## Abstract

The problem of multi-scale modelling of damage development in a SiC ceramic fibre-reinforced SiC matrix ceramic composite tube is addressed, with the objective of demonstrating the ability of the finite-element microstructure meshfree (FEMME) model to introduce important aspects of the microstructure into a larger scale model of the component. These are particularly the location, orientation and geometry of significant porosity and the load-carrying capability and quasi-brittle failure behaviour of the fibre tows. The FEMME model uses finite-element and cellular automata layers, connected by a meshfree layer, to efficiently couple the damage in the microstructure with the strain field at the component level. Comparison is made with experimental observations of damage development in an axially loaded composite tube, studied by X-ray computed tomography and digital volume correlation. Recommendations are made for further development of the model to achieve greater fidelity to the microstructure.

This article is part of the themed issue ‘Multiscale modelling of the structural integrity of composite materials’.

## 1. Introduction

SiC–SiC_{fibre} ceramic matrix composites are candidate materials for fuel cladding in Generation IV nuclear fission reactor concepts such as the gas-cooled fast reactor (GFR) [1]. With excellent high-temperature capability and damage tolerance, they may have future applications for accident-tolerant fuel cladding for current (Generation II/III) light-water reactors (LWRs) [2] and as structural materials in the tritium-breeding ‘blanket’ for nuclear fusion power generation [3]. SiC–SiC_{fibre} composites have been developed largely within the international fusion materials’ programmes, which have optimized their microstructures for thermo-mechanical properties in composites with unidirectional fibres or two-dimensional weaves. Within the European nuclear fission programme, a tubular pin-type GFR fuel-clad geometry with a filament-wound and braided architecture has been designed [1] and layered monolithic SiC and SiC–SiC_{fibre} composite structures have also been proposed for accident-tolerant LWR fuel cladding [4].

Component testing is critical for composites, in order to identify the potential role of the composite architecture and its defects. For these nuclear applications testing in realistic conditions of fast neutron flux and temperature will be essential eventually to evaluate fully these structures. Macroscopic mechanical tests (e.g. [5]) can be used to assess the average properties of the composite, and provide data for use in design. The effects of irradiation and oxidation, batch–batch variations in properties and also the sensitivity to composite fabrication can also be examined by such tests. Component tests will be critical also to evaluate the properties of complex structures of joints [6]. Stochastic property variations may occur due to heterogeneities in the weave [7], and this also needs to be quantified by testing. Test data are also needed for design, and a possible structural design criterion for SiC–SiC_{fibre} composites may be some fraction of the proportional limit stress [8]. This is the critical stress, in tensile loading, above which nonlinear stress/strain behaviour of the composite is observed. This nonlinearity is caused by the development of significant matrix cracking [9], while the propagation of damage is resisted by the ‘fibre pull-out’ toughening mechanism from the continuous, high-strength SiC fibres [10,11]. Micro-cracking damage also affects the heat transfer properties of the cladding [12].

An understanding of failure mechanisms is needed to design tests that will qualify materials for service, and also accelerated tests to understand the potential degradation of the mechanical properties of composites due to environmental factors. Predictive modelling of these materials in engineering components must be able to predict their performance under different states of loading, using data obtained in such tests. Continuum-mechanics methods can be insufficient for this, as they may not take into account the effects of the heterogeneous fabric architecture on the strain distribution that develops within the composite microstructure [13,14]. The strain distributions in woven fibre-reinforced composites are influenced by factors that include the orientation of the fibre tow with respect to the loading axis [15], the arrangement and structure of neighbouring tows, matrix and pores [16], and also the mechanical properties of the composite’s constituents [17,18]. Together, these may affect the statistical distribution of failure strength under different states of loading.

