You are on page 1of 9

8as JORNADAS DE FRACTURA - 2002

LIFE PREDICTION USING FINITE ELEMENTS IN COMPLEX GEOMETRIES


R.A. CLUDIO*, R. BAPTISTA*, V. INFANTE**, C.M. BRANCO ** * Department of Mechanical Engineering, EST/Instituto Politcnico de Setbal Campus do IPS, Estefanilha, 2914-508 Setbal ** Department of Mechanical Engineering, Instituto Superior Tcnico Av. Rovisco Pais, 1096 Lisboa Codex
Abstract. One of the major problems in numerical determination of crack propagation parameters is the mesh generation around the crack tip. Most finite element programs dont have, yet, tools for crack meshing using collapsed quarter point elements. This paper reports some of the principal difficulties that can be found modelling 3D cracks in complex geometries. The most frequent tasks, like finite element program choosing, elements that should be used, the mesh generation process and nonlinear calculus are described. Some tricks and a list of complete solutions are given for the most frequent problems. Finally are presented references where the presented instructions were applied successfully. Resumo. Um dos maiores problemas na determinao numrica dos parmetros da Mecnica da Fractura a gerao de malha junto fenda. A maioria dos programas de elementos finitos no tem capacidade para gerao de malhas usando elementos colapsados e singulares. Este artigo faz referncia s principais dificuldades que se encontraram em modelao 3D de fendas, em geometrias complexas. So descritas as principais etapas a seguir, tais como a escolha do programa de elementos finitos, elementos a usar, processo de gerao de malhas e clculo no linear, apontando algumas pistas para a resoluo da maioria dos problemas encontrados. Finalmente so apresentadas referncias para trabalhos onde estas instrues foram aplicadas com sucesso.

INTRODUCTION

Fatigue material properties can be obtained from small laboratory specimens at low cost. This paper describes the procedures used to find K solution for two cases. In both cases 3D cracks were modelled. Due to the geometry complexity with the cracks and because singular collapsed elements should be used, to get better solutions, sometimes problems appeared. The majority of these problems are here reported, including solutions that had been taken. First of all this paper describes the 3D elements normally used in fracture mechanical problems and the procedures used to modify these elements to improve solution quality. The next step was to choose FE code. In this case ANSYS and ABAQUS were used and the reasons why are reported. Mesh generation was when more time was spent. Its presented some features used to improve mesh quality and reduce time, like the volume subdivision and the use of parameters to define geometry and mesh. It is also explained how conversion from ANSYS to ABAQUS was done.

The Finite Element Method (FEM) is a very powerful technique, frequently used in design to predict stress distributions. However, sometimes a simple stress analysis is not sufficient. Low cycle fatigue crack propagation predictions are an essential part in the life assessment of critical components working in hard conditions and when a fail can cause catastrophic loses. Testing real size components or event test components in simulated conditions can be very expensive and time consuming. Predicting operating life can be an option, but to do that is essential to determine fracture mechanic parameters, like for example K. Most components in service have geometries where K solutions arent derived. Numerical procedures like the FE technique are commonly used to solve these solutions. One of the most difficult tasks in K evaluation using FEM is modelling cracks in geometries that in most situations are complex. Some commercial programs are well suited for 2D were isnt difficult to introduce a crack in a shape using collapsed quarter point elements. Generating a 3D fracture model is considerably more involved than a 2D model.

8as JORNADAS DE FRACTURA - 2002

Some problems with processing are also reported including how convergence in some elastic-plastic problems was obtained. The technique used to get K parameters was to obtain J integral directly from FE postprocessor. With K solutions calculated by FE method, together with appropriate fatigue data and crack propagation laws, its possible to obtain component life. All that work resulted from two cases studied by the authors a Gas Turbines disc and a T Joint improveded by Hammer Peening. A complete description and results for the two cases are published in ref. [1], [2] and [3].

