃𝐿0
𝑀𝑧 ( ) =
2 4

At distance 𝑦 from neutral axis, longitudinal strain is:

𝑀𝑧 𝑦 𝑃𝑥 𝑦
𝜀𝑥 (𝑥, 𝑦) = − = − ×
𝐸𝐼𝑧𝑧 2 𝐸𝐼𝑧𝑧

We deduce strain at the surface, where 𝑦 = 𝑇 = ℎ/2:

𝑃𝑥 𝑇
𝜀𝑥 (𝑥, 𝑇) = − ×
2 𝐸𝐼𝑧𝑧

For a beam with rectangular cross-section of “width” b and “height/thickness” h:

𝑏ℎ3
𝐼𝑥 =
12

Numerical application: 𝑇 = ℎ/2 then strain at the upper beam surface is:


𝑃𝑥 3𝑃𝑥
𝜀𝑥 (𝑥, 𝑇 = ℎ/2) = − × 23 = − (15)
2 𝐸𝑏ℎ 𝐸𝑏ℎ2
12

With equation (15), compressive strain of the upper beam surface, and of the
bonded carbon fiber, can be directly deduced from easily obtainable measurements:
Load, position of the fiber damage limit, beam material Modulus, beam cross-section
dimensions.

We can deduce maximum compressive strain which occurs at beam length


center 𝐵 where the load 𝑃 is applied:

3 𝑃𝐿
𝜀𝑚𝑎𝑥 = 𝜀𝑥 (𝑥 = 𝐿0 /2, 𝑇 = ℎ/2) = − × (16)
2 𝐸𝑏ℎ2

i.e.

127
2 𝐸𝑏ℎ2
𝑃=− 𝜀𝑚𝑎𝑥 (17)
3 𝐿

2𝑥
𝜀𝑥 = 𝜀𝑚𝑎𝑥 × (18)
𝐿

Equation (16) provides an engineering requirement for the Beam material, since
the material has to accommodate at least the critical compressive strain of the fiber.
Equation (17) provides the requirement for the Load capacity of the 3-point testing
machine. Equation (18) provides a quick relation for the calculation of bonded fiber
critical compressive strain based on the position of its damage limit.

Finally, we want to verify maximum beam deflection, so that the condition of


“small beam deflection” is satisfied. Maximum deflection happens at the middle of beam
length (Load point B, Figure A2):

𝑃𝐿3 𝑃𝐿3 𝑃𝐿3 1 L2


𝑤𝐿 = = = = − 𝜀
2 48𝐸𝐼 48𝐸𝑏ℎ3 4𝐸𝑏ℎ3 6 𝑚𝑎𝑥 h
12

Therefore, we have maximum compressive strain: 𝜀𝑚𝑎𝑥 = 6 𝑤𝑚𝑎𝑥
𝐿2

For small beam deflection, the ratio of deflection over corresponding length
should be less than 10%:

𝑊𝐿
2 1 𝐿
𝐶𝑠𝑚𝑎𝑙𝑙 = | |= 𝜀𝑚𝑎𝑥 < 0.1 (19)
𝐿/2 3 ℎ

The inequation (19) provides criteria to select Bending Beam dimensions of


length 𝐿 and thickness ℎ.

Now we study the forces acting on the bonded fiber. Local beam strain is
transmitted into the fiber through shear force at the interface. The analysis provides
estimation of the maximum shear stress acting on the fiber, so a proper bonding

128
adhesive can be selected. Figure A3 shows forces acting on a short section 𝑑𝑙 of
bonded fiber.

Figure A 3. Forces acting on a short section 𝑑𝑙 of bonded fiber in Bending Beam test.

