Department of Aeronautics

Modelling the mechanical response of multiphase recycled CFRP

Maciej Lembke maciej.lembke09@imperial.ac.uk

Supervisors: Soraia Pimenta Silvestre T. Pinho

London, February 2010

Table of Contents
Table of Figures .............................................................. - 3 Table of Tables ................................................................ - 4 Abstract ......................................................................... - 5 1. 2. Introduction ............................................................. - 6 Literature review ....................................................... - 9 -

2.1. Influence of fibre bundles on the response of the materials ......................................................................... - 9 2.2. 3. 3.1. 3.2. Crack propagation techniques in Abaqus ............... - 12 Choice of the modelling technique ........................ - 13 Modelling properties ........................................... - 14 Modelling and Numerical Analysis .............................. - 13 -

3.2.1 Geometrical properties ..................................... - 14 3.2.2 Physical properties ........................................... - 15 3.2.3 Numerical properties ........................................ - 17 3.2.4 Element type .................................................. - 18 3.3. Different approaches to modelling the idealised rCFRP model............................................................................ - 18 3.3.1 2D single-instance model ................................. - 18 3.3.2 2D model with cohesive elements modelling interface..... ................................................................ - 23 3.3.3 2D cohesive contact model ............................... - 27 3.3.4 3D model ....................................................... - 31 3.4. Mesh sensitiveness ............................................. - 33 4. 2D cohesive contact model approach for different geometries.. ..................................................................... - 35 4.1. 4.2. 5. Geometry of the fibre bundle ............................... - 35 Inclination of the fibre bundle .............................. - 38 -

Conclusions and further work.................................... - 41 -

Appendix A: Fractals as a tool for analysis of materials properties......................................................................... - 43 References ................................................................... - 46 -

-2-

Table of Figures
Figure 1 rCFRP microstructure (bundle highlighted)..................- 6 Figure 2 Bundle fracture.......................................................- 7 Figure 3 In-plane geometry and dimension of the 2D model .... - 14 Figure 4 In-plane geometry and dimensions of the 3D mode........... .......................................................................................... - 15 Figure 5 Boundary conditions .............................................. - 17 Figure 6 Crack propagation arrest before the bundle (von Mises avg. stresses) .............................................................................. - 19 Figure 7 Convergence problems after cracking most of the bundle (von Mises avg. stresses) ....................................................... - 20 Figure 8 Damage of crack-path elements (STATUSXFEM status (damaged, failed) of the XFEM elements) ................................. - 20 Figure 9 ETOTAL (total energy) and ALLSD (dissipated energy) for 2D single instance model........................................................ - 21 Figure 10 Magnification of a vertical drop of energy for 2D single instance model ..................................................................... - 21 Figure 11 Crack propagation in a thin-bundle model. a) crack propagates towards the bundle; b) Failed matrix elements before and after the bundle; c) completely failed bundle; (STATUSXFEM)... .. - 22 Figure 12 Energy plot for thin-bundle model ......................... - 22 Figure 13 Layer of cohesive elements around the bundle (SDEG degradation (damaged, failed) of cohesive elements) ................. - 23 Figure 14 Crack propagating towards the bundle (von Mises avg. stresses) .............................................................................. - 24 Figure 15 Crack propagation through the bundle (von Mises avg. stresses) .............................................................................. - 25 Figure 16 Damaged cohesive elements (SDEG) ..................... - 26 Figure 17 Model behaviour with additional precracks .............. - 26 Figure 19 Crack propagating towards bundle ......................... - 27 Figure 18 Partially failed interface ........................................ - 27 Figure 20 Fractured bundle (STATUSXFEM) ........................... - 27 Figure 21 Pulled-out bundle (von Mises avg. stresses) ............ - 27 Figure 22 Delamination of the matrix-bundle interface case I weakened interface (CSDEG degradation (damaged, failed) of cohesive contact) .................................................................. - 28 Figure 23 Delamination of the matrix-bundle interface case II weakened interface (CSDEG) .................................................. - 29 -

-3-

Figure 24 Delamination of the matrix-bundle interface case III weakened interface and bundle (CSDEG) .................................. - 30 Figure 25 Pull-out of the bundle embedded into the matrix ..... - 31 Figure 26 Pull-out of the bundle positioned at the surface of the specimen ............................................................................. - 31 Figure 27 Cross-section of a 3D plate ................................... - 32 Figure 28 3D fibre bundle ................................................... - 32 Figure 29 Cracked 3D model ............................................... - 32 Figure 30 Non-expected crack-path (von Mises avg. stresses)......... .......................................................................................... - 33 Figure 31 Too coarse mesh causing stress concentration (von Mises avg. stresses) ....................................................................... - 33 Figure 32 2D cohesive contact model for different mesh sizes (CSDEG) .............................................................................. - 34 Figure 33 Crack propagation for baseline geometry (CSDEG, von Mises avg. stresses) .............................................................. - 35 Figure 34 Crack propagation for a thick bundle ...................... - 36 Figure 35 Crack propagation for a thin bundle (CSDEG, von Mises avg. stresses) ....................................................................... - 36 Figure 36 Crack propagation for a long bundle (CSDEG, von Mises avg. stresses) ....................................................................... - 37 Figure 37 Crack propagation for a short bundle (CSDEG, von Mises avg. stresses) ....................................................................... - 37 Figure 38 Crack propagation for bundle inclined at 10 (von Mises avg. stresses, CSDEG) ........................................................... - 39 Figure 39 Crack propagation for bundle inclined at 30 (von Mises avg. stresses, CSDEG) ........................................................... - 39 Figure 40 Crack propagation for bundle inclined at 45 (von Mises avg. stresses, CSDEG) ........................................................... - 40 -