FINITE USED

ELEMENT

PROGRAMS

One thing to do when working with FE is to choose the program. Part of the FE programs available arent able to calculate fracture mechanic parameters. Some programs could be good in some modules but bad in others. The best thing to do is to take a look for the program manuals and for the samples that are given, to verify if that program can really do what is wanted without too much effort. In both cases that were modelled it hasnt possible to find any program that was good enough, in all modules, for fracture mechanics analyses. The authors only had available for use ABAQUS and ANSYS. ABAQUS is a very powerful program, which is one of the most used software by scientific community. In our work group we already had some experience with ABAQUS. ABAQUS preprocessor, for us, is the horst part of the program. At two years ago, when this work was started, ABAQUS had a module called ABAQUS/Pre for pre-processing purposes. ABAQUS/Pre was a graphical interface for problem definition that generates a text file for ABAQUS/Standard module. The ABAQUS/Pre module had some tools for mesh generation around the crack tip but this tools didnt work well for 3D purposes, as Miranda [4] observed. Now instead of ABAQUS/Pre, the ABAQUS program has a new product called ABAQUS/CAE. This is a complete ABAQUS environment that provides interface for creating, submitting, monitoring, and evaluating results from ABAQUS/Standard and ABAQUS/Explicit simulations. But ABAQUS/CAE hasnt yet any tools for mesh generation around crack tip, which makes this program not appropriated for fracture

mechanic problems. Its also possible to write the input file by hand without any graphical interface, but is a very difficult task when its a complex geometry. ABAQUS/Standard is a module that reads an input file, prepares the problem for solving and solves the equations. This module has also some post-processor capacities including solving the most common fracture mechanics parameters. With ABAQUS is possible to solve the problem, but without a good pre-processor is difficult to model a complex geometry with a crack. Then, it was necessary to search for another program. The next step was to verify what ANSYS could do. ANSYS is able to calculate Fracture Mechanical parameters like stress intensity factors, the J-integral, and the energy release rate. In meshing capacities was found that ANSYS have only tools for mesh generation around crack tip in 2D geometries. After, some tests made, it was verified that if a volume was generated from a 2D mesh in one area and using the sweep function, its possible to obtain a 3D crack tip mesh. After that was verified that ANSYS has a very good graphical interface with powerful tools for mesh generation. With ANSYS its also possible parameterise the problem, very important when the crack size has to be updated. When the authors tried to solve the problem, a lot of errors with the mesh appeared without apparent solution. The option taken was to model the problem using ANSYS, then exporting the data to ABAQUS and calculating the solution with ABAQUS. For summarise, the FE programs used were: - Pre-processing, ANSYS 5.5 Due to the complex geometry, ANSYS code was used instead of ABAQUS. Some programs had to be made to convert node and element data to the ABAQUS code. - Processing, ABAQUS 5.8.14 code, with the STANDARD module. - Post-processing, ABAQUS 5.8.14 code, with the POST module.

ELEMENTS USED

3.1 Isoparametric Singular Elements The isoparametric singular elements are obtained from the classical 2D 8 node and 3D 20 node isoparametric elements, positioning the intermediate nodes near to the crack at

8as JORNADAS DE FRACTURA - 2002

quarter-point positions. Special crack tip elements are unnecessary. Crack tip elements can be obtained from standard isoparametric elements, changing the node positions to quarter-point [5]. Henshell et al [5] and Barsoum [6] found that in an 8 node quadratic isoparametric element the strain field becomes singular at the corner node if the mid-side nodes are placed at the quarter points of the sides emanating from this corner node. The change in the positioning of the mid-side nodes in isoparametric elements for achieving the desired singularity was extended to 3D. The region of square root singularity behaviour exists in every cross sectional plane orthogonal to the crack edge.

uniformly through the thickness in its plane, J and G are equivalent [9], so K can be obtained from J using equation (1).

K=

EG 1 2

(1)

A universal optimum size does not exist and it is difficult to give general guidelines for proper use of singular elements for all cases of crack problems. But, using appropriate transition element sizes, a universal optimum size lies within a 15-25% crack length for a 5% error bound. A convergence study is recommended to identify the lower bound of the error, when accurate results are searched. If the results are being used to calculate the stress intensity factor and if the method for calculating this parameter is being on the far-field FE solution, the results can be insensitive to the singular element size. Murti et al [10] concluded that the maximum span angle for each singular triangle is /2 and hence acute singular elements should be used. So, it is required that the minimum number of singular elements enclosing the crack tip is five (preferably six). The non-uniformity of the singular elements can be measured by the ratio of maximum to minimum side length. This ratio should be as close as possible to unity for best results [10]. This behaviour is partly due to inherent properties of the isoparametric finite elements (FE) itself. With large aspect ratio and distortion, there is a considerable increase in the degree of non-linearity of isoparametric mapping, which leads to a decrease in accuracy. More importantly, the radial variation within the element is a function of element size. Thus, when non-uniform singular elements are used, the local radial variation from one element to another varies, leading to an improper modelling of crack-tip behaviour, causing a decrease in accuracy [10].

