CEAS Aeronaut J (2011) 1:83103 DOI 10.

1007/s13272-011-0004-x

ORIGINAL PAPER

Virtual testing of aircraft structures

Received: 28 April 2010 / Revised: 23 March 2011 / Accepted: 30 May 2011 / Published online: 26 July 2011 Deutsches Zentrum fur Luft- und Raumfahrt e.V. 2011

Abstract This paper will focus on the prediction of aircraft structural strength using virtual testing analysis methods. Virtual testing is a concept with several attributes and is to be understood as the simulation of aircraft structure using advanced nonlinear nite element analysis. It will involve the combination of analysis software, methods, people skills and experience to predict the actual aircraft structural strength with a high level of condence. This is achieved through the creation and execution of a detailed nonlinear nite element analysis model of an aircraft structure, which represents as accurately as possible the actual physical behaviour when subjected to a wide range of loading scenarios. Creating a virtual representation of an aircraft structure presents the analysts with several signicant challenges, including the creation of the complex nite element model that accurately represents the global aircraft structure, and then adding the signicant detail in terms of material and construction required to make accurate failure predictions with condence. An overview will be provided of the general principles used in the process of virtual testing of both metallic and composite aircraft structures. The paper will focus on the key

challenges and enablers for future successful virtual testing demonstrations in an industrial context. Keywords Virtual testing Aircraft structures Non-linear analysis Strength predictions Industrial requirements The wishbone analysis framework

1 Introduction Historically, the use of structural analysis in commercial aircraft design and certication has been focussed on linear nite element analysis for the calculation of internal load distributions and on the use of analytical stressing methods, both for initial sizing and then more detailed calculations for nal certication. This stressing approach, when combined with structural testing both to demonstrate the aircraft structure integrity and to demonstrate the adequacy of stressing methods, has proven itself to be highly reliable in the development of safe aircraft structures. The above approach is based on demonstrating the adequate strength of the aircraft structure, which is ensured through conservative assumptions in both the methods and material properties used. In recent years, advanced nonlinear analysis methods have been used increasingly to obtain more accurate assessments of the actual strength of aircraft test structures, both for risk mitigation prior to test and subsequent to a failure event [1]. Nonlinear nite element analysis has been employed with great effect to increase condence in the large-scale and expensive structural tests that are required before certication, as well as to understand in more detail the likelihood, causes and consequences of structural failure. There is an important distinction between predicting actual and adequate structural strength. For the design and

M. G. Ostergaard (&) A. R. Ibbotson Airbus in the UK, Bristol BS997AR, UK e-mail: [email protected] A. R. Ibbotson e-mail: [email protected] O. L. Roux Airbus in France, 31060 Toulouse Cedex 9, France e-mail: [email protected] A. M. Prior ` Dassault Systemes Simulia Limited, Warrington WA3 7PB, UK e-mail: [email protected]

123

84

M. G. Ostergaard et al.

certication of aircraft structures, adequate and conservative strength assumptions must be employed, irrespective of the methods used; whereas, for predictions of the actual strength of an aircraft structure, the analyst must make use of methods that are as accurate as possible. Due to the highly competitive nature of the aircraft manufacturing industry and the need to meet customer expectations in terms of efciency, aircraft structures are highly optimised for weight and strength. Most of an aircraft structure is typically constructed of thin-walled stiffened panels. Predicting the strength and failure mode of such structures which, especially for metallic structures, are often designed to allow buckling, presents the analyst with many challenges. Failure can occur due to buckling alone, but it is usually the consequences of buckling that can lead to critical failure modes in joints and materials and interactions between these failures. In addition to the complex design and nonlinear deformation behaviour of the aircraft structure, the analyst is also faced with the problem of understanding and analysing both metallic and composite aircraft constructions, where each new type of material presents new and different challenges with respect to detailed failure predictions [2, 3]. The increased use of composite materials has presented the analyst with a raft of new difculties, largely due to the highly complex failure modes of composite materials and associated adhesive joints [4]. The analysis of composite materials has undoubtedly also increased awareness of the many uncertainties that can exist in component manufacture, uncertainties which may signicantly affect the reliability of actual strength predictions based on analysis methods. It is inevitable therefore that today composite structures are designed with more conservatism than metallic aircraft structures. Advanced nonlinear analysis has in recent years been used very successfully to predict the actual strength of metallic and composite aircraft structures in Airbus. Examples of detailed nonlinear nite element models of composite fuselage and wing box structures are shown in Figs. 1 and 2. To provide a structured overview of virtual testing, the following topics will be discussed in the following sections: (a) (b) (c) (d) (e) (f) (g) Best practice in virtual testing Analysis software Multi-scale analysis Composites Modelling details and structural idealisations Detailed failure predictions Analysis framework for virtual testing using nonlinear analysis (h) Implicit and explicit nite element methods

(i) Robust analysis (j) A380 wing certication (k) Summary The principles outlined in this paper will be illustrated using the A380 wing certication experience, where advanced nonlinear nite element analysis was used successfully in the process to certify the wing structure. More recent models used to support Airbus aircraft programmes will illustrate todays best practice in Airbus and the rapid and organic evolution in the nonlinear analysis technology made possible through innovative approaches and improved software and hardware capabilities.

2 Best practice in virtual testing Finite element models used for virtual testing can be extremely complex. The application of best practice is therefore paramount in order to instil condence at all levels. The analyst must be condent in the methods and software being used; the principal FEA engineer must be condent in the skills, expertise and experience of the analysts; and the aircraft manufacturer (and ultimately the Airworthiness Authorities) must be condent that the virtual testing approach is valid and safe and therefore can be used for the purpose of demonstrating the actual strength of the aircraft structure. In principle, condence must be ensured in all of the following three areas: 1. 2. 3. analysis software, methods and analysis processes, people skills and experience.

