This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

5 Theory Page 1

1st Page

Back

Main Menu

FE-Fatigue theory – Table of contents

**1. Introduction to fatigue analysis from finite element models ................. 3 2. Overview of new features in FE-Fatigue Rel. 5........................................ 6
**

2.1 Auto elimination by Group ............................................................................................ 2.2 Virtual strain gauges...................................................................................................... 2.3 Seam weld analysis........................................................................................................ 2.4 High temperature fatigue analysis................................................................................ 2.5 Vibration fatigue analysis.............................................................................................. 6 6 6 6 6

3. Fatigue Theory............................................................................................ 7

3.1. The S-N (High Cycle Fatigue) Approach ..................................................................... 7 3.2. The Local Strain theory of fatigue ............................................................................... 8 3.3. Multiaxial strain-life ...................................................................................................... 9 3.3.1 Introduction ............................................................................................................ 9 3.3.2. Co-ordinate System Rotation............................................................................... 9 3.3.3. Assessing Multiaxiality....................................................................................... 10 3.3.4. Critical Plane Analysis........................................................................................ 11 3.3.5. The Hoffmann-Seeger Method ........................................................................... 12 3.3.6. The Klann-Tipton-Cordes Method ..................................................................... 13 3.3.7 Non-proportional multi-axial methods ............................................................... 15 3.3.8 Multiaxial notch correction procedure ............................................................... 16 3.3.9 Cyclic Plasticity Modelling .................................................................................. 17 3.3.10 Multiaxial Rainflow Counting ............................................................................ 20 3.3.11 Fatigue Damage Calculation ............................................................................. 21 3.3.12 Critical Plane Methods....................................................................................... 22 3.3.13 Types of Cracks ................................................................................................. 23 3.3.14 Critical Plane Rainflow Counting...................................................................... 24 3.3.15 References.......................................................................................................... 25 3.4. Safety factor................................................................................................................. 26 3.4.1 Life based factor of safety................................................................................... 26 3.4.2 Stress based factor of safety .............................................................................. 26 3.4.3. Multiaxial safety factor ....................................................................................... 26 3.5. Multiaxial Safety Factor .............................................................................................. 27 3.5.1. The Definition of High-Cycle Multiaxial Fatigue............................................... 27 3.5.2. High-Cycle Multiaxial Fatigue Theories ............................................................ 27 3.5.3. McDiarmid Criterion............................................................................................ 28 3.5.4. Dang Van Criterion.............................................................................................. 30 3.6. Theoretical background to Spot Weld analysis ....................................................... 34 3.6.1. Introduction ......................................................................................................... 34 3.6.2. The fatigue analysis of spot welds - general description ............................... 35 3.6.3. Structural stress calculation.............................................................................. 35 3.6.4. Material properties .............................................................................................. 38 3.6.5. Damage Calculation............................................................................................ 40 3.6.6. A note on the subject of modelling spotwelds................................................. 41 3.6.7. References........................................................................................................... 42

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. 5 Theory Page 2

1st Page

Back

Main Menu

44 44 44 46 53 62 63 63 66 66 67 69 71 76 77 82 82

3.7 FE-based fatigue analysis of welded structures ....................................................... 3.7.1 Introduction .......................................................................................................... 3.7.2 Background .......................................................................................................... 3.7.3 BS7608 and other simple S-N methods ............................................................. 3.7.4 The “Volvo” approach for thin-sheet structures............................................... 3.7.5 References............................................................................................................ 3.8 Multiple mean stress curve analysis .......................................................................... 3.8.1 Interpolating life from a set of multiple mean stress life curves ..................... 3.9 Vibration fatigue ........................................................................................................... 3.9.1 Introduction .......................................................................................................... 3.9.2 Review of S-N analysis in the time domain ....................................................... 3.9.3 Simple fatigue analysis using time history recreation ..................................... 3.9.4 Fast fatigue analysis methods in the frequency domain ................................. 3.9.5 Comparison between fatigue analysis techniques ........................................... 3.9.6 FE based vibration analysis in the frequency domain ..................................... 3.9.7 Conclusions.......................................................................................................... 3.9.8 References............................................................................................................

4. Time Histories........................................................................................... 83

4.1 Why time histories are required.................................................................................. 4.2 Constant amplitude versus variable amplitude......................................................... 4.3 Obtaining and creating variable amplitude time histories ....................................... 4.4 Multiple load cases ...................................................................................................... 4.5 Using nSoft time history processing to increase calculation speed ...................... 4.6 Transient or time-step analysis .................................................................................. 83 83 85 86 86 89

5. Materials Data ........................................................................................... 90

5.1 Materials database ....................................................................................................... 5.2 S-N data......................................................................................................................... 5.3 E-N data......................................................................................................................... 5.4 Multiaxial data............................................................................................................... 5.5 Auto-generated data .................................................................................................... 5.6. Surface finish correction ............................................................................................ 5.7. Fatigue strength reduction factor.............................................................................. 90 90 91 91 92 94 95

**6. Principles of Linear Superposition ......................................................... 96 7. Calculation of fatigue life from rainflow matrices ................................. 98
**

7.1. Strain-life Fatigue Analysis from a Rainflow Matrix................................................. 98 7.2. Stress-life Fatigue Analysis from a Rainflow Matrix.............................................. 104

8. Auto elimination ..................................................................................... 105 9. Virtual strain gauges .............................................................................. 107

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. 5 Theory Page 3

1st Page

Back

Main Menu

**1. Introduction to fatigue analysis from finite element models
**

FE-Fatigue performs fatigue analysis from FE results. CAE environments include, for example, HyperMesh, which interfaces to FE-Fatigue by exporting stress/strain information in nCode's FES file format. This is comparable to exporting data for other solvers, such as NASTRAN. FE-Fatigue provides stress-life and strain-life analysis, the theoretical basis of which is described in this document. FE-Fatigue is part of the nSoft suite of products and a basic nSoft-E license provides the functionality for data display and manipulation. FATIMAS products are not essential to use FEFatigue however they provide additional useful functionality. Time histories of stress/strain (DAC files) can be exported from FE-Fatigue for use in FATIMAS modules. An FE-Fatigue installation comprises: • • • • • • • • fatfe (FE-Fatigue solver) fatres (FE-Fatigue results listing and sorting) mdm (Materials Database) fatduty (Duty Cycle Analyser) pvxmul (Multiple Channel Peak Valley Slicing) remdac, dacrem (RPC translators) sdrc2fes (.UNV to .FES conversion) fe2fes (Generic FE results to .FES conversion)

• Additional files include utilities and example files. • On-line documentation is also available electronically in Adobe Acrobat PDF format along with a printed user guide of worked examples. The documentation includes documents which detail the internal structure of FES and DAC files. The following pages show the process of analysing FE data with FE-Fatigue – the important files are as follows: • Partial FES file • Full FES file • Results file – created by the translator – file required by nCode analysis package – file in a format suitable for direct import into the appropriate pre and post processor.

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. 5 Theory Page 4 1st Page Back Main Menu Figure 1 FE Fatigue’s basic system diagram nCode n 1st Page Back Main Menu .

the following system diagram applies: Figure 2 System diagram with translation (if a translator is required it must be run first to create the partial FES file).FE-Fatigue Rel. 5 Theory Page 5 1st Page Back Main Menu Where the pre-and post-processor does not support the FES file. nCode n 1st Page Back Main Menu .

5 Theory Page 6 1st Page Back Main Menu 2. See Auto elimination for more details.FE-Fatigue Rel. See Virtual strain gauges for more information. See Seam weld analysis for more information.4 High temperature fatigue analysis Release 5 of FE-Fatigue supports the fatigue analysis of components at constant high temperature.2 Virtual strain gauges This feature allows the user to place a virtual strain gauge on the part being analysed. nCode n 1st Page Back Main Menu . 5 2. 2. See Vibration fatigue analysis for more details.3 Seam weld analysis Release 5 supports seam weld analysis from finite element stresses. which is in the form of Power Spectral Density (PSD) stress output. 2. 2.5 Vibration fatigue analysis Fatigue calculations may be done based on the results of a random vibration FE analysis. This provides an improved facility for measurement to CAE correlation and calibration to occur and improves the confidence of FE models by a more direct comparison of strain results. Element or node strains that are calculated in the FE solver are exported such that the strain output will emulate a strain gauge rosette at the same location on the actual part. 2.1 Auto elimination by Group The auto elimination feature of FATFE has been enhanced to allow the user to retain the top n percent of results for each material group present in the model. Overview of new features in FE-Fatigue Rel.

The S-N (High Cycle Fatigue) Approach It has been recognised since 1830 that a metal subjected to a repetitive or fluctuating load will fail at a stress level lower than that required to cause fracture on a single application of the load. see the nCode book of Fatigue Theory. Fatigue Theory 3.FE-Fatigue Rel.1. is best suited to that area of the fatigue process known as high cycle fatigue. In this region a strain based methodology must be used. For more information. The nominal stress method does not work well in the low cycle region where the applied strains have a significant plastic component. the nominal stress approach. The nominal stress method was the first approach developed to try to understand this failure process and is still widely used in applications where the applied stress is nominally within the elastic range of the material and the number of cycles to failure is large. nCode n 1st Page Back Main Menu . From this point of view. 5 Theory Page 7 1st Page Back Main Menu 3.

e. i. the cyclic loads are relatively large and have significant amounts of plastic deformation associated with them together with relatively short lives. the magnitude of the observed cyclic stresses were observed to be less than the tensile elastic limit and the lives long.FE-Fatigue Rel. This pattern of behaviour has classically been referred to as high-cycle fatigue. Traditionally. In this regime. 5 Theory Page 8 1st Page Back Main Menu 3. As duty cycles have became more severe and components more complicated. nCode n 1st Page Back Main Menu . The transition from low-cycle to high-cycle fatigue behaviour generally occurs in the range 104 to 105 cycles. This type of behaviour has been commonly referred to as low-cycle fatigue or more recently strain-controlled fatigue. see the nCode book of Fatigue Theory. For more information. greater than about 105 cycles. The Local Strain theory of fatigue The nominal stress approach has been used extensively in the study of premature failures of components subjected to fluctuating loads. The analytical procedure evolved to deal with strain-controlled fatigue is called the strain-life. local stress-strain or critical location approach. another pattern of fatigue behaviour has emerged.2.

and there are only two principal stresses which may be non-zero. These are implemented in FE-Fatigue. At any node or element in an FE-Fatigue analysis. no special algorithm corrections need to be applied to convert elastic stresses and strains to elastic-plastic. A simple critical plane approach is also included which can take into account the mobility of the stress tensor – this is described on page 11. and with the increasing detail comes a cost of analysis time and interpretation of results. Results from highly multiaxial elements with high levels of strain should be treated with caution. the local stress state can be assessed to determine the level of multiaxiality. Multiaxial strain-life 3. Damage models have been developed which can account for non-proportional and multiaxial loading conditions. alone are sufficient. procedures should be used to take into account the fact that the loading is nonuniaxial.3. 3. a full multiaxial notch correction procedure should be used. The angle remains constant also. If we assume that fatigue damage initiates at the surface of a structure then any direct or shear stresses normal to the surface are zero. FE-Fatigue requires stress or strain data in the plane of the surface. Hence there are three basic cases: • Case 1 Uniaxial stress state – there is one principal stress which is significantly larger than the second for the whole of the load history and whose angle does not change. They apply only to the local strain approach. In the current release of FE-Fatigue there is no multiaxial notch correction procedure available.FE-Fatigue Rel. • Case 2 • Case 3 For case 1. in which case a header flag will be set to indicate that only 2D stresses are present. In particular. most translators do not support this feature and therefore most data files will nCode n 1st Page Back Main Menu . Two such procedures are due to Hoffmann and Seeger [25] and Klann-Tipton-Cordes [26].1 Introduction The essential elements of any FE-based fatigue analysis are: • Identification of the loading environment • Establishment of the relationship between applied loads and local stresses and/or strains • The relationship between local stresses/strains and fatigue damage For multiaxial analysis.3. Either the biaxiality ratio or the angle of the maximum principal changes significantly through the time history. these three elements can be analysed at various levels of detail. This can be done by the software creating the partial FES file. but remains constant for the duration of loading. with the Z axis normal to that plane.3. The Neuber correction. Non-proportional stress state. it requires that the X-Y axis pair defines the plane tangential to the surface. Co-ordinate System Rotation In order to do any multiaxial assessment or analysis. or similar methods. These are described briefly in the following pages.2. Examples are Fatemi-Socie. In conjunction with regular uniaxial stress-strain predictions and damage models it can provide a life estimate in non-uniaxial cases and can be applied in both strain-life and stress-life damage models. For case 2. Bannantine and Wang-Brown. 5 Theory Page 9 1st Page Back Main Menu 3. However. Proportional biaxial stress state – the ratio of the two principal stresses is nonzero. For case 3.

