Fatigue Testing of a Unidirectional Carbon Fiber Reinforced Polymer

IN

DEGREE PROJECT MECHANICAL ENGINEERING, SECOND CYCLE, 30 CREDITS

STOCKHOLM SWEDEN 2020 ,

Fatigue Testing of a Unidirectional Carbon Fiber Reinforced Polymer

Investigation of damage development using Digital Image Correlation

LARS JOHAN WENNER BERG

KTH ROYAL INSTITUTE OF TECHNOLOGY

SCHOOL OF ENGINEERING SCIENCES

Utmattningstest av en enkelriktad

kolfiberförstärkt polymer

LARS BERG

Aerospace Engineering Date: December 14, 2020 Supervisor: Per Wennhage Examiner: Zuheir Barsoum

School of Industrial Engineering and Management Host company: Scania CV AB

Swedish title: Utmatting av komposit

iii

Abstract

Carbon Fibre Reinforced Plastic (CFRP) is a material with high specific proper- ties and good fatigue and vibration dampening characteristics, and can potentially be used instead of steel and aluminium in heavy duty vehicles. This work fo- cuses on testing methodology and the fatigue properties of a unidirectional (UD) material in the 0° and 90°orientation, reproducing and validating the method de- veloped by Wanner[1]. While conducting a fatigue test of a CFRP composite in tension-tension fatigue, in-situ strain measurements were performed to examine the gradual elongation of the specimen (as this relates to stiffness loss, i.e. fatigue damage). An imaging methodology capturing the specimen at peak loading has been developed, including a trigger mechanism that activates the camera at the de- sired time and cycle count, as well as a method of extracting the photograph of the specimen at maximum displacement, allowing for peak-to-peak comparison.

A method improving specimen production output and consistency has been de- veloped. SN-curves have been produced for both 0° and 90° fibre orientations.

Micrography of sectioned specimen has been conducted. The study finds the fa-

tigue limit of the 0° specimen to be as high as 80 % of the material tensile fail-

ure strength, while results from the 90° study indicate a lower but inconclusive

value. An attempt at qualitatively determining the factors causing the material be-

haviour has been made and is deliberated upon.

iv

Sammanfattning

Kolfiberförstärkt plast (CFRP) är ett material med höga specifika mekaniska egen-

skaper och goda utmattnings- och vibrationsdämpande egenskaper, och kan poten-

tiellt ersätta stål och aluminium i fordonsstrukturer. Detta arbete fokuserar på test-

metodik och utmattningsegenskaperna för ett enkelriktat material (UD) i 0° och

90° orientering, reproduktion, validering och utveckling av metoden utvecklad av

Wanner [1]. Under genomförande av utmattningsprovning av en CFRP-komposit i

drag belastning utfördes töjningsmätningar på plats för att undersöka den gradvisa

töjningen av provet (eftersom detta avser styvhetsförlust, dvs utmattningsskada). En

avbildningsmetodik som fångar provet vid toppbelastning har utvecklats, inklusive

en utlösningsmekanism som aktiverar kameran vid önskad tid och cykelantal, samt

en metod för att extrahera fotografiet av provet vid maximal förskjutning, vilket

möjliggör jämförelse vid olika cykelantal. En metod som förbättrar provstavspro-

duktion och kvalitet har utvecklats. S-N kurvor har skapats för både 0° och 90° fiber-

orientering. Mikrografi av snittprov har utförts. Studien indikerar at utmattnings-

gränsen för provet i 0° kan vara så hög som 80 % av materialets dragfasthållfasthet,

medan resultaten från 90° studie indikerar ett lägre men tvetydigt värde. Ett försök

att kvalitativt bestämma de faktorer som orsakar det materiella beteendet har gjorts.

v

Acknowledgements

I wish to express deep gratitude to Sara Eliasson for her invaluable cooperation, patience and tutelage during this project, as well as to Associate Prof. Per Wennhage and Prof. Zuheir Barzoum for their insight and leadership.

I wish to thank Scania CV AB and KTH for providing me the opportunity to conduct this study and for helping in its realisation.

Lastly I wish to to thank Anders Beckman and Monica Norrby for their

assistance and solutions in the LWS lab, and Johan Nygren, Johan Larsson

as well as the entire Weld and Composite Mechanics Group for their

inclusion and input on this challenging topic.

Contents

1 Introduction 1

1.1 Purpose of study . . . . 1

1.2 Fatigue . . . . 2

2 Literature review 3 2.1 Reference litterature . . . . 3

2.2 Digital image correlation . . . . 7

3 Methodology 10 3.1 Introduction . . . . 10

3.2 Test material . . . . 11

3.3 Specimen manufacturing . . . . 12

3.4 Static testing . . . . 15

3.5 Fatigue testing . . . . 17

3.5.1 Out-of-plane twist . . . . 17

3.5.2 Clamping grips . . . . 18

3.5.3 Ramp testing . . . . 18

3.5.4 Fatigue process . . . . 19

3.6 DIC capture . . . . 19

3.6.1 Photography . . . . 21

3.6.2 Verification . . . . 22

3.6.3 Failure types . . . . 23

3.7 Micrography . . . . 23

4 Results 25 4.1 Overview . . . . 25

4.1.1 Camera trigger mechanism . . . . 25

4.2 DIC post processing . . . . 30

4.2.1 Peak determination . . . . 30

vi

CONTENTS vii

4.3 Fatigue test results . . . . 32

4.4 Fatigue S-N Curves . . . . 34

4.5 0

DIC fatigue measurements . . . . 35

4.6 90

DIC fatigue measurement . . . . 36

4.7 1-FA-6 displacement and strain . . . . 36

4.8 Migrographs . . . . 37

5 Discussion 38 5.1 Overview . . . . 38

5.1.1 Fatigue results . . . . 38

5.1.2 DIC . . . . 39

5.1.3 Micrography . . . . 41

5.1.4 Sources of error and uncertainty . . . . 41

6 Conclusion 43 6.1 Overall impressions . . . . 43

6.1.1 On the viability of 2D-DIC . . . . 43

6.2 Suggested improvements . . . . 44

Bibliography 47 A Test manual 50 A.1 Notes on 2D-DIC testing . . . . 50

A.1.1 Procedure . . . . 50

A.1.2 Camera placement and RBM verification . . . . 52

A.1.3 Camera Trigger . . . . 54

A.1.4 Camera control software . . . . 56

A.1.5 Specimen preparation and speckle pattern . . . . 56

A.1.6 Illumination . . . . 57

B Arduino Setup 59

C Arduino Code 60

Nomenclature

AOI Area of Interest AWJ Abrasive Water Jet

CFRP Carbon Fibre Reinforced Plastic FLD Fatigue Life Diagram

GFRP Glass Fibre Reinforced Plastic HDV Heavy Duty Vehicle

LVDT Linear Variable Differential Transformer RBM Rigid Body Motion

UD Unidirectional (Fibre Orientation) UI User Interface

USB Universal Serial Bus UTS Ultimate Tensile Stress

viii

Introduction

1.1 Purpose of study

CFRP is a lightweight yet strong, stiff and durable material that has a growing application in the automotive industry. Lower vehicle mass equates to lowered energy requirements and fuel consumption, increased payload, and better perfor- mance. In addition to its high specific strength and stiffness[2], it can also exhibit a higher fatigue resistance than current alternatives in certain configurations[3].