Table of Tables
Table 1 Material properties of the bundle .......................... - 16 Table 2 Material properties of the matrix with applied XFEM - 16 Table 3 Material properties of matrix-bundle interface ........ - 16 Table 4 Step properties .................................................. - 17 Table 5 Interface properties ............................................ - 18 -4-

Abstract
Technologies for recycling carbon-fibre reinforced-polymers (CFRP) have been developed. The mechanical properties of the recyclates show they can be used in non-critical structural applications, e.g. in secondary structures of aircraft and automobiles. The presence of fibre bundles, groups of aligned fibres held together by residual virgin-matrix, is very common. It was already verified experimentally that these bundles have a significant influence on the mechanical properties of the rCFRP and that they add a considerable degree of complexity to the failure mechanisms. The objective of this project was to develop numerical models of idealised rCFRP, focusing on the failure mechanisms in the presence of fibre bundles bundle pull-out and bundle fracture. Different models were created in order to meet the objectives, both 2D and 3D. In all models, Extended Finite Element Method (XFEM) was used to model crack propagation. Several different tools were tried to model the interface, namely cohesive elements and surfaceto-surface cohesive contact. A 2D cohesive contact model that meets the objectives was developed; using this model, a study of the influence of bundle geometry and position on the failure mode was run. A 3D model based on the previous 2D model was also developed. There are strong premises that, after further work, these models can be a very good representation of rCFRP.

-5-

1. Introduction
Technologies for recycling carbon-fibre reinforced-polymers (CFRP) have been developed in the past years and are now mature. Generically, the process involves two main stages. The first one is reclaiming the carbon-fibres (CF) from an existent CFRP. It is done by matrix removal, either from an end-of-life part or manufacture scrap. The second step is the re-impregnation of the recycled carbon-fibres (rCF) with new resin, producing a recycled (r-) CFRP. The mechanical properties of the recyclates show they can be used in non-critical structural applications, e.g. in secondary structures of aircraft and automobiles. For rCFRP to be used in real structural applications, it is necessary to understand its mechanical behaviour and to develop design methods. However, recycled composites are considerably different from their virgin precursors: the properties of the CF undergo (little) degradation during the reclamation, and more important the typical architecture of the rCFRP is very peculiar [Figure 1].

Figure 1 rCFRP microstructure (bundle highlighted) The presence of fibre bundles groups of aligned fibres held together by residual virgin-matrix which was not entirely removed -6-

during the recycling process in addition to individual fibre filaments dispersed within the resin this dispersed composite phase will be hereby referred as matrix confers a multiphase / hierarchical / fractal characteristic to the microstructure of the recyclates. For information about fractals as a tool for analysis of materials properties see Appendix A. It was already verified experimentally that these bundles have a significant influence on the mechanical properties of the rCFRP [1]. They can increase the local toughness of the material up to 3 times. They also add a considerable degree of complexity to the failure mechanisms [Figure 2].

Figure 2 Bundle fracture The overall aim of this project is to develop numerical models of idealised microstructures of rCFRP, focusing on the failure mechanism in presence of fibre bundles. Models are expected to reproduce the two basic behaviour templates: fibre bundle fracture and fibre bundle pull-out. Fibre bundle fracture can be observed when, after crack propagation through the bundle, it is broken into two parts, with its crack-path generally in-line with the matrix crack-path. Fibre bundle pull-out can be observed when, after crack -7-

propagation through the bundle, it is not fractured but part or the entire bundle slides out of its original position; such behaviour is caused by the failure of matrix-bundle interface. This report opens with chapter 2 which covers a literature review: to select a suitable modelling technique for the rCFRP, it was necessary to understand the mechanical response of materials in presence of fibre bundles. Comparison of different crack propagation methods available in Abaqus 6.9 is also presented. The development of the numerical model turned out to be much more challenging than initially considered. Chapter 3 explains the development and discusses over the results presented by four different modelling strategies. The following chapter 4 presents the differences in crack propagation for different geometries; both geometry and inclination of the bundle are changed. The model used in this chapter is a best performing 2D model from chapter 3. The project is summarised by chapter 5 covering conclusions and recommended future work.

-8-

2. Literature review
2.1. Influence of fibre bundles on the response of the materials
In fibre composites every component of the material and its production is very important. High-performance fibres, matrix which not only binds the material, but also transfers stresses, also the way the two components are connected highly influences the way the material behaves. By choosing suitable fibres we can obtain very tough composites. With recycled CFRP everything is more complex than with the virgin material. Length and orientation of the fibres is not uniform. In rCFRP we find many features specific of the recyclates: variable fibre length, broken fibre segments, fibre waviness, fibres aggregated in bundles and high void content [1]. All of these features influence the mechanical behaviour of the material in different ways. In this paper, the influence of fibre bundles will be analysed. Presence of bundles results in reduced distance between fibres within the composite, local fibre alignment, end synchronisation, increase in the variation of local volume fraction over the composite, and presence of resin-rich areas [2]. Mulligan et al. [2] reviews models for toughness of a material with bundles using different characteristics as a main factor e.g. local change in volume fraction or aspect ratio of the bundle. According to Piggott [3], in cases when the elastic work is similar to the work of fracture, a low aspect ratio helps reducing the amount of elastic stress transfer and increases toughness. He also points out that for a low aspect ratio even when a fibre bundle is two times longer than its critical length (xo), still more than half of the fibres will pull out, by what they still contribute to fracture toughness. -9-