3.2 Collapsed Elements The collapse of one side in the 8 node 2D element generates, at the collapsed side, the strain needed. In the 20 node 3D brick element, the singularity is obtained at the crack front when a face is collapsed. Once the corner nodes have been positioned, only the mid-side node on crack front doesnt have its position fixed. In the majority of the problems, the most efficient mesh design at the crack tip has proven to be the spider web configuration (fig. 1), which consists of concentric rings of elements that are focused toward the crack front [7]. The inner-most ring of elements are degenerated to triangle prisms. The spider web design facilities a smooth transition from a fine mesh at the tip to a coarser mesh remote from it. Elastic analysis of K can be accomplished with relatively coarse meshes, since modern methods eliminate the need to solve local crack tip fields accurately.

MESH GENERATION

Figure 1 Spider Web configuration. In the linear elastic case, G can be expressed as the scalar product of the J integral vector and a vector in the plane of the crack and normal to the crack front [8]. So, for a 2D plane strain tensile crack, with the crack advancing

Mesh generation is one of the most difficult tasks when modelling a FE problem. In fracture mechanics problems, the most important region to model is around the edge of the crack. Generating a 3-D fracture model is considerably more involved than a 2-D model, especially when the geometry is not regular. The use of hexahedral elements instead of tetrahedral provides more accurate solutions

8as JORNADAS DE FRACTURA - 2002

[11] and [12], but was consumed more time and promote more problems in mesh generation. With tetrahedral elements the mesh generation in ANSYS is totally automatic, including in irregular volumes. The use of hexahedral elements requires regular volumes and the mesh generation must be done with simpler functions like extrude and sweep. In all simulations made hexahedral elements were always used. The first step was to define the geometry. The software used was the MECHANICAL DESKTOP (a CAD program), because is a parametric based program with superior 3D functions for geometry definition. This program was also chosen because is possible to can be use the explode function. With this AUTOCAD function its possible to change from, parametric to nom parametric, volumes to surfaces and surfaces to lines easily. This is very important, if a volume has to be subdivided in small regular volumes or to made some repairs. Geometry subdivision is a very important task if the objective is to generate good quality meshes. To use hexahedral elements in ANSYS the region has to be divided in regular volumes (fig. 2).

the same topology and the volume cant have any discontinuities in the middle. As said before were used some typical AUTOCAD functions to subdivide the exploded geometry from MECHANICAL DESKTOP. To export the geometry for ANSYS, for crack definition and mesh purposes, the IGES standard interface was used.

Figure 3 The sweep process in mesh generation. Careful must be taken dividing the volume; sometimes small volumes needed to be created near larger ones resulting in meshes too distorted. The cracks were defined parametrically, in ANSYS, as function of crack depth and half crack length at the surface, a and c. Parametric modelling is available in most common FE codes. In the cases studied a base model was already meshed without the crack tip region defined. When the problem was loaded in ANSYS two dialogs appear to enter the crack size. With crack size defined, the geometry and mesh were updated for the size entered. The parameterisation was one of the most time consumption in all process. Some problems appeared when meshing with crack tip elements. ANSYS generated well the collapsed quarter point elements in areas, but when the sweep function was used, the nodes changed gradually to position until the mesh reached the target area. To correct this problem a macro was made, with ANSYS functions, to change the nodes that were in wrong position to position. The elements, in cracks with a circular front shape, have the same appearance along the front. But when modelling cracks with elliptical shape the elements along the front have different aspect ratios resulting in more distorted meshes, making the a/c relation a limitation for crack generation process. Another thing that users should be careful about is with the total number degrees of freedom (DOF). With the available hardware it

Figure 2 Subdivisions made to help the generation of regular meshes.

Regular volumes are those witch can be generated using a base area that is extruded or sweeped. A regular volume has only 6 areas (fig. 3). The source and final area must have

8as JORNADAS DE FRACTURA - 2002

was only possible to solve problems with less than 200.000 DOF. The hardware used was a PC, Pentium III 500MHz processor with 320MB of RAM and 10GB. Our limitation in solving large problems was only the size of the disk. For the most refined meshes, submodelling was carried out in the area near the crack, using the medium mesh solution to establish the boundary conditions (fig. 4).

something similar to what the old PREABAQUS program did, that was writing a node and element file and then adding some information like loads, boundary conditions, material properties and some commands for data requesting. This operation consists in building the INP (input) file. A program was made in VISUAL FORTRAN to change the list format, of nodes and elements, to ABAQUS input file format. The nodes and elements are used for mesh definition.