It might be argued that the most important of these condence factors is the skills, expertise and experience of people, without which the process would not work and the detailed analysis understanding could not be gained. However, the functionality of the analysis software is also critical and condence must exist in the methods and analysis processes used. The analysis software must be both capable and feature rich to enable the analyst to predict actual strength for a wide range of potential failure modes. In addition, the solver must be highly efcient, given the tendency in recent years towards increasingly large analysis models. Computations running over several days on high-performance computing platforms are now commonplace. It is of fundamental importance, and an airworthiness authority requirement [5], that the analysis methods and processes used have been fully validated against test data from similar aircraft structures and materials. As for conventional stressing techniques, the methods used for virtual

123

Virtual testing of aircraft structures Fig. 1 Detailed nonlinear FE model of composite fuselage structure

85

Fig. 2 Detailed nonlinear FE model of composite wing box structure

testing based on nonlinear analysis must also be demonstrated to provide accurate predictions of actual behaviour for various levels of structural testing, from component or system level (e.g. wing or fuselage sections), to detailed coupon level (e.g. material specimens and fasteners). In order to build condence, the nonlinear nite element methods must be fully validated against test data at all levels of the testing pyramid. For virtual testing purposes, where actual strength predictions are considered, a close

correlation between methods and test data must be demonstrated. It should be noted that virtual testing is and always will be an approximation to reality. Actual aircraft structures will always have imperfections in materials, variations in manufacturing processes, and a range of assembly tolerances that will give rise to built-in stresses and variability in stiffness and strength. Composite materials, in particular, can exhibit signicant variability in the actual lay-up and

123

86

M. G. Ostergaard et al.

quality of the laminate. These variabilities lead to uncertainty regarding the actual response of the structure. Tests on several identical physical structures would not result in identical results. In a virtual testing program, however, uncertainty and variability need to be addedthey do not arise naturally. For virtual testing of a baseline structure, such as a Type Certicate Aircraft, the analysis model would normally be constructed to the nominal geometry and property; subsequent strength assessments should then take into account the likely variation in properties and construction. Analytical or conventional stress analysis methods are usually based on a set of assumptions with respect to structural behaviour, loading and constraint systems. These assumptions make conventional methods less exible, and less able to reect the details of the actual structural design. They also require that the results are treated as approximations with a degree of conservatism. The progression to detailed and advanced nonlinear nite element methods has allowed some of that conservatism to be addressed. Modern nite element analysis methods are general-purpose and highly exible. By validating the fundamental modelling and analysis methods, (i.e. the idealisation principles and the models for materials and joints), these building blocks can be used to construct full models of most types of aircraft structure. The validation of numerical models of materials and structures at detailed structural levels is more efcient and less costly than validating against more complex structural tests. This means that less validation against actual test data will be required at complex and large-scale structural levels than is the case for analytical methods. This is one of the most important advantages of advanced nonlinear nite element analysis compared to conventional stressing methods. A signicant challenge in the coming years will be to validate all aspects of virtual testing methods against all levels of structural testing, from coupon to component and full aircraft scales. This validation will need to address the increased levels of structural complexity and the full range of materials in use today and in the near future. The objective should be to increase progressively the level of condence in virtual testing methods, to the extent that they can be used for reliable up-front predictions of structural testing and, ultimately, can be considered for use in the certication of aircraft structures. This will only be possible through the continued application of best practice principles.

` products (Abaqus software, Dassault Systemes SIMULIA, Providence, USA). The tools allow for a wide range of nonlinear nite element analysis, which means that simulations can include nonlinear material responses (plasticity, damage and failure), nonlinear boundary conditions (contact), and nonlinear geometric effects (stress-stiffening, large rotations and displacements). The majority of analyses are carried out using the implicit FE method, which provides an incremental-iterative solution to a quasi-static loading problem. There are some applications where explicit solutions are appropriate, such as transient dynamic events, including birdstrike, debris impact and crash, though these are carried out less frequently. There are advantages in offering both implicit and explicit solvers with the same model denition used for both analysis techniques for complex failure simulations of static test structures. The direct solver technology provided with Abaqus/ Standard has improved signicantly in recent versions and, today, the direct sparse solver remains the standard solution for this type of nite element model. The thin-walled, stiffened structure common to modern aircraft is prone to buckling and also exhibits highly nonlinear material behaviour together with rapid changes in both geometry and boundary condition due to mechanical contact. Direct solvers have proven to be more effective than iterative solvers for the solution of this type of ill-conditioned system.

4 Multi-scale analysis In the context of virtual testing of aircraft structures, the term multi-scale analysis describes the process of sequentially coupling different analysis models at different scales and levels of delity. This multi-scale approach requires a Level 1 prediction of the behaviour of the complete structure through a nonlinear nite element model. This is then used to dene the driving boundary conditions for the next models at the more rened modelling scales, as illustrated in Fig. 3. Modelling detail is increased as successive analyses zoom in on structural regions, identied as being potentially strength-critical. At each level of model renement, different modelling idealisation principles, element types and even material and joint models might be employed. However, the underlying principle is to maintain a consistent interface and link between the different modelling scales used. All modelling and analysis methods used must be fully validated against structural testing. It is important to understand that, unlike traditional modelling techniques, where direct links between model scales are provided using built-in detail meshes or super-elements, the multi-scale

3 Analysis tools The virtual testing methods and simulations discussed and presented in this paper are carried out using the Abaqus

123

Virtual testing of aircraft structures

87

analysis process discussed in this article are based on submodelling technology available in the Abaqus software. This technology enables model data to be transferred between modelling scales through a parentchild type relationship, enabling far-reaching future opportunities for integration of CAD and CAE processes. It is important that the Level 1 model is nonlinear. Previous work using a linear Level 1 model has shown that the assumptions and approximations inherent in the linear approach will not provide a sufciently accurate base level from which to launch more detailed nonlinear analysis models. In principle two different approaches exist: (a) The analysis zooms in on predetermined structural zones that are then modelled to the required detail (such as in the A380 wing example). The purpose of the global model is purely to provide the denition of boundary conditions for lower scale models. (b) The high level analysis results, using the Level 1 model, are used to predict zones of interest for more detailed analysis, so that at each modelling scale the results are screened in order to identify regions for subsequent strength analysis.

Both approaches have advantages and disadvantages. The rst (predetermined) multi-scale analysis approach can be used only where the critical area of interest is fairly well known, but can benet from a relatively coarse Level 1 model. However, the coarser the Level 1 model, the bigger the lower scale sub-models must be to ensure that the correct loading is applied to the area of interest. This

approach is also of use when the Level 1 model is used to dene input data to other simulation techniques such as parametric modelling or analytical methods. However, from a virtual testing point of view, it is the second approach that has by far the most signicance. For this approach, it is not necessary to have a predetermined understanding of the critical structural response; instead, screening methods are applied to the analysis results in a systematic process in order to identify the critical structural regions to be analysed subsequently in more detail. A fundamental requirement for this analysis approach is that the Level 1 model must be capable of predicting the overall nonlinear behaviour and have sufcient detail to calculate the local nonlinear behaviour, such as panel buckling, nonlinear deformation due to structural eccentricities, and locations of joint failure. The geometrical size of a family of sequentially rened sub-models used within the multi-scale analysis process is directly linked to the accuracy of the Level 1 model. As a general guideline, the structural domain covered by these must ensure that the interfaces between the Level 1 model and subsequent sub-models are sufciently away to avoid inuencing the accuracy of analysis predictions. This is a particular concern where the analysis objective is to predict the propagation of structural damage and failure. Here the analyst must ensure that local stiffness change due to local damage and failure propagation does not invalidate the sub-modelling analysis process. This presents a very signicant challenge to the analyst, the analysis software, and the computational resources.