Piggott [3] presented this equation for critical fibre length:

x 0 - critical length
d - fibre diameter

u - tension strength
s - strength transfer
y - shear strength

x0

d ( u s ) 2 y

This will make 2x0 the best fibre length for maximum fracture toughness. If the crack propagates further the stresses between fibres and matrix will break the fibres that do not pull out [4]. Piggott [3] noticed also that during fibre-bridging, the maximum stress appears on the plane of the crack, making it the place where the strongest fibres, which are not pulled out, will finally fail. Trapeznikov [5] considered the influence of the number of fibres within the fibre bundle and noticed that the fewer fibres in the bundle, the smaller stress is required to pull them out, which leads to failure. From that he concluded that the number of fibres per bundle is counter proportional to the strength of the material. Piggott [3] developed an expression for fracture toughness, in which elastic stress transfer from matrix to the fibre is not taken into account. It is derived based on the examination of the process, where fracture of the fibre causes plastic flow in the matrix while relaxing. The same article presents an equation for the value of fracture surface energy and toughness of composites however, there are limitations that apply to these equations. Elastic stress transfer near cracks for fibres bridging the cracks should be small compared with stresses transferred by plastic flow at the interface between matrix and fibres. It is also worth noting that fracture energy is much more

- 10 -

dependent on the properties of the fibre than on those of the matrix [6]. Kim and Mai [7] suggested that, depending on the fibre volume fraction and impregnated material, the fracture toughness of a bundled material may increase up to 100% while maintaining its flexural strength, compared to composites without bundle impregnation. Kim and Mai [7] based their suggestions on studies of a composite, in which fibre bundles were impregnated with polymer before they were connected with the matrix in a range of small volume fraction. The process of impregnating fibres into the matrix is very important: it is best to use the matrix with shear flow properties. When the process is strictly controlled and the bonding is effective, a much better toughness can be achieved. According to Fila et al. [4], the fracture toughness is linearly proportional to the volume fraction of fibre bundle. Fibre bundling also significantly improves the stability of cracking. It is also worth mentioning that shear deformation in the fibre bundle is often not taken into account when calculating fracture toughness [4]. Fila et al. [4] describes how one of the abilities of bundles is to toughen significantly very brittle materials. The degree of their influence is not only dependant on their strength. It can also increase with their diameter or decrease with the modulus, which is counter proportional to fibre cluster volume fraction [2]. Large diameter fibres (as bundles are often considered [2, 6, 7]) are likely to significantly increase fracture toughness of the composite. This is caused by increased shear stresses transferred to the matrix, which are proportional to the diameter of the fibre [6, 7]. Additionally, Piggott [6] concludes that fibre diameter should be maximized to get the largest value of toughness only for moderately ductile and brittle fibres, while length of the fibre should be as large as possible for ductile fibres to achieve the same effect.

- 11 -

2.2. Crack propagation techniques in Abaqus

There are four main techniques for modelling crack propagation available in Abaqus 6.9: progressive damage models, cohesive elements method, contour integral estimates, and Extended Finite Element Method (XFEM) [8]. Progressive damage models are mainly used for ductile materials or UD composites. They offer a general capability for modelling progressive damage and failure of the material. After damage initiation, the material stiffness is degraded progressively according to the specified damage evolution response. The progressive damage models allow for a smooth degradation of the material stiffness, which makes them suitable for both quasi-static and dynamic situations. The cohesive elements method uses cohesive elements to model crack propagation. It requires the estimation of possible crack-paths at the stage of modelling. A relatively very thin line of cohesive elements needs to be placed in all the places that the crack is expected to propagate throuthg. The crack propagation analysis consists of analysis of failed or damaged cohesive elements on the crack-path. Contour integral estimates are used to study damaged initiation in quasi-static problems. They cannot be used to predict crack propagation. The Extended Finite Element Method (XFEM) is used to study both damage initiation and crack propagation. The crack-path does not need to be predicted as crack propagation is found by the method itself. XFEM uses the partition of unity framework to model strong and weak discontinuities independent of finite element mesh.

- 12 -

3. Modelling and Numerical Analysis

3.1. Choice of the modelling technique
The idealised rCFRP consists of four parts: matrix, which is an homogenised representation of the composite phase of individual fibres dispersed within the polymeric resin; bundle, which is an homogenised representation of the composite bundle; matrix-bundle interface, which connects the bundle to the matrix; precrack, which is used to initiate failure. The model should be able to model crack propagation from the direction of the precrack towards the bundle while loaded in tension. The crucial part is to capture the behaviour of the crack in the area of the bundle. There are two possible ways in which the bundle can behave. The first is bundle fracture; in this case, the crack propagates through the matrix, then fractures the bundle, and continues cracking the matrix afterwards. The other option is for the bundle to pull-out; in that case, the crack similarly propagates from the precrack towards the bundle; the next step, instead of bundle fracture, is the delamination of the matrix-bundle interface; after that, the bundle is only partly attached to the matrix and the crack propagates further through the matrix. The crack propagation modelling technique chosen for all models is the Extended Final Element Method (XFEM). It has an advantage over other available techniques. Contour integral cannot be used to predict crack propagation. Progressive damage model can only be used for ductile materials or UD composites and cannot represent a sharp crack. Cohesive element method requires a crack-path to be predicted a priori. XFEM enables sharp crack propagation, allows

- 13 -

new cracks to appear in all areas of the model where XFEM is applied to, and enables crack propagation in all directions.

3.2. Modelling properties 3.2.1 Geometrical properties

3.2.1.1 Two-dimensional (2D) model All the 2D models are planar and deformable with plane stress, thickness of 1mm. Figure 3 represents the in-plane geometry and dimensions.