NONLINEAR SOLUTIONS

Figure 4 Sub-modelling technique Another problem that sometimes appears in mesh generation is when the crack size becomes near an edge (fig. 5). In that case its impossible to do the mesh. The solution that was taken was to extrapolate the K results for the crack sizes that couldnt be done.

There are three basic notions to regard when executing a FE analysis: Analysis Step: the basic division of analysis, consisting on a specific loading situation, boundary conditions, type of analysis and output variables. A fully completed analysis is composed of several steps, which are executed by the processor following a previous defined order. Step Increment: if necessary each step is divided into several increments. When executing a nonlinear analysis the processor has to follow a law, which implies the total load division into smaller portions, in order to simulate the nonlinear behaviour. The initial value of the step increment may be suggested by the user, or calculated by the processor, but influences tremendously the final result. Iteration: it is an attempt to obtain a solution for the equilibrium equations, in each increment. When the processor doesnt reach the equilibrium for a given increment in a single iteration, then it begins a new iteration until the equilibrium is found. If the equilibrium is not obtained in a predefined number of iterations, the size is reduced to the original increment. 6.1 Iterations increment and convergence of an

Figure 5 A large centre crack near an edge.

5 EXPORTING DATA FROM ANSYS TO ABAQUS As said before ABAQUS program was used to solve the problem and ANSYS to meshed the geometry. One additional task was to do a program to export data from ANSYS to ABAQUS because ANSYS doesnt have any export function. What was done was

In a nonlinear analysis the structure reply can be represented in the fig. 6. FE processor attempts to solve each increment using the initial structure stiffness K 0 = dP , based upon du the structure initial configuration u0 and updated load increment P, to calculate an initial displacement correction ca = ua u0 .

8as JORNADAS DE FRACTURA - 2002

iteration, repeating the process until the necessary convergence is obtained.

Figure 6 - Nonlinear load-displacement curve Based on these results the processor can determine if the solution has reached the equilibrium. In order to do this a new structure stiffness is calculated, based on the new configuration ua, which allows it to obtain the internal forces for this iteration Ia. The difference between the total applied load and these internal forces are named as the residual force (fig. 7):

Figure 8 Second iteration. On the second iteration the processor uses a new structure stiffness, already calculated, and the same load increment value, in order to calculate a new configuration ub. The determination of the new value for the residual force Rb, a new displacement correction cb = ub u a and a new incremental displacement u b = u b u 0 is allowed (fig. 8). All these values are evaluated in order to verify the iteration equilibrium and increment convergence. The process is repeated until these two are verified. This way, in each iteration is necessary to form a model stiffness matrix and to solve the equilibrium equations. Each iteration is equivalent, in computational effort, to conduct a complete linear analysis, which is solved in a single increment and iteration. The computational effort to solve a nonlinear problem is several times superior than for analysing a linear one, without the guarantee of the solution convergence. The initial increment value of each step is essential. For ABAQUS if this value is not given to the processor, itll begin from the largest possible reducing the increment value to 25 %. If after 16 iterations the solution havent reach the equilibrium, this process is repeated until the convergence is obtained. From this point on, the processor will adapt this increment in the running simulation. Being so all initial effort of searching the best increment value may be speared if the user suggests a good estimation.

Ra = P I a

(2)

If Ra is null at every degree of freedom in the model, point a in the next figure would lie on the load-deflection curve, and the structure would be in equilibrium.

Figure 7 First iteration in an increment. In a nonlinear problem this value is always different of zero, therefore the processor as to compare with a defined tolerance in a way to verify the iteration equilibrium. By definition this tolerance is 0.5 % of the averaged force applied to all the structure DOF along the time. The processor throughout the simulation computes this average. If the residual force is less then the current tolerance value, the processor accepts ua as an equilibrium configuration, but not before verifying the increment convergence. In order to do this the processor calculates the incremental displacement value:

POST PROCESSING

u a = u a u0

(3)

and verifies if ca is not superior to 1 % of this value. If so, the processor executes another