Main forces are the longitudinal normal stress 𝜎11 (𝑥) and adhesive shear stress
𝜎𝑆 (𝑥). The section is static so the total longitudinal force is zero:

Σ𝐹 = 𝜋𝑅𝑓2 (𝜎11 (𝑥 + 𝑑𝑙) − 𝜎11 (𝑥)) + 2𝜋𝑅𝑓 𝑑𝑙. 𝜎𝑆 (𝑥) = 0

We deduce the local shear stress:

𝑅𝑓 𝜕𝜎11 (𝑥 )
𝜎𝑆 (𝑥 ) = − (20)
2 𝜕𝑥

3𝑃𝑥
From equation (15): 𝜀11 (𝑥) = −
𝐸𝑏ℎ2

and we assume the fiber to be perfectly elastic, with longitudinal modulus 𝐸11 :

3𝑃𝑥 E11
𝜎11 (𝑥) = 𝐸11 𝜀11 (𝑥) = −
𝑏ℎ2 E

𝜕𝜎11 (𝑥 ) 3𝑃 𝐸11
→ =− 2 (21)
𝜕𝑥 𝑏ℎ 𝐸

From equations (20) and (21), we can calculate adhesive shear force acting on
the fiber:

3 𝑃 𝐸11 𝐸11
𝜎𝑆 (𝑥 ) = 𝑅𝑓 2 = −𝑅𝑓 𝜀 (22)
2 𝑏ℎ 𝐸 𝐿 𝑚𝑎𝑥

It is interesting to note that under given simplifications, shear force is uniform


along bonded fiber length.

129
Example of numerical application for engineering requirements

Assuming an example configuration for the Bending Beam of:

- Dimensions: (𝑙𝑒𝑛𝑔𝑡ℎ, 𝑡ℎ𝑖𝑐𝑘𝑛𝑒𝑠𝑠, 𝑤𝑖𝑑𝑡ℎ) = (100 𝑚𝑚, 10 𝑚𝑚, 20 𝑚𝑚). The ratio
𝑙𝑒𝑛𝑔𝑡ℎ
should be higher than ten (10) for a valid slender beam assumption.
𝑡ℎ𝑖𝑐𝑘𝑛𝑒𝑠𝑠

𝐿
- Objective maximum fiber compressive strain 𝜀11 ( ) = 2%.
2

The Bending Beam is proposed to be made of PolyCarbonate, which allows high


elastic strains and is ductile (details presented in Table A3).

The maximum shear stress at fiber-adhesive interface is calculated to be


0.141 𝑀𝑃𝑎, well below shear strength of the Acrylic and Urethane adhesives (Table A4).

The maximum Load required by the 3-point bending heads is 629 Newtons, well
within the capacity of the popular 2kN bending beam machine [122].

Other considerations

To avoid interference with fiber damage development, the head of the 3-point
bending should avoid touching/pushing on the bonded fiber area. This can be achieved
using a head with a middle groove allowing the fiber area “passing through”.

The method can be combined with DIC and FEM analysis to confirm strains
developed and compare with analytical predictions.

During the specimen preparation, the fiber is typically straightened using a small
weight, creating a pre-tension. The pre-tension as well as the weight should not be too
high to significantly interfere with experimental results. Regarding the reference IM7
carbon fiber, one individual fiber can withstand approximately 10 𝑔𝑟𝑎𝑚 tension load. A
hanging mass of 0.1 𝑔𝑟𝑎𝑚 would lead to a pre-tension of 𝜀𝑝𝑟𝑒 = 0.017% on the fiber. For
𝑜
comparison, the critical tension strain of the IM7 fiber is 𝜀𝑡11 = 2%.

130
Bending Beam Material

The table A3 below presents some potential materials for Bending Beam’s
material, along with their important properties: Young Modulus 𝐸, Yield Stress 𝜎𝑦 and
estimated Yield Strain 𝜀𝑦 = 𝜎𝑦 /𝐸. For critical yield strain estimation, Compressive
Modulus is used when available. Otherwise, the more frequently provided Tensile
Modulus is used for estimation.

Amongst the listed materials, only PolyCarbonate (PC) possesses Yield Strains
in both tension and compression that respond to our criterion (𝜀𝑦 > 2%), and is ductile.
Finally, the PolyCarbonate material is inexpensive and widely available.