High tensile strength steels typically have a fatigue limit in the range of 30 to 60 % of the Ultimate Tensile Strength (UTS)[4]. Unidirectional (UD) Carbon Fibre Reinforced Plastics (CFRP) regularly perform significantly better than this, and given that a modern truck is designed to a great extent against fatigue failure, its potential implementation in a Heavy Duty Vehicle (HDV) is of interest to Scania.

Whereas our understanding and modelling of the fatigue behaviour of metals has reached a mature level, modelling this behaviour for CFRP is still challenging, and more physical testing is required in order to develop and validate fatigue models. Other challenges also remain to be solved before widespread adoption of CFRP can occur in HDVs. This pertains to several phases in the life cycle of such a vehicle, from design and production, to maintenance and recycling. This work however, focuses on measuring and validating numerical fatigue models and the fatigue performance of a certain CFRP material already used in an automotive application, incorporating 2D Digital Image Correlation (2D-DIC) to measure strain in the sample over the course of testing, and investigating the damage that occurs.

validate numerical fatigue models

1

2 CHAPTER 1. INTRODUCTION

1.2 Fatigue

Fatigue is the process by which mechanical properties of a material degrade due to repeated loading. The magnitude of this cyclic loading and resulting stress may be much lower than that which would induce plastic changes or failure for a single loading. For instance, a truck axle routinely experiencing stresses half its failure stress, may experience fatigue failure after a certain amount of cycles. For an isotropic material such as a metal, this can occur due to imperfections present in the material, which either lowers the local material strength, or creates a stress concentration such that it is exceeded. These preconditions are necessary for what is known as crack growth to occur, and when it does, a physical discontinuity or crack, initiates. If conditions allow, this crack will propagate, the stress will be redistributed (and increase), ensuring further crack growth, typically in a direction perpendicular to the main tensile axis[4]. There are more or less two views on this phenomenon; one microscopic, focused on modelling of the crack geometry and conditions in an attempt to predict the rate of crack growth, as well as a macroscopic view - one in which statistical data is produced based on serial tests of standard geometries. This thesis work is conducting the latter.

Whereas fatigue damage to an isotropic material may be largely characterised by

a single parameter (crack growth), a composite exhibits a multitude of damage

mechanisms due to its inhomogenous nature. For a CFRP the material stiffness

and strength properties of each constituent typically differ by an order of magni-

tude in favour of the fibre, and the failure strain of the matrix is usually significantly

higher. The prevailing damage mechanisms differ in each constituent and at their

boundaries. A local failure in one constituent (for instance a single fibre break-

age) and the resulting redistribution of load may lead to a wholly different mode of

failure occurring elsewhere in the composite[5]. On the contrary, the redistributed

load may lead to a load scenario where damage growth is restrained, an example of

this would be the scenario where a progressing matrix crack is restrained (or pre-

vented from coalescing) by intersecting fibres[5]. This behaviour is dependent on

a multitude of factors that are difficult to control from a modelling and production

perspective, although its complexity is reduced somewhat in UD laminates.

Literature review

2.1 Reference litterature

Figure 2.1: Fatigue-life diagram according to Talreja et al[6]

Talreja and Singh[6] characterise the prevailing damage mechanism(s) and life as a function of strain level (or load level) in a fatigue life diagram (FLD). This diagram has three distinct regions, as shown in fig. 2.1

The first region is at high load, near the failure strain of the fibres (but still far below the failure strain of a typical matrix) low cycle fatigue range, where some fibres fail outright.

Each fibre failure redistributes load on

the remaining fibres, potentially causing more fibres to fail . If enough fibres fail, the stress in the remaining fibres will exceed the tensile capacity and fail. The rate at which this process occurs varies greatly, as is evident in large horizontal spread in the diagram. Testing in this region is characterised by "sudden death"

(unstable successive fibre failure) and by being non-progressive (i.e. apparent lack of material degradation).

3

4 CHAPTER 2. LITERATURE REVIEW

Figure 2.2: Types of fatigue damage, a ) fibre breakage, b ) fibre bridged matrix cracking, c ) matrix cracking without crack coalescence [6], reproduced.

In the case where the gradual process of individual fibre failure halts, the process might not resume again until the matrix has degraded sufficiently to redistribute load sufficiently to continue the process. This second region is a sloped area in the middle to high cycle fatigue range that is characterised by matrix fatigue cracking, with a crack forming at a point of local fibre failure or traversing the fibre after debonding with it (see fig. 2.2). The resulting load redistribution may cause an arresting fibre to fail, or further accelerate matrix damage. The horizontal spread in this region is less than in the first region, but still significant.

The third region is the high-cycle, fatigue limit region, where either the stresses in the matrix are insufficient to cause significant crack formation, or that growth of cracks traversing the fibre is effectively arrested. Any load redistribution occurring is insufficient to cause fibre failure within a predefined number of cycles. This number is referred to as the composite fatigue limit, in this study initially set to 2 ∗ 10

cycles (later increased to 5 ∗ 10

cycles).

The location of region one and three (and subsequently the span of region two), is governed by the material properties of the constituent materials. The region one scatter band is centred on the failure strain of the fibres, whereas region three is at the fatigue limit of the matrix. This indicates that the fatigue performance of a composite with a matrix fatigue limit close to the failure strain of the fibres, may produce fatigue test data in a very narrow region in terms of loading (region one spread, with a very narrow region two, and run-outs below this).

One may casually observe that the expected spread of fatigue life results in the

SN-curve for a composite is the opposite of that of a metal (which has a low

spread at high load levels, but increased spread with lowered load and increased

life). The spread in fibre failure (which dominates region one) is subject to many

CHAPTER 2. LITERATURE REVIEW 5

factors, particularly material surface imperfections. Due to the high anisotropy of the composite, even small deviations in fibre orientation or load alignment has a large impact on the fatigue performance[7], apart from increased material strength in off-axis directions, misalignment may introduce undesired shear loads and damage effects as a result. Talreja and Singh[6] refer to the matrix fatigue limit as a constituent property, however Brunbauer and Pinter[8] has compared specimen with different fibre volume fraction (V

) in 90° (where fatigue performance is as- sumed to be matrix dominated) as well as homogeneous matrix material specimen and found that the composite specimen had a significantly poorer performance, illustrating that the fatigue limit of the matrix also has dependencies. Further- more, void content in the matrix also plays a large role in its fatigue performance[9].

Figure 2.3: Nominal stress amplitude vs life of 0° UD specimen of V

= 55% and 30%, reproduced[10].

Brunnbauer and Pinter[10] tested UD CFRP fatigue performance with regards to V

, load ratio and nominal stress amplitude (analogous to load level in this case). Two SN curves of UD CFRP 0° are reproduced in fig. 2.3. Here, results (particularly for the V

= 30% specimen) are distributed in a fairly small range, but with large horizontal spread at the load level corresponding to region one, as well as a slope corresponding to region two.