Near the free boundaries, when K solution is derived from FE, the solution isnt good. This happens because the stress field singularity changes from 1/ r to K.r , where is a parameter that can change from 1 to 0.5,

8as JORNADAS DE FRACTURA - 2002

depending on the Possion coefficient and angle between crack and surface [13], [14], [15] e [16]. Since the elements are only well suited for 1/ r singularity some errors appear because the singularity is different. When the angle between the crack and surface is 90 (the most common situation) Benthen [17] found an analytical solution for the singularity. In ref. [15] and [16] its possible to find solutions for some angles when = 0.3 . Some investigators solve this problem using fine meshes and extrapolating the solution from inside the body to the surface [16] and [18].

equations have some limitations because they can only apply to semi-elliptical crack in semiinfinite bodies. The stress field must be interpolated by a 3rd degree polynomial. In [2] and [1] is published a case where the Pommier equations were used to check K solutions.

THE STATE OF ART

Each time a new version of a finite element program comes to market new things appear related with fracture mechanics parameters evaluation. So probably soon are able programs to mesh in an automatic way geometries with crack tip elements. There are some programs like ZENCRACK (the most known) that are able to model automatically cracks in general shapes. Theses programs operate on an existing finite element mesh and introduce the crack by replacing some of the existing brick elements by a crack block. The only thing that the user has to do is to mesh the geometry without any crack. The crack shape can be fully arbitrary and can be used in mixed mode propagation. Since the crack front can be automatically updated, applying a propagation law with a simple algorithm, is possible to predict component lifetime and crack geometry. Dhondt [23] and [24] presents some works where he uses this methodology successfully.

VALIDATION OF RESULTS

Sometimes it s possible to find in literature K solutions for geometries and load case similars to the situation that is being studied. One of the most complete references in this area is Murakami [19] and [20] and the PD6493 document [21]. The most frequent situation is that there is nothing in literature similar to the case that is being studied. So direct comparation cant be made. The best thing to do is to be cautious generating the mesh, applying the right boundary conditions and analyse carefully the finite element solution in a post-processor. A very good indicator of mesh quality is J integral. J integral is an energy parameter, independent from contour to contour. In the finite element codes, J integral is a quantity that can be computed along different contours. If the result is not the same in differents contours this is an indicator that mesh quality is not good. This indicates a need for mesh refinement near the crack tip. Numerical tests suggest that the estimate from the first ring of elements abutting the crack front does not provide a high accuracy result, so at least two contours are recommended. As well contours far from the crack tip normally give unsatisfactory results. Only convergence in different contours is not sufficient to guarantee that the solution is good. The correct thing to do, is test also different meshes to ensure that the solution remains constant. Another way to check solution is using Pommier solutions [22]. These equations apply to engineering problems, and it is possible to obtain K knowing only the stress field near the crack zone. They are very useful to use with a finite element post-processor code. These

10

CONCLUSIONS

1. Finite element analysis is a powerful method to get fracture mechanic parameters. Can be used to replace part of the expensive tests in real components by fatigue tests in small laboratory test pieces. 2. Mesh generation still continues one of the most difficult tasks to perform calculus with finite elements when a crack has to be simulated. This is because finite element programs arent yet well suited to generate crack tip elements. 3. Parameterisation of the crack front size is recommended. Its difficult to do but as the first crack is modelled the next ones are much easier to do. This is because the crack size must be updated a lot of times to get a reasonable curve K vs a. 4. Determining the initial value for the load increment is fundamental when performing a nonlinear simulation. This value has a

8as JORNADAS DE FRACTURA - 2002

large influence in the final results and also makes a difference in the simulation running time. Therefore using the correct value provides good results with low computational effort. 5. Programs like ZENCRACK seams to be one of the most practical alternatives that exist today for meshing cracks in complex geometries.