Since the FES file does not contain co-ordinate system information. points whose maximum principal stress is less than a specified gate are excluded from the calculations. then we have a transformation matrix which will transform any stress tensor to a co-ordinate system X’-Y’-Z’ in which X’ is aligned with the maximum principal stress. The angle calculated by FE-Fatigue describes the angle of the absolute maximum principal stress vector with the X-axis of the planar co-ordinate system. This allows plots of the time histories of stress. the second column the second principal and the third the principal closest to zero. calculate S’ = TSTT where TT is the transpose of T.3. strain. 3. There are cases where this is not true. dominant angle and angle spread are calculated for the entire model and can be contour plotted. To remove the effect of low stress values. global measures of mean biaxiality ratio. In practice this value is in itself of little practical use.FE-Fatigue Rel. If we have a surface stress tensor of 6 components in a co-ordinate system X-Y-Z. In order to derive a transformation matrix. where zero indicates a uniaxial condition. Assessing Multiaxiality FE-Fatigue provides tools to assess whether uniaxial techniques are valid at any part of the model. Because the value of the biaxiality ratio and the angle can change for each point in a time history. a method is applied which can derive co-ordinate transformation information from the stress data itself. The Node/Element option on the Utilities menu in FE-Fatigue can be used to look at the state of stress at any particular node or element in detail. standard deviation of biaxiality ratio. 5 Theory Page 10 1st Page Back Main Menu contain stresses oriented in a co-ordinate system not compatible with the above requirement. The eigenvalues give the principal stresses and the eigenvectors the corresponding direction cosines. If the third principal is not zero. for example where an external or internal pressure is applied or where the coarseness of the FE model leads to inaccuracies in the extrapolation of results from element centroid or Gauss point location to surface nodes. absolute nCode n 1st Page Back Main Menu . we need to define a system X’-Y’-Z’ where X’-Y’ defines the plane tangential to the surface and Z’ is normal to the surface.3. The biaxiality ratio lies between –1 and +1. especially where stress levels are high.g. The eigenvectors can be stored in a 3x3 matrix T. To obtain principal stress values and directions. it is assumed that the two largest principal stress tensors lie in the plane of the surface and that the third principal is normal to the surface and has a value of zero. can be calculated and plotted. element by element basis. The biaxiality ratio. To transform a tensor in X-Y-Z to X’-Y’-Z’. either as a mean and spread over the whole model or on a detailed. which is the transformation matrix from the original co-ordinate system X-Y-Z to the co-ordinate system defined by the principals. Syx= Sxy ). defined as the ratio of the smaller absolute in-plane principal to the larger. then the results of the transformation may not lie in the surface tangential plane and the software will issue a warning. If we re-order the columns of T to make the first column correspond to the absolute maximum principal. This method relies upon the integrity of the data and will not work if any of the assumptions are violated. we can calculate the eigenvalues and eigenvectors of the following matrix: S = [ Sxx Sxy Sxz Syx Syy Syz Szx Szy Szz ] where the cross-terms are identical ( e. but the spread of the angle for all significant stress tensors shows the mobility of the principal vector. or near zero. biaxiality ratio and angle as well as cross-plots of biaxiality ratio and angle vs.

The local strain approach does not model cracks explicitly but attempts to predict initiation life on the basis of bulk stress and strain parameters. Critical Plane Analysis When the stress tensor is mobile it becomes necessary to use a critical plane approach.4. The critical plane is simply the plane on which the stress and strain parameters are calculated. The dominant value and the degree of scatter in the plot clearly show whether the data is uniaxial. proportional or non-proportional.FE-Fatigue Rel. Cracks typically start in shear mode and after a transition period grow in opening mode. The critical plane should not be confused with the crack plane.e. The method used to rotate the co-ordinate system of the stresses and strains by means of a tensor rotation such that the x-y plane of the new co-ordinate system lies in the critical plane. 3. This information can be used to determine the best choice of analysis route for a particular component.3. one which references specific planes rather than using combined stress or strain parameters such as Von Mises or maximum principal strain. 5 Theory Page 11 1st Page Back Main Menu principal stress. The period we call crack initiation life may include growth in a number of planes. i. For planes with an angle of intersection at 90 degrees. the equations for the strain and stress nCode n 1st Page Back Main Menu .

5 Theory Page 12 1st Page Back Main Menu normal to the plane and at a rotated angle φ (phi) is as follows: ε n = ε x cos φ + ε y sin φ + γ xy cosφsinφ 2 2 2 2 σ n = σ x cos φ + σy sin φ + 2τ xy cosφsinφ FE-Fatigue calculates the damage using the strain (E-N calculations) or stress (S-N calculations) at 18 angles from 0 to 170 degrees inclusive. The largest damage is then used as the result. and the principal axes not only rotate. Hoffmann-Seeger is used in conjunction with Neuber to correct the elastic stresses and strains to elastic-plastic. by solving the equations: 1 -n 2 σq σq ε q = ----. The Hoffmann-Seeger Method This method applies when loading is proportional. If the loading is non-proportional then most of the assumptions implicit in the method are violated. or additional calculation in FATIMAS. Only the rainflow matrix for the worst case angle can be extracted. Hencky’s flow rules and Masing material memory behaviour are also assumed.FE-Fatigue Rel.5.+ æ -----ö . which is assumed to be constant and equal to the elastic value. Neuber alone is insufficient as the cyclic stress-strain curve is from uniaxial data. If loading is biaxial. subject to the following assumptions: • The principal stress and strain axes are fixed in orientation • The ratio of the in-plane principal strains is constant • The uniaxial stress-strain curve can be extended for use with suitable equivalent stress and strain parameters such as the Von Mises parameters.è -ø E K′ And σ q ε q = σ eq ε eq σ eq = ---------E nCode n 1st Page Back Main Menu . Then the Neuber correction is made. First the elastic values of the signed Von Mises stress and strain (from the FE analysis) are computed: σ eq σ e1 2 2 = ---------.σ e1 – σ e1 σ e2 + σ e2 σ e1 and σ eq ε eq = ------E together with the strain biaxiality ratio e2/e1.3. but the principal stress and strain axes are no longer necessarily aligned with each other. 3. but time histories for all angles can be exported for preview. Hoffmann and Seeger suggest a method for extending use of the Neuber correction to multiaxial loading. hardening is in reality kinematic and not isotropic.

è -ø K' E is used to calculate a series of corresponding values of σq and εq. ν’ is calculated from: σq ν' = 1 – æ 1 – ν eö --------.FE-Fatigue Rel.+ æ -----ö .6.3. Essentially.+ ν' ε1 σ2 a = ----.ö ′ ø è 1–va 1 σ 1 = σ q ----------------------------2 1 – ae + ae ε2 ---. Tipton and Cordes makes a one-time modification of the cyclic stress strain curve according to the biaxiality ratio and then performs a normal Neuber calculation using the modified parameters. the method proposed by Klann.+ ν' ε1 a = -----------------ε2 1 + ν' ---ε1 nCode n 1st Page Back Main Menu .= -----------------ε2 σ1 1 + ν' ---ε1 3.è -ø Eε q 2 2 and a from: ε2 ---. The Klann-Tipton-Cordes Method Whilst the Hoffmann-Seeger method adjusts the stresses and strains cycle by cycle. 5 Theory Page 13 1st Page Back Main Menu Then the principal strains and stresses can be calculated as follows: σq ν' = 1 – æ 1 – ν eö --------. the equation: 1 --- σ q n' σq ε q = ----. This method is also known as the parameter modification method.-ø Eε q 2 è2 1 – ν' a ε 1 = ε --------------------------q 2 1–a+a ′ 1+a ε 3 = -ε 1 v æ --------------. For each pair of values.

5 Theory Page 14 where: 1st Page Back Main Menu ε2 ae – νe ---. E” is calculated explicitly from: E E'' = ---------------1 – νe a The modified parameters are used to carry out the Neuber correction in the normal way.FE-Fatigue Rel.+ æ -----ö E'' è K''ø to the calculated points.= ------------------1 – ae νe ε1 Pairs of values of ε1 and σ1 are then computed as follows: 1 – ν'a ε 1 = ε q ----------------------------2 1 – ae + a e 1 σ 1 = σ q ----------------------------2 1 – ae + ae These pairs of values represent a cyclic stress strain curve for a particular biaxiality ratio ae. nCode n 1st Page Back Main Menu . A new set of Ramberg Osgood parameters K” and n” is obtained by fitting the equation: 1 ---- σ 1 n'' σ1 ε 1 = ----.

incorporating rainflow cycle counting and material memory [12] and Miners rule [13]. through application of a cyclic plasticity model.3. 5 Theory Page 15 1st Page Back Main Menu 3. FATIMAS [14] addresses problems where the strains can be measured with a rosette. Such methods are available in FATIMAS [14] and FE-Fatigue. σy and σxy. For these problems the process can be summarised by the flow chart in: 3 strain histories from an elastic-plastic FE run in the plane of the surface OR cyclic plasticity modelling 3 strain histories from a strain gauge rosette OR cyclic cycle plasticity 3 stress component counting modelling histories and 4 strains and damage accumulation LIFE 3 elastic strain histories from linear superposition cyclic plasticity modelling + notch correction Figure 3 Outline of strain-based life prediction process. The model includes a notch correction procedure when strains are elastic.7 Non-proportional multi-axial methods Uniaxial methods for life prediction using the local strain approach have been in use for some time. The essential calculations made by the software are as follows: 1.6]. there are many other components where a combination of loads and geometric effects generates local loadings which are proportional or non-proportional multiaxial. i. FE-Fatigue uses the same model to calculate lives from elastic-plastic FE strain tensors resolved to the surface of the model. having their roots in the work of Basquin [8] Manson [9] and Coffin [10. Within the well known limitations of these methods they work quite well for a variety of components where the local loading in the critical area is uniaxial or near-uniaxial. It processes the resulting 7 components of stress and strain either by the conventional critical plane methods. This class of components includes many that are subject to complex multiaxial loading environments [5. It feeds these 3 strain channels into the Mròz-Garud cyclic plasticity model. For elastic inputs this also involves a notch correction procedure. or from elastic strain inputs. However. the outputs of which are the remaining non-zero strain component εz and the in-plane stresses σx. 2.FE-Fatigue Rel. The second step is to carry out cycle and damage accumulation.11]. or by multiaxial rainflow counting and then accumulating damage using the nCode n 1st Page Back Main Menu . It uses a notch correction procedure when the strains or stresses are linear elastic. The life prediction process from measured strains. either directly from FE or via linear superposition. The first step is to determine the relationship between the strains and all the stress and strain components required for the damage calculation. It takes three components of strain εx. 3. εy and εxy.e. can be divided into two steps. biaxial loading on a free surface.

e. plane stress assumption are utilized. Figure 4 Overall strain energy density equivalence. 5 Theory Page 16 Wang-Brown methods. is necessary. figure 4.FE-Fatigue Rel. which utilizes results from simple static linear-elastic FE analyses and the stress-superposition principle. The cyclic plasticity model. figure 5.3. Inputs to the procedure are “pseudo” stresses (from the linear-elastic computations). is the MrozGarud model. from which the elasticplastic strains and related stresses are estimated using a Neuber-type analysis. 3. The procedure is confined to the free surface conditions. i. nCode n 1st Page Back Main Menu . Multiaxial notch correction procedure. i.8 Multiaxial notch correction procedure Transient elastic-plastic FE analyses can be prohibitively expensive for realistic FE models and input loading histories.e. 1st Page Back Main Menu These calculations are described in more detail in the following sections. necessary for the notch correction procedure. the overall strain energy density equivalence between linear-elastic and elastic plastic stress-strain states.