Table A 3. Potential materials for the Bending Beam.


Beam Material Et (GPa) Ec (GPa) σyt (MPa) σyc (MPa) εyt εyc Ductile/ Brittle? Ref

Steel, general 200 350 0.18% Ductile [123]


Aluminum 69 95 0.14% Ductile [124]
Copper 110 33 0.03% Ductile [125]
PolyCarbonate 2.36 63.3 65.7 2.68% 2.78% Ductile [126]
PETE 3.10 63.6 45.2 2.05% 1.46% Ductile [127]
Polystyrene 2.98 36.1 1.21% Ductile [128]
Acrylic 3.10 2.96 65 120 2.10% 4.05% Brittle [129]

Adhesive for Fiber-Beam bonding

Deteresa et al. [46] bonded fibers to the PC beam using a clear Acrylic spray
coating. Fidan [59] utilized Urethane spray coating (MS470). Table A4 presents some
strength properties of the Acrylic and Urethane adhesives.

Using the analytical formula for shear stress at bonded fiber interface (Equation
22), adhesive shear stress in the experiment by Deteresa et al. [46] is estimated at
0.126 𝑀𝑃𝑎, well below the shear strength of the adhesive that was used in their
experiment.

131
Table A 4. Shear strength of suggested adhesive for fiber bonding on Bending Beam.
Adhesive Shear Strength at room temperature, film (MPa)
Acrylic 6.9 [130] , 8.0 [131] - depending on specific type
Urethane 5.5 [132]

Conclusions on the Bending Beam method

The Bending Beam method is proposed to measure carbon fiber critical


compression strain. An analytical analysis was performed to demonstrate the feasibility
and provide some preliminary engineering requirements of the experiment. An example
of the Bending Beam features was proposed and demonstrated to satisfy all
requirements of the experiment.

132
Appendix C: Review: longitudinal properties of some carbon fibers

Table A5 presents single-fiber compressive properties of several notable Carbon


Fibers frequently mentioned in articles from UTA and elsewhere from the literature.
𝑜
The Compressive Strength 𝜎𝑐11 , when available, is cited along with experimental
measurement method.

Table A 5. Longitudinal properties of selected carbon fibers from literature. MC*: obtained using Micro-
Compression method. BB*: obtained through Bending Beam method. For critical strain values: Pr*:
provided by datasheet, if available.
Fiber Tensile Tensile Critical Compressive Critical
Modulus Strength Tensile Strain Strength Compressive Strain
𝐸(GPa) 𝑜
𝜎𝑡11 (GPa) 𝑜
𝜀𝑡11 𝑜
= 𝜎𝑡11 /𝐸𝑡 𝝈𝒐𝒄𝟏𝟏 (GPa) 𝜺𝒐𝒄𝟏𝟏 = 𝝈𝒐𝒄𝟏𝟏 /𝑬𝒕
T300 230 [41] 3.53 [41] 1.53% 1.8 (±16.9%) [52] (MC*) 0.78%
T50 393 [59] 2.9 [133] 0.74% 4.09 [59] (BB*) 1.04% ± .08% [59]
390 [133] 0.6% (Pr*)
[133]
T700S 230 [134] 4.9 [134] 2.13% 2.4 (±15.6%) [52] (MC*) 1.04%
T800 294 [42] 5.49 [42] 1.87% 2.3 (±14.6%) [52] (MC*) 0.78%
H
T1000 294 [135] 6.37 [135] 2.17% 2.8 (±18.8%) [52] (MC*) 0.95%
M40J 377 [136] 4.4 [136] 1.17% 1.8 (±13%) [52] (MC*) 0.48%
P75S 517 [59] N/A 2.69 [59] (BB*) 0.52% ±.07% [59]
M50J 475 [137] 4.12 [137] 0.87% 1.3 (±15.5%) [52](MC*) 0.27%
M60J 588 [138] 3.82 [138] 0.65% 1.0 (±13.9%) [52] (MC*) 0.17%
IM7 276 [40] 5.516 [40] 2.00% N/A N/A
HM63 435 [139] 4.826 [139] 1.11% N/A N/A
HS40 455 [113] 4.61 [113] 1.01% N/A N/A
MR70 325 [140] 7.0 [140] 2.16% N/A N/A