Figure 3 In-plane geometry and dimension of the 2D model

- 14 -

3.2.1.2 Three-dimensional (3D) model

Figure 4 In-plane geometry and dimensions of the 3D model

3.2.2

Physical properties

Table 1 to Table 3 present the material properties of bundle, matrix and bundle-matrix interface. Table 4 and Table 5 present numerical properties of the step and the matrix-bundle interface. 3.2.2.1 Material properties All the bundle properties were calculated based on data from [9], considering the bundle as a UD composite with fibre volume fraction of 65%. All matrix and interface properties were taken from [9], and correspond to experimental testing of a rCFRP.

- 15 -

Table 1 Material properties of the bundle Symbol E1 E2 Nu12 G12 G13 G23 X G Value 142700 8900 0.33 4000 4000 1200 2704 100 Unit MPa MPa MPa MPa MPa MPa kJ/m2 Property Young modulus in direction x Young modulus in direction y Poissons ratio Shear modulus in direction x Shear modulus in direction y Shear modulus in direction z Maximum principal stress Fracture Energy / Toughness

Table 2 Material properties of the matrix with applied XFEM Symbol E1 E2 Nu12 G12 G13 G23 X G Value 28100 16000 0.42 7000 1200 1200 194.5 2.79 Unit MPa MPa MPa MPa MPa MPa Property Young modulus in direction x Young modulus in direction y Poissons ratio Shear modulus in direction x Shear modulus in direction y Shear modulus in direction z Maximum principal stress

kJ/m2 Fracture Energy / Toughness

Table 3 Material properties of matrix-bundle interface Symbol E/Knn G1/Kss G2/Ktt G Value 2810 700 120 194.5 86.9 86.9 0.2 Unit GPa GPa GPa MPa MPa MPa kJ/m2 Young modulus Shear modulus in direction x Shear modulus in direction y Maximum nominal stress in normal direct. Maximum nominal stress in shear direct. 1 Maximum nominal stress in shear direct. 2 Fracture Energy / Toughness - 16 Property

3.2.2.2 Boundary conditions The displacement in direction x is applied on the top right part of the model as shown on the figure. The top left part of the model is fixed in x direction. Both left and right parts are fixed in direction y [Figure 5]. The maximum displacement applied was 0.67mm.

Figure 5 Boundary conditions

3.2.3

Numerical properties

The analyses were run in implicit, static mode. The most important step properties are presented in Table 4. The numerical properties of the interface are presented in Table 5. Table 4 Step properties Value 0.0002 - 0.2 20 0.025 Property Damping dissipated energy fraction Maximum number of iterations per increment Convergence criterion for the ratio of the largest residual to the corresponding average flux norm for convergence (5 times larger than the default value) 0.05 Convergence criterion for the ratio of the largest solution correction to the largest corresponding incremental solution value (5 times larger than the default value) - 17 -

Table 5 Interface properties Symbol Knn Kss Ktt Value 0.0027 281000 500 1000 Property Tolerance for adjustment zone Stiffness in normal direction Stiffness in shear direction 1 Stiffness in shear direction 2

3.2.4

Element type

All 2D models have a structured mesh, with each element approximately 0.15mm large. The element type used is CPS4I; it is a linear Quad element shape with incompatible modes to prevent hourglassing. In majority of the 3D models the mesh size was 0.45mm. Parts of the fibre bundle were meshed using Hex element shape with a sweep (medial axis) technique of meshing. The element type is C3D8I: An 8-node linear brick. It is a Hex element shape with a structured technique of meshing. In all models Incompatible modes were used.

3.3. Different approaches to modelling the idealised rCFRP model 3.3.1 2D single-instance model

The 2D single-instance model is a very simple model. Only one part/instance is used. It has a sketched bundle shaped area that has different material properties assigned. No specific features are used to model the bundle-matrix interface. When the crack approaches the bundle, the propagation is arrested because of rapid change of material properties [Figure 6]. Both fracture energy and maximal principal stress of the bundle are - 18 -

much bigger than those of the matrix. Crack is arrested for the second time when it is propagating back into the matrix [Figure 7], the reason this time are problems with convergence.

Figure 6 Crack propagation arrest before the bundle (von Mises avg. stresses)

- 19 -

Figure 7 Convergence problems after cracking most of the bundle (von Mises avg. stresses) The first problem with the model appears at that point. When the crack is about to entirely fracture the bundle, the job aborts due to too low increment time needed for further iterations. Figure 8 presents how elements on the crack-path below the bundle are damaged, but the job fails to converge before the bundle fails completely.

Figure 8 Damage of crack-path elements (STATUSXFEM status (damaged, failed) of the XFEM elements) - 20 -

After analysis of the Total Energy plot [Figure 9][Figure 10], it is observed that the energy drops vertically, hence the very low increment time requested by the job.

Figure 9 ETOTAL (total energy) and ALLSD (dissipated energy) for 2D single instance model

Figure 10 Magnification of a vertical drop of energy for 2D single instance model

The only reason for such behaviour is a physical instability: the next increment would cause the bundle to fail completely and, with the amount of accumulated energy, the entire plate would also fail. The confirmation of that is a different model. It is very similar to the one described above with only one difference: the width of the bundle is two times smaller. Figure 11 shows how the bundle fractures completely and the crack propagates. In spite of similarities, the model with the thinner bundle behaves differently, because the elastic energy stored in the plate is not so high. This allows it to be more stable and converge.

- 21 -

a)

b)

c)

Figure 11 Crack propagation in a thin-bundle model. a) crack propagates towards the bundle; b) Failed matrix elements before and after the bundle; c) completely failed bundle; (STATUSXFEM) In Figure 12, we can see the Total Energy plot of the model with a thin bundle. A very similar effect of a sudden energy drop appears, only that in this case it is not vertical but very steep.