Numerical modelling is a staple ingredient of industrial component design and the assurance of structural integrity, particularly in civil, automobile, aerospace and nuclear engineering. These all need numerical methods that can deal with complex geometries and loading conditions, particularly to simulate the performance of real components under service loadings that cannot be reproduced experimentally. The need for sophisticated, efficient and reliable numerical models is demonstrated by their large expansion in the last decade, employing a wide range of approaches to advance their complexity and speed via alternative concepts and mathematical solutions that are frequently optimized to the problem under consideration. Probabilistic methods of structural integrity assessment require numerical tools to model damage, so numerical modelling is a necessary tool to simulate the behaviour of structural components that are made of ceramic composites [19].

All numerical computational models that have serious potential for engineering design employ discretization to break the problem into smaller units, e.g. the finite-element (FE) method; their coupled behaviour describes the final result. In models that compute damage by fracture, and its interaction at the larger scale with the engineering structure, the discretization must be sufficiently fine to describe the critical micromechanical processes [20]. Damage may be introduced through a cohesive law to describe the microstructure properties, or via an explicit representation of the microstructure and its behaviour, so that the energy released by damage within the microstructure is defined or calculated accurately [21,22]. Predicting the likelihood of failure and the patterns of damage propagation requires stochastic approaches, since the variability of microstructure can produce a range of larger scale responses [23] that depend on the size and geometry of the component. Consequently, the goal is to develop multi-scale computational engineering models that have the necessary discretization to reproduce accurately the effects of microstructural damage [24] and yet have sufficient computational speed to simulate the stochastic behaviour of the component that will arise from microstructure heterogeneity [25].

One approach to computationally efficient multi-scale modelling is to locally insert the microstructure into a coarser FE mesh, only where such refinement is needed (e.g. [19,26,27]). For instance, cohesive FE models (FEMs) may use a mesh that is adapted to key features of the microstructure such as the composite plies [28], with some models representing each tow of fibres explicitly [29]. The finite-element microstructure meshfree (FEMME) model for quasi-brittle materials [30] uses the microstructure as a local enrichment of a coarse-cohesive FE mesh. This approach has been applied to simulate damage development in a range of quasi-brittle materials including SiC–SiC composites [31] and thermal barrier coatings [32]. In this paper, it is applied to the problem of damage development in a ceramic fibre-reinforced ceramic matrix composite tube. The objective is to demonstrate the ability of the FEMME model to introduce high fidelity to important aspects of the microstructure into a larger scale model of the component. This allows the sensitivity to microstructure heterogeneity to be explored. The characteristics that emerge from the model simulations can then be compared with experimental data that describe the bulk properties of the composite and the development of damage within its structure.

## 2. Experimental observation of damage development by tomography and digital volume correlation

Damage development within a SiC–SiC ceramic composite tube has been studied recently by digital volume correlation (DVC) of laboratory X-ray computed tomographs. Brief details only are included here of an experiment that was conducted to examine damage development within a ceramic composite tube; full details are given in [33]. DVC is the three-dimensional analogue of digital image correlation (DIC), which is a displacement measurement method [34] that tracks the local movements within successive images that are due to deformation; it is well suited to the detection of damage [35] and has been applied previously to study deformation and damage in ceramic composites [9]. X-ray computed tomography can be used to observe damage development *within* materials, including composites [31,36]. It can be combined with three-dimensional DVC [37] to measure the displacements that arise from deformation and damage [31,38,39].

The SiC–SiC material, provided by the MatISSe consortium [40], was fabricated by the French Alternative Energies and Atomic Energy Commission, as described in [9]; it comprised an inner SiC fibre filament-wound layer (±45°) with two layers of two-dimensional braided structure (±45°). The inner and outer diameters were 7.80 and 9.75 mm, respectively. A 55 mm length of the composite tube was tested in tension. Medium resolution computed tomography scans with a voxel size of 17 μm were obtained using a laboratory X-ray instrument. Higher resolution tomographs were used to characterize the porosity in the microstructure; an example image is shown in figure 1. After a reference scan in the ‘unloaded’ state (at 20 N), the tube sample was mechanically loaded axially in tension; scans were obtained at loads of 400 N and 800 N, which correspond to axial tensile stresses of 17.3 MPa and 34.6 MPa. The sample was unloaded to 20 N and scanned by tomography between these loads. DVC analysis was applied to the tomographs, using the LaVision Davis software, to measure the three-dimensional full-field displacements between successive tomographs. The reference tomograph was the initial unloaded specimen in each case. An anisotropic interrogation subset (128×128×256 voxels at 75% overlap, with the elongated dimension of the subset aligned parallel to the tube axis) was used.