and the user is forced to make arbitrary choice between them. sometimes observed with the Mroz-Garud cyclic plasticity model.9 Cyclic Plasticity Modelling The stresses and strains required by the damage models can be calculated if the relation between the equivalent plastic strain increment Deeq and the equivalent stress increment ∆σ eq is known during the application of a given load increment. However.3. A powerful feature of the implemented notch correction procedure is that a set of additional conditions. is dramatically reduced or disappears altogether with the implemented multiaxial notch correction routine. Several models are available in the literature [1. Other often used conditions. Equation 1. like ratios of principal strain or stress increments. In the case of a two dimensional stress state.FE-Fatigue Rel. the plasticity surfaces reduce to ellipses on the plane of principal stresses described by: nCode n 1st Page Back Main Menu . 3. is based on ratios of strain energy density increments contributed by each pair of corresponding stress and strain increments (“proportional work path”). since the solution process is essentially “energy bound”.2.17] of which the model proposed by Mròz [1] and recently modified by Garud [2] are the most popular. Mròz [1] has proposed that the uniaxial stress-strain material curve be represented by a set of plasticity surfaces in three dimensional stress space.16. it is known that the current relation depends on the previous load path and therefore the plasticity model must deal with loading path dependent material constitutive behaviour. can make the results strongly dependent on the notch constraint conditions. This assumption allows for significantly reduced dependency of the results on geometry and constraint conditions at the notch tip. from proportional to significantly non-proportional. 5 Theory Page 17 1st Page Back Main Menu Figure 5 Coordinate system definition. necessary for the complete formulation of a multiaxial stress state problem. e e e e a a a a S 22 ∆e22 + e22 ∆S 22 = S 22 ∆e22 + e22 ∆S 22 e e e e a a a a S 23 ∆e23 + e23 ∆S 23 = S 23 ∆e23 + e23 ∆S 23 e e e e a a a a S 33 ∆e33 + e33 ∆S 33 = S 33 ∆e33 + e33 ∆S 33 (1) The predictions of elastic-plastic strains and related stresses from linear-elastic inputs compare very favorably with transient elastic-plastic FEA results for a wide range of loads. Uncontrolled ratcheting.

The translation rule proposed by Garud [2] avoids the intersection of the ellipses that could occur in some cases in the original Mròz [1] model. of only two plastic surfaces (ellipses). 5 Theory Page 18 1st Page 2 Back Main Menu σ eq = σ1 – σ1 ⋅ σ2 + σ2 2 (1) and illustrated in figure 4. It is also assumed that the ellipses move inside each other and they do not intersect. for simplicity.FE-Fatigue Rel. nCode n 1st Page Back Main Menu . If the ellipses come in contact with one another they move together as a rigid body. The load path dependent memory effects are modelled by prescribing a translation rule for the ellipses moving with respect to each other over distances given by the stress increments. σ2 σ 2 1 0 ε f 1 f 2 f 3 σ1 Figure 6 Linearisation of the material σ-e curve and corresponding plasticity surfaces. The Garud translation rule is illustrated in and can be described by a model consisting.

The principal of the method is described here for two surfaces. Connect the conjugate points B1 and B2 by the line B1B2 5. the following steps are made: 1.FE-Fatigue Rel. 5 Theory Page 19 1st Page Back Main Menu σ2 dσ B2 2 4 1 B1 f2 3 5 O1 O1 f’1 σ1 f1 Figure 7 Geometrical interpretation of the Mròz-Garud incremental plasticity model. Connect point B2 and the centre of the intersected plastic surface f2 3. The FATIMAS module MLF and FE-Fatigue use 40. Extend the stress increment dσ to intersect the first external non-active plastic surface f2 at point B2 2. with the corresponding linear sequences being fitted to the Ramberg-Osgood equation to give a more or less smooth curve. Translate the ellipse f1 in the direction of B1B2 from point O1 until the end of the vector dσ lands on the moving ellipse f1 The translation rule assures that the two ellipses are tangential with the common point B1B2 without intersecting each other. In order to predict material response due to the stress increment ds. More details of nCode n 1st Page Back Main Menu . Two or more tangential ellipses translate as a rigid body and the largest moving ellipse (figure 7) indicates the proper constitutive relation (linear segment) to be used for a given stress increment. Find point B1 on the active plastic surface f1 by drawing a line parallel to the line O2B2 through the centre O1 of the surface f1 4.

5 Theory Page 20 this method can be found in reference 3. analogous to rainflow counting in the uniaxial case. and the portions of the strain hardening curve for the reversal are selected by a traditional rainflow procedure [3]. The cycle counting method is illustrated by the following example. where the equivalent relative strain rises monotonically to a peak value. Relative stresses and strains were introduced so that a pair of turning points define the start and end points of a reversal. which is followed by reloading and plastic strain hardening up to the next turning point. a reversal can be defined starting from 0. nCode n 1st Page Back Main Menu . 1 STRAIN (%) 0. Starting from the most significant turning point. The region of unloading and reloading within that reversal is counted in the next step. Since plastic deformation generates the driving force for small fatigue cracks. showing applied tensile (epsilon) and torsional (gamma) strains with the absolute equivalent strain.FE-Fatigue Rel. 10 and 20 seconds. The equivalent strain is defined as the von Mises strain.5 -1 epsilon gamma equivalent 0 20 40 TIME (secs) 60 80 Figure 8 A variable amplitude non-proportional strain history. where relative strain ε A * = ε ij – ε ij ij represents the change of strain since time A. figure 9 shows the relative equivalent strain. hysteresis hardening provides a physical parameter for cycle counting. The most significant turning point occurs at the highest value of equivalent strain. To obtain the second reversal the relative strain is re-plotted starting from the next turning point where unloading commences (at 10 seconds). a graph is drawn for the loading block of relative equivalent strain. which shows a repeating block of a combined tension/torsion non-proportional load history.10 Multiaxial Rainflow Counting Wang and Brown [18] proposed a multiaxial cycle counting method on the basis of strain hardening behaviour under non-proportional variable amplitude loading. This is illustrated at time 0 in figure 8.5 0 -0.3. 1st Page Back Main Menu 3. up to the maximum value 10 seconds. with respect to times 0. Each reversal commences with elastic unloading. Using the relative strain.

The recursive nature of this process makes this module significantly slower than other analysis modules such as CLF and uniaxial methods in FE-Fatigue. showing relative equivalent strains plotted with respect to times 0. Mean stress is measured as the average of the maximum and minimum stress values over the reversal. nCode n 1st Page Back Main Menu . and σn.0 relative . relative strain is re-plotted with respect to 20 seconds for the subsequent continuous fragment of strain history.5 0 20 40 TIME (secs) 60 80 epsilon gamma relative .5 -1 -1.20 Figure 9 The variable amplitude history.5 1 STRAIN (%) 0. from Reference [4]. This procedure is repeated for each turning point in chronological order. yielding the third reversal in figure 9. ′ σ – 2σ n. If the counted reversals are non-proportional. δεn is the normal strain excursion between the two turning points of the maximum shear strain (that is the range of normal strain experienced on the maximum shear plane over the interval from start to end of the reversal).5 0 -0. 5 Theory Page 21 1st Page Back Main Menu 1. being based on hysteresis deformation behaviour.3.= ----------------------------. this is equivalent to the morrow method. a fatigue damage parameter that accounts for non-proportional straining effects is required. 10 and 20 seconds respectively. The path-independent damage parameter proposed by Wang and Brown [19] has been shown to provide good correlation for several materials under proportional and non-proportional loading.mean γ max + S ⋅ δε n ′ f b c ˆ ε ≡ ---------------------------------------.11 Fatigue Damage Calculation The counting method described above is independent of fatigue damage parameters.FE-Fatigue Rel. with a Morrow mean stress correction [20]. If mean stress is included. it can be integrated with any multiaxial fatigue damage model. Using the next turning point. The right hand side of the equation is the same as the uniaxial strain life equation. The total damage induced by a loading history is calculated using Miner’s rule [13]. 3. The term S is a material constant determined from a multiaxial test (typically between 1 and 2 for Case A and around 0 for Case B) and v‘ is the effective Poisson’s ratio. This can be omitted if no mean stress correlation is required.( 2N f ) + ε ( 2Nf ) ′ ′ E f 1 + v + S(1 – v ) (2) where γmax is the maximum shear strain amplitude on a critical plane (proportional or nonproportional).10 relative . Being unrelated to material properties. The plot in figure 10 is a plot of predicted life against experimental life for a variety of proportional and non-proportional tests on laboratory specimens.mean is the mean stress normal to the maximum shear plane. until every fragment of strain history has been counted.

3.= -------------------------. Normal strain (θ = 90 degrees only): σ′ f ∆ε n b c -------. The plane with the largest accumulated damage is said to be the critical plane.( 2N f ) + ( 1 + V p )ε′ f ( 2N f ) E 2 (4) 3. inclined at an angle θ = 90 degrees (Case A) and/or θ = 45 degrees (Case B) to the free surface.⋅ σ n. Otherwise this is the Smith-Topper-Watson method [22].12 Critical Plane Methods The other multiaxial damage parameters considered are the more conventional critical plane parameters. Otherwise this is the usual Coffin-Mansonn Basquin equation. 5 Theory Page 22 1st Page Back Main Menu 10.FE-Fatigue Rel. The models are: 1. Smith-Watson-Topper/Bannantine (θ = 90 degrees only) [21]: ∆ε n σ′ f -------. ∆ε 2 2.000 Predicted lifetime (block) 1000 non proportional proportional ideal factor 2 1. Shear strain: ( 1 + V e )σ′ f b c ∆γ ----. Cycle counting (uniaxial) and damage parameter calculation is carried out on the critical plane and the damage accumulated.= -----. max is the maximum normal strain on the critical plane which occurs during each rainflow cycle. max = ------. stresses and strains are resolved onto a particular plane.( 2N f )2b + σ′ f ⋅ ε′ f ( 2Nf ) b + c 2 E 2 (5) where σn.000 10. 3..( 2Nf ) + ε′ f ( 2N f ) E 2 (3) is the strain amplitude normal to the critical plane.000 100 100 measured lifetime (blocks) Figure 10 Comparison of experimental and predicted results for the Wang-Brown method. The orientation φ of the projection of the normal to the damage plane is increased by 10 degree increments from 0 to 170 degrees. nCode n 1st Page Back Main Menu . In these methods. where -------.

For positive biaxiality ratios.1 + n ----------------ö = --------------------------f ( 2Nf ) + -----------------------------. These are named type B cracks.e. depending on the biaxiality ratio. when a = 0.( 2Nf ) + ( 1 + V p )ε′ ( 2N f ) + -----------------------------------f ( 2N f ) f σy ø E 2Eσ y 2σ y 2è (6) 3. biaxial tension when a > 0. e. max ∆γ æ b 2b c n ( 1 + v p )ε′ f σ′ b+c ----. torsion loading when a = -1. For negative biaxiality ratios. Fatemi-Socie [23]: ( 1 + V e )σ′ n ( 1 + V e )σ′ f σ n. cracks tend to grow at 45 degrees into the surface and tend to be deep and short. as shown in Figure 9. they can be both type A and B. 5 Theory Page 23 1st Page Back Main Menu 4.13 Types of Cracks Figure 11 Types of Cracks In multiaxial fatigue conditions cracks can grow in different directions into the surface. cracks tend to grow at 90 degrees into the surface and tend to be shallow and long. e. These are named type A cracks.FE-Fatigue Rel. nCode n 1st Page Back Main Menu .g.g. i. In uniaxial loading conditions.3. cracks can grow either way.

it should not be performed in multiaxial cases. the maximum stress during the cycle would not necessarily occur at point F. Common rainflow counting algorithms are not able to do this correctly. nCode n 1st Page Back Main Menu . and "track" the secondary damage parameter. Thus.g.3. Recommended procedure for critical plane methods is to rainflow count the primary damage parameter. For example. shear strain parallel to the crack plane. but anywhere from E to E'. however. A related issue is worth noting: while peak-valley editing (extracting peaks and valleys from a time history) is perfectly valid in uniaxial case. component of stress perpendicular to the crack plane. because important information about stress magnitudes can be lost. e.g. 5 Theory Page 24 1st Page Back Main Menu 3. while identifying strain cycle E-F.14 Critical Plane Rainflow Counting Figure 12 Uniaxial Rainflow Counting The method of identifying cycles in variable amplitude multiaxial fatigue conditions is crucial for obtaining correct life predictions. the stress has to be "tracked" between points E-F-E' in order to properly determine its maximum value. e. In multiaxial fatigue conditions.FE-Fatigue Rel. common rainflow counters would identify cycle E-F in the above diagram with associated maximum stress at point F.