[8] DeLorenzi, H.G., Energy Release Rate Calculations by the Finite Element Method, Eng. Fracture Mechanics, Vol. 21, n 1, pp. 129-143, 1985. [9] Li, F.Z., Shih, C.F., Needleman, , A Comparison of Methods for Calculating Energy Release Rates, Eng. Fracture Mechanics, Vol. 21, n 2, pp. 405-421, 1985. [10] Murti, V., Valliappan, S., A Universal Optimum Quarter Point Element, Eng. Fracture Mechanics, Vol. 25, n 2, pp. 237-258, 1986. [11] ABAQUS/Standard Users Manual, Vol I, II e III; ABAQUS/Theory Manual; ABAQUS/Standard Example Problems Manual, Vol I e II, ABAQUS/Standard Verification Manual; ABAQUS/Post Users Manual, Version 5.8; Hibbitt, Karlsson & Sorensen, Inc; 1999. [12] ANSYS manuals, Version 5.5, September 1998. [13] Dhondt, G., Analysis of the Boundary Layer at the Free Surface of a Half Circular Crack, Eng. Fracture Mech., Vol. 60, n 3, pp 273-290, 1998. [14] Fonte, Manuel A., Anlise da propagao de fendas semi-elpticas em veios sob flexo e toro, Tese de Doutoramento em Eng. Mecnica, Universidade Tcnica de Lisboa, Instituto Superior Tcnico, 1997. [15] Pook, L. P.; Crack Profiles and Corner Point Singularities; Fatigue Fracture Eng. Mater. Struct, 23, pp.141-150, Setembro 1999. [16] Bazant, Z. P.; Estenssoro, L. F.; Surface Singularity and Crack Propagation, Int. J. Solids and Structures, vol. 15, pp. 1411-1426, 1979. [17] Benthen, J. P., State of Stress at the Vertex of a Quarter-Infinite Crack in a Half-Space, Int. J. Solid and Structures, Vol. 13, pp. 479-492, 1977. [18] Carpinteri, A., Elliptical-Arc Surface Cracks in Round Bars, Fatigue Eng. Mater. Struc. Vol. 15 n 11, pp. 11411153, 1992.

REFERENCES

[1] Cludio, R. A.., "Previso de vida fadiga/fluncia em discos de turbina de turborreactores", MSc thesis, Technical University of Lisbon, Instituto Superior Tcnico, April 2001. Also available in electronic version at http://ltodi.est.ips.pt/rclaudio/ (in Portuguese). [2] Claudio, R., Branco, C. M., Gomes, E., Byrne, J., Life Prediction of a Gas Turbine Disc Using the Finite Element Method, 8th Portuguese Conference on Fracture, Vila Real, 20-22 February 2002. [3] Infante, V., Branco, C. M., Baptista, R., Gomes, E., Fracture Mechanics Analysis of Welded Joints Repaired by Hammer Peening, 8th Portuguese Conference on Fracture, Vila Real, 20-22 February 2002. [4] Miranda, Rui M. A.; Anlise de Fissuras de Canto pelo Mtodo dos Elementos Finitos Linear Elstico e Elastoplstico; Tese de Mestrado, Instituto Superior Tcnico, Universidade Tcnica de Lisboa, Dezembro de 1996. [5] Henshell, R.D., Shaw, K.G., Crack Tip Finite Elements are Unnecessary, Int. Journal for Num. Methods in Eng., Vol. 9, pp. 495-507, 1975. [6] Barsoum, R.S., On the Use of Isoparametric Finite Elements in Linear Fracture Mechanics,. Int. Journal for Num. Methods in Eng., Vol. 10, pp. 2537, 1976. [7] Antunes, Fernando J. V., Propagao de Fendas por Fadiga a Alta Temperatura em Inconel 718. Tese de Doutoramento, Faculdade de Cincias e Tecnologia da Universidade de Coimbra, 1999. (in Portuguese)

8as JORNADAS DE FRACTURA - 2002

[19] Murakami, Y.; Stress Intensity Handbook, Vols. 1 e 2, Pergamon Press, Oxford, 1987. [20] Murakami, Y.; Stress Intensity Handbook, Vol. 3, The Society of Materials Science, Japan, 1992. [21] PD6493 Document. British Standard BS 7910. Guide on Methods for Assessing the Acceptability of Flaws in Structures, 1997. [22] Pommier, S.; Sakae, C.; Murakami, Y.; An Empirical Stress Intensity Factor Set of Equations For a Semi-Elliptical Crack

in a Semi-Infinite Body Subjected to a Polynomial Stress Distribution; Int. J. of Fatigue, Vol. 21, pp. 243-251, 1999. [23] Dhondt, G.; Automatic ThreeDimensional Cyclic Carck Propagation Predictions with Finite Elements at the Design Stage of an Aircraft Engine, NATO RTO-MP-8, May 1998. [24] Dhondt, G., Automatic 3-D mode I crack propagation calculations with finite elements, Int. J. Numer. Meth. Engng., vol. 41, 739-757. 1998.

You might also like