The gradients of the measured displacement field can be used to obtain three-dimensional visualizations of the maximum principal strain, which are shown in figure 2, for the tube under tensile load and after unloading. These illustrate the progressive development of localized strains that are distributed non-uniformly throughout the sample. The magnitudes of the strains decrease on unloading, but the pattern of strains remains. There is a greater degree of deformation towards the upper end of the tube, which is attributable to the method of sample mounting. The strains tend to occur on one side of the tube, indicating that there was some flexure introduced during the loading. These strains are due to damage in the microstructure, and arise from the opening displacements of cracks that tend to initiate from the significant porosity in the braided microstructure [33]. The DVC analysis is not sufficiently precise to measure elastic strains at this length scale. The reported strains are due to the accumulated displacements from micro-cracking damage. An effective strain of 0.01 is equivalent to approximately 1 μm net displacement between adjacent displacement vectors, which have a separation of approximately 100 μm in the axial direction.

## 3. Modelling of damage development

### (a) The finite-element microstructure meshfree model for damage development in quasi-brittle materials

The FEMME model [30] is a multi-scale model for fracture simulation that is designed to deal with the full scale of an engineering component and also the fine-scale microstructural processes of damage efficiently. This is done through a combination of three layers that interact between each other and are located at different scales. The large scale is simulated through a standard FEM with a coarse mesh. If the strain of any element in the FEM reaches a critical strain, arbitrarily chosen and close to the critical strain of the material, the element is subdivided to create the cells and the other two layers are then activated locally. The first of these is a microstructural adaptive meshfree (MAM) layer, which is a meshfree model with exponential shape functions [41,42] that are geometrically adapted to the microstructural features, i.e. pores and particles that are defined as ellipsoids. The MAM model uses the nodal displacements of the subdivided elements of the FE mesh as its boundary conditions to calculate the displacements applied to the microstructural features. Those are interpolated into the third cellular layer to calculate the strain of each cell; these strains are evaluated against the cell’s critical strain. If the strain of a cell is greater than its critical strain, the cell is damaged, releasing its elastic energy into fracture energy. The sum of the fracture energies of all the cells damaged at each iterative step of the simulation is used, through energy homogenization, to reduce the mechanical properties of the other layers. The effect is analogous to the cohesive laws employed in traditional FEM; cohesive behaviour emerges from the operation of the cellular layer. Depending on where each cell is located, it represents the matrix, the pores or the particles of the material and its critical strain is assigned accordingly. With this, the heterogeneity of a microstructure is reproduced within a large-scale FEM, without significant homogenization of the microstructure.

One of the main potentialities of the FEMME model when compared with other fracture methodologies is that it is able to release the spurious strain energy that is calculated using a coarse FE mesh during fracture simulation [30]. When a fracture process is modelled using a coarse FE mesh, the calculation of the fracture energy dissipated depends on the size of the mesh. This can lead to an overestimation of the peak load and a subsequent generation of spurious strain energy; this energy is not real and makes the FEM calculation completely invalid. The spurious fracture energy generated by a too-coarse mesh is a common drawback for the modelling of damage in large-scale models because it forces that FE mesh to be sufficiently fine to reduce this effect. The refined mesh increases exponentially the computational cost of the model and reduces its applicability. Through its multi-layer approach, the FEMME model is able to release this spurious fracture energy by inserting the microstructure as a local enrichment that substitutes for the mesh refinement traditional FE models require to solve this problem. Mesh refinements or predefinition of the fracture path are therefore not required. This decreases drastically the number of operations needed to solve the problem, compared with FEM at an equivalent level of discretization. This is extremely beneficial to fracture problems where considerable numbers of iterations are needed to reach convergence at every loading step.