. J. (1945).. American Society for Testing and Materials. D. (1996). (1966). Proceedings of the Symposium on International Automotive Technology. Journal of Engineering Materials and Technology.. TIPTON. Journal of Mechanics and Physics of Solids. and TOPPER T. The Problem of Thermal Stress Fatigue in Austenitic Steels at Elevated Temperatures. (1996). 16. (1996). and SEEGER. (1996). Journal of Materials. H. and DAKIN J. vol. An Analysis of Mean Stress in Multiaxial Random Fatigue. pp. 261-277 (25) HOFFMANN. 118-125 (3) WANG C. H. M. and BROWN M. A Path-Independent Parameter for Fatigue under Proportional and Non-Proportional Loading. W. 2/3. (1993).. pp. Presented to Kyushu District Meeting. CORDES. Berkeley Nuclear Laboratories (18) WANG C. W. Proceedings of the 12th International Conference on Structural Mechanics in Reactor Technology. 19. pp. Fatigue of Metals Subjected to Varying Stress.. 163-175 (2) GARUD Y. Journal of Engineering Materials and Technology. 767-778 (23) FATEMI A. H. 118. pp. O. EGF3 (Edited by M. F. Ph. nSoft-E FATIMAS software manual (15) GLINKA G. and BROWN M.” SAE technical Paper 930401 nCode n 1st Page Back Main Menu . 159-170 (19) WANG C. (1997).. (1993) “Notch stress and strain estimation considering multiaxial constraint. pp. vol. 76. vol. pp. Journal of Applied Mechanics. 22. H. pp. 165.. vol. pp. 15. W.Miller). pp.. Tools for Fatigue Evaluation of Non-Proportional Loading. Proceedings of Fatigue Design 1995 (Edited by Gary Marquis and Jussi Solin). C. J.. F. and BROWN M.JOHN C. N. A Study of the Effects of Cyclic Thermal Stresses on a Ductile Metal. 3. Proceedings of the 5th International Conference on Biaxial/Multiaxial Fatigue and Fracture.M.. pp. pp. (1981).S. Poland (16) CHU C. pp. Cumulative Damage in Fatigue. (1995). ESIS 21 (Edited by A. Behaviour of Materials under Conditions of Thermal Stress. University of Michigan Engineering Research Institute.. S. A Program for the Fatigue Analysis of Automotive Spot-Welds Based on Finite Element Calculations. Fatigue and Fracture of Engineering Materials and Structures. vol. and KÖTZLE H. (1953). and BUCZYNSKI A. pp 3-24. vol. (13) MINER M. Journal of Engineering Materials and Technology. JSME. 1989. 9-75 (10) COFFIN L. S. KÖTTGEN V.. (1910). vol. (1995). 4. (1988). and FREDERIC C. 12. Proceedings of the American Society for Testing Materials. ASTM STP No.W. 367-370 (4) WANG C. B. Heat Transfer Symposium.A. pp.. University of Illinois at Urbana-Champaign (22) SMITH K. 118. 189-199 (7) HEYES P.. and SOCIE D. CEGB Report RD/B/M731... A Critical Plane Approach to Multiaxial Fatigue Damage Including Out-ofPhase Loading. A Variable Amplitude Multiaxial Fatigue Life Prediction Method. K. MILSTED M. A159-A164 (14) nCode International Ltd. London. 1. Life Prediction Techniques for Variable Amplitude Multiaxial Fatigue Part 1: Theories.. Lindley) MEP London. (1954).... L.. WATSON P.J.D.15 References (1) MRÒZ Z. (26) KLANN. (1996). no. H. (1997). T. 103. 5 Theory Page 25 1st Page Back Main Menu 3.” Biaxial and Multiaxial Fatigue. A. W. p.. Krakow. no.... (1954). (1970). 625-630 (9) MANSON S.. pp. and BROWN M.149-165 (24) DREßLER K. vol. The Exponential Law of Endurance Tests. T. 5. vol. 1285-1298 (20) BROWN M. C. F. Inelastic Deformation and Fatigue under Complex Loading. no. thesis. and WANG C.31 (11) COFFIN L. 10. Life Prediction Techniques for Variable Amplitude Multiaxial Fatigue Part 2: Comparison with Experimental Results.. pp. H. 3. 323-333 (21) BANNANTINE J. Journal of Mechanics and Physics of Solids. and ENDO T. G. A Three-Dimensional Model of Anisotropic Hardening in Metals and its Application to the Analysis of Sheet Metal Formability.. pp. vol. “Estimating multiaxial elastic-plastic notch stresses and strains in combined loading. 461-475 (6) HEYES P. The Assessment and Use of Linear Static FE Stress Analyses for Durability Calculations.Brown and K. Proceedings of the Ninth International Conference on Vehicle Structural Mechanics and CAE.. A Mathematical Representation of the Multiaxial Bauschinger Effect. S. 197-212 (17) ARMSTRONG P. H. vol. 371-374 (5) HEYES P. Mechanical Engineering Publications. G. On the Description of Anisotropic Work Hardening. A. SUKER D.FE-Fatigue Rel. Multiaxial Fatigue and Design.. vol. Cailletaud and T. 931-950 (12) MATSUISHI M. Multiaxial Fatigue Assessment of Automotive Chassis Components on the basis of Finite-Element Models. A Stress-Strain function for the Fatigue of Metals.. and FERMÉR M. (1989). no. (1989). 11. (1967).3. (1993). DAKIN J. Fatigue and Fracture of Engineering Materials and Structures. Fatigue and Fracture of Engineering Materials and Structures. A New Approach to the Evaluation of Fatigue under Multiaxial Loading. vol. Pineau. Elastic-Plastic Stress-Strain Analysis of Notches under Non-Proportional Cyclic Loading Paths.. Trans. vol. SAE Technical Paper 962507 (8) BASQUIN O. W. pp. (1968). and ST.

The reasoning behind this is that the mean stress and alternating stress often have different origins. 3.3. with amplitude σa and mean σm. life based. User preferences determine the maximum factor to be attempted and the tolerance for ending the iteration process. The definition of safety factor using the Goodman correction is illustrated below. The life is often an endurance limit. there are three ways in which a safety factor can be calculated. 3. adjusting the linear scaling factor on stress until the life is within a tolerance factor of the specified life. An alternative definition is A'/σa which will give more pessimistic predictions in the "safe" regime. a target life is specified.4. It considers only the largest stress cycle developed during the loading history. These methods are discussed separately. The safety factor calculated in FE-Fatigue is defined as A/σa. 3.4. Safety factor using Goodman Alternating Stress Reference Stress A A’ σa σm Mean stress UTS Only the largest stress cycle is considered. Multiaxial safety factor FE-Fatigue includes options to apply the Dang Van and McDiarmid methods to create a multiaxial safety factor plot. taking into account the mean stress. nCode n 1st Page Back Main Menu . The method used gives a safety factor on alternating stress for a given mean stress.1 Life based factor of safety In this method. stress based and multiaxial. or a level of stress above the endurance limit which can be tolerated by the structure. Safety factor Safety factor calculations are designed to provide a single scalar quantity which is used to determine a level of confidence that a component will achieve a given life.4. but may also be a finite life. FEFatigue does not use this method. In FE-Fatigue. Factors < 1 are considered safe and >1 unsafe.FE-Fatigue Rel. 5 Theory Page 26 1st Page Back Main Menu 3. The program then performs iterative calculations. applicable for strain-life and stress-life.4.2 Stress based factor of safety The stress-based safety factor method calculates a safety factor relative to a reference stress which will typically be the fatigue limit of the material.

5. and therefore this type of method will not be considered further. A simple uniaxial fatigue method is not sufficient and can produce non-conservative results. but his modified approach proposed in 1987 is easier to use and provides better correlation. The above observations lead to the more intuitive Tresca-based high-cycle fatigue criteria. the Sines and Crossland criteria being quite well-known. microscopic stresses in calculations of fatigue damage. but McDiarmid’s Tresca-Based criterion introduced in 1987 appears to be widely accepted and well-known. For high-cycle multiaxial fatigue. Instead. nCode n 1st Page Back Main Menu .2. but just to consider if any fatigue damage will occur during the load history. 3.FE-Fatigue Rel. If damage does occur. Findley reported early (1959) on Tresca-based predictions. Dang Van first introduced this type of approach in 1973. and rear axles. The high-cycle criteria considered can also be used to calculate a safety factor of a component if infinite life is determined. Such components include crank-shafts. propeller shafts. In these components. Microstructural approaches to high-cycle fatigue have also become more common in recent years. the complex stress states in the loading history are important to consider in order to accurately predict fatigue life. The Definition of High-Cycle Multiaxial Fatigue Multiaxial fatigue is a term referring to the fatigue of components under complex stress states. High-cycle multiaxial fatigue deals with components that experience large number of cycles and are therefore designed for infinite life.1. Multiaxial Safety Factor 3. the loading history is examined to determine only if damage occurs at any moment. then the component does not have infinite life. which does not consider the plane of maximum shear stress range. Diboine (3) classifies these methods according to the basis of calculation: • Empirical • Von Mises • Tresca • Microstructural The empirically-based fatigue criteria require testing of a specific material in a loading configuration similar to that of interest in order to produce a numerical fit to the data and make further predictions. This defeats the purpose of using fatigue predictions. Von Mises-based criteria were prevalent early in the studies of high-cycle fatigue.5. it is not necessary to quantify the amount of fatigue damage. actual quantities of fatigue damage are not considered. High-Cycle Multiaxial Fatigue Theories Several methods have been developed to calculate a fatigue limit in the high-cycle fatigue (HCF) regime with significant multiaxial loadings.5. 5 Theory Page 27 1st Page Back Main Menu 3. In cases of high-cycle multiaxial fatigue. Fatigue cracks have been observed to initiate and grow initially along these planes of maximum shear stress. These approaches consider the stabilised. Von Mises is a deformation criterion. However.

The fatigue limit according to McDiarmid can be expressed in the following form: τa / tA. the shear stress amplitude (half of the maximum range) on the critical plane and the maximum normal stress (for the entire loading history) on the plane of maximum shear stress range. Figure 13 Any loading that goes above the line is considered to cause damage and the fatigue life will not be infinite. when there is a torsional loading. As with many calculation methods. The case B value would require the fatigue limit to be determined under conditions that would generate Case B cracking. McDiarmid Criterion Description The McDiarmid method is a critical plane based fatigue limit criterion which is applicable to cases where crack initiation is at free surfaces.B + σn.B vs. The critical plane is defined as the plane of maximum shear stress range. e. This fatigue limit can be displayed graphically in a plot of τa / tA.5. The McDiarmid method considers Case A and Case B cracking separately in calculating the fatigue damage. 5 Theory Page 28 1st Page Back Main Menu 3.g. Case A cracking when cracks initiate by shear along the surface.FE-Fatigue Rel. max / 2σT (see below). nCode n 1st Page Back Main Menu .3. one problem is to obtain suitable materials properties.B= reversed shear fatigue strengths (amplitude) for Case A and B crack growth σn. In this case we require the fully reversed shear fatigue strength for Case A and Case B cracking. damage is predicted to occur. max / 2σT = 1 where: τa= shear stress amplitude tA. σn. Case B cracking occurs when the maximum shear plane is inclined at 45 degrees to the free surface with shear cracks being driven into the surface. max= maximum normal stress for the entire loading history on the plane of maximum shear stress amplitude σT = tensile strength (1) If the value on the left hand side of this equation exceeds 1. the Case A value should simply be the shear fatigue stress amplitude at the fatigue limit for a pure torsion case. The fatigue criterion is based on two loading parameters. According to the McDiarmid theory.