Gamstedt and Talreja[11] are able to observe and document the aforementioned

damage types (fibre bridged cracking and debonding) at the microscopic level, and

also correlate them to the proposed damage regions in the FLD.

6 CHAPTER 2. LITERATURE REVIEW

Figure 2.4: Load level vs life for six 0° UD specimen, by Wanner[1], reproduced.

Wanner’s[1] work with the same CFRP material that is used in this study, aimed to develop a testing methodology w.r.t. specimen adhesive, tab configuration, tab material, and clamping force and its conclusions formed a framework for this study. Wanner was also able to produce a preliminary SN curve containing six data points(fig. 2.4). These results fall in the region proposed by Talreja [6], and have low spread in the high-load, low-cycle fatigue region, but the results are too few to determine a fatigue limit conclusively. Wanner’s work formed the outset for this thesis, which continues the work using much of its methodology

Figure 2.5: Overview of test setup used by Sanjay et al[12]

Sanjay et al[12] focused on in-situ full field strain development in a notched specimen, and also conducted micrography of sections of tests that were stopped pre-failure. The specimen was photographed at peak load and DIC analysis performed.

The study inspired this study in its methodology, but was performed under different circumstances.

The specimen, being notched, sustained localised damage at a predictable location, allowing for ideal

placement of the high-speed camera. This again allows for better resolution and

CHAPTER 2. LITERATURE REVIEW 7

opportunities to capture local events at higher detail. The article features DIC colour-plots of the strain for the notched region, as opposed to average strains, see fig. 2.5.

2.2 Digital image correlation

Digital Image correlation (DIC), is an image processing method of tracking surface displacements. It is commonly used in static testing of materials, and has the benefit of being non-contacting, full field (i.e. it can capture the entire surface of the specimen, unlike a strain gauge or linear variable differential transformer (LVDT) extensometer, which produce an "average" by default), and time efficient once the method is established.

Figure 2.6: Principle of operation of 2D-DIC in a tension test, from digitalimagecorrelation.org [13]

(CC-BY-4.0).

DIC is used in this project to track the developing fatigue induced deformation between cycles at constant load. In this application the so called 2D-DIC method is used, which relies on capture from a single camera. The method is subject to one important condition: It assumes in-plane displacement. This poses challenges for the testing as it requires the camera axis to be perfectly normal to the specimen surface, which in turn must be completely flat [14].

The testing machine used in this experiment has a rotational degree of freedom which is problematic due to low torsional rigidity of the specimen,

to counter this an aluminium bracket was installed on the piston, which effectively limits the out of plane rotation tendency to near zero (one side of the bracket is in constant contact with the support column of the test machine). fig. 2.8 illustrates the magnitude of error that can result from out-of-plane translation.

Other sources of error include lens distortion, an optical aberration that can

produce a fish-eye effect in the images. This problem can be dealt with in software

by introducing a lens correction factor which would have to be measured for this

specific setup. Due to time constraints and reasonable results in preliminary tests

8 CHAPTER 2. LITERATURE REVIEW

using an aluminium specimen, this has been neglected in this study.

The DIC algorithm used works in the following manner: a sequence of images are compared with a reference (initial image). Each image contains the Area of Interest (AOI) that is selected in advance (a speckle pattern fills this area completely). Within this area the algorithm selects a number of subsets which it attempts to match to the corresponding subsets in each image in the sequence. If a match is made, the centre-to-centre distance can be measured from the reference image to the current image.

Figure 2.7: Subset matching and displacement

The consistency of the algorithms ability to match subsets and yield results is dependent on numerous factors; speckle pattern, camera resolution, specimen illumination (strength and homogeneity). The speckle pattern should consist of speckles with roughly the same size, and the area coverage should be roughly 50 %. Each speckle should be between 10 to 20 pixels in size, and the "grey tone area"

(gamma gradient) between completely white and completely black should ideally not be more than 3-4 pixels [13]. Uneven illumination of the specimen makes matching less likely as it affects the thresh-

olding process used by the algorithm to differentiate between white and black areas

CHAPTER 2. LITERATURE REVIEW 9

Figure 2.8: Example of pseudo-strain caused by out-of-plane displacement of 1 mm, at an imaging distance of 61 cm



=

 g

g

− ∆g − 1



=

 610 610 − 1 − 1



= 0.1652% (2.1)

Methodology

3.1 Introduction

By comparing the sample at a constant load at different times (cycle counts, or

"specimen life"), the gradual degradation in the specimen can be investigated, as done previously by Sanjay et al[12]. This study aims to combine the rigorous testing scheme developed by Wanner[1], while additionally comparing peak to peak displacement using DIC.

All tests have been conducted in tension-tension, meaning that the sample is sub- jected to loading causing stress to vary sinusoidally between a low and high state, as in shown fig. 3.1 a. The applied load at both levels is constant, i.e. load controlled.

Fatigue damage coincides with elongation of the specimen, as shown in subfigure c.

.

Figure 3.1: a) Sinusoidal load, b) constant maximum load level resulting in increasing maximum strain over the course of the test (c).

σ = F

A

R = σ

σ

S = σ

σ

(3.1) The ratio of minimum to maximum stress is known as the load ratio R

10

CHAPTER 3. METHODOLOGY 11

(eq. (3.1)). Positive ratios means non-alternating stress, i.e. tension-tension or compression-compression; subsequently negative R-ratios indicate stresses alternating between tension and compression. The load ratio has been demonstrated to impact fatigue performance[7], and it was originally an ambition of this work to perform testing at three load ratios R = (0.1, 0.5, 0.8) and compare. However this goal was abandoned due to time constraints.

Figure 3.2: Stress-time curve for a specimen under different load ratios.

The ratio of maximum stress to UTS is known as load level S. The relationship between load level and cycles to failure for a given sample and material type can be plotted in a Wöhler diagram, or S-N curve. Here, the abscissa constitutes number of cycles endured prior to failure on a logarithmic scale, and the load level is plotted on the ordinate, examples of which are shown in fig. 2.3 and fig. 2.4.

3.2 Test material

The CFRP material tested is a stitch-bonded UD non-crimp, epoxy-carbon composite produced with a resin transfer moulding (RTM) technique. The constituent properties of the material are given in table 3.1.

Property Zoltec UD Dow Epoxy Composite

Tensile strength 4137 MPa 68 MPa 1290 MPa*

Young’s modulus 242 GPa 2.8 GPa 118 GPa*

Elongation at

break 1.7 % 7 % 1.18 %*

Density 1810 kg m

1160 kg m

1400 kg m

V

na na 48.6 %

Table 3.1: Material properties, * measured

12 CHAPTER 3. METHODOLOGY

The material was received as 300 mm by 650 mm plates, which were subsequently cut. All the available plates exhibited a slight tendency of fibre waviness near the edges of the plate, and care was taken to use material farther from the edges where this effect was less pronounced. As a result of the closed-mould manufac- turing method, both front and back surfaces of the plate feature a glossy, even finish.