Fig. 3 Multi-scale analysis processes

123

88

M. G. Ostergaard et al.

5 Composites For composite aircraft structures in particular, it is important to consider the uncertainties introduced as a result of manufacturing processes. Many of the failure mechanisms that occur in composites are so localised that it is not possible to capture them at a global model scale. Engineering judgement and best practice is therefore required. In addition, the manufacturing processes used today will introduce variability in the composite lay-ups in terms of resin/bre volume fractions, ply waviness, resin pockets, inconsistent adhesive layer thicknesses, etc. Such factors must be considered in the analysis either through the imposition of conservative assumptions or the use of robust analysis methods. These particular aspects of composite construction methods mean that the accurate prediction of the strength of composite structures will remain a signicant challenge for many years to come. New advanced constitutive models for laminated composites which include coupled damage and failure capabilities are being developed and may provide a framework for the screening process for material failure at global model scale. Material models available today that are based on physical composite failure modes and fracture mechanics principles are computationally too expensive for use at most model scales, but are necessary for the accurate simulation of complex failure modes at lower scale levels, where they can be used to consider through- thickness failure and in-plane interaction modes. The maturity of reliable composite failure modes for all modelling scales and, in particular, for detailed failure predictions, is still to be demonstrated.
Fig. 4 Use of Abaqus cohesive contact modelling

The following example (Fig. 4) illustrates the use of screening processes to check for interfacial forces in adhesive joints between stringers and skins in a detailed virtual testing process applied to a composite wing structure. This enables areas with high interface forces in the adhesive joints to be identied so that more rened analysis can be carried out. The adhesive joints are modelled using the cohesive contact capability and the fasteners are modelled using the mesh-independent fastener feature available in Abaqus/Standard.

6 Modelling details and structural idealisations The importance of the detailed modelling techniques to be used, even at the Level 1 modelling scale, must not be underestimated: successful simulation in a virtual testing framework is entirely dependent on the way the structure is modelled and how the interactions between structural parts are represented. The nite element model must represent as closely as possible the actual structure being investigated, and all approximations and idealisations must be carefully considered. For example, it is not sufcient to make an accurate representation of the in-plane stiffness of an aircraft panel: the local and global torsion and bending stiffness of the panel must also be represented accurately. For local buckling to be predicted, it is essential that the support provided by stringers to suppress skin buckling is modelled accurately. For metallic aircraft structures, it is therefore necessary to include fasteners (including pre-tension) and mechanical contact in order to accurately simulate

123

Virtual testing of aircraft structures

89

buckling. To some extent, composite structures that are adhesively bonded are easier to model and to analyse, because the bonds can be represented using geometrical constraints or cohesive contact models. It is the responsibility of the analyst to fully understand the level of certainty associated with a given strategy to be used for the idealisation and modelling of an aircraft structure. Existing constraints on solvers and high-performance computing mean that there is always a compromise between model renement and analysis efciency. At one extreme the mesh might be too coarse to capture any useful response, and at the other the model may be too large to run on even the largest computers. The analyst must therefore deploy a consistent and wellunderstood strategy for meshing and modelling all the standard aircraft structures that will be encountered, so that the solution of the large-scale and multi-scale analysis program can be completed effectively. Numerous studies must be carried out to fully dene the best practice for modelling these structures. The best practice will dene the types of element to be used, the use of element off-sets, the use of structure mid-planes for meshing, the number of elements in part segments such as stringer webs, anges and between stringers. For virtual testing of large-scale aircraft structures it is essential to ensure that such best practice is followed precisely and consistently, particularly where many individuals in several teams and even external suppliers are involved. In such cases, the adherence to best practice for meshing and modelling quality can only be controlled through the rigorous application of detailed specications and stringent quality checks. Although the nal assembly and execution of the largescale models will typically be carried out by highly experienced analysts, it will not be possible to assign any level of condence to the nal results unless the modelling approach for all the systems, subassemblies and components has been consistently checked against the best practice guidelines. Figures 5 and 6 illustrate the above principle, through the example of a composite wing top cover. It is important to note the highly consistent meshing approach, which is applied irrespective of who built the particular model components. 6.1 Elements Another challenge facing the analyst is the choice of element to use for a given analysis problem. The principal choice is between beam, shell and solid elements. For most virtual testing purposes, where the objective is to determine the accurate strength of the aircraft structures,

Fig. 5 Wing boxlower cover removed, ribs and stringers

Fig. 6 Stringers and skin mesh

beam elements are of less practical importance, even at global model scales. Shell elements have many different formulations and not all are suitable for nonlinear calculations. The Abaqus software includes efcient and robust element types such as S4 and S4R which are recommended for most applications of nonlinear analysis on aircraft structures. Continuum solid elements, such as C3D8, C3D8I, C3D10M and C3D10I are used mostly at the lower modelling scales. An element that is of particular interest for modelling composite structures from CAD geometry data is the continuum shell element (SC8R). This element is based on standard shell theory but has the 3D topology of a solid hexahedral element. This offers certain advantages when checking and visualising complex assemblies of aircraft structure parts, in particular when dening and checking contact interactions. The continuum shell has been shown to be efcient when modelling composite parts from CAD geometry (CATIA V5) as this data contains all the lay-up information dened from the tooling surface, which can be assigned to the continuum shell element properties using the element thickness direction vector orientation (stack direction).

123

90

M. G. Ostergaard et al.

Fig. 7 Stringers on panel

Another notable advantage with the continuum shell element is that it makes the transition between shell-like structures and continuum solid elements relatively straightforward. This is advantageous when using Abaqus/ Standard sub-modelling methods for multi-scale analysis. Figures 7 and 8 show the usage of continuum shell elements for modelling of composite aircraft structures. There is a substantial time, effort and cost involved in creating t-for-purpose virtual testing models. A future challenge to both the aircraft manufacturers and the suppliers of the analysis software is to make such processes as automatic as possible. This requires much more than just automatic mesh generation, which is in any case available in most commercial modelling packages today. The successful implementation of best practice principles also requires modelling of CAD parts based on appropriate idealisations, which are consistently applied.