Figure 12 Energy plot for thin-bundle model This problem can be minimised with launch of Abaqus 6.10, which will enable XFEM in implicit dynamics and explicit code. This will - 22 -

allow models to capture the dynamic effect which is the cause of the problem. The second problem with the single-instance model is that it is not able to model the pull-out of the fibre bundle. The reason for this is XFEM not being able to effectively make a 90 turn at the point of touching the bundle and therefore following the interface. In order to make pull-out possible, the next model was developed.

3.3.2 2D model with cohesive elements modelling interface

This model was designed to enable fibre bundle pull-out. For this to be possible, a very thin layer (0.01mm thickness) of cohesive elements was added around the bundle [Figure 13]. Cohesive elements are modelling the interface between fibre bundle and matrix, which enables the pull-out without directly damaging any of them. The main problem with the model is crack propagation through the interface. The cohesive elements can only open in one direction. As in this case they are to model the interface, they are set to open in the direction normal to bundle outline; for that reason, crack propagation across a cohesive element is not possible. In order to deal with this problem, three different approaches were taken.

Figure 13 Layer of cohesive elements around the bundle (SDEG degradation (damaged, failed) of cohesive elements) - 23 -

3.3.2.1 Cohesive element on the crack path This model has a continuous layer of cohesive elements around the fibre bundle. As XFEM cannot be applied to cohesive elements, it is impossible for the crack to propagate into the bundle [Figure 14]. As result a pull-out is also impossible, because for it to happen all the elements between the precrack and the bundle must be failed.

Figure 14 Crack propagating towards the bundle (von Mises avg. stresses) 3.3.2.2 Matrix element on crack-path The next approach was to substitute the two cohesive elements on the predicted crack-path with two matrix elements . This enabled crack propagation across the interface and through the bundle [Figure 15].

- 24 -

Figure 15 Crack propagation through the bundle (von Mises avg. stresses) The fibre bundle pull-out was still not possible. The replaced elements cracked vertically, so a horizontal crack, which would correspond to delamination, was not possible. This restriction is caused by the numerical possibilities of XFEM (only one crack per element allowed).

3.3.2.3 No element on crack-path The last approach taken to this model was to delete the two cohesive elements on the predicted crack-path. Here again, similarly to the model with cohesive elements on the crack-path, problems appeared on the stage of crack getting near the bundle. An obvious problem with this model is present before starting the analysis. The lack of the cohesive elements is not realistic, even though the elements are relatively small. This causes cohesive elements to fail in an unexpected location and in an unexpected mode. Although some hope for a failure of cohesive elements appear as the crack draws nearer to the bundle [Figure 16], the model has problems with further converging.

- 25 -

Figure 16 Damaged cohesive elements (SDEG) Even with help of additional precracks in the area of the deleted elements, the model does not produce results it is expected to [Figure 17].

Figure 17 Model behaviour with additional precracks All the models where cohesive elements were modelling the interface between fibre bundle and matrix have not brought much improvement. As with the 2D single instance model, it was possible to model the bundle fracture but still impossible to model a bundle pull-out. The most promising results came from the model with matrix like elements substituted for cohesive elements. It was possible to see the model leaning towards the pull-out, but unfortunately, because of the models structure, it was impossible to completely do so. Based on all that, it is safe to conclude that the idea to model an interface between bundle and matrix turned out to be good. - 26 -

3.3.3

2D cohesive contact model

This model, in contrary to the two models described above, has two separate instances for the bundle and the matrix plate. The interface between these two elements is modelled with use of surface-to-surface cohesive contact [Figure 18][Figure 19]. 1819

Figure

18

Partially

failed

Figure 19 Crack propagating towards bundle

interface

This type of interface solved the problems with crack propagation that were present in the previous model. Modelling of both bundle fracture [Figure 20] and fibre bundle pull-out [Figure 21] was possible, by changing the properties of the bundle and/or interface.

Figure

20

Fractured

bundle Figure 21 Pulled-out bundle (von Mises avg. stresses)

(STATUSXFEM)

- 27 -

Figure 22

to Figure 24 present different sequences of events of

matrix-bundle interface failure.

a) The first degradation of the b) Delamination propagates up interface appears near the left and around in the direction of the end of the bundle main crack

c) After almost half of the bundle d) Crack propagates downwards is delaminated a new crack appears near the left end of the bundle Figure 22 Delamination of the matrix-bundle interface case I weakened interface (CSDEG degradation (damaged, failed) of cohesive contact)

- 28 -

a) The first degradation of the b) Another delamination begins interface appears near the left near the crack tip end of the bundle

c)

Two

delamination

are d)Almost half of the bundle is delaminated

propagating towards each other

Figure 23 Delamination of the matrix-bundle interface case II weakened interface (CSDEG)

- 29 -

a) Bundle is damaged, but not b) Only after bundle is damaged fractured the delamination begins

c) Delamination proceeds around d)New cracks appear after the the bundle bundle is pulled out

Figure 24 Delamination of the matrix-bundle interface case III weakened interface and bundle (CSDEG) The 2D two-instance model enables visualisation of both pull-out and crack propagation through fibre bundle. There is only one problem with this model apart from difficulties with convergence. The path of the crack after either pulling-out or breaking the bundle is on the side of the bundle. The cause of such behaviour is the nature of the 2D planar model. The experimental results behave in the same way only when the bundle is on the surface of the specimen [Figure 25]. When the bundle is embedded within the matrix, the crack should propagate after the bundle in the same direction it propagated towards the bundle, either fracturing or - 30 -

pulling it out [Figure 26]. In order to achieve crack continuity a new model had to be created. 25 26

Figure Figure 25 Pull-out of the bundle bundle positioned at the surface of the matrix specimen