The model presented in the paper is an extension of the original FEMME model, and uses the same relationships between layers and method for the insertion of pores in the microstructure. The development here is the insertion of fibres, which influence both the FEM and MAM layers. An algorithm inserts the fibres using the mathematical description of each fibre tow to calculate how it crosses the element that it intersects. Mechanical properties are assigned to each fibre to describe the stress–strain behaviour and critical failure strain. A relationship between the strain of the element and the strain of the fibre is then established, so a reaction force from the fibre is obtained that can be applied at the nodes of the element that is crossed by the fibre as a boundary condition. This needs to be done iteratively, and, in some cases, relaxed to ensure the convergence and stability of the problem.

### (b) Extension of the finite-element microstructure meshfree model by insertion of fibres

This paper describes the extension of the FEMME model to consider damage in fibre-reinforced material with direct representation of the fibres. Previously, only the pores between fibre tows have been considered [31]. In this development, the fibres are inserted as a separate layer of the multi-scale calculation, influencing both the FEM and MAM layers. Each fibre tow is first inserted in the FEM as a single solid fibre; the individual fibres of the tow are then represented in the MAM model once the FE is subdivided and this region of the microstructure is activated. An alternative case would be where single fibres are not grouped in tows, such as in steel-fibre-reinforced concrete, and here the same fibre could be inserted in both the FEM and MAM layers. The relationship between the FEM and MAM layers and the fibres is obtained through the strain field of those models and the equivalent load that is inserted from the fibres as a reaction force. For this each FE that is crossed by a fibre tow has associated with it a simplified straight fibre tow segment that connects the crossing points where the fibre tow enters and leaves the element (figure 3*a*).

To calculate the positions of these crossing points, each fibre tow is defined through the parametric mathematical expression of the locus of its central axis. Then, the problem is reduced to the calculation of the crossing point between the plane formed by each side of the tetrahedral element and the mathematical expression of the fibre tow. The crossing points between the parametric description of the fibre tow and the FE are shown in figure 3*a*, with the simplified straight fibre that represents the fibre tow within the element. In figure 3*b* the relationship between the fibres and the microstructure is presented, with the region where the fibres are inserted in the mechanical model of the MAM layer; this shows the inter-particle domains that are defined between pores. All the crossing points are evaluated and the pair that lies at the borders of the tetrahedron is selected to define the fibre path [43]. The same algorithm is also used with the tetrahedrons that define the inter-particle domains of the MAM layer [30].

Any fibre typology can be implemented using its mathematical description. In the case of the SiC–SiC fibre composite that is considered in this paper, the fibres can be approximately described by equations (3.1)–(3.4). As noted, this composite comprises an inner SiC fibre filament-wound layer (±45°) with two layers of two-dimensional braided structure (±45°). There are flat pores located in the intersections between the fibre tows and longitudinal pores (long and short) located along the fibre direction [33] (figure 1). The braiding effect is implemented through a combination of sine and cosine curves, with the parameter *u* that grows in the Cartesian *x* direction (figure 4*a*). With these equations, each fibre tow of the three superimposed braided layers can be characterized:
3.1
3.2
3.3
3.4In equations (3.1)–(3.4), *x*, *y* and *z* are the Cartesian coordinates of the parametric plot. The parameter *c* is used to create different parallel fibres or fibre tows starting in different points. *R*_{tube} is the average radius of the layer; *ΔR* is the maximum variation of this radius given by the braiding, where *d* controls the amplitude of the braiding. *A* is the amplitude of the main fibre bundle (i.e. each tow describes a spiral, so *A* is the distance between peaks of a complete loop). The plus and minus signs are for the two sets of fibre directions, depending in which direction the spirals are growing (figure 4*b*).