Fig. 8 Continuum shell stringer detail

This section will discuss the usage of material and joint failure models. 7.1 Material damagemetals The modelling of the elasticplastic behaviour of metals is well established and most modern simulation tools offer a variety of plasticity models for specic applications. For example, some components made of high-strength aluminium may have orthotropic properties because they are milled from rolled billets which have an orientated grain structure within the material. This processing method, which is used for aluminium wing spars, can lead to a variation in yield stress in the principal material directions that can be accounted for using Hills plasticity model (a non-cylindrical 3D yield surface) rather than a standard von Mises plasticity model that is based on the assumption of isotropic material properties. Capabilities to model material behaviour beyond initial yield, taking into account ductile or brittle damage and failure, have progressed in recent years. A generalized framework for damage and failure modelling is built into the Abaqus tools (Fig. 9), and provides a relatively straightforward approach. It allows the denition of a damage initiation state, followed by some form of degradation in the yield stress which is accompanied by a reduction in elastic modulus, down to a failure state where the material can carry no further load. This approach is an approximation to the physical response of the material, but allows for relatively straightforward tting to test data and can be implemented in such a way as to minimize mesh dependence. More complex physically based material models are available, for example those that include a consideration of

7 Detailed failure predictions As outlined in the previous sections, the purpose of the virtual testing and multi-scale analysis processes is to enable reliable strength assessments to be made, and this requires accurate, reliable and robust failure models for materials and joints. It is accepted that currently some modelling techniques are more mature than others and that more condence exist in those failure models used with metallic components than those in composite aircraft structures where signicant development and research still is to be carried out. For most static strength analyses, predicting the initiation of failure is adequate. However, more recent failure modelling capabilities like the X-FEM method is allowing for more accurate simulation of the propagation of failure within a material [13, 1820]. In addition, this capability is potentially providing a bridging capability across different types of materials and failure modes.

123

Virtual testing of aircraft structures

91

Fig. 9 Abaqus damage and failure framework

void nucleation and coalescence for ductile failure. However, it is inevitable that the more complex the model becomes, the more parameters are required and the more difcult it is to obtain the necessary data from material coupon tests. 7.2 Material damagenon-metals For aerospace applications, the majority of non-metallic materials are laminated composites of carbon-bre and the modelling of such materials is still very much an area of development. Analysis models of laminated composites are usually built-up in modern pre-processing tools that are capable of constructing complex lay-ups with different thicknesses, materials and orientations at each ply. These can either be condensed into a single equivalent anisotropic elastic behaviour, or kept as a distinct set of ply properties. The latter approach aids ply-based post-processing and also offers the extension of ply-by-ply damage modelling during a nonlinear solution. The initiation and evolution of material damage in laminated composites is highly complex. It depends not only on the behaviour of the individual constituents but also the interfaces between them. Damage can occur in the bres themselves, either in a compressive buckling mode or as a tensile failure. Compressive or tensile damage can also occur in the matrix surrounding the bres, leaving the bres intact and able to carry tensile load, but likely to buckle under compressive load. Additional failure mechanisms include the fracture of the bond between the bre and the matrix, as well as the delamination of adjacent plies.

This complex set of potential damage and failure mechanisms could in theory occur simultaneously in any number of combinations. This makes it very difcult to produce a constitutive model that is capable of simulating the overall behaviour, and very difcult to test the material in order to derive suitable parameters for the constitutive model. A result of these difculties is that the development of comprehensive constitutive models for laminated composites continues to attract signicant effort in academia. Currently, constitutive models can include most of the bre and matrix failure modes [1012, 14, 15, 17]. However, because many structural models employ plane stress shell theory, the delamination effect has to be taken into account separately. This means that the inter-laminar bond strength must be modelled explicitly, either with cohesivetype elements [6, 16], a cohesive contact formulation, or fracture mechanics techniques such as the virtual crack closure technique (VCCT) [79]. Again, it is difcult and expensive to derive appropriate parameters for a delamination model from a suitable test; however, obtaining accurate test data can lead to a more sophisticated and accurate model enabling reliable failure assessments to be made. 7.3 Joint damagefasteners For metallic airframe structures, the most common fastener is the rivet. Rivets are relatively straightforward to model in a nite element analysis: they are often idealized to a simple constraint between two plates, with no preload, no stiffness of their own, and no potential for damage or failure. Similar techniques have been used for many years in the automotive industry for the simulation of spot-welds. A natural extension to this approach is to model the rivet as a point-to-point constraint but to augment the behaviour with some elastic stiffness, a plasticity response and, ultimately, damage initiation and damage evolution to failure. A combination of axial, shear and bending loads can be included in the failure envelope for this type of constraint. At a high level, therefore, the rivet can be modelled as a point-to-point connection with relatively complex behaviour. The overall behaviour can be implemented through the generalized framework of damage initiation and evolution to failure as described above. This is useful for models that might contain many thousands of rivets, where it is important to consider the state of the connection as the load increases, but where it is not possible to model every rivet as a 3D component. For some types of fastener, a point-based connection may not be sufcient, so some form of coupling is needed to simulate the effect of the ends of the fastener (bolt head or nut) on the surrounding material. Modern analysis tools include capabilities to construct large numbers of fasteners

123

92

M. G. Ostergaard et al.

at varying levels of complexity, based on the fastener map for the structure, and including the coupling between the bolt head and plate material where appropriate. At a more detailed level, full 3D continuum models of riveted joints, including the rivet, the plates, the holes and perhaps even the tools used in the riveting process itself can be used to simulate the local behaviour under various loading conditions. Ideally, the study of the fastener behaviour should include pure shear, pure tension and various combinations of loading angle in between. Such models are important for correlating the fastener failure envelope described previously with physical test results. An important consideration is the degree to which the high-level constraint method can take into account the local behaviour of the joint. In general, it is not possible to include effects such as pre-load, hole-deformation, fastener rotation, pull-through etc., unless the specic combination of fastener and plates is correlated carefully with test. For complex fasteners which might have countersunk heads, inserts and pre-loads, even this level of correlation is unlikely to be sufcient. The combination of the fastener, the hole and the parent material is a complex system. The behaviour can depend on the relative strengths of the fastener and the surrounding material, as well as on the form of the loading. It is not uncommon to see a transition in the failure mode of a joint as the angle of loading varies from normal to pure shear, leading to fastener failure in tension through to failure of the local material around the hole. The complexity is increased yet further if a single fastener is used to join more than two plates, because additional failure modes can arise. Much has been achieved in recent years, with notable success in several large scale simulations. However, the detailed modelling of fastener failure continues to evolve, with aircraft manufacturers carrying out more extensive investigations into the correlation with test and software developers seeking more efcient ways to replicate the physical behaviour. 7.4 Joint damageadhesives Adhesive joints are becoming more common in aerospace structures because of the increasing use of laminated composites. In some cases the joint is made entirely with adhesive, while in others the adhesive is used to augment a joint fastened with bolts. As with other forms of fastening, the analyst can employ a range of techniques to model the effect of the adhesive bond, depending on the level of delity required in the particular simulation. At its simplest, an adhesive joint might be considered as a tied surface constraint with zero thickness between the tied surfaces. In general this is adequate for all but the most detailed analyses, since the elastic stiffness of a thin