3.3 Specimen manufacturing

The 0° and 90° specimen were manufactured in accordance with ASTM- D3039/D3039M[15], with 5 mm wider tabs to compensate for any potential lack of coverage due to tab misalignment during gluing, as recommended by Wanner[1].

Figure 3.3: 0° Specimen dimensions, all measurements in mm

The CFRP Material was cut into strips using a diamond bladed band-saw, whereas the aluminium tabs were cut with either an abrasive water jet (AWJ) or a hydraulic precision metal shear to the dimension as per fig. 3.3 and fig. 3.4 for the 0° and 90° specimen respectively.

In order to develop the DIC-methodology, and with the test material being in lim-

ited supply, around two dozen additional specimen were manufactured using left-

over scrap material made earlier by Wanner for this purpose. These had been

manufactured in-house using the vacuum-assisted resin infusion (VARTM) tech-

nique, resulting in a carbon-vinyl-ester composite that was cut to similar dimen-

sions as the Audi specimen, except with a thickness of 2.4 mm, and with an infe-

rior surface quality on one side (a so called "B-side"[5], as a result of the manu-

facturing technique). These tests were only occasionally carried out uninterrupted,

and data collection on the samples was done for the purpose of developing the

test, consequently they are not treated here.

CHAPTER 3. METHODOLOGY 13

Figure 3.4: 90° Specimen dimensions, all measurements in mm

Each individual specimen was measured at three locations to determine width and thickness, these values where then averaged and used to calculate a nominal cross sectional area. This area in turn determined the appropriate loads (σ

and σ

) for the desired load level S (eq. (3.1).

Figure 3.5: Specimen alignment and gluing jig

To ensure specimen consistency and efficient production, a gluing template tool was manufactured out of a 20 mm aluminium plate, using an abrasive water jet cutter. The purpose of this tool is to ensure proper alignment of tab and test material, and can be seen in figure fig. 3.5. This jig consists of a lower slab embedding 4 mm round steel pins to constrain any movement of the specimen material and end tabs during gluing. A top slab with cuts allowing the steel pins to pass through

is placed on top to apply even pressure to the six specimen inside. The jig is then placed in an oven for post curing of the glue, and the semi-fixed steel pins allows adequate tolerance for thermal expansion during the curing process and easy removal afterwards.

Initially an AWJ process was considered for cutting the specimen themselves,

but this was later decided against (in favour of a diamond blade saw), as the

material is susceptible to delamination with this method[16, 17]. The change to

using a saw for cutting precluded the option to cut specimen post-gluing, as the

constituent materials require different saw blades. The specimen glue area (50

x 15 mm) and was subjected to a light grind with 400 and 800 grit sandpaper,

14 CHAPTER 3. METHODOLOGY

washed with dishing soap and rinsed, wiped and left to dry for a minimum of one hour. The tab area (50 x 20 mm) was machine ground with 400 grit sandpaper and cleaned with pressurised air, and rinsed with acetone prior to gluing.

To ensure consistent glue thickness, simple, serrated glue spreading pads were 3D

printed, reducing material waste while providing a controlled glue thickness. The

adhesive used is the two component epoxy DP420M, determined by Wanner[1] to

be well suited for this use. Immediately after adhesive application, the assembled

jig is placed in an oven and allowed to cure at 50

C for one hour, as per the

DP420M product specification[18], which lists a lap shear strength of 24 MPa

bond to aluminium, and 35 MPa to CFRP. Metal weights were distributed atop

the jig prior to the glue hardening and left there during curing process to ensure

adequate contact and consolidation of glued surfaces.

CHAPTER 3. METHODOLOGY 15

3.4 Static testing

Figure 3.6: Static tension specimen immediately prior to failure, note that tilted appearance is due to the view being from a single (in this case the left) camera.

To verify the static UTS obtained by Wanner[1] for the test material, two static

tests were conducted. The specimen were made with spare pre-cut material

16 CHAPTER 3. METHODOLOGY

and featured 15 mm wide glass fibre reinforced plastic (GFRP) tabs but were otherwise similar to the fatigue specimen as given infig. 3.4. The move to GFRP tabs for static testing is done to lower the chance of slippage and pullout, as they exhibit greater resistance to pull-out compared to aluminium tabs, at the cost of having lower thermal conductivity (which is a considerable problem in fatigue testing, see section 3.5). As thermal issues were of no concern in static testing consisting of one slow cycle, GFRP tabs were successfully used.

Equipment Item

Tension machine Instron 4505

Grips Mechanical

Camera and lens 50 mm Aramis 5M LT 3D system

Software GOM ARAMIS PRO

Table 3.2: Equipment used in static testing

An Instron 4505 100 kN machine was for used for the static tests, and each specimen was loaded under displacement control at a rate of 2 mm min

, as per ASTM D3039/D3039M[5]. The displacement and strain capture and computation was conducted with a Aramis 3D-DIC setup, producing a strain-time dataset.

The test machine logged displacement and load over the same duration, and the

combination of this data yielded the stress-strain curve given in fig. 3.7. The two

test results fell within the range found by Wanner[1], who tested five specimen,

yielding an average UTS of 1295 MPa, with a S.D. (standard deviation) of

108 MPa.

CHAPTER 3. METHODOLOGY 17

Figure 3.7: 3D-DIC Stress vs 

measurement on the first of two static tension tests.

3.5 Fatigue testing

Fatigue testing has been conducted at the Lightweight Structures Laboratory at KTH Stockholm

, on a test setup as shown in fig. 3.9 and listed in table 3.3. The machine has a degree of freedom about its operating axis (the lower actuator piston is free to rotate), which, for a stiffer metal specimen is uncontroversial. A step-by-step instruction on how to correctly initiate a test is provided in appendix A

3.5.1 Out-of-plane twist

Due to the low torsional stiffness of the CFRP specimen considered here, the specimen twisted, causing out of plane motion (see section 2.2), so an aluminium guide bracket (fig. 3.8) was installed on the piston to mitigate this phenomenon by

1

KTH Lightweight Structures Laboratory website

18 CHAPTER 3. METHODOLOGY

resting and sliding against a support column on the machine.

Figure 3.8: Aluminium guiding bracket to prevent out-of-plane torsion

The twisting motion exhibited by specimen prior to the instalment of this feature can be characterised as drift; there are no apparent forces causing it, subsequently the friction force exerted on the piston is deemed negligible. A lubricant was applied to prevent cosmetic damage to the support column. The mass of the bracket is low compared to that of the piston assembly, and any offset inertial loads caused by this addition is considered to be negligible. The instalment of the bracket was required

to mitigate out of plane pseudo-strain in testing, and any detrimental effect of the bracket solution on testing accuracy were disregarded.

3.5.2 Clamping grips

The specimen is held in place by hydraulically closed grips which provide a clamping force independent of the applied axial load (unlike many mechanical grips where clamp force is generated by the interaction of the enclosed clamping wedges and the specimen itself). The clamping was adjustable, and a linear relationship between hydraulic pressure and clamping force for this device was determined empirically by Wanner[1] earlier. Too high clamp force resulted in the specimen being crushed and sustain damage conductive to near or in-tab failure, too low and the sample would be prone to adhesive failure and pullout.