The maximum shear range for each plane is calculated by the difference of the maximum and minimum values of shear stress for the loading history.5σysin2φ + τxycosφsinφ τyz = -σxcos2φ . Case A or B cracking may occur. In the case of uniaxial loading. (7) (6) (4) (5) (2) (3) nCode n 1st Page Back Main Menu .B)+ (σn. 5 Theory Page 29 1st Page Back Main Menu In the absence of this information.τxycosφsinφ Because the fatigue limit criterion is expressed as τa / tA. max / 2σT)) which is reported in the results file for contour plotting. Case A planes intersect the surface at 90 degrees and case B at 45 degrees.σysin2φ .5*σL)/(2*σT) = 1 So the necessary case A and B fatigue limits are given by: tA. the orientation of the case A and B planes with respect to the X axis is varied in 10 degree increments. the values of tA.(σL/2*σT)) In practice. Rather than searching infinite planes.5*σL)/(tA.B + σn. so we can rewrite the McDiarmid criterion at the fatigue limit under uniaxial loading conditions as: (0.B ) + (0.5σx cos2φ + 0.B = σL/ (2 . max / 2σT = 1 a safety factor can be defined as: Safety factor = 1 / ((τa / tA. The McDiarmid criterion calculations are then performed for the planes of maximum shear stress range. The software requires that the stresses are resolved to the plane of the surface such that the X-Y plane is the plane of the surface. both the case A and case B planes are searched to find the plane with the maximum value of shear stress amplitude.B may be estimated from the uniaxial fatigue limit σL as follows.FE-Fatigue Rel. The stresses at these angles (angle indicated as φ) are calculated as follows: Case A Cracking (θ =90 degrees): σn = σxcos2φ + σysin2φ + 2τxycosφsinφ σxz = 2σxsinφcosφ + 2σysinφcosφ + τxy (-sin2φ + cos2φ) Case B Cracking (θ = 45 Degrees): σn = 0.

5. This very localised cyclic plasticity will lead to relaxation of mean stresses on the microscopic scale. the Dang Van criterion (1. whatever their orientation). cyclic plasticity will occur on a microscopic scale due to the inhomogeneity of the material on the scale of individual grains. the Dang Van method assumes that around the fatigue limit. fillet) V(M)= an elementary representative volume around M P= a point in V(M) where a grain(s) is critically oriented σ(P. f{σ(P. Once this stabilised state (elastic shakedown) has occurred. Essentially. is greater than or equal to zero at any time in a stabilised cycle.4.FE-Fatigue Rel. or another way of putting this is the development of microscopic residual stresses.t)} ≥ 0 for P∈V(M) where: M = a point of interest/stress concentration on a component (notch. Dang Van Criterion Description The Dang Van Criterion is a multiaxial fatigue limit criterion which can be classified as a ‘microstructural’ method because it is based on the concepts of stress and crack initiation at a microscopic level.2) states that crack initiation will occur whenever a function f of the microscopic stress. 5 Theory Page 30 1st Page Back Main Menu 3. The function is a simple linear one: f(σ) = τ + a*ph – b and f(σ) = τ + a*ph + b where: τ ph b a = microscopic shear stress at a given time = microscopic hydrostatic tension at a given time = shear stress amplitude at the fatigue limit in pure torsion = hydrostatic stress sensitivity (9) (8) nCode n 1st Page Back Main Menu .t)= a microscopic stress tensor at location P at time t Dang Van proposes that f is a function of the microscopic shear stress (which can cause plasticity) and the hydrostatic tension (which can open microscopic cracks.

f/2)/f (10) If the fatigue limit is known under only one condition. and bending. and the parameter a is known as the hydrostatic stress sensitivity.5774f and a=0. ph and the deviatoric part. only the upper part of the diagram need be considered. Sij Sij = Eij – ph. b. To determine the hydrostatic stress sensitivity factor.23. The macroscopic stress Eij can be divided into a hydrostatic part. The microscopic deviatoric stress is calculated from: dev σij = Sij + dev ρij* Then the microscopic maximum shear stress is calculated from: τmax = 1/2(Tresca(dev σij)) The microscopic residual stress is calculated using an iterative procedure which attempts to simulate the process of elastic shakedown. For instance. a can be estimated from a = 3(b. 5 Theory Page 31 1st Page Back Main Menu This function establishes the Dang Van Endurance Domain shown below: Figure 14 A loading path Γ is illustrated in Figure 14 and damage is predicted to occur only if the loading path crosses the lines delimiting the endurance domain. and the part of the loading path below the abscissa is reflected upwards ( Γ’ ).FE-Fatigue Rel. the fatigue limit must be known under at least two loading conditions.δij σij = Eij + ρij* where ρij* is the stabilised residual stress. For instance if you only had the bending fatigue limit f. b is taken to be the shear stress amplitude at the fatigue limit under conditions of pure torsion. Considering again equation (9). Note that if the microscopic shear stress is replaced by the microscopic Tresca or maximum shear stress. some assumptions could be made. All this is relatively simple. [where δij is the Kroneker delta] The microscopic stress in a stabilised condition is calculated from: nCode n 1st Page Back Main Menu . f. you could assume that the Von Mises stress was a reasonable criterion to apply which gives b=0. The complexity of the Dang Van method lies in the calculation of the microscopic shear stress. if the fatigue limit is known under fully reversed torsion.

M. A. Practically this is achieved by simulating a combination of isotropic and kinematic hardening as illustrated (in two dimensions!) below: Figure 15 2D representation of hypersphere A small initial yield surface is defined. CAILLETAUD. but there may be overshoot leading to loss of accuracy.FE-Fatigue Rel. Biaxial and Multiaxial Fatigue. A. pp. Biaxial and Multiaxial Fatigue.. The default value of this parameter is 0. FLAVENOT. Mechanical Engineering Publications. STJOHN. LIEURADE. For the purposes of contour plotting. but which can be set as a preference. pp. 2] DANG VAN. The displacement of the centre of the yield surface from the origin then defines the microscoopic residual stress. 479496. A hardening parameter controls the rate of isotropic hardening.P. 5 Theory Page 32 1st Page Back Main Menu Its aim is to find the centre of the nine-dimensional hypersphere which can encompass the loading path. O. (1989) On a New Multiaxial Fatigue Limit Criterion: Theory and Application. 459-478. When the loading path contacts the yield surface.05 which is a good compromise for most loadings. GRIVEAU. pp... London. with the default being 100... cycling through the loading history until there is no further expansion or movement of the yield surface. ESIS 21. MILSTED. MESSAGE. Multiaxial Fatigue and Design. the yield surface is allowed to move (kinematic hardening) and expand (isotropic hardening) and the process continues. Mechanical Engineering Publications. London. J. but a smaller value will give greater accuracy for very short loading histories. H. LE DOUARON.ph) except when the denominator is less than or equal to zero. when the life is predicted to be infinite and the safety factor is effectively infinite. Mechanical Engineering Publications. P. B. It is convenient when using the Dang Van criterion to define a safety factor which may be used to compare the worst loading condition to the fatigue limit criterion. an upper limit for safety factors is used in results files. C. The centre of this hypersphere defines the microscopic residual stress.. London. The definition of safety factor used in FE-Fatigue is: Safety factor = b/(Max(τmax) + a. EGF 3. Report for Ford Motor Company. (1996) Fatigue Assessment of Components under Complex Loadings. EGF 3..F. 4] HEYES. G. nCode n 1st Page Back Main Menu . (1989) Criterion for High Cycle Fatigue Failure Under Multiaxial Loading. K. 425-443.. (1993) Application of Finite-Element Based Life Prediction to Chassis Engineering. denoted C0. K. Using a larger value of this parameter causes the stabilised condition to be approached more rapidly. 3] DIBOINE. References 1] DANG VAN.

. 7] PAPADOPOULOS.. 4. No. Vol. (1997) A Comparative Study of Multiaxial High-Cycle Fatigue Criteria for Metals.V. Fatigue and Fracture of Engineering Materials and Structures. DAVOLI. No 3. Fatigue and Fracture of Engineering Materials and Structures. Vol. 19.. (1991) A General Criterion for High Cycle Multiaxial Fatigue Failure. 17.FE-Fatigue Rel. pp. 14. D. 6] MCDIARMID. nCode n 1st Page Back Main Menu . C.219-235. and BERNASCONI. M. No. D. Vol.. pp. 429-453. (1994) A Shear Stress Based Critical Plane Criterion of Multiaxial Fatigue Failure for Design and Life Prediction.L. P. pp. 5 Theory Page 33 1st Page Back Main Menu 5] MCDIARMID. FILIPPINI. GORLA. 1475-1484.L. 12. I. A. International Journal of Fatigue.

Design analysis based on fatigue life calculations results in designs that are less conservative (i. The method used in FE-Fatigue is more or less identical to that described by Rupp et al [1]. low cost of manufacture. transient and modal (using modal superposition approach) structural analysis results [7]. short development cycles.considered to be a circular solid surrounded by a deep circumferential crack. Calculations based on fatigue life and realistic loading histories permit structures and components to be optimised for durability without the need for the expensive and time-consuming testing of a series of prototypes. In fact.6] note that durability of spot welds of a variety of configurations and loadings can be better understood through numerical analysis of the local stresses at the weld spot edge on the inside of the plate . Introduction A key feature of the modern durability design process is the use of computer-based finite element methods to predict durability at an early stage in the design cycle.e. They noted that a spot weld could be ". These costs may be minimised if the life of spot welds can be predicted at an early stage in the design process. Resistance spot welds are very commonly used in the automotive industry in the fabrication of all manner of components and structures. The method was based on detailed finite element modelling of simple spot-welded lap-joints loaded in shear. and good durability. It is therefore possible to make spotweld fatigue calculations in FE-Fatigue based on static. load is a rather poor parameter for correlating the fatigue strength of spot-welds under different loading conditions.000. This method would need further development in order to cover all the possible weld configurations used in automotive structures and to deal with the variable amplitude out-of-phase loadings to which they are subject..1. Rupp. Smith and Cooper [2] addressed the problem of life prediction of shear spot welds using a fracture mechanics approach. Radaj [4] and Sheppard [5. better optimised) than those based on traditional criteria such as maximum load or stress for a series of standard load cases. They showed that good predictions of life could be made on the basis of calculated crack growth rates. would grow a branch crack in the direction of maximum local Mode I". This process is driven by the need for designs with low weight. and used their calculations to generate some simple design curves.FE-Fatigue Rel. and the durability of such structures is very often controlled by the strength of the spot welds.the structural stresses around the weld. Störzel and Grubisic [1] describe the calculation of these structural stresses. nCode n 1st Page Back Main Menu . The results of such a development might be a simple design code for spot-welds along the lines of the British Standard BS 7608 [3] with families of load-life curves for different classes of spot-weld. which when loaded in a combination of Mode I and Mode II. Theoretical background to Spot Weld analysis 3..6. The cost of tooling up for a single weld spot as part of an automated manufacturing process may be as much as $30. and this can more than double if a weld spot has to be added during production to remedy a problem [1]. and combines their method for structural stress calculation with the usual techniques for calculation of dynamic stresses.. and also carry out fatigue life predictions based on maximum and minimum stresses and a load spectrum.6. 5 Theory Page 34 1st Page Back Main Menu 3. though the reduction in development time and improvement in quality is likely to be more significant. in practical FE models of automotive structures detailed modelling of individual spot welds is not a desirable option.