adhesive layer is unlikely to be a key variable in the overall structural response. For more complex cases, it is important to consider the possible failure of adhesive joints, particularly in areas where peeling can be initiated. A common area for attention is at stringer run-outs where failure may also be driven by high interfacial shear stresses in the adhesive joint. The failure of the adhesive bond requires a modication of the standard tied constraint, either via an extension to a basic contact algorithm, or through the use of some kind of zero-thickness cohesive-type element, so that the bond between the surfaces can be progressively weakened under increasing load. The most straightforward approach is to use another form of the generalized damage framework described previously: the bond has an elastic stiffness which is applicable up to a limit value of stress or strain, at which point the properties are degraded down to ultimate failure. In general, for zero-thickness adhesive bonds, these properties are dened through a tractionseparation law with a damage phase evolving into failure. Such approaches can be correlated to peel tests, but it is not always straightforward to separate the normal and shear responses, which in the physical bond are very closely coupled. The adhesive joint might also be modelled as a layer of material which has a nite thickness and which has its own constitutive law that includes an elasticplastic response with failure. The layer is then modelled as a 3D continuum using conventional solid nite elements. This can be effective, but has several potential difculties. The elements are 3D continuum, but may need to be thin in comparison to other structural dimensions, which presents problems both in meshing and in obtaining a converged nite element solution. Also the material behaviour of the adhesive layer can be extremely complexthe strength properties of the adhesive may vary signicantly with the thickness of the layer and may also be highly dependent on the curing process. As a consequence, it is not easy to develop an analysis approach that can produce reliable results for a nite thickness adhesive modelled with continuum elements and a full constitutive model. Another analysis method available in Abaqus for analysis of failure propagation in adhesive joints is the VCCT. This methodology is based on linear fracture mechanics theory and can be used to predict the stability of existing cracks in adhesive joints and for simulation of crack propagation but will not cope with more complex failure propagation like crack ply-jumping or complex failure interaction between adhesive joints and adjacent adherents. Notably, the method can be used with signicantly larger element sizes at the crack front compared to cohesive element methods typically requiring much smaller element

123

Virtual testing of aircraft structures

93

sizes to provide accurate results but cannot be used to predict the initiation of failure. For an accurate assessment of adhesive joint damage and failure, the interaction between failure modes, or mode openings I, II and III must be included in the analysis. Again, such interactions must be well correlated against test data. Several types of interaction models are available in Abaqus for both cohesive contact and VCCT analysis methods, where the BenzeggaghKenane (BK) mixed mode fracture criterion is often used [21].

8 Analysis framework for virtual testing using nonlinear analysis The preceding discussion on detailed methods leads to the requirement to ensure that the modelling and analysis methods are t-for-purpose and validated against test. The topic can therefore be addressed through the following three nonlinear analysis building blocks: 1. 2. 3. material modelling (metallic and composite), fastener modelling, adhesive joint modelling.

To build condence in each of these analysis categories, a range of structural coupon tests with increased complexity must be carried out and close correlation with analysis models must be demonstrated. This implies using exactly the same type of element types, mesh topology and

density, methods and properties at each level of coupon test/analysis correlation. This methods validation framework is illustrated in Fig. 10, using the example of adhesive joint modelling. The process starts with simple coupon tests where the actual loading and crack opening mode is well understood; extends to a more complex 7-point bend test where the initial aw in the bondline is subjected to complex loading and mixed mode crack behaviour; and concludes with a coupon test of an actual aircraft structure. A fundamental requirement is to demonstrate that the detailed analysis and modelling methods employed will provide consistent accuracy and correlation with test results at all three levels of structural complexity. Combining the multi-scale analysis process as shown in Fig. 3 with the methods validation framework in Fig. 10 provides a general analysis framework for advanced nonlinear analysis of aircraft structures, as depicted in Fig. 11. The analysis framework is named after the shape of the wishbone found in common birds. It is vital to ensure that at the point of conuence between the upper and lower arms of the wishbone analysis framework there is a consistent set of analysis and modelling methods and processes that are used. In practice this means that for detailed failure predictions in the multi-scale analysis framework, validated methods from the lower arm of the wishbone must be used. By validated we mean that the methods have been demonstrated to provide accurate results at different structure complexity levels.

Fig. 10 Methods validation framework

123

94 Fig. 11 General analysis frameworkthe wishbone

M. G. Ostergaard et al.

In order to enable this integrated analysis framework to be developed and deployed, there is a requirement for it to be based on the consistent use of a common, feature-rich analysis tool that in turn creates the opportunity for future analysis developments and wide collaboration with external partners to Airbus. The wishbone analysis concept has proven to provide a robust analysis framework for the development and deployment of advanced nonlinear analysis methods and processes. Furthermore, and perhaps more importantly, the analysis framework provides a structured, logical and unied basis for exploitation and deployment of the virtual testing technology in an industrial context. 9 Implicit and explicit nite element methods Implicit nonlinear nite element analysis methods are currently the standard for static virtual testing simulations. Explicit methods are normally too expensive computationally for use in quasi-static type analysis problems. If dynamic effects and numerical noise are to be eliminated then run times become unmanageable. There are, however, certain cases where explicit nite element analysis methods are of signicant importance in static type calculations and where implicit and explicit methods can be used together. The Abaqus software has interfaces between the implicit and explicit solvers that enable the same analysis model to be used for both types and analysis. Examples include:

(a) manufacturing process simulations, (b) residual strength calculations where the static strength of impact damage initially can be assessed using Abaqus/Explicit for impact damage and subsequently using Abaqus/Standard for residual strength, (c) simulation of failure propagation and assessment of local failure. Abaqus/Explicit can be used to understand the likely failure sequence and nal result after the initial failure has been predicted using Abaqus/Standard. It is possible to simulate the structural failure by taking the implicit solution close to the predicted failure load level and, using compatible material, fastener and contact models in both solvers, to import the solution to Abaqus/ Explicit in order to complete the ultimate failure prediction. This is illustrated in Fig. 12 for a metallic box beam structure used for the testing of wing compression panels [22]. It can be difcult to use implicit solvers to model the progressive damage and failure of both materials and joints, because of the instabilities and bifurcations that can occur in the solution. Some form of damping or stabilisation is frequently required. Explicit solvers, on the other hand, do not have any such instability issues. In the near future, explicit nite element methods are likely to play an increasingly important role in the simulation of progressive structural failure events. The explicit technique provides an improved understanding of the effect of initial local failures which might result only in local load

123

Virtual testing of aircraft structures Fig. 12 Box-beam test and analysis

95

redistribution effects rather than catastrophic failure of the aircraft structure.