These equations describe the position of the central axis of each fibre tow, and then the cross-section area needs to be inserted as a surface orthogonal to the tow axis. For the micromechanical model, the crossings of each individual fibre and the tetrahedrons of the MAM layer are computed after the FE that hosts this part of the microstructure is subdivided into domains. Each individual fibre is considered as a straight segment that is parallel to the path formed by the two crossing points of the FE that is associated with the fibre tow. As the fibre tows are not fully dense, the individual fibres are positioned randomly within the associated cross-section area of the fibre tow that is inserted in the FE model (figure 3*c*). The individual fibres are not continuous across the elements, and do not overlap. They are inserted randomly until the sum cross-sectional area of the fibres is equal to a pre-determined proportion (80%) of the cross section of the fibre tow. The remainder of the cross section represents the matrix and pores that are within the fibre tow. Once all the crossings between the fibres and tetrahedrons are computed, this information is used for the mechanical model. In each tetrahedron of the FE and MAM layers that is crossed by a fibre or fibre tow, there is a segment that is defined by the two crossing points and has a description of its orientation, solid area and mechanical properties.

The interaction between the fibre and the tetrahedron needs to reproduce the real behaviour of the material. Initially, the strain state of the fibre needs to be determined by considering its stress–strain relationship. Assuming a perfect bond between the fibre tow and the matrix, the strain state of the tetrahedron in the FE layer is projected into the fibre direction to determine the effective strain along the fibre. In future developments of the model, through the damage of the cellular layer, the state of the bond can be evaluated and the adherence of the fibre will be reproduced, increasing the accuracy of the model. Presently, this is addressed by tuning the nonlinear stress–strain behaviour of each fibre to produce a stress–strain function that represents the mechanical response at the length scale of the cellular description of the microstructure due to matrix damage and fibre pull-out. Once the strain of the fibre is determined, through its stress–strain behaviour and the area of the fibre, the reaction force applied by the fibre into the tetrahedron is determined. This force is shared between the nodes of the tetrahedron with an opposite sign, i.e. if a tensile strain is applied to the fibre, the forces to be applied in the nodes will make compression. With this the influence of the fibres on the matrix is reproduced. Then those forces are inserted in the FE model as Neumann (or second type) boundary conditions. These need to be relaxed iteratively. For a material with a low content of fibres (i.e. reinforced concrete), the problem is relaxed by simply applying the strains of the fibres in the FE model, which should be run until the convergence of the strains in the fibres. For the case where the material is mainly composed by fibres (e.g. SiC–SiC fibre composites), the convergence is not reached through the direct iteration of the model, and the fibre forces need to be inserted gradually in different iterations to ensure the gradual convergence of the material. This is feasible due to the computational efficiency of the FEMME model.

### (c) Simulation of a SiC–SiC fibre composite under tensile axial load

In this example, we use the model to simulate an experiment in which a tube with the SiC–SiC fibre composite described is loaded in tension. In the experiment described above, distributed damage was observed within the composite. The detailed study of Bernachy-Barbe *et al.* [9] measured the bulk properties of the same material, and studied damage development using DIC of the external tube surface together with acoustic emission. Together, these provide some observations against which the model simulations may be compared.

The geometry of the tube was reproduced by the FE layer, and the microstructure of the material was described in the MAM and cellular layers. The FE mesh that describes the composite tube is composed of only 3228 tetrahedral elements, with an element size of 1.5 mm (figure 5). The total tube length is 68 mm and all the applied displacements are null except the axial displacement (i.e. *x*-direction) of the top of the tube that is applied uniformly to represent the experiment loading. The internal diameter is 8 mm and the external is 10 mm. The three superimposed braided layers occupy the wall thickness of 1 mm. Each FE may be subdivided into 4096 cells to simulate its damage, so the discretization size (i.e. cell size) is 50 μm. This is larger than the 12 μm diameter of the fibres, so their representation is an approximation.