but this discussion will primarily focus on using CBAR elements with NASTRAN [8]. 0.more than twice the diameter of the weld nuggets. These stresses can then be used to make fatigue life predictions on the spot weld using a S-N (total life) method. nCode n 1st Page Back Main Menu . the weld is modelled as a stiff bar element joining the mid-planes of 2 sheets of thin shell elements. most realistic joint stiffnesses) are achieved when the dimensions of the shell elements are quite large .general description The method described below has been typically applied using bar elements to represent spot welds joining two sheets of shell elements. The shells are positioned at the mid-planes of the sheet metal and the length of the CBAR element and the separation of the shells should therefore be half the sum of the sheet thicknesses. In a finite element analysis. Point 3 is on the axis of the weld nugget and at the interface of the 2 sheets. the spot-welds should be represented by stiff CBAR elements joining 2 sheets of shell elements. In fact it seems that best results (i. 5 Theory Page 35 1st Page Back Main Menu 3. The steps in the fatigue calculation are now described in more detail. and perpendicular to both.e. All forces and moments are taken to be in the co-ordinate system illustrated in Figure 16. This is different both from the arrangement used by Rupp et al [1] and that used in NASTRAN for CBAR elements. The fatigue analysis of spot welds . In the FE model. The method can be used with other element types and FE analysis codes.6.3.e. The CBAR elements should be sufficiently stiff that the stiffness of the joint is not sensitive to the CBAR properties. 3. Structural stress calculation Figure 16 Typical spot weld The shaded part is the spot weld "nugget".6. The relationship between the FE-Fatigue and NASTRAN CBAR coordinate systems is illustrated in Figure 17. i.FE-Fatigue Rel.The length of the bar element is 0. but a little simpler.5 s1 from Point 1.2.5(s1+s2) where s1 and s2 are the thicknesses of sheets 1 and 2 respectively. and dimensions of the cross section approximately equal to the weld nugget it represents. There is no need for any refinement of the mesh around the spot-welds. This is a Cartesian system with the Z axis going from Point 1 to Point 2. The only requirement for the shell elements used to model the sheets is that they transmit the correct loads to the bar elements. In practice this can readily be achieved by giving the CBAR element the same modulus of elasticity as steel. The forces transmitted through these CBAR elements are used to calculate the structural (nominal or "hot-spot") stresses in the weld "nugget" and the adjoining sheet metal at intervals around the perimeter of the nugget.

z are written to the .MZ2*t1) / (t1 + t2) Where t1 is the shell thickness at point 1 and t2 is the shell thick at point 2 The CBAR element forces and moments Fx. the following transformation is required: Point 1 FE-FatigueNASTRAN FX1= -Z Force (plane 2 shear) FY1= Y Force (plane 1 shear) FZ1=X Force (axial) MX1= -Z Moment 1 (plane 1 bend A) MY1= -Y Moment 1 (plane 2 bend A) MZ1= X Moment (torque) Point 2 FE-FatigueNASTRAN FX2= Z Force (plane 2 shear) FY2= -Y Force (plane 1 shear) FZ2= -X Force (axial) MX2= Z Moment 2 (plane 1 bend B) MY2= Y Moment 2 (plane 2 bend B) MZ2= -X Moment (torque) Point 3 Nugget values (using above FE-Fatigue values) FX3 = FX1 FY3 = FY1 FZ3 = FZ1 MX3 = (MX1*t2 .y. for each of the three specified points.MY2*t1) / (t1 + t2) MX3 = (MZ1*t2 .MX2*t1) / (t1 + t2) MY3 = (MY1*t2 . These forces and moments (except Mz) are used to calculate nominal stresses (structural stresses) on the inner surface of sheet 1 and nCode n 1st Page Back Main Menu . 5 Theory Page 36 1st Page Back Main Menu Figure 17 In order to account for the local CBAR coordinate system in NASTRAN.FE-Fatigue Rel.z and Mx.y.FES file in FE-Fatigue co-ordinate system.

. nominal stresses are calculated at intervals around the circumference of the weld nugget....... say at 10 degree intervals...6√s1 (compensating for the size/stress-gradient effect in bending) and d is the diameter of the weld nugget....(7) These equations are based on analytical expressions for stress... all dimensions being in mm.(2) Fy1 πds1 . The forces and moments at points 1 and 2 are those applied by the spot welds on the sheets.......... spot welds do not usually fail by cracking through the nugget unless the nugget diameter is too small compared to the sheet thickness. i... at intervals around the circumference of the spot weld (by default.. Forces are in N and moments in Nmm.e.(6) è ds 1 2 ø æ 1.. From the forces calculated for point 3.. and the forces and moments at point 3 will be those applied by the upper section (between point 3 and point 2) on the lower section (between point 1 and point 3) This statement defines the directions and signs of the forces and moments... For this reason...... The stresses in the nugget are treated differently... using a stress-based critical plane method... θ = 0o to 360o by increments of 10o)...FE-Fatigue Rel.872M x1ö σ max ( M x1 ) = K 1 ç ---------------------... modified with empirical factors taking into account the observed fatigue behaviour from a variety of spotwelded components and loading conditions.. as commonly used in FE-Fatigue (only 36 calculations)..... two faster options are included in the software: to ignore the possibility of nugget failure altogether. The method of Rupp et al then suggests that the direct stress be calculated on multiple planes at 10 degree intervals...744F z1ö σ ( F z1 ) = K 1 ç ---------------------÷ forF z1 >0 .(3) σ max ( Fy1 ) = æ 1. The equations for point 2 are similar.. and: æ 1. and to use the absolute maximum principal stress as the damage parameter.÷ ....(5) so that only the tensile component of the axial force in the nugget contributes to damage.. This would mean 36 x 18 = 648 calculations for each weld nugget. This is nCode n 1st Page Back Main Menu ..(4) è s12 ø σ( Fz1 ) = 0 for Fz1 ≤ 0 .÷ è ds 1 1 ø ..872M y1ö σ max ( M y1 ) = K 1 ç ---------------------.. especially in view of the fact that the normal failure mode is by crack growth through the sheet metal.. and in the weld nugget at the interface of the two sheets.. 5 Theory Page 37 1st Page Back Main Menu sheet 2. Note that K1=0. This is very computationally intensive.. The stresses are calculated as follows: For point 1 the equivalent stress on the inner surface of the sheet as a function of angle θ around the circumference of the spot weld is: σ v1 = – σ max ( F x1 ) cos θ – σ max ( F y1 ) sin θ + σ max ( F z1 ) + σ max ( M x1 ) sin θ – σ max ( M y1 ) cos θ (1) where: σ max ( Fx1 ) = Fx1 πds1 ....

..(16) for number of cycles to failure Nf < Nc1 the transition life.(14a) max y3 ..(15) 2 2 2 The principal stress with the greatest magnitude is taken as the damage parameter.... plus a mean stress sensitivity factor and a standard error parameter.(14b) From the shear and direct stresses on the nugget....... and slopes b1 and b2 are material constants. the in-plane principal stresses can be calculated from: σ 1.. 3 σ σ .(12) σ( Fz 3 ) = 0 when Fz 3 ≤ 0 .4........(10) τ max ( Fy 3 ) = σ ( Fz 3 ) = 16 Fy 3 3πd 2 ..........(13) σ (M 32M y3 ) = ----------------2 πd σ max ( M x 3 ) = 32 M x 3 πd 3 .......± ----. The formulation of the S-N curve is as normal. These S-N curves are specifically spot-weld S-N curves and are quite distinct from the parent plate material S-N curves.(8) σ = σ ( Fz 3 ) + σ max ( M x 3 ) sin θ − σ max ( M y 3 ) cos θ ... Specimens that have been used for spotweld testing include nCode n 1st Page Back Main Menu .. 5 Theory Page 38 calculated as follows: τ = τ max ( Fx 3 )sin 2 θ + τ max ( Fy 3 ) cos 2 θ . 3.....+τ .2 = -....... For Nf > Nc1 a second slope b 2 is used.... The Stress Range Intercept SRI1........ that is: ∆S = SRI 1( N f ) b1 .....6...(17) where M is the mean stress sensitivity (typically 0..............(9) 1st Page Back Main Menu where: τ max ( Fx 3 ) = 16 Fx 3 3 πd 2 .. with different spotweld sizes and sheet diameters.... Material properties The system requires an S-N curve for each metal sheet and for the weld nugget at load ratio R=0. Each cycle with amplitude S and mean stress Sm is corrected to calculate an equivalent stress amplitude S 0 at R=0 using the following equation: S0 = S + MS m M + 1 .1) S-N curves for spot weld analysis must be obtained from tests on spot welded specimens..(11) when Fz 3 > 0 4 Fz 3 πd 2 .............FE-Fatigue Rel.. The datasets included in the material database are taken from the work of Rupp et al [1] and represent the results of tests on a variety of different specimen and loading configurations......

or even a material specification is not available. with a load ratio R=0 For each specimen and loading configuration. In practice. hat-profile (torsion). Such S-N curves are provided in the material database. the easiest way to determine the structural stresses in the fatigue tests is to simulate the test in FE-Fatigue. including mild and high strength. double cup. Rupp et al [1] noted that. Plot the data and carry out regression to determine the S-N curve parameters (equation (16)) as normal. Figure 18 H Shear specimen For more details of the testing and different specimen types. especially when more precise S-N data. Calculate the stresses using equations (1) to (15) – FE-Fatigue provides access to the structural stress results. the theory is not perfect. The S-N curves in the database represent an average from a variety of different component types. The stress “S” in the spot weld S-N curves is that defined in equations (1) to (15) above. H-peel. and S-N curves derived from specimens which load the spot welds in a different way (e.g. H-shear and H-peel) may be slightly different in slope and intercept. the procedure for determining spotweld S-N curves is as follows: Carry out constant amplitude fatigue tests with chosen specimen and loading configurations. However the errors are not sufficiently large to be a problem in a design analysis environment. is illustrated in figure 18. 5 Theory Page 39 1st Page Back Main Menu H-shear. There is quite a wide scatter band. hat-profile (internal pressure). for a range of automotive steels. One such specimen – the H-shear specimen – so called because it has the profile of the letter H and the spotwelds are loaded in shear. and unless the specimen and loading is extremely simple. its calculation requires that the specimen first be modelled in FE to determine the spotweld cross-sectional forces and moments. and so-on. see reference [1]. carry out finite element modelling and analysis to determine the spotweld cross-sectional forces and moments. which is partly a reflection of the fact that this data represents spot-welds in a variety of steels. nCode n 1st Page Back Main Menu . Better predictions may be possible if S-N data specific to the materials being used is available. To summarise.FE-Fatigue Rel. so in general. there were actually modest differences between the S-N curves and so the use of generic S-N curves for sheet steel and weld nuggets is quite reasonable.

Computation time is roughly proportional to the number of data points in the load histories. In practice. • Therefore when modelling three sheet spotwelds for analysis in FE-Fatigue. there are therefore 108 fatigue calculations per spot-weld. ignore results calculated at this node. Damage Calculation By default. substantial reductions in computation time can therefore be achieved by judicious filtering of the loading inputs. damage calculations are carried out at 10 degree intervals around the spot weld in both sheets and in the weld nugget (though the number of calculation points can be modified by increasing the angle interval to 20 or 30 degrees and by omitting the weld nugget from the calculation). Rainflow cycles are converted to equivalent stress amplitude for R=0 using equation (17). by scaling and superimposing the force and moment results of a number of static load cases according to the quasi-static method [7]. Further research was carried out to extend the method to three sheet welds. Where three sheets are spotwelded together.g. then rainflow cycle counted to form a range-mean histogram. e. the structural stresses in the outer two sheets. no cracks grew in the middle sheet. This will detect the common node shared by the two CBAR elements. By default. The basic findings of this research were: • Where three sheets are spotwelded together. then damage is calculated and summed using Miner's rule. using PVXMUL. during tests on a variety of specimen shapes and loading conditions. The method for life prediction of spot-welds described here is somewhat computationally intensive. modified by empirical factors which adjust the theoretical stress to take into account size and loading type effects. and also provides an indication as to which sheet the crack is likely to appear in. Mx. in fact. the best approach is to model them using two CBAR elements.y. it is unusual for the crack to grow in the middle sheet. At each calculation point the effective stress history is calculated either directly from the force and moment results from a transient FE analysis. The shortest predicted life from all the calculation points predicts the life of the spotweld.y and Fz. of the following general form: nCode n 1st Page Back Main Menu .6√s in equation (4) is such a factor. some spotwelds are often created which join three sheets or even four (though this is undesirable and likely to lead to poor welds). or using a modal superposition method. In addition to reducing the number of calculation points. The method as originally developed by Rupp et al [1] only applied to spotwelds joining two sheets together.5. and its initial direction. This approach can be generalised to include a stress factor applied to components of the structural stress due to Fx. The spotweld analysis option in FE-Fatigue uses the cross-sectional forces and moments in CBAR elements to calculate the structural stresses around the spotweld. and report for both CBAR elements the shortest life from either sheet.FE-Fatigue Rel. as if they were separate spotwelds. and 1 for constant amplitude loadings. The factor K1=0. and to check the three-sheet correction option when running the software. and ignoring the possibility of failure at the middle sheet. Rupp et al [1] recommend that a Miner’s sum of 0.5 be used when the loading is of variable amplitude. and the resulting fatigue lives can be reasonably predicted by treating the weld as two two-sheet welds.6. The structural stresses are based on analytical expressions for stress. The stress history at each calculation point is calculated from the force and moment results using equations (1) to (15). Note: In FE-Fatigue the equivalent units can be used to simulate a Miner’s Sum value not equal to 1. 5 Theory Page 40 1st Page Back Main Menu 3.