10 Robust analysis It is important to understand that increasing the complexity of material models by adding capabilities to simulate plasticity, damage and failure, does not necessarily by itself improve the accuracy of the simulation; nor does increasing the precision of the input data. Improved simulations arise through careful construction of a realistic analysis model that represents as closely as possible the real structure under real-world conditions. The preceding sections have highlighted the importance of using appropriate test data, appropriate levels of abstraction, and correlating models against experiment before embarking on predictive virtual tests. Another major consideration for the analyst is the level of uncertainty in the model. Uncertainty arises because many aspects of the real structure, including material properties, dimensions, and the initial state of the assembly, cannot be known with absolute certainty, and also because the physical structure will have some variability in both properties and state, from location to location within the structure, and from batch to batch. It is unlikely that an analysis model could be constructed that replicates the true property and state of the physical structure at every point, even if that data could be measured in the rst place. Therefore, it is important for the analyst to take account of uncertainty and variability in the simulation work. This is usually achieved by running several analyses to explore the effects of uncertainty, rather than running one single deterministic analysis.

Unfortunately, in some cases, small variations in material properties can have a signicant effect on the response of the model as well as the actual structure, particularly if the effects of plasticity, damage and failure are included. This high level of sensitivity to key failure parameters means it is very important to carry out a range of analyses to fully explore the effect of variability. Simulation of a structural test up to and including damage and failure requires extensive modelling of the behaviour of both materials and joints. The more complex the model, the more data is required to represent the complexity and the more correlation work is needed. Another consequence of the addition of more complex failure behaviour is that the initial state of the structure becomes more important. There is little benet in modelling the failure of fastened joint more accurately, if the starting point of the analysis differs markedly from the initial state of the real structure. Therefore, when increasing the accuracy of analysis models it is necessary, as far as is practical, to take into account initial stresses arising from component manufacture and assembly, together with the dimensional tolerances, variation in material properties and imperfections arising from the assembly process. Simple sensitivity studies can be used to gain adequate insight into the likely sensitivity of the structural response and failure mode to variations in properties and geometric imperfections. However, it is important to note that before considering any type of robust analysis, be it based on stochastic or other types of probabilistic methods or simple sensitivity studies, best practice principles must be followed in the modelling process. The baseline analysis model should be constructed to nominal, or if fully known, actual geometry data and actual measured material and fastener properties. If the baseline analysis model is not t

123

96

M. G. Ostergaard et al.

Fig. 14 Sub-model with loading pads Fig. 13 A380 wing in test structure

for purpose, then not even the most advanced probabilistic methods will deliver adequate results.

11 A380 wing certication Advanced nonlinear nite element analysis methods were used to solve several issues during A380 certication. The most signicant was to identify the root cause of the wing structural failure during the nal ultimate static test trial in Toulouse in 2006. The aircraft wing test structure is shown in Fig. 13. In order to illustrate the scales of deformation involved in this test, it is interesting to note that the maximum deection of the wing tip at ultimate load level is approximately 8 m. Both wings broke simultaneously, at the same location, at about 3% below the ultimate design load. The ultimate design load is dened as 1.5 times the maximum load that the aircraft structure will experience during in-service ight conditions (in turn dened as the Limit Load). Despite the fact that the test was so close to demonstrating the ultimate load strength capability, a large investigation was launched to identify the reason for the wing failure and to design a structural modication to achieve the certication of the A380 aircraft structure. Amongst other efforts launched, it was decided to create a detailed nonlinear nite element model of a section of the A380 wing box, bounded by both spars and with a span of 7 rib bays. There was no detailed nonlinear model of the complete A380 wing structure available, so it was decided to translate a global, but relatively coarse, linear MSC Nastran model (MSC Nastran, MSC Software Corporation, Santa Ana, CA, USA) of the wing structure into an Abaqus nite element model suitable for nonlinear analysis. This was

then used to drive the boundaries of the detailed model at the in-board and out-board cut-sections. The Abaqus model of the detailed wing box section is shown in Fig. 14 and includes all discrete load inputs (rubber loading pads) within the domain of the wing box section. The global analysis process used is illustrated in Fig. 15. This gure also shows that due to the coarseness of the global wing model, the detailed model had to include additional structure away from the zone of interest. This was required in order to introduce the loading correctly into the detailed model of the section where the failure was expected to have initiated (a zone of approximately 3 rib bays and from rear to front spar). Very little information was available about likely cause or location of the failure other than that the failure was unlikely to have occurred in the lower cover as this was largely intact at the end of the test. Every structural part within the zone of the wing box was modelled from nominal CAD geometry, mostly using Abaqus shell elements (S4 or S4R). For the top cover about 8,000 fasteners (rivets and bolts) were included in the model using the Abaqus mesh-independent fastener element as depicted in Fig. 16. The mechanical contact between skins and stringers and between top cover and rib feet was also modelled. Signicant effort was put into meshing the upper skin and stringers as consistently as possible using best practice modelling techniques as illustrated in Fig. 17. Likewise Figs. 18 and 19 show the meshing details used for spar and rib panels. Material and fastener properties were modelled using actual measured properties for the wing structure. Extensive material coupon testing was carried out to fully characterise the properties for the upper skin and stringer materials in particular. Detail coupon tests were machined from the test structure and used to characterise the stringer

123

Virtual testing of aircraft structures Fig. 15 Analysis process global and sub-model of wing

97

Fig. 16 Rivets between skin and stringers

Fig. 18 Spar and detail meshing

Fig. 17 Mesh of cleat, rib-feet and Stringers

and rivet properties as shown in Fig. 20. Detailed Abaqus models were used to correlate the analysis properties against the coupon test data using the same modelling and

meshing strategy as used in the detailed sub-model of the wing box structure. The coupon test programme also allowed the physical nonlinear shear and tension stiffness characteristics of the rivets to be determined. These were subsequently included in the detailed wing box model together with an interaction to describe the relationship between the shear and tension failure behaviour. As explained in previous sections, it is a fundamental requirement that all analysis methods are fully validated against test data. Extensive correlations were therefore carried out between measured and calculated wing deections