The microstructure is inserted to represent the porosity of the matrix and the fibres. To create the different types of pores, we consider their size as ellipsoids and their orientation [43]. Pores between fibre tows that are located where tows cross each other are defined by their centre, and are aligned tangentially to the fibre tow layer. The longitudinal pores are also ellipsoidal, and are aligned with the fibre direction. Both pore shapes are introduced using equations (3.1)–(3.4) to locate the position of each pore’s centre and orientation with respect to the fibre tows, and the shape of the tube to align them orthogonal to the tube radius. The total porosity measured by experiment (approx. 5%) [33] is reproduced in the model: by volume, 35% are flat pores and 65% are longitudinal; and by number, 6% are flat pores and 94% are longitudinal. A comparison between the numerical and experimental microstructures is shown in figure 6. The cross section of the numerical pores (figure 6*b*) agrees with the segmented tomograph of the real material where the thin pores are distributed across the thickness and the flat pores are between the braided layers. Visualizations of the numerical fibres and the material surface are shown in figure 6*c*. With this, the computational time of the model is 8 h running in serial in a Pentium i7 processor.

Having developed a model description of the microstructure, the aim is to fit the simulation to the experimental results of Bernachy-Barbe *et al.* [9] and to examine the predicted development of damage. In that experiment, the authors applied a tensile load to a SiC–SiC composite tube of length 68 mm, with careful observation by extensometry of the axial and radial strains of the tube. In the FEMME simulation, the mechanical properties used are: for the fibres, a Young modulus of 400 GPa, an elastic strain limit of 0.0125% and a failure strain of 2%, with linear strain softening from the elastic limit; for the matrix, Young’s modulus is 100 GPa, the critical elastic failure strain is 0.025% and Poisson’s ratio is 0.26. Those material properties are an optimized fit that reproduces the bulk behaviour of the material (figure 7).

The predicted development of damage for these properties is shown in figure 8, which presents the maximum principal strains within the tube at different applied axial strains (figure 7*a*). Damage initiates from a significant pore from an early stage (from a tensile stress of approximately 70 MPa) with nonlinearity apparent above approximately 110 MPa. As the applied strain is increased, the damage propagates through the central region and causes gradual failure of fibre tows. The stress field around the damaged region is shown in figure 8*b*.

### (d) Simulation of a SiC–SiC fibre composite under internal pressure

Having demonstrated the capabilities of the model to reproduce the bulk behaviour of the material in one stress state, it may be applied to examine the expected performance under different stress states, such as internal pressure. In this case, the microstructure of the simulated sample is the same as that used in the example of tensile loading. The predicted nucleation and growth of damage in the deformed tube (without magnification of displacements) is shown in figure 9, as pressure is increased progressively to 150 MPa. The simulation is an approximation, as the applied internal pressure is not relieved by damage to the tube. Damage first appears at 50 MPa internal pressure, i.e. at a hoop stress of approximately 200 MPa and axial stress of 100 MPa. The developed damage is aligned longitudinally, due to the hoop stress, but the critical damage nucleates at a different location from the tensile loaded case.

### (e) Simulation of variability in fibre tow properties

The described model framework is suited to examine the influence of defects in the material, and also of variability of the properties of the fibres and matrix since the material properties are taken from statistical distributions of the strength of every phase and the geometry of the microstructure is described in detail. Owing to the low computational cost of the model, it may be used to perform a sensitivity analysis to identify the key material factors. As an example, figure 10 shows the same model of tensile loading (i.e. figure 8) in which the effects of random defects that affect the fibres tows have been studied. This is done (i) by decreasing the elastic limit stress of 20% of a random selection of the fibre segments by 10 MPa (i.e. by 20%) and (ii) by additionally reducing in the same segments the fibre failure strain to 0.02%. The effect of this on the bulk behaviour of the composite tube is quite apparent; the weakening of the fibre tows leads to more distributed damage and earlier softening of the composite as the applied strain increases.

## 4. Discussion

The aim of the presented model is to address the gap between the modelling of the full-scale component and micromechanical damage, by introducing the influence of the fibres. The fibres inserted in the FEMME model have the potential to arrest a crack that is propagating, increase the stiffness of the component and increase the load that is transferred across the crack faces. The model therefore approximates the influence of the fibres on composite behaviour.