This requires that when meshing the flanges to be joined. SFFZ.. A note on the subject of modelling spotwelds One criticism that has been levelled at the simple method of modelling spotwelds using CBAR elements is that it tends to lead to a slight under-estimation of the global stiffness of spotwelded structures. TEFXY. Component Fx. The bending stiffness of this connection is a little too low compared with reality. This lack of stiffness is due to the fact that each spotweld is modelled by CBAR elements which are connected to the surrounding mesh at a single node..0 Thickness Exponent e2 TEFXY = -0. 5 Theory Page 41 SF = F x diametere1 x thicknesse2 .6 SFFZ = 0. If the material type is an aluminium alloy.e. SFMXY.y Mx.0 Thickness Exponent e2 TEFXY = 0. and that they be of the correct length.0 DEFZ = 0.0 Diameter Exponent e1 DEFXY = 0.FE-Fatigue Rel. TEFZ.5 DEMXY = 0. The problem is minimised by using large shell elements around the spotweld..6.25 TEMXY = -0. DEFXY. This is done by using ENM to set the environment keyword SPOTWPAR.0 It is also possible for the user to override these automatically selected values and define the parameters. Comparing this with the original structural stress equations from Rupp et al. Another issue is that of the meshing requirement. DEFZ.0 SFMXY = 0..0 TEMXY = 0..y Mx.0 DEMXY = 0. and is more apparent as the sheet thickness increases (leading to larger bending moments)... The value of the keyword should be the nine parameters.5 These are the default parameters used in the software for spotwelds in steel.4 SFMXY = 0.4 SFFZ = 1. i. comma separated. TEMXY..y Fz Factor F SFFXY = 0.5 DEFZ = 0...(18) 1st Page Back Main Menu Where F is a factor and e1 and e2 are the diameter and sheet thickness exponents. WARNING: This keyword should not be set unless the user has sufficient evidence to justify using values different to those set as defaults in the software! 3. In general there are therefore 9 parameters which describe the complete set of stress factors. DEMXY.25 TEFZ = 1.. especially if the parts to be joined are meshed separately.6 Diameter Exponent e1 DEFXY = 0.6. SPOTWPAR = SFFXY. This is not always easy to achieve.5 TEFZ = 0. it can be seen that the parameters for steel are as follows: Component Fx.y Fz Factor F SFFXY = 1. The method requires that the CBAR elements be perpendicular to the sheets of shells they join. the size and thickness effects are somewhat different and the following set of parameters are automatically selected. nodes exist directly opposite each other at the location of the spot weld. Some FE pre-processors do however include automated spotweld nCode n 1st Page Back Main Menu .

For these reasons. ease modelling requirements and satisfy the requirements of groups analysing other vehicle attributes. The MPCFORCES on the solid elements are used to calculate an equivalent set of forces and moments at each end of the weld. negative mass elements have to be added to compensate. which is supported in FE-Fatigue. Some of the different methods that have been tried are listed below: • Selective thickening of shell elements around the spotweld – this improves the stiffness of the structure. Durability. the simple CBAR method tends slightly to underestimate the global stiffness of a spotwelded structure. the ability to calculate reasonably accurate local stresses is paramount. Support has been added for some of these and other new methods are being reviewed. uses 8-noded CHEXA elements in NASTRAN to connect the sheets of shell elements. 5 Theory Page 42 1st Page Back Main Menu modelling capability to generate bar elements of the required type at appropriate locations in the model. the global stiffness and modal behaviour is most important. and so can effectively double the model size for a typical car body. whereas for durability. • • An alternative method. a large number of different strategies for modelling spotwelds have been explored. Connecting the CBAR elements to the surround shells using a “spider” of rigid bar elements to distribute the load – this can improve the stiffness but adds complexity. Crash) whilst allowing rapid meshing and easy positioning of spotwelds.FE-Fatigue Rel. A number of other methods have been suggested in order to improve global stiffness. If dynamic calculations are being carried out. with the ultimate aim of achieving a strategy which will satisfy the requirements of different groups (NVH. Using shell elements to represent each spotweld – this method adds 32 shell elements for each spotweld. but has proved its worth over some years as a practical and effective way of modelling spotwelds for durability calculations. The solid is connected to the sheets using MPC equations (RBE3 elements) and congruent meshes on the two flanges are not required. For NVH. nCode n 1st Page Back Main Menu . In summary.

250. The Welding Institute. Landgraf. S. pp. (1995) "The Assessment and Use of Linear Static FE Stress Analyses for Durability Calculations. D. A. (1993) "Code of Practice for Fatigue Design and Assessment of Steel Structures.. Maddox. 169 . Dakin. S. 5 Theory Page 43 1st Page Back Main Menu 3." BS 7608. J. R. W. [4]Radaj." ASTM STP 1211. and Strange.185." Fatigue and Fracture of Engineering Materials and Structures. 6. D. Störzel. P. J. pp." Fatigue of Welded Structures. (1992) "Fatigue Life Estimation in Resistance Spot Welds: Initiation and Early Growth Phase. 15." SAE Technical Paper 951101. The MacNeal Schwendler Corporation.FE-Fatigue Rel. V. 37. 1994. Advances in Fatigue Life Prediction Techniques.7. S. [6]Sheppard. CA. (1993) "Estimation of Fatigue Propagation Life in Resistance Spot Welds. Los Angeles.293. [7]Heyes. and Grubisic. 531 . No.. K.6. J. M. Mitchell and R. (1995) "Computer Aided Dimensioning of Spot-Welded Automotive Structures". No. Version 68. 245 . A. C. pp." Engineering Fracture Mechanics. [3]British Standards Institution.549. D. ASTM Philadelphia. 287 . [5]Sheppard. E. nCode n 1st Page Back Main Menu . pp. [2]Smith. Ed. Vol.. F. (1990) "Local Fatigue Strength Characteristic Values for Spot Welded Joints. References [1]Rupp. 1. Vol. [8]MSC/NASTRAN Quick Reference Guide.. (1988) "Theoretical predictions of the fatigue life of shear spot welds. SAE Technical Paper 950711. Eds. R. and Cooper. and StJohn. M.

1 Introduction Welding is used in many industries as an effective and economical method for making structural joints between metal parts. is in general much less than that of the parts which are welded together.7. leading to a stress concentration. welded structures will normally have residual stresses. All these factors cause the fatigue behaviour of a welded joint to be quite different from that of the parent materials from which it is constructed. notches. it is not reasonable to model the behaviour of the welded joint based on the parent plate properties. which may be of the order of the yield stress of the material. This may cause major changes to the microstructure and properties in this region. and expect to get accurate predictions! nCode n 1st Page Back Main Menu .7 FE-based fatigue analysis of welded structures 3. Even in a well-designed structure. There are a number of reasons for this: • Welding typically introduces geometric features. The stress will typically be highest at the toe of the weld. For this reason. it is often the welded joints that are most likely to fail by fatigue. showing fusion and heat affected zone (HAZ) and most likely failure location The fatigue strength of welded joints. • The welding process will very often produce defects which can act as crack initiation sites – slag inclusions. incomplete fusion. However. • Around the fusion zone there is a heat affected zone (HAZ) where the parent material has been heated to a high temperature and allowed to cool fairly rapidly. • Unless steps are taken to relieve them. 5 Theory Page 44 1st Page Back Main Menu 3. residual stresses. This section provides a brief overview of: • some of the issues concerning fatigue of welds • some different methods • implementation in FE-Fatigue 3. welds naturally tend to be made at geometric features or changes in section in the structure. with modifications to take into account the defects.FE-Fatigue Rel.7. and a fatigue assessment of any such structure must place a high priority on the evaluation of the durability of the welds. the nature of the welding process means that welded joints generally have a fatigue strength that is inferior to that of the parts being joined together. and material property changes. and the shape in this area may not be well controlled. porosity etc.2 Background WELD WELD TOE HAZ PLATE Figure 19 Weld cross section. At the same time. or of the “parent plate”.

and life may be well predicted by a linear elastic fracture mechanics (LEFM) approach. • There may be large differences between the fatigue resistance of different grades of steel. These curves incorporate all the effects of defects. residual stresses. Note that the fatigue strength of the weld is much less than that of the parent plate. 5 Theory Page 45 1st Page Back Main Menu All the approaches described in this section are therefore based on characterisation of the fatigue behaviour of the welded joints themselves.123 b2: 0 E: 1. This is typical of weld S-N curves and is indicative of a situation where fatigue damage is dominated by crack growth. Consider the S-N curves shown below: S-N Data Plot classF SRI1: 1. and material property changes which are introduced when the weld is made. Note also that the slope of the weld S-N curve is much greater. This is because the processes occurring in the fusion and heat affected zones completely change the properties of the material. There are a couple of other points to note about fatigue of welds.914E5 UTS: 480 1E3 Str ess Ra ng e (M Pa) 1E2 1E1 1E3 1E4 1E5 Life (Cycles) 1E6 1E7 1E8 This figure compares the S-N curves for a typical structural steel (BS4360-50D) and one of the weld classes from BS7608.3333 b2: -0.07E5 UTS: 500 BS4360-50D SRI1: 1903 b1: -0. nCode n 1st Page Back Main Menu . • Fatigue in welds is comparatively insensitive to mean stresses. normally in the form of S-N curves. these differences may become very small. it is not surprising that the effects of mean stress are often neglected when predicting the fatigue life of welds. particularly in large structures will use a crack growth approach.FE-Fatigue Rel.2 E: 2. including negating the effect of any heat treatment. When you consider that welds are often subject to large and unknown residual stresses. notches.201E4 b1: -0. When welded however. except at very short lives where the curves converge. In fact many assessments of the structural integrity of welds.

From this name you can tell that the methods were initially developed for application of civil engineering structures constructed from thick plates and beams. The main steps The main steps in a BS7608 fatigue analysis are as follows: 1.3 BS7608 and other simple S-N methods Introduction One typical (and very useful) S-N based method for life prediction of welds is described in the British Standard BS7608:1993 “Code of practice for fatigue design and assessment of steel structures” [1]. if the reader intends to use this method. This standard is very closely based on BS5400 Pt 10 “Code of practice for fatigue design of steel concrete and composite bridges” [2]. joint geometry b. and welds.7. use S-N curve for the appropriate weld class c. or part of a joint based on a. it is strongly recommended that he read the standard document first.e. in air or seawater in the sub-creep regime (i. However. it has been successfully applied to a variety of problems in the automotive and other industries. Choose a weld class for each joint.FE-Fatigue Rel. the likely failure location d. bolted and riveted joints. together with some hints on application in FE-Fatigue. with some surprisingly good results. the loading direction c. 5 Theory Page 46 1st Page Back Main Menu 3. not at high temperatures). on much thinner structures than envisaged by the writers of the standards. rainflow count the stress history b. It allows the fatigue assessment of plates. use Miner’s rule with small cycle correction to calculate damage and life d. However. the type of weld (full penetration etc) 2. A brief summary of the main points is given here. Assess the validity of the results nCode n 1st Page Back Main Menu . apply size effect correction (for plates > 16 mm only) 4. Determine the required stress history due to the loading 3. The standard is applicable to structures made from wrought structural steel with a yield strength of less than 700 MPa. Here we will only consider its application to welded joints. Calculate the life: a.