3.5.3 Ramp testing

The fatigue machine is operated in displacement control during specimen

installation, extraction or replacement, and set to load control once the specimen

is fastened just before testing. Before to the initiation of a fatigue test, a ramp test

was performed in which the specimen is loaded from near zero to the maximum

load of the fatigue test (typically in the range 30 to 38 kN). The load was

increased in constant increments of 2 kN s

, and subsequently unloaded. For a

ramp test the camera would be manually triggered near simultaneously as the

CHAPTER 3. METHODOLOGY 19

fatigue machine, taking a predefined number of images (typically 300 images at 25 frames per second). This was done to provide the later peak-to-peak analysis with a reference point, as well as a load-displacement plot for the first cycle. The fatigue machine would read piston displacement and load cell data, typically at a sampling rate of 10 Hz, to complement the DIC measurements.

3.5.4 Fatigue process

After completion of the ramp test, the machine would be set to conduct a sinusoidal load variation between the desired maximum and minimum, at a frequency of 5 Hz, and with safety domain defined in terms of load and displacement limits. The data acquisition for the machine is for this set to

"peak / valley" ensuring that maximum and minimum load and the correspond- ing displacement is logged every fifth cycle. When a specimen breaks the displacement or load domain limits would be exceeded, causing the machine to halt.

Equipment Item

Fatigue machine Schenck Hydropuls 100 kN

Grips Hydraulic

Clamp force 50 to 90 kN (Depending on tension load)

Camera DantecDynamics NanoSense mk3 MKIII

Lens Nikon AF-S 60/2.8 ED

Capture distance 610 mm

Software Motionpro X studio

Table 3.3: Equipment used in static testing

3.6 DIC capture

The imaging setup used during fatigue testing consists of a single high

speed camera operated by a laptop, lighting for specimen illumination, and

a proprietary system developed to trigger the camera. Under the assumption

that maximum displacement is only achieved close to instantaneously (or

over a very small timespan), any photograph of this occurrence will logically

have some margin of error associated with it. Photographs are again not

created instantaneously, but are a result of a sensor being receptive over its

exposure time. Consequently the camera system is subject to both missing

the peak by some amount, and by "smearing out" the displacement over the

20 CHAPTER 3. METHODOLOGY

Figure 3.9: Schenk Hydropuls 100 kN, with high speed camera and lighting in place.

CHAPTER 3. METHODOLOGY 21

Figure 3.10: Fatigue testing and photo acquisition in progress

exposure duration. Furthermore there are time delays associated with the various system components, making a definitive single-photo trigger approach impractical.

3.6.1 Photography

The high speed camera is operated at a capture rate of 1000 frames per second and has a sensor resolution of 1280 by 1024 pixels. This system is configured via a laptop running MotionProX Studio, which stored the captured images on an external hard drive. The camera operates in a "circular mode", meaning that images are taken and stored in flash memory continuously, with the oldest images being continuously overwritten. A circuit formed by a synchronised output and the trigger input is closed by a transistor switch (an electro-mechanical relay was used initially) which is in turn operated by the microcontroller. Once triggered the camera transfers a pre-selected number of images from its flash history (i.e.

the pre-trigger event frames) and captures a selected number of images after the triggering event. Because both these values are adjustable, the circular mode offers a means of tuning the timing capture of relative to the trigger event and load threshold setting.

Each imaging event resulted in 150 pictures, consisting of 40 pre-trigger and 110

22 CHAPTER 3. METHODOLOGY

post-trigger images. These images were subsequently uploaded to the external hard-drive via a USB 2.0 link, initially resulting in a transfer time of about 20 seconds, during which the camera would be unable to capture new images, and as such created a limited factor in the form of a gap between each capture cycle.

Due to the elongated shape of the target area, and the higher horizontal pixel count, the camera was rotated 90° about its viewing axis to minimise the capture distance. MotionProX features settings that allows for a rotation of the input image, as well as reduced capture area in each picture, producing cropped, rotated images (250 x 1280). This way the transfer time is reduced to 5 seconds, and the apparent resolution is maximised, while still allowing capture of the entire gauge length of the specimen.

3.6.2 Verification

In order to verify that the camera setup had proper alignment with the sample, a rigid body motion (RBM) test (where the sample is not attached at the top, i.e. no strain in specimen) would be conducted and processed in NCORR[19].

The results would then be examined, and the quality of the strain measurement determined. This was done by looking at the noise level of individual frames, and its development frame by frame. Firstly, too much noise would indicate improper speckle pattern, specimen illumination or camera settings, secondly, any increase in strain as a result of RBM would be psuedo-strain as a result of improper camera alignment. The lower the noise, and the more constant (and near zero) the strain over the course of the RBM-test, the better the DIC test setup. To verify the DIC measurement technique a specimen was tested and measured using a calibrated clip-on LVDT extensometer as well as the 2D-DIC. The results of which is given in fig. 3.11 and fig. 3.12 respectively.

Figure 3.11: Extensometer strain Figure 3.12: DIC obtained strain

CHAPTER 3. METHODOLOGY 23

1st. Character 2nd. Character 3rd. Character

Failure type Code Area Code Location Code

Angled A Inside tab I Bottom B

Edge delamination D At tab A Top T

Tab G <1 width from tab W Left L

Lateral L Gage G Right R

Multi-mode M(xyz) Multiple M Middle M

Long. splitting S Various V Various V

Explosive X Unknown U Unknown U

Other O ∼ ∼ ∼ ∼

Table 3.4: Specimen failure codes, reproduced from [15]

3.6.3 Failure types

ASTM 3039D[15] categorises different failure types in a three char- acter scheme that covers all the recorded failure types. Categorisation is done according to table 3.4.

3.7 Micrography

In order to investigate the effect of fatigue loading on the composite micro-structure, a few specimen were sectioned in order to be studied using optical microscopy. In this procedure a piece of the test specimen on the order of 10 mm was removed, embedded inside cylindrical mould using a thermoset polymer cement.

The resulting cylindrical specimen resembles a hockey puck with a diameter of 30 mm. The initial grind is performed to uncover the specimen cross-section, and is performed with coarse sandpaper. The specimen surface is inspected after each round of grinding to determine the next step, the grooves resulting from the grinding process should be uniform in roughness (indicating a fresh layer has been uncovered), before proceeding to a finer grit sandpaper until it has a polished finish. For this process sandpapers 300, 600, 800, 1200 and 2000 were used.

Once adequately prepared, the specimen was examined with a optical microscope,

magnified in the range 5, 20 and 50 times, and photographed.

24 CHAPTER 3. METHODOLOGY

Figure 3.13: Failure types as outlined in ASTM D3039/D3039M — 17 [15]

Results

4.1 Overview

The results consist the developed camera triggering system, the post processing procedure as well as all the data gathered over the course of the 24 successful tests conducted.