In its current form, the FEMME model allows the effects of the significant pores between fibre tows to be simulated, and the effects of the stress state on damage development and propagation in the composite component to be examined. The predicted patterns of damage are consistent with experimental observations (i.e. figures 2 and 8). The stresses that develop damage in the tube under tensile and pressure loading are also described; the critical stresses that cause damage with internal pressure are comparable to those measured experimentally [9]. The change in the location of the onset of significant damage between tensile and pressure loading demonstrates the importance of the geometry and location of defects relative to the local stresses in the composite microstructure. The simulation of material variability (figure 10) also shows that variations in the ability to sustain load by fibre pull-out may have a significant effect on the bulk properties of the composite.

In an experimental study [44] of the tensile behaviour of single tows of aligned SiC fibres with a SiC matrix, the Young modulus of a single fibre tow was reported as 375 GPa and the elastic limit approximately 500 MPa. Matrix cracking around the fibre tow occurred above 0.1% strain (i.e. close to the elastic limit), with ultimate failure at around 0.7% strain. Compared with this, the properties applied in the FEMME simulation that give good agreement with the experimental properties of the composite tube appear quite low. This is because the simulation is designed to address the effects of the significant stress concentrating pores and defects at the interfaces of the fibre tows. The simulation does not directly reproduce the pull-out of the fibres, and the input properties describe the influence of strain on the load that is carried by the fibres that are represented within the component. The elastic limit for the fibre tows in the FEMME model is thus the strain (0.0125%) at which interface damage is assumed to begin at the fibre tow, and the failure strain represents the fibre pull-out limit when it cannot carry load. Higher resolution modelling that is more fundamentally related to the structure of the fibre tows and their interfaces would be needed to directly relate the FEMME model input properties to the physical properties of the composite constituents.

Future extensions of this model will allow one to address the development of damage of the composite at high temperature, which will influence its thermal properties [32] and simulate the effects of fast neutron irradiation and oxidation on composite performance due to changes in local properties. The FEMME model may also be inserted into other numerical frameworks for modelling of fibrous structures (e.g. [45]), which would provide more representative descriptions of composite weaves and their defects [46], and also of non-uniformity of their deformation under shear loading, for example.

## 5. Conclusion

A numerical modelling framework has been applied which aims to capture the key features of the ceramic composite microstructure that controls damage development, particularly the location, orientation and geometry of significant porosity and the load-carrying capability and quasi-brittle failure behaviour of ceramic fibre tows. This allows a high-fidelity numerical simulation to be performed that describes damage development at a length scale that can be studied experimentally. The simulations performed can describe the failure of the component under different states of stress and can obtain good agreement with bulk properties. Sensitivity to the potential degradation of material properties is also obtained. The general characteristics of the predicted damage are also consistent with experimental observations. The model requires further development to achieve high fidelity to microstructural processes such as matrix cracking and fibre pull-out.

## Authors' contributions

L.S.-M. conducted the modelling simulations that are described in this paper, under the academic supervision of T.J.M. Both authors jointly conceived the analysis, interpreted the data and wrote the paper.

## Competing interests

The authors have no competing interests.

## Funding

L.S.-M. was funded by the UK EPSRC project ‘QUBE: Quasi-Brittle fracture: a 3D Experimentally-validated approach’ (EP/J019992/1, principal investigator T.J.M.). T.J.M. acknowledges the support of the Oxford Martin School. This work also contributes to Work Package 3 of the MatISSe Collaborative Project ‘Materials’ Innovations for Safe and Sustainable Nuclear’ (European Commission Seventh Framework Programme grant no. 604862, FP7-Fission-2013).

## Footnotes

One contribution of 22 to a Theo Murphy meeting issue ‘Multiscale modelling of the structural integrity of composite materials’.

- Accepted March 18, 2016.

- © 2016 The Author(s)

Published by the Royal Society. All rights reserved.