26 Pull-out of embedded into

the the

3.3.4

3D model

This model was developed to ensure the crack continuity around the bundle. It is very similar to the 2D two-instance model. There are two different elements: a fibre bundle [Figure 27] and a plate with a bundle-shaped hollow space [Figure 28]. The two are also connected with surface-to-surface cohesive contact. 27 28

- 31 -

Figure 27 3D fibre bundle

Figure 28 Cross-section of a 3D plate

As predicted, the 3D model enabled crack propagation around the bundle and bundle pull-out [Figure 29].

a) Crack propagating around the b) Cross-section of a half-cracked bundle, pulled-out stresses) Figure 29 Cracked 3D model There are some problems with the model. One of them is a difficult convergence. There are two causes of this problem. The first is the current limitations of the XFEM tool [Figure 30]. The second reason is a too coarse mesh [Figure 31]. Unfortunately, when the mesh is finer, the jobs become much more time-consuming and powerdemanding. which (von is Mises being plate (STATUSXFEM) avg.

- 32 -

Figure 30 Non-expected crack-path (von Mises avg. stresses)

Figure

31

Too

coarse

mesh

causing stress concentration (von Mises avg. stresses)

Basing on the 2D two-instance model performance, it is safe to assume that this 3D model is also able to model a fibre bundle pullout [Figure 29]. When combined with the crack continuity that the 3D approach delivers, it presents an interesting possibility to be a very accurate failure model of idealised rCFRP.

3.4. Mesh sensitiveness

To validate the 2D cohesive contact model, it was run with different mesh sizes, and the results of these jobs were compared. Five different mesh sizes were used: 0.45mm, 0.3mm, 0.15mm, 0.075mm.

- 33 -

a) Mesh size 0.45mm

b) Mesh size 0.3mm

c) Mesh size 0.15mm

d) Mesh size 0.075mm

Figure 32 2D cohesive contact model for different mesh sizes (CSDEG) The resemblance between different meshes is clearly visible in Figure 32. The variations between directions of crack propagation depend on small deviations of the point in which the crack propagating from the top touches the fibre bundle. Also on the force/displacement graphs the same trends are present.

- 34 -

4. 2D cohesive contact different geometries

model

approach

for

Different modelling techniques of idealised rCFRP were shown in the previous chapter. The best-performing one was the 2D cohesive contact model. In this chapter, it is going to be used to study the influence of both fibre bundle geometry and inclination on crack propagation.

4.1. Geometry of the fibre bundle

The baseline geometry of the bundle is presented on Figure 3. It is 2.3 mm long and the bundle is 0.7 mm thick. Figure 33 presents a model with baseline geometry of the fibre bundle. After the crack propagated to the bundle, the matrix-bundle interface delaminated at both ends of the bundle and new cracks started appearing in the matrix below the bundle, near both ends.

Figure 33 Crack propagation for baseline geometry (CSDEG, von Mises avg. stresses) Model presented on Figure 34 has doubled bundle thickeness now it is 1.4 mm. In this case, crack propagated by delaminating the matrix-bundle interface, which was expected. If the bundle was pulled-out for the baseline-geometry model, then it should do the same when the thickness of the bundle is larger. - 35 -

Figure 34 Crack propagation for a thick bundle (von Mises avg. stresses) The next model was changed in an opposite way the bundle was made thinner (0.35 mm bundle thickness). In this case, fibre bundle pull-out was also present [Figure 35]. This was not as likely as in the case of the thick bundle: the bigger the length to thickness ratio, the more prone to fracture the bundle should be.

Figure 35 Crack propagation for a thin bundle (CSDEG, von Mises avg. stresses) The next presented model has doubled value of length to 4.6 mm, and thickness left at the baseline value of 0.7 mm. Figure 36 presents neither fibre bundle pullout nor fibre bundle fracture, because the model had problems to converge further. Nevertheless - 36 -

a prediction can be made based on this figure; as the matrix-bundle interface is not damaged and high stress concentrations appear in the middle of the bundle (and not at its ends), it can be predicted that bundle fracture is the most likely mechanism for this model. As written before, this pattern of behaviour is expected from models with high length to thickness ratio of the fibre bundle.

Figure 36 Crack propagation for a long bundle (CSDEG, von Mises avg. stresses) In the next model, length has been made two times smaller than the baseline value (1.15 mm). Results from this model are presented on Figure 37. As expected, bundle pull-out is present in this case.

Figure 37 Crack propagation for a short bundle (CSDEG, von Mises avg. stresses) - 37 -

Drawing conclusion from the presented four cases, it is suggested that bundle length has more influence on crack propagation behaviour than bundle thickness. Changes in fibre bundle length translate directly to change of the amount of interface that needs to fail when loaded in shear in order for the bundle to pull out. Changes in fibre bundle thickness translate to change of strength of the bundle which can prevent or allow a bundle fracture. Such geometrical dependencies are only valid for 2D model: in case of a 3D bundle, a change of length or thickness (diameter) translates to change of both volume of the bundle and area of interface.

4.2. Inclination of the fibre bundle

The baseline angle of inclination of the longitudinal axis of the bundle is 0 [Figure 3, page - 14 -]. As stated above, the crack propagation behaviour for this type of geometry is the following: after the crack propagated to the bundle, the matrix-bundle interface delaminated at both ends of the bundle and new cracks started appearing in the matrix below the bundle, near both ends. [Figure 33]. The inclination of the longitudinal axis of fibre bundle in the model presented in Figure 38 is equal to 10. The bundle pulled-out and the crack propagated further into the matrix, going around the bundle upwards.