There are fifteen 0°-tests and nine 90°-tests with viable outcomes. Of the fifteen 0°-tests, seven have both ramp and fatigue DIC data. There were no ramp photography performed for the 90° tests. The type specimen failure type has been noted, and several micrographs have been produced.

4.1.1 Camera trigger mechanism

A way to capture the definite peak displacement is to take multiple photos at short intervals with low exposure times, and then post-process the resulting series of photos. In order to time the imaging event, a camera trigger operating on fatigue machine output, was developed and used.

As an input for the camera trigger, the 10 V output of the load cell was selected.

The load cell is a piezo-electric variable linear transistor with a ±100 kN range.

To read and process the load signal, an Arduino Uno microcontroller was used.

During testing this device was connected to a laptop via a universal serial bus (USB) connector, and programmed via the Arduino IDE[20] through which the Arduino serial output can be monitored continuously.

The device is equipped with an analogue to digital converter (ADC) capability that converts a 0 - 5 V analogue signal to a 10-bit (0-1023) variable. Figure 4.2 and fig. 4.3 depict a typical serial monitor and plotter output in the Arduino IDE.

25

26 CHAPTER 4. RESULTS

Figure 4.1: Laptop operating the camera software and reading the arduino serial output

The load signal analog readout is read using the analogRead(); function, which on an Arduino UNO has a theoretical maximum sample rate of approximately 9600 Hz (in reality reduced due to other tasks in the code loop), and is subjected to a running average of 20 measurements to reduce signal noise.

Figure 4.2: Example output from the

Arduino IDE serial plotter function Figure 4.3: Serial monitor output

example, as viewed in the Arduino IDE

The microcontroller is programmed with a number of constants defining all inputs,

most importantly the load threshold which defines the level above which the

displacement is close to the peak. The servohydraulic fatigue machine is subject to

CHAPTER 4. RESULTS 27

a short ramp-up phase (approximately 150 cycles when operating at 5 Hz) before reaching its set maximum load. This causes some difficulty setting the correct load threshold in the program initially, but it can easily be updated after the test has begun. As a rule of thumb, the threshold should be in the region 70 to 80 % of the maximum readout observed in the serial monitor. An attempt at making the load level a function of the readout (i.e. dynamic), introduced unforeseen timing issues that proved difficult to solve, and was subsequently abandoned.

Each time the load signal crossed the threshold, a boolean variable was switched, allowing every fatigue cycle to be counted and stored in a variable. The resulting integer value was displayed in the serial monitor, and proved to be very accurate compared to the cycle count given by the fatigue machine interface, a discrepancy of less than a dozen cycles was common for a test running for several million cycles, and the discrepancy seemed solely to depend on the accuracy of the initial threshold value.

(A) Load is above threshold (B) Interval is full

(C) Trigger is not active (D) Specimen has yet to break List 4.1: Main conditions required

for the camera trigger to be engaged. Figure 4.4: A load cycle, with threshold and imaging sequence coverage indicated.

The software uses several conditions that govern the trigger activation scheme.

The process is described in the system flowchart given in fig. 4.6. After the analogue signal has been read and "cleaned", it is subjected to several logic gates that either triggers the load signal variable (meaning a 5 volt output on the trigger pin is made) or alters the value of variables which alters the functionality of the program. In general, a trigger signal is performed if conditions A through D in list 4.1 are fulfilled.

In terms of fatigue damage, measurements during the first few hundred cycles and late (prior to failure) in the process are of more interest than measurements in the "middle", where activity was assumed to be lower (see region B in fig. 4.5).

This, and data volume considerations made it interesting to make image capture

intervals a function of cycle count, ideally with lower intervals in region A and C

28 CHAPTER 4. RESULTS

in fig. 4.5.

Figure 4.5: Paris’ Law[21] for fatigue, with the stable crack growth phenomenon in the middle

To solve this, an interval variable (IV) was introduced to set the amount of cycles between each imaging instance.

The interval variable might be set to 100 during the first 5000 cycles, whereas it might be set to 50 000 cycles once the sample has endured one million cycles.

The problem with this approach however is that it is difficult to predict when the specimen is going to fail, effectively meaning that unless the IV is sufficiently low, rapid developments in fatigue damage immediately prior to failure of the specimen are likely to be lost to the camera.

The IV and the cycle counts at which it was changed was selected by guessing. It quickly became apparent

that a test sample at 70 % load would live at least 2 ∗ 10

cycles, and that a lower temporal resolution after the first 10

cycles was desirable.

Conversely a load of 80 % would be prone to sudden death from the start, meaning that a constant low IV was necessary. A decreased IV where the sample would transition from region B to region C in fig. 4.5 was aimed for, but proved difficult to realise.

Specimen failure was detected by comparing the measured load to a variable containing a very low value, below which the specimen necessarily must have failed. Upon measuring this occurrence, the trigger would change to a different state and notify a second microcontroller (MKR WIFI1010), which had network capabilities that allowed for failure notification.

A pinout schematic of the microcontroller setup as well as the additional MKR

WIFI1010 notification device is given in appendix A.

CHAPTER 4. RESULTS 29

Figure 4.6: Flowchart diagram describing the functionality of the trigger system

(A) Analogue signal from test machine input (B) Running average to "clean" signal of noisiness (C) Is the load zero? has the specimen failed?

(D) If so, has test started yet?

(E) Is the signal above the threshold?

(F) If so, has the peak been counted?

(G) Increment cycle and interval counts

(H) Is the interval count equal to the set interval?

(I) Activate trigger if the specimen has broken

(J) Dead state: stop counting, notify and enter sleepmode if applicable (K) Hardware: MKR WIFI1010

(L) Hardware: High Speed Camera

List 4.2: Functionality of core components of the software

30 CHAPTER 4. RESULTS

4.2 DIC post processing

In order to compare peak-to-peak displacement and strain for a single fatigue test, the images depicting the sample at maximum displacement must first be found.

This is done by performing a DIC-analysis of each 150-image sequence.

4.2.1 Peak determination

Vertical displacement of each individual subset is computed for the entire area of interest, covering 95 % of one face of the gauge length of the specimen. To lessen the impact of artefacts - local spikes in displacement, i.e. noise - the average displacement is taken for all the subsets. Because the images are stored and numbered sequentially upon capture, and processed in that order, the resulting displacement value corresponds to the image number in the sequence. The displacement plot is saved (see fig. 4.7), as well as the image in the sequence corresponding to maximum displacement in two separate folders.

Figure 4.7: Matlab average displacement plot of a single peak

This process is looped for all sequences for each test specimen (see fig. 4.8) producing one folder containing the displacement plots (for verification) while another stores the actual images.

Filenames for each maximum includes the number in the sequence, as well as its sequence number, identifying its position within each fatigue test. Once all the peaks have been determined and aggregated, a separate 2D-DIC analysis can be made, resulting in a strain vs. cycle plot, an early example of which is given in fig. 4.10.