123

98

M. G. Ostergaard et al.

Rib-Foot & Boom

data, measured strain levels at all strain gauge locations in the wing box section of interest and other information available such as known permanent deformation. Figure 21 shows the correlation between the measured wing deections along the wing front spar and the results of the global nonlinear nite element model that was used to drive the detailed sub-model. It illustrates that the global wing model provides an excellent representation of the global wing stiffness. All strain gauges used on the test structure were modelled explicitly as depicted in Fig. 22, which enables a

Rib-Web
8000

Front Spar Vertical Deflection

2000
1000 0 0 5000 10000 15000 20000 25000 30000 35000 40000

Stiffeners

Distance along wing from wing -root (mm)

Fig. 19 Mesh of rib-foot and beam, rib web and stiffeners

Fig. 21 Front spar vertical deection, test and analysis comparison

Fig. 20 Rivet shear and tension coupon modelling, plus photo of test rig

123

Virtual testing of aircraft structures Fig. 22 Modelling method for strain gauges

99

Fig. 23 Outer skin strain correlationglobal FEM

Fig. 24 Outer skin and stringer strain correlationdetailed submodel

straight forward correlation between measured and calculated strain levels. The following two gures (Figs. 23, 24) show an example of correlation against strain gauges for the global

and detailed model, respectively, located on the outer skin surface and stringer ange as shown in Fig. 22. The strain gauge correlation clearly shows that the global nite element model is capable of calculating the skin strain levels accurately up to the point where signicant buckling occurs in the top cover. However, after that point the model is not capable of capturing the detailed local post-buckling response. The detailed sub-model, however, does calculate the buckling correctly and is able to predict the strain levels very well in the outer skin surface as well as on the stringer free ange. It should be noted that due to the highly complex buckling taking place in the top cover near ultimate load, the detailed strain correlation is sensitive to the actual location of the strain gauge and some deviation from intended location is possible during installation. Overall, all the available evidence veried that the detailed sub-model and global wing nonlinear nite element model represented the actual A380 wing box structure very well and that the detailed model could be used to calculate the nonlinear deformation behaviour, including the effects of post-buckling. Once it had been established that the detailed wing box model was t for purpose, the investigation focussed on the identication of the root cause of the structural rupture. Every possible stress concentration in the structure modelled was identied and investigated using rened meshes and detail. An example is shown in Fig. 25 for one of the rib panels. However, all evidence suggested that the rupture had occurred in the top cover as this was where the highest stress and plastic strain levels were present. The top cover was subjected to extensive skin buckling resulting in very complex post-buckling behaviour. This is shown in Fig. 26 at ultimate load level, i.e. 3% above the actual wing structural failure load level.

123

100 Fig. 25 Local stress concentration on rib stiffener

M. G. Ostergaard et al.

Fig. 26 Top cover buckling, sub-model

The buckling calculated is shown in more detail in Fig. 27 using a scale factor equal to 5. The analysis carried out conrmed that global panel buckling (from rib to rib) would occur at ultimate load exactly, which was as predicted by the conventional stressing methods used for design of the structure. As it was now conrmed that the root cause could not be explained by material rupture or global buckling, the attention was now focussed on the rivets and bolts used in the top cover. Figure 28 shows that as a consequence of the skin buckling, very localised separation or gapping occurred between skin and stringers (shown at 1.449 Limit Load) and for one zone in the top cover in particular. Screening all the fasteners in this region for shear and tension loads, it became evident that the local skin buckling resulted in additional nonlinear shear and tension forces in

Fig. 27 Detail of top cover buckling at 1.5 9 LL

the rivets, as shown in Fig. 29. It can be seen that initially the shearing force carried by the rivet is increasing linearly with the load as the rivet resist in-plane shear forces in the panel (wing torsion) and that the tension force is constant and equal to the small preload dened in the rivet model. At the onset of initial skin buckling, the local buckling results in separation forces between skin and stringers, which can be seen as a sudden increase in the rivet tension. Additional shearing forces in the rivets are also

123

Virtual testing of aircraft structures Fig. 28 Contact opening in stringer attached ange

101

Tension and shear forces for a failed rivet

1.2
1.01

Tension

Force (N)

Shear Circular Interaction

Onset of local skin buckling

1 0.8 0.6 0.4 0.2 0

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

Load (LL)

Fig. 29 Tension and shear forces for failed rivet

observed after onset of buckling, which are caused by the complicated buckle patterns resulting in curvature changes both span-wise and chord-wise in the top cover panel. These local changes in curvature will sometimes increase the rivet shear forces and sometimes reduce them, depending on the rivet position in relation to the local buckles in the panel. Screening all fasteners showed that more than one rivet, on one stringer, was predicted to fail at about 1.44 to 1.459 Limit Load. This was in good agreement with the actual wing rupture at 1.459 Limit Load. No other plausible failure mode was predicted below 1.59 Limit Load and the root cause of the rupture had therefore been identied as being caused by rivet failures. Not only was it possible to identify the cause of the rupture but it was also possible to fully understand the underlying issues that had to be considered for the structural modication to demonstrate adequate strength. Figure 30 shows the local panel deformation in cross section AA indicated in Fig. 28 and illustrates the level of detail considered and captured in the analysis.

It clearly shows that the local buckling resulted in gapping between skin and stringer. A design modication, using straps along the stringer feet both sides, was therefore designed to avoid the separation and the rivets were replaced locally with bolts. The preceding section provides only a brief description of the analysis models and processes used to identify the reason for the A380 wing structural failure. It is not possible to fully detail, in this paper, the many different analyses and sensitivity studies carried out during the intense investigations. However, it is important to state that advanced nonlinear nite element analysis had been used successfully to identify and explain an extremely complicated industrial structural analysis problem and to contribute to the process to achieve the certication of the A380 wing structure.

Circular interaction

12 Summary This paper has provided an overview of the virtual testing technology of aircraft structures in Airbus subjected to static loading conditions. The importance of condence and best practice associated with a virtual testing approach has been discussed. A general frameworkthe wishbonefor working with multi-scale analysis methods and the various challenges facing the analyst have been presented, with particular focus on detailed failure prediction methods for materials and fasteners. The construction of FE models to include plasticity, damage initiation, damage evolution and ultimate failure requires careful consideration of the behaviour of the underlying materials as well as of the joints and fasteners between components. Modern FE tools are capable of

123

102 Fig. 30 Cross-section showing fasteners

M. G. Ostergaard et al.

simulating complex damage processes, but increasing the level of complexity requires additional parameters, which need to be obtained from tests and correlated against experimental results. There are therefore signicant trade-offs to be considered in relation to the expediency of running many relatively simple analyses, the difculty of obtaining and correlating complex material behaviour, and the potential risk of generating misleading results from apparently advanced simulations that have not been properly validated. However, there is no doubt that when used carefully, with due regard to the derivation and validation of model data and the trade-offs of complexity against efciency, the use of nonlinear FE analysis can have a signicant impact on the development and structural strength assessment of advanced aircraft structures. A number of key enablers have been identied in order to improve virtual testing capabilities still further. These include: detailed modelling and meshing methods, automatic meshing and modelling methods from CAD denition based on consistent meshing rules, automatic composite property and lay-up capabilities from CAD to CAE, large-scale computations and increased use of detailed modelling, based on continuous improvements to highperformance-computing capabilities, efcient multi-scale analysis methods and screening processes to identify critical structures,

t-for-purpose detailed failure models for materials and joints and in particular for composite materials, robust quality processes.