- 38 -

Figure 38 Crack propagation for bundle inclined at 10 (von Mises avg. stresses, CSDEG) The next model presents an inclination of the longitudinal axis of the bundle of 30 [Figure 39]. Because of convergence problems, it is not obvious what happens with the crack. Based on failure of interface on one end of the bundle and little stress concentration in the middle of the bundle, it is safe to assume that the bundle pulls out. Unfortunately it is impossible to state on which side of the bundle the crack propagated further in the matrix.

Figure 39 Crack propagation for bundle inclined at 30 (von Mises avg. stresses, CSDEG) Figure 40 presents a model with an inclination of the longitudinal axis of the fibre bundle of 45. Similarly to the previous two models, also in this model the bundle pulls out. Unfortunately, again due to convergence problems the model was not able to complete - 39 -

the analysis. Luckily enough, it is possible to state that delamination is propagating towards the lower end of the bundle and a new crack in the matrix started appearing also at the lower end of the bundle.

Figure 40 Crack propagation for bundle inclined at 45 (von Mises avg. stresses, CSDEG) Drawing conclusions from the three models with different

inclinations of the longitudinal axis of the fibre bundle described, above two assumptions can be made. Change of the said inclination does not change the crack propagation behaviour from pulling out to fracturing the bundle. Direction of crack propagation in the bundle depends of the value of inclination. For small values definitely not higher than 45 - crack propagates after the bundle into the matrix from the higher end of the bundle. For inclinations bigger than 45 the crack propagates after the bundle into the matrix from the lower end of the bundle; this was valid for the geometries analysed

- 40 -

5. Conclusions and further work

Numerical models of idealised recycled carbon fibre-reinforced polymer (rCFRP) were developed. Using Extended Finite Element Method (XFEM) a primary objective was achieved. The two dimensional (2D) cohesive contact model is able to model failure mechanisms existing in the presence of fibre bundle: fibre bundle fracture and fibre bundle pull-out. Three different 2D models were created to capture these mechanisms. First one was a single instance model. It was able to model crack propagation towards and through the bundle, but aborted when the bundle was about to fracture;it was also not able to model fibre bundle pull-out. The next model had cohesive elements modelling the matrix-bundle interface. It was possible to propagate the crack into, through and past the bundle. Unfortunately, similarly to the single instance model, it was unable to model fibre bundle pull-out. The third model created used cohesive contact to model the matrix-bundle interface. It made it possible to model both mentioned failure mechanisms. It was a good representation of bundle present at the surface of the specimen, because of the way crack propagated back into the matrix after pulling out the bundle. In order to model a situation when fibre bundle is embedded into the matrix a three dimensional (3D) model, based on the 2D model using cohesive contact, was created. It was possible to achieve fibre bundle pull-out with a continuous crack propagation using this model. With further development this model can be a very accurate representation of idealised rCFRP. The 2D cohesive model was used to conduct a study on the influence of bundle geometry and position on the failure mode. It was concluded that length of the fibre bundle has influence on the failure mode of the bundle, whereas no variation of the failure mode - 41 -

with the thickness (within the geometries tested) was noticed. When the bundle was made longer, the failure mode changed from the fibre bundle pull-out to fracture. There were also two conclusions from the study of different inclinations of the longitudinal axis of the bundle. First is that, within the geometries tested, the change of the inclination does not influence the failure mode (pull-out or fracture of the bundle). Second was that the magnitude of the inclination has influence on crack propagation back into the matrix after fibre bundle pull-out. For the geometries tested, when the angle was lower than 45 the crack propagated from the higher end of the bundle and when it was higher the crack propagated from the lower end of the bundle. Further work, as stated earlier, should involve further development of the 3D model. As the second point, a more extensive study on the influence of geometry and position of the bundle should be conducted, involving both 2D and 3D cohesive contact models. This should help to understand in detail the mechanisms involved in each failure mode. After completion of two points listed above work should progress out of numerical modelling. Based on developed numerical models and analytical theories as well as multiphase / hierarchical / fractal theories for materials properties analytical models for strength and toughness are to be developed. Further, analytical results should be compared with numerical and experimental data to validate the analytical models.

- 42 -

Appendix A: Fractals as a tool for analysis of materials properties

Similarly to many of biological, natural materials, there are man-made materials that manifest complex, irregular and often at the first sight chaotic patterns [10]. There are multiple examples of such materials and objects e.g.: human lungs, veins, trees, fluid turbulence, polymers, fractures and coastline shapes [11]. All of these materials have one thing in common they are hard to be described by means of Euclidean geometry. That is the situation when fractal geometry applies - it makes use of the multi-scale nature of the materials. Different mathematical descriptions, formulas and patterns can be easily found in appropriate literature [10]. One of the amazing fractals characteristics is described as ability to fit an effectively infinite length within a finite area and an effectively infinite area within a finite volume. Fractals geometry can have very broad applications. It is widely uses in mathematics, physics but less in mechanics and even less in engineering [11, 12]. The key to understanding mechanics of materials representing fractal properties is to understand the correlation between their structure and mechanical performance. There are two approaches as to the range of steps in which fractal theory can be applicable it can be infinite or finite. When the first one is taken into account new properties such as scale-invariant quantities or abnormal physical dimensions must be defined. For many cases only the second, finite range, approach is applicable [13]. For example, sea shells have been found to have only 2 3 orders of lamellar structures [12]. However, despite the fact that some materials show self-similarities only in a small range, it is possible to model their scaling properties at each level by means of a fractal approach, using recursive relationships [13].

- 43 -

Carpinteri

et

al.