Figure 4.8: Peak detection process, applicable to both NCORR and Python approaches

To perform these analyses, a setup incorporating an open-source program for

CHAPTER 4. RESULTS 31

Figure 4.9: Peak detection output from python script

matlab (NCORR)[19]. This software allows for computation of full-field 2D displacements and strains, given a number of inputs. It is operated via a series of Matlab commands in addition to a internal user interface (UI), allowing for a wide range of adjustment of the analysis.

Automation of the Matlab based software to this use was after considerable effort made possible, but its performance was deemed insufficient as a test con- sisting of some 300 sequences took over a week of continuous operation to process.

An alternative solution was discovered in Pydic[22], a python based 2D DIC

software that offered much of the same functionality. This sotware proved

to be much easier to loop, and allowed for much faster processing and peak

determination. Both programs share the same basic functionality outlined in

fig. 4.8. An example of the output from each loop from this software is given

in fig. 4.9. For the final peak-to-peak analysis, NCORR was used and the peak

sequences were analysed in both NCORR (which works great for final analysis),

and to some degree GOM Correlate 2019[23] for verification.

32 CHAPTER 4. RESULTS

Figure 4.10: The first average peak displacement readout from an NCORR displacement analysis. Note the varying capture rate.

4.3 Fatigue test results

The results from the fatigue tests are given for each specimen in table 4.1 and 4.2 for the 0° and 90° specimen respectively. In the first table the column ”DIC”

indicate whether or not a ramp (”r”) or fatigue (”f”) capture was conducted, failure code is defined from the remains of the sample, and according to the scheme outlined in table 3.4.

Results for measured displacement and strain from the peak-to-peak DIC analysis is given for three of the specimen in section 4.5. These plots feature the averaged displacement and strain measured at the bottom of the specimen.

In fig. 4.18 and fig. 4.19, micrographs of transverse cross sections of

different specimen are presented, the first is of an un-cycled specimen, and the

second is of a specimen that has undergone 2.1∗10

cycles at a load level of 75.8 %.

CHAPTER 4. RESULTS 33

Specimen Load % Cycles DIC Failure Code

1_FA_1 70 2000000 f Run-out

1_FA_2 75.8 2000000 r+f Run-out

1_FA_3 80 785746 r+f GIT

1_FA_4 80 500366 r+f SGM

1_FA_5 80 412526 r+f LIT

1_FA_6 80 5000000 r+f Run-out

1_FA_7 - - N/A N/A

1_FA_8 90 853 N/A LGM

1_FA_9 90 2113876 fracture LAV

1_FA_10 90 171 N/A LAT

1_FA_11 90 150 N/A LIT

1_FA_12 85 7807 r+f GAT

1_FA_13 85 397036 r+f LAV

1_FA_14 85 1621 r LAT

1_FA_15 85 366 N/A LIT

1_FA_16 85 386 N/A GAT

Table 4.1: Fatigue data for the 0° specimen

Specimen Load % Cycles DIC

10_FA_90_5 70 4726 f

10_FA_90_6 65 11361 f

10_FA_90_7 60 140296 f

10_FA_90_8 65 12081 f

10_FA_90_9 65 25631 f

1_FA_90_10 60 6686 f

1_FA_90_11 60 4786 f

1_FA_90_12 60 8681 f

1_FA_90_13 60 5991 f

Table 4.2: Fatigue data for the 90° specimen, prefix ”10” denotes that these specimen were

produced earlier by Wanner [1].

34 CHAPTER 4. RESULTS

4.4 Fatigue S-N Curves

Figure 4.11: S-N data for specimen 1-FA-1 through 16

Figure 4.12: S-N data for 90° specimen

CHAPTER 4. RESULTS 35

4.5 0 ^◦ DIC fatigue measurements

(a) Average strain over time for a 1-FA-3 (b) Average displacement over time for a 1-FA-3

(c) Average strain over time for a 1-FA-6 (d) Average displacement over time for a 1-FA-6

(e) Average strain over time for a 1-FA-5 (f) Average displacement over time for a 1-FA-5

Figure 4.13: Comparison of different strain plots

36 CHAPTER 4. RESULTS

4.6 90 ^◦ DIC fatigue measurement

Figure 4.14 and fig. 4.15 are example plots of displacement and strain respectively, from the testing conducted on the 90° specimen.

Figure 4.14: Displacement plot for 10_FA_90_7.

Figure 4.15: Strain plot for 10_FA_90_7.

4.7 1-FA-6 displacement and strain

A plot of the displacement output from the fatigue machine for specimen 1-FA-6 for all 5 ∗ 10

cycles is given in fig. 4.17, note that this output covers the entirety of the test, whereas DIC capture was stopped after 2 ∗ 10

cycles. In fig. 4.16 the resulting strain plot is superimposed on the displacement from the fatigue machine readout.

Figure 4.16: DIC strain (2 ∗ 10

) superpositioned on fatigue machine displacement (5 ∗ 10

cycles), for specimen 1-FA-6.

Figure 4.17: Fatigue machine

displacement output vs cycle count for

specimen 1-FA-6.

CHAPTER 4. RESULTS 37

4.8 Migrographs

Figure 4.18: Micrograph of an non-tested Audi-specimen

Figure 4.19: Micrograph of 1_FA_2 ( 2.1 ∗ 10

cycles at 75.8 % load).

Discussion

5.1 Overview

The aim of the study has been to continue the development of a testing methodology for UD CFRP - building on the earlier work by Wanner - by testing more specimen, and also incorporating in-situ 2D-DIC to measure displacement and strain over the course of testing.

In terms of fatigue testing methodology, the steps taken in this study and earlier by Wanner[1] w.r.t. specimen preparation - choice of glue and post-cure, tab material, cooling, clamping solution and more - have proven to be adequate in testing and practical in implementation. This process and the manner in which it is conducted is of high importance, as the testing regime is inhabiting a small but viable region, between the limits of adhesive failure, in-tab compressive failure, temperature-induced adhesive failure, shear failure induced by unintended off-axis specimen placement or cutting. all of the above have occurred to varying degrees, underscoring the difficulty in testing the material, and the importance of a rigorous methodology.

5.1.1 Fatigue results

The test result of the 0° material (table 4.1 and figure 4.11) indicate a highly fatigue resistant material up to load levels of 75 % of UTS. Above this limit the spread in the test data becomes very high, exemplified by the comparison of the results from 1_FA_9 and 1_FA_10, where the former survived over 2 million cycles, the latter failed after 171 only cycles, despite being loaded at the same level (90 % of UTS).

This large discrepancy is to some degree expected (it is arguably too large);

38

CHAPTER 5. DISCUSSION 39

whereas successive fibre failure halted in 1_FA_9 while the matrix caught up fatigue wise, the fibre breakage progressed unchecked in 1_FA_10 from the beginning. The factors that determine this behaviour are many (see section 2.1), but largely they can be described as fibre irregularities such as off-axis orientation, discontinuity, weak fibre-matrix bonding, local low fibre volume content, production defects as well as various polymer defects such as for example voids, inclusions of foreign material, damage or other causes of low matrix material strength.