Airbus has developed strong partnerships with both software providers and research institutes, including various European universities, in order to make progress on the above key enablers. The EU FP7 research project MAAXIMUS (more affordable aircraft structure through extended, integrated, and mature numerical sizing) is an example of a major project designed to make progress on virtual testing ` methods, which has both Airbus and Dassault Systemes SIMULIA as partners [23]. Many of the analysis short-comings discussed in previous sections are addressed in the frame of MAAXIMUS. It is expected that within the timeframe of the project, the size of the models that can be handled in a nonlinear nite element approach can be increased by between 1 and 2 orders of magnitude. The Giga-DOF model (10^9 DOF) is the gure being used as a target for the developers in the project. In recent years, signicant progress has been made in exploiting virtual testing methods for the solution of complex industrial structural issues, such as the A380 wing certication described in this paper. However, it must be acknowledged that virtual testing methods of composite aircraft structure are still being developed and will continue to provide the analyst, the software developers and academia with signicant challenges.

123

Virtual testing of aircraft structures

103 in notched CFRP laminates under longitudinal compression. Compos. Sci. Technol. 70, 1223 (2010) Pinho, S. T., Davila, C. G., Camanho, P. P., Iannucci, L., Robinson, P.: NASA/TM-2005-213530 NASA (2005) Camanho, P.P., Davila, C.G., Pinho, S.T., Iannucci, L., Robinson, P.: Prediction of in situ strengths and matrix cracking in composites under transverse tension and in-plane shear. Compos. Part A: Appl. Sci. Manuf. 37, 165 (2006) Moes, N., Dolbow, J., Belytschko, T.: A nite element method for crack growth without remeshing. Int. J. Numer. Methods Eng. 46(1), 131150 (1999) Ladeveze, P., Le Dantec, E.: Damage modelling of the elementary ply for laminated composites. Compos. Sci. Technol. 43, 257 (1992) ` Ladeveze, P., Lubineau, G., Marsal, D.: Towards a bridge between the micro- and mesomechanics of delamination for laminated composites. Compos. Sci. Technol. 66(6), 698712 (2006) Camanho, P.P., Davila, C.G., Pinho, S.T.: Fracture analysis of composite co-cured structural joints using decohesion elements. Fatigue Fracture Eng. Mater. Struct. 27(9), 745757 (2004) Pinho, S.T., Davila, C.G., Camanho, P.P., Iannucci, L., Robinson, P.: Failure models and criteria for FRP under in-plane or threedimensional stress states including shear non-linearity, NASA/ TM-2005-213530. NASA Langley Research Center, Hampton, (2005) Moes, N., Belytschko, T.: Extended nite element method for cohesive crack growth. Eng. Fract. Mech. 69(7), 813833 (2002) Meschke, G., Dumstorff, P.: Energy-based modeling of cohesive and cohesionless cracks via X-FEM. Comput. Methods Appl. Mech. Eng. 196(2124), 23382357 (2007) Hettich, T., Hund, A., Ramm, E.: Modeling of failure in composites by X-FEM and level sets within a multiscale framework. Comput. Methods Appl. Mech. Eng. 197(5), 414424 (2008) Kenane, M., Benzeggagh, M.L.: Mixed-mode delamination fracture toughness of unidirectional glass/epoxy composites under fatigue loading. Compos. Sci. Technol. 57(5), 597605 (1997) FP6 MUSCA non-linear static multi-scale analysis of large aerostructures: http://cordis.europa.eu/fetch?CALLER=FP6_PROJ& ACTION=D&DOC=15&CAT=PROJ&QUERY=0125d43f7184: 23ab:79c5c95f&RCN=75782 FP7 MAAXIMUSmore affordable aircraft through extended, integrated and mature numerical sizing: http://www.maaximus.eu/

In addition, it is important that particular attention is paid to the development of best practice in methods and processes in order to enable industrial deployment. The correct combination of skills, tools and processes used within the wishbone analysis framework can then be used to maximise the benet of the virtual testing technology in an industrial context and to provide a shared platform for future collaboration between industry, academic partners and software providers.

11. 12.

13.

14.

References
1. Imbert, J.F.: Airbus, challenges in aircraft structure analysis. ESA/NAFEMS Seminar on Engineering Quality, verication and validation. Noordwijk, 6 December 2007 ` 2. Prior, A.: Dassault Systemes, nonlinear simulation of large scale aircraft structuresimplications for certication methodology and high performance computing infrastructure, NAFEMS World Congress, June 2009 ` 3. Prior, A.: Dassault Systemes, simulating damage and failure in aircraft structures, RAeS Conference: challenges for the next generation-concept to disposal, October 1416, 2008 4. Brown, T.: Airbus, working to meet the challenges of next generation composite wing structural design. RAeS Conference: challenges for the next generationconcept to disposal, 1416 Oct 2008 5. European Aviation Safety Agency, Certication Specications for Large Aeroplanes CS-25 Amendment 5. 5 Sept 2008 6. Davila, C., Camanho, P.P., Turon, A.: Effective simulation of delamination in aeronautical structures using shells and cohesive elements. J. Aircr. 45, 663672 (2008) 7. Krueger, R.: Virtual crack closure technique: history, approach, and applications. Appl. Mech. Rev. 57(2), 109143 (2004) 8. Krueger, R., Ratcliffe, J.G., Minguet, P.J.: Panel stiffener debonding analysis using A shell/3D modeling technique. Compos. Sci. Technol. 69, 23522362 (2009) 9. Krueger, R.: An approach to assess delamination propagation simulation capabilities in commercial nite element codes, NASA/TM-2008-215123 (2008) 10. Gutkin, R., Pinho, S.T., Robinson, P., Curtis, P.T.: On the transition from shear-driven bre compressive failure to bre kinking

15.

16.

17.

18. 19.

20.

21.

22.

23.

123