[14]

described

size

effect

as

common

phenomenon in multiscale materials. With change of structural size some of the mechanical properties, which would be constant in classical mechanics, change. A good example of such properties is tensile strength, which decreases with an increase of the size of a structure [14]. Unfortunately the size effect has not been interpreted yet, due to not knowing its source [10]. Soare [15], in turn attempted to model the size effect in the deformation and fracture process in a heterogeneous material by reformulating the field equations using a non-Euclidean metric. He suggested that for the processes that take place in the fractal space it should be assumed that the deformation is localized on a fractal. Newman and Gabrielov [16] tested failure properties for multiscale fibre bundles with equal load sharing. What they found is a universal asymptotic scaling law which links system size to failure stress threshold, what they believe will bring advantages to constructing fibre bundles in engineering applications. Cohesive crack model as a most commonly used in purpose of damage localization is described by Carpinteri [14]. In this model the material is characterised by a stress-strain relationship, valid for undamaged zones, and by a stress-crack opening relationship, describing how the stress decreases from its maximum value to zero as the distance between the crack lips increase from zero to the critical displacement. Piggott [17] points out that vast majority of available models for fibre composites have straight, regular, aligned fibres, what is far from reality and can lead to unexpected results. Presented by he author concept of mesostructures includes possible mentioned defects and is applicable to composites with randomly distributed short fibres. Wang and Pan [18] in their work on numerical approach for elastic properties prediction of multiphase composites also emphasised that if not taken into account, the small amounts - 44 -

of voids in a composite may lead to relevant differences between theoretical and experimental values. In their study, when calculating the material characteristics with voids taken into account, all Youngs modulus, shear modulus and Poissons ratio values coincided much better with experimental data. Two main sources of common difficulties with modelling behaviour of multiscale materials or structures are pinpointed by Soare [15]. First of them is the complexity of geometry in a micro scale and difficulties selecting the constitutive behaviour. The second problem concerns the macro scale often the deformation leads to the selforganization of internal structures, which control the constitutive behaviour. An efficient approach for solving boundary value problems on domains with complex microstructures based on multiple scales was proposed. Solving this problem with use of finite element method seemed too expensive. Soare proposed to use enriched shape function to capture the complexity of the geometry and using the outcome as a parametrically defined boundary value problem. After that all refinement levels can be obtained without any additional cost. Oshmyan et al. [19] studied composite materials containing rigid inclusions/voids with distributions similar to the Sierpiski-like carpet. They used the finite element method to determine the effective elastic constants for the first steps of the iteration and renormalization group techniques to identify the scaling properties. The scaling exponents turn out to be functions of the fractal dimension of the microstructure. Similar results were obtained by others in similar cases.

- 45 -

References
1. 2. Pimenta, S., Experimental study performance of recycled CFRP. 2009. on the mechanical

Mulligan, D., et al., Fibre-bundling in a short-fibre composite: 1. Review of literature and development of a method for controlling the degree of bundling. Composites Science and Technology, 2003. 63(5): p. 715-725. Piggott, M., Theoretical estimation of fracture toughness of fibrous composites. Journal of Materials Science, 1970. 5(8): p. 669-675. Fila, M., C. Bredin, and M. Piggott, Work of fracture of fibrereinforced polymers. Journal of Materials Science, 1972. 7(9): p. 983-988. Trapeznikov, D., A. Toropov, and O. Loskutov, Modelling approach to optimization of mechanical properties of discontinuous fibre-reinforced C/C composites. Composites, 1992. 23(3): p. 174-182. Piggott, M., The effect of aspect ratio on toughness in composites. Journal of Materials Science, 1974. 9(3): p. 494502. Kim, J. and Y. Mai, Fracture of CFRP containing impregnated fibre bundles. Composites Science and Technology, 1993. 49(1): p. 51-60. Abaqus, Inc. Abaqus version 6.9-1 Documentation. Simulia, 2009. Pimenta, S., Mechanical performance of recycled CFRP Transfer Report. 2010, Imperial College London, Departent of Aeronautics. Khezrzadeh, H. and M. Mofid, Tensile fracture behavior of heterogeneous materials based on fractal geometry. Theoretical and Applied Fracture Mechanics, 2006. 46(1): p. 46-56. Laizet, S. and J.C. Vassilicos, Multiscale generation of turbulence. Journal of Multiscale Modelling, 2009. 1(1): p. 177-196. Carpinteri, A. and N. Pugno, Mechanics of hierarchical materials. International journal of fracture, 2008. 150(1): p. 221-226. Carpinteri, A., Strength of hierarchical materials. Microsystem technologies, 2009. 15(1): p. 27-31. - 46 -

3.

4.

5.

6.

7.

8. 9.

10.

11.

12.

13.

14.

Carpinteri, A., B. Chiaia, and P. Cornetti, On the mechanics of quasi-brittle materials with a fractal microstructure. Engineering Fracture Mechanics, 2003. 70(16): p. 2321-2349. Soare, M. and R. Picu, An approach to solving mechanics problems for materials with multiscale self-similar microstructure. International Journal of Solids and Structures, 2007. 44(24): p. 7877-7890. Newman, W. and A. Gabrielov, Failure of hierarchical distributions of fibre bundles. I. International journal of fracture, 1991. 50(1): p. 1-14. Piggott, M., Mesostructures and their mechanics in fibre composites. Advanced Composite Materials, 1996. 6(1): p. 75-81. Wang, M. and N. Pan, Elastic property of multiphase composites with random microstructures. Journal of Computational Physics, 2009. 228(16): p. 5978-5988. Oshmyan, V., S. Patlazhan, and S. Timan, Elastic properties of Sierpinski-like carpets: Finite-element-based simulation. Physical Review E, 2001. 64(5): p. 56108.

15.

16.

17.

18.

19.

- 47 -