Fiber Critical Tensile Strain is estimated from frequently provided Tensile


𝑜 𝑜
Strength and Tensile Modulus: 𝜀𝑐11 = 𝜎𝑐11 /𝐸𝑡 . In fact, this Critical Tensile Strain is very
close to the value provided by the Fiber manufacturer when this parameter is available.
This confirms the fragile nature of the Carbon Fiber, at least in tension.

There is currently no established method to measure directly a single fiber’s


Compressive Modulus 𝐸𝑐 . Therefore, when Critical Compressive Strength 𝝈𝒐𝒄𝟏𝟏 is
directly measured (as the case with Micro-Compressive Method), its Critical
𝑜 𝑜
Compressive Strain is estimated with Tensile Modulus 𝐸𝑡 : 𝜀𝑐11 = 𝜎𝑐11 /𝐸𝑡 .

133
Inversely, in the Bending Beam method, the experiment actually determines
Critical Compressive Strain 𝜺𝒄𝒓𝒊𝒕
𝒄 . Then the authors estimate Critical Compressive

stress using Fiber Tensile modulus:


𝑜 𝑜
𝜎𝑐11 = 𝜀𝑐11 × 𝐸𝑡

In some cases, datasheets provide Fiber Critical Tensile Strain. In such cases, in
addition to the estimated value using the above formulae, the value is noted (Pr*) along
with citation to the document.

Table A6 compares Composite critical strain with its corresponding Base Fiber’s
critical strain, for the fiber types in Table A5. Fiber Critical Compressive Strain from
Table A5 is provided in the last column of Table A6 for quick reference and comparison.
Typically, Composite compressive strength is provided. Composite critical compressive
strain is estimated by dividing strength over modulus:

𝜀𝑐 = 𝜎𝑐 /𝐸

If Compressive Modulus 𝐸𝑐 is provided, the above formula applies 𝐸 = 𝐸𝑐 , and in


Table A6 the Modulus is also marked with “(comp)”.

Else if Compressive Modulus 𝐸𝑐 is unavailable and only Tensile Modulus 𝐸𝑡 is


available, the above formula applies 𝐸 = 𝐸𝑡 , and in Table A6 the Modulus is marked
with “(tens)”.

134
Table A 6. Longitudinal properties of selected CFRP composites, corresponding to fibers from Table A5.
“comp”: compressive property; “tens”: tensile property.
Base Composite Composite Composite Base Fiber
Fiber Compressive Modulus Critical Critical
Strength Compressive Strain Compressive Strain
𝜎𝑐 (GPa) 𝐸 (GPa) 𝜀𝑐 = 𝜎𝑐 /𝐸
T300 1.57 [41] 125 (comp) [41] 1.18% 0.78%
T50 N/A N/A N/A 1.04%
T700S 1.47 [134] 125 (tens) [134] 1.18% 1.04%
T800H 1.57 [42] 145 (comp) [42] 1.08% 0.78%
T1000 1.57 [135] 165 (tens) [135] 0.95% 0.95%
M40J 1.27 [136] 226 (tens) [136] 0.56% 0.48%
P75S N/A N/A N/A 0.52%
M50J 0.98 [137] 295 (tens) [137] 0.33% 0.27%
M60J 0.79 [138] 360 (tens) [138] 0.22% 0.17%
IM7 1.689 [40] 150 (comp) [40] 1.13% N/A
HM63 1.31 [139] 221 (comp) [139] 0.59% N/A
HS40 1.31 [113] 241 (comp) [113] 0.54% N/A
MR70 N/A N/A N/A N/A

135
Appendix D: Mesh and size convergence studies for Residual Stress
Analysis using Fiber Push-Out experiment [109]

The mesh and size convergence analyses utilizes the reference FE geometry
(Figure A5). Reference in-plane size is 30 𝜇𝑚 × 40 𝜇𝑚. Convergence studies utilize the
measurement of matrix sink-in following fiber push-out.