For the 90° specimen the results are to a lesser degree in accordance with expectations, and probably the subject of some experimental error or test material artefact. All tested specimen fail after 10

or so cycles (with the exception of 1_FA_90_7), and the expected relationship between life and load level is not observed conclusively. Note that specimen named "10_FA_.." are leftover specimen produced (as in cut, glued and post-cured) by Wanner[1] some eight months prior to the later specimen (1_FA_90_10 through 13) made for the current project. Care was taken to follow the described methodology as closely as possible, yet subtle differences could exist. Both series of specimen were cut with the same method, and post cured similarly. The material plates are however different, so there is a possibility that the specimen material is at fault. Further testing of both would have do be done to get a definitive answer.

The failure types encountered for the 0° specimen are a mix of combined modes such as SGM, LAV, and single mode types types like LGM, LIT and GAT. LGM and LIT is considered the ideal failure mode (lateral failure in middle of the gauge length), after which comes LAT (lateral failure at tab). Failures occuring in-tab (LIT) are unfortunately common, and proved hard to avoid. SGM is a type of longitudinal splitting coupled with two lateral failures that also occurred, and for which the primary failure type is hard to determine (i.e. did it fail first in the gauge length or at the tab). Overall there seems to be no discernible pattern to the failure types encountered in the study. It is likely that many of the in-tab-failure types could have been avoided with a lowered clamp pressure, but such a move would likely cause an increase in GAT failures (adhesive failure in tab, i.e. pull-out). Of the 0° specimen, all failure types at lateral either near tab or in the gauge length.

5.1.2 DIC

The DIC results from the 0° test (figure 4.13) show the sample vertical strain in

sub-figure (a),(c), and (d), and displacement in sub-figure (b),(d) and (f), all as

40 CHAPTER 5. DISCUSSION

functions of cycle count. The image sequence for each sample is analysed against a reference image of the sample at no load, to which all the points in the data is compared. These points represent displacement and strain at peak loading over the course of testing.

The observed trend for all three specimen is that of high initial stiffness loss until a cycle count on the order of 10

-10

cycles, after which the development slows almost to a halt (as is the case for 1-FA-6), or slows down considerably (as in the case of 1-FA-3 and 5). A change in the strain trend immediately prior to failure was hypothesised but not conclusively measured.

The initial peak strain in each sample vary significantly. Despite the load level being the same (80 % of UTS), 1-FA-3, 5 and 6 have initial peak strain values of 0.96, 1.05, 0.72 % respectively. The initial peak strains for 1-FA-3 and 5 seem reasonable considering that the failure strain found in static testing is 1.18 %, whereas the initial strain for 1-FA-6 seems a bit low, at 0.72 %.

The three specimen had a fatigue life of roughly 785, 412 and > 5000 ∗10

cycles, respectively. 1-FA-5 showed the greatest initial strain of the three and also had the shortest fatigue life. 1-FA-3 is an intermediary, whereas 1-FA-6 showed both a very low initial peak strain and a fatigue run-out at 5 million cycles. Considering the strain development after the initial phase, both 1-FA-3 and 5 show weak to moderate positive trends, whereas 1-FA-6 seems to have stopped elongating, which is consistent with the fatigue life observation of the same specimen. 1-FA-5 both began at higher initial points and increased more rapidly in the later phase.

Camera capture for 1-FA-6 was halted at 2 ∗ 10

cycles, but the test continued.

Displacement readout for 1-FA-6 from the machine matches the DIC data for strain reasonably well for the first 2 ∗ 10

cycles, and continues until the test was halted at 5∗10

cycles (see fig. 4.17 and with strain superimposed in fig. 4.16) . The initial stiffness difference indicates that the either the material in 1-FA-6 was much stiffer than that of 1-FA-5 and 3, or that there was an error in machine load calibration, camera capture, image analysis, or that the measurement of the cross sectional area was performed or calculated erroneously, or a combination of all of the above.

The 90° specimen show only a minuscule initial stiffness loss, and no clear trend

from there on due to excessive spread in results. These specimen are challenging to

test due to their low yield strength and stiffness. It may be argued that the process

of measuring an average displacement is less accurate in a 90°-test considering

CHAPTER 5. DISCUSSION 41

that the matrix material dominates in this direction, meaning that the sam- ple is less homogeneous in its tested direction, leading to uneven strain distribution.

Several of the ramp tests of the 0°-specimen had timing as well as camera settings issues (capture number was too high, but this was later solved in post-processing by decimation), and there were also a number of fatigue tests in which the image cap- ture sequence missed the peak entirely. This fault could have been easily rectified by increasing the image count, or increasing the precision of the trigger software.

The DIC implementation, with its current trigger, peak finding software and general practical setup, is capable of tracking average strains over time in a way that is practical (compared to using an extensometer for the same task), but with a significant room for improvement (see section 6.2).

5.1.3 Micrography

The produced micrographs (see section 4.6) were all of a transverse cross-section of the specimen (i.e. fiber direction out of the image), and the only observable damage was longitudinal splitting of the specimen near the corners of the cross- section. It was expected to at least find some evidence of void nucleation in the matrix, but this could not be identified in 1-FA-2. A possible result of the choice of cross section, as voids were expected to form as cracks propogating transversely.

Ideally a longitudinal section would be made, but due to time constraints and the relative labour and time intensiveness of the process this was not performed.

5.1.4 Sources of error and uncertainty

1. Fatigue testing and DIC Capture

• It may be argued that the measurement setup is prone to low accu- racy due to its sensitivity camera-specimen misalignment. Out of plane movement as shown in fig. 2.8 may easily give rise to psuedo-strains that are relatively large to the small strains measured post ramp, for example. The results do however indicate high precision.

• Low camera resolution, speckle pattern inconsistency, and variation in lighting quality causes variation, noise, in the measured displacement.

• The testing has been performed under the assumption that the fatigue

machine is holding constant maximum load - and machine output pro-

duced by the same machine does not suggest otherwise - but any drift

or fluctuation in the load on the specimen would cause a different dis-

placement and subsequently affect the resulting DIC analysis.

42 CHAPTER 5. DISCUSSION

• The clamping of the specimen introduces high compressive loads in the tab area that is conductive to near-tab failure, potentially shortening the fatigue life.

• The ability of the system to capture the instantaneous peak displace- ment is instance limited by the capture frame-rate of the camera, which effectively amounts to resolution in the time domain. 1000 fps at 5 Hz means 200 images per cycle. The actual ultimate displacement may consequently fall between two images (whereas the next capture may not), introducing noise in the final result given in section 4.5.

2. Material defects and specimen production

• As previously mentioned the fatigue performance is greatly affected by the presence of irregularities in the test material, be that fibre mis- alignment, void content, improper thermal treatment, and so forth (see section 2.1 and section 5.1.1).

• The specimen production produces two new surfaces (the specimen sides), with uneven surface quality (see section 3.3), potentially cre- ating crack initiation points, shortening fatigue life.

3. Post processing and final analysis

• The analysis process itself, as well as averaging full field strains intro-

duce a degree of uncertainty in the results. The displacement measure-

ment involves filtering out non-viable or outlandish results - inability to

match subsets, and measurement artefacts that produce more displace-

ment than intended.