The FE model meshing utilizes three parameters (Figure A4): in-plane inner
mesh density; in-plane outer mesh density; and through-thickness number of elements.
A denser inner mesh around the push-out fibers allows efficient evaluation of the
interested area. Mesh convergence is considered achieved when a doubled mesh
density leads to less than 2% variation in matrix sink-in. Figures A6.a, b, c present the
results of the mesh convergence studies. The following reference configuration was
found to satisfy the convergence criterion: four (4) elements per 𝜇𝑚2 for the inner mesh,
one (1) element per 𝜇𝑚2 for the outer mesh, and ten (10) elements through the
membrane thickness.

Figure A 4. Mesh parameters of the FE Model.

To evaluate model size convergence, each smaller-sized geometry is a rectangle


central slice from a bigger-sized geometry, as illustrated in Figure A5. Size convergence
analysis results are presented in Figure A6, suggesting that the presented FE model
size convergence is achieved at a relatively small reference size that covers the push-
out fibers and an outward in-plane expansion by a few fiber-diameters [109].

136
Figure A 5. Some FE models of different sizes for the convergence study.

Figure A7 reports the computation time for the mesh and size convergence
studies. 11 CPUs of the Intel Xeon Processor X5650 were utilized in these analyses.
The reference model took approximately 8 minutes, while the largest model took
approximately 100 minutes.

Figure A 6. Convergence analysis results: a) Inner mesh density; b) Outer mesh density; c) Through-
thickness mesh; d) Model size. [109]

137
Figure A 7. Computation time for the convergence analyses: a) Inner mesh density; b) Outer mesh
density; c) Through-thickness mesh; d) Model size. [109]

138
Appendix E: Size convergence analysis of Residual Stress to validate the
application of far-field Boundary Conditions on micromodels [109]

To validate the application of far-field constraint-free boundary conditions at the


in-plane limits of the FE micromodel during the curing stage, a size convergence
analysis is conducted on the residual stress at the end of the curing process. The
residual stress is examined across models with varying in-plane dimensions. The study
analyzes the average residual stress generated in a Probed matrix volume containing
matrix elements situated beneath the probed area (Figure A8.a), and the average
residual stress across the entire matrix volume (Figure A8.c). The volume-averaged in-
plane stresses 𝑆𝑥𝑥, 𝑆𝑦𝑦 and the through-thickness stress 𝑆𝑧𝑧 are presented in Figure
A8.b for the probed volume, and in Figure A8.d for the entire matrix volume. For some
models of different sizes, on an in-plane cross-section cut through the midpoint of
membrane thickness, Figure A9 visualizes the distribution of von Mises residual stress
in the local fiber push-out area.

Within both the local probed volume and the entire matrix volume, convergence
of residual stress is observed as the model size increases. The size of the FE model
shows only a weak influence on the distribution of von Mises stress in the locality of the
push-out area. These results suggest that with the application of the proposed boundary
conditions (BCs), generated residual stress would be consistent irrespective of the
model size. Through inductive inference, applying the far-field in-plane BCs of the
macroscale specimen to the micromodel, starting from a certain minimum size, is valid
for the purpose of residual stress analysis.

139
Figure A 8. a) Probed matrix Volume: matrix elements beneath the probed area; b) Average residual
stress after curing within Probed Volume; c) Entire matrix volume; d) Average residual stress after curing
within entire matrix volume. [109]

Figure A 9. After curing: von Mises residual stress within the central push-out region, observed on an in-
plane cross-section positioned at the midpoint of the membrane thickness. Results are shown for several
evaluated model sizes. [109]

140

You might also like