Most welds tend to be classified F.FE-Fatigue Rel. or considered to lie between two possibilities. Because the standard was designed essentially for civil engineering structures. some judgement and imagination is called for when applying it to different situations such as automotive components. F2 or G. However. and where the loading direction may not be simple. 5 Theory Page 47 Weld classification 1st Page Back Main Menu BS7608 includes a system for classifying welds as briefly described above. Figure 20 Weld classification in BS7608 nCode n 1st Page Back Main Menu . where the configurations of welds may not closely match any of those illustrated in the standards. welds can normally be given a reasonable classification. The figure below shows some of the pictures and notes that are useful in classifying the welds.

where t is the sheet thickness. but excluding stresses from locations closer than 0.1t from the weld toe.FE-Fatigue Rel. with its edge 0. but more often it is a “hot spot” or structural stress. 5 Theory Page 48 Determine the stress history 1st Page Back Main Menu The location and nature of the stress that must be used to calculate the fatigue damage is described in detail in the standard. F Stress Structural stress at toe of weld t 0. • The stress may be the nominal stress in a plate. It is also allowable to determine the stress by experimental means.4t Figure 21 Determination of structural stress The required stress is determined by extrapolating the stress in the plate to the weld toe. This corresponds in most cases with the absolute maximum principal stress. usually: • The stress required is the principal stress with the largest range. and may vary for each case.4t to the weld toe. However. The method of determining the structural stress is illustrated in Fig. See Figure 20. Figure 22 Experimental determination of weld toe stresses nCode n 1st Page Back Main Menu . close to the weld toe. 21. In that case a resistance strain gauge of 3 to 5 mm in length should be applied to the plate.

especially when many welds must be considered in a rather simply meshed structure. The mesh should be arranged so that it has a row of elements with their centroids at the location of the weld toe as illustrated in Figure 22. the first method is rather cumbersome. F Stress t 0. and the experimental method is not useful. nCode n 1st Page Back Main Menu . the stresses from these elements should give a good approximation to the weld toe structural stress. 5 Theory Page 49 1st Page Back Main Menu Clearly. A rather simpler strategy is required. where the elements represent the mid-planes of the sheet metal. but neglecting the detailed effect of the stress concentration. and the weld is not modelled in detail.FE-Fatigue Rel. One simple approach that gives satisfactory results for welded sheet structures is to make a mesh from thin shell elements. The essence of the stress determination is that we need to find the structural stress at the weld toe.4t Element centroids positioned at weld toe Figure 23 Simple mesh to support BS7608 calculation So long as the mesh is of good quality. in a design analysis environment.

and if all cycles are below this stress level. open the material database manager and copy your chosen weld class data to a new dataset and edit the data to set the second slope b2 to zero. The stress level at this point represents the fatigue limit.FE-Fatigue Rel. when carrying out the fatigue calculation in FE-Fatigue. from -1/3 to -1/5. and your calculation may be somewhat conservative. For a more rigorous application of the standard. However. on the Advanced Options form. In BS7608 welded analysis.2 standard deviation curves).e. However. nCode n 1st Page Back Main Menu . i. the second slope of –1/5 is used for the whole calculation. if one or more cycles is above this stress level. If you want to make a BS7608 calculation using FE-Fatigue. S-N curves such as that illustrated above are used (the dashed lines represent the +/. select the Small Cycle (Haibach) correction. The other feature of BS7608 is the way that it considers the effect of small cycles. in the normal way. you can simply use the S-N curves as defined in the database. no damage should be predicted. No mean stress correction is used. At a life of 107 cycles there is a change in slope. the second slope is changed from –1/ 5 to 0. Then. 5 Theory Page 50 Calculate the life 1st Page Back Main Menu Figure 24 BS7608 S-N curve for Class F2 welds Life predictions in BS7608 use rainflow counting of the largest principal stress (absolute maximum principal) and Miner’s rule. in this case you will not be considering the case where all cycles lie below the fatigue limit.

there is a requirement in BS7608 that the principal stress directions should not vary by more than 45 degrees. so results were calculated for Class G and Class F2. In particular. • standard S-N curves were used with no further consideration of material variation • the component was made of thin sheets no more than 3 mm thick. Part 1:Code of practice for design” [3] which has a short section on fatigue. • there is good correlation between analysis and test Figure 25 Typical analysis results for an automotive suspension component subjected to a constant amplitude fatigue sign-off test. a multiaxial assessment should be carried out. 5 Theory Page 51 Assess the results 1st Page Back Main Menu Some typical fatigue analysis results are illustrated in Figure 24.FE-Fatigue Rel. In fact the S-N curves look very similar except that the fatigue strength is reduced by a factor of 3. nCode n 1st Page Back Main Menu . This is described in BS8118 Part 1: 1991 – “Structural use of aluminium. Where there is multi-axis loading. • there was some doubt about the correct classification of the weld. Note that: • results are only calculated for the strip of elements representing the weld toe. the analyst will have to exercise some judgement as to the most appropriate method. If this limit is exceeded. Aluminium alloys There is a very similar procedure available for use with welded aluminium structures. This section is very closely based on the methods described in BS5400 Pt 10. In the example illustrated above. the loading was quite simple.

FE-Fatigue Rel. 5 Theory Page 52 Other simple approaches for welds

1st Page

Back

Main Menu

Of course, the general S-N based procedure described above may be used even if the analyst does not want to use the weld classification and S-N curves found in the British or other standards; there is nothing to prevent the use of user defined S-N curves. There are a number of other approaches which might also be applied in FE-Fatigue. One such method developed by a group of the JSAE is summarised in the diagram below.

Figure 26 JSAE method This method is based on a parameter called εa,0.3t which is simply the strain amplitude determined at a distance of 0.3t from the weld toe, which, with some modifications to take into account the effect of thickness and flank angle variations, can be used to correlate fatigue lives from a variety of different weld configurations.

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. 5 Theory Page 53

1st Page

Back

Main Menu

**3.7.4 The “Volvo” approach for thin-sheet structures
**

Introduction Approaches such as the BS7608 method are not universally popular for automotive applications, for the following reasons: • They were mainly developed for thick sheet structures, whereas the majority of automotive welds join sheets of thickness 3mm or less. • Weld classification systems designed for civil engineering structures such as bridges can be difficult to apply to many automotive structures where many different weld geometries may be seen. • They do not always lend themselves to use in conjunction with a fairly unrefined finite element model such as is typically used for automotive body analysis. These criticisms were the motivation behind a new method developed at Chalmers University in Gothenburg, at the behest of the Volvo Car Corporation. The method as originally developed and described in Reference [5,6] was based on similar concepts to the LBF spot weld method. In this original concept, structural stresses at the weld toe were calculated based on the nodal forces acting on the weld toe elements. The method has been refined and modified to use node-atelement stresses from the weld toe, and some practical examples of the use of the refined method are described in Reference [7]. The method implemented in FE-Fatigue is based on that originally proposed by Volvo Car Corporation and Chalmers University of Technology in Gothenburg and developed in co-operation with nCode. This method has been developed specifically for automotive components welded from thin sheets (1-3 mm in thickness). It can be summarised briefly as follows: 1. The method was originally developed for welds modelled and analysed in NASTRAN. However, it should also work with any FE code, provided node-at-element stress results are available. There are some specific rules for meshing the welds, but essentially the structure must be meshed predominantly with CQUAD4 elements (or equivalent) representing the mid-planes of the metal sheets, and the weld bead represented by a single row of shells with thickness corresponding to the thickness of the weld throat. The nodes of these elements should lie at the weld toe. The mesh around the weld should be regular with elements of around 5 mm in size, and triangular elements must be avoided. See Figure 26. Unaveraged node-at-element stresses are recovered for the nodes at the weld toes, based on the elements adjacent to the weld (not the weld elements themselves) and used to make the fatigue calculations. The worst result for each node is reported to the results file. The method determines the contribution of bending to the total stress and from this determines whether the weld is essentially “stiff” or “flexible”. Each of these cases requires a different S-N curve.

2.

3.

4.

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. 5 Theory Page 54

1st Page

Back

Main Menu

Figure 27 Seam weld modelling of a simple T-joint The method is now described in more detail. Meshing and FE analysis rules In order to get good results, some simple meshing rules need to be observed. • The sheets are modelled by thin shell elements located at the mid-planes of the sheets. The element thickness should be equal to the plate thickness • A row of nodes should be positioned (at the mid planes) adjacent to the each weld toe. (See figure 27) • The weld bead should be modelled using 4 or 3 noded shell elements (CQUAD4 and CTRIA3 if using NASTRAN). The thickness of these elements should be equal to that of the effective weld throat. • The elements representing the part of the plate immediately adjacent to the weld toe MUST be 4 noded shell elements (CQUAD4). These are the elements from which the stress results are recovered for use in the fatigue calculation, so it is particularly important that these are the correct type and of good shape. The element length should be approximately 5 mm.

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. Note the well-shaped elements at the weld toe and the way the corner is radiused. 5 Theory Page 55 1st Page Back Main Menu Figure 28 Basic meshing practice for seam welds. nCode n 1st Page Back Main Menu . and weld starts that should be observed. Figure 29 a) Overview of continuously welded T-joint. b) Detail of corner weld. These are detailed in the following figures for different joint geometries. There are a number of additional requirements for weld corners.

Note that the modelling of the weld still is according to the guidelines. b) Detail of corner showing shape of weld toe elements. Note that the weld elements in the edge to edge connection are modelled in the same plane as the elements representing the section sides. 5 Theory Page 56 1st Page Back Main Menu Figure 30 a) Overview of intermittently welded T-joint. b) Side view of weld corner (see direction marked A). c) Detail of corner – note the angles used to radius the weld end. nCode n 1st Page Back Main Menu . The weld is modelled with 4-node and 3-node shell elements. Figure 31 Modelling of an edge to edge thin walled section.FE-Fatigue Rel. a) Overview of joint.

FE-Fatigue Rel.PLOT.SNORM.1. nCode n 1st Page Back Main Menu . PARAM.NO PARAM.g. 5 Theory Page 57 1st Page Back Main Menu Figure 32 Intermittently welded overlap joint. $ $ GRID Data $ GRID 1 0.PRINT)=all DISPLACEMENT(PLOT)=all BEGIN BULK PARAM. Note that the weld end is modelled in the same was as for the intermittently welded T-joint except that the triangular element at the end of the weld bead should have its free edge normal to the sheets.55.-1 PARAM.AUTOSPC.K6ROT. PARAM.1. a) Overview of joint.0 and stresses recovered to the output2 file using the STRESS(CUBIC) option.EPZERO.0 0.0 The SNORM parameter defines a common surface normal at each node. the parameter SNORM should be set e. b) and c) Detail of weld end.SNORM.E-5 PARAM.PRGPST.0 0.POST. A simple NASTRAN file header might be as follows: SOL 101 CEND $ TITLE = static analysis SPC=1 SUBCASE 1 SUBTITLE = Simple loadcase 100 N LOAD = 2 STRESS(CUBIC. and the CUBIC stress recovery option uses a strain gauge approach with cubic bending correction to recover the nodal and element centroid stresses. FE Stresses When running the analysis in NASTRAN.55.YES PARAM.

Results are recovered from top and bottom surfaces. mode or timestep. which we should not be using and may lead to more conservative predictions. absolute maximum principal stresses for each static loadcase. Note that. more than one set of stress results will generally be recovered. unlike a normal FE-Fatigue analysis.FES file are the top and bottom surface absolute maximum principal stress values. Figure 33 Results recovery nCode n 1st Page Back Main Menu . The results are node-at-element results from the elements in the sheets adjacent to the weld toe only – not from the weld elements themselves. which we should be using to calculate the life. the stress results stored in the . The stresses recovered from the FE results are the unaveraged. At each node along the weld toe therefore. the results translator FE2FES is unable to distinguish between the elements in front of the weld toe. FE-Fatigue will make separate fatigue analyses based on these unaveraged stresses and report the worst (shortest life or greatest damage) result. at the location of the weld toe. and this prediction should be based on stresses recovered from the weld toe elements at the nodes along the weld toe. The user can avoid this by controlling the set of elements for which stresses are recovered to exclude the unwanted elements. the full stress tensor is not recovered. and those “underneath” the weld toe.FE-Fatigue Rel. The user should also be aware that in this version. 5 Theory Page 58 1st Page Back Main Menu The fatigue analysis method as formulated in this version of FE-Fatigue has been developed and validated to predict fatigue failure of welds from the weld toe.

The most distinctive feature of this plot is that the data correlates quite well giving rise to two distinct S-N curves. Closer inspection of the results revealed that the the stresses at the weld toe for the test results on the upper curve were characterised by a predominance of bending.5(σ top + σ btm ) nCode n 1st Page Back Main Menu . Let us call these σtop and σbtm. Now the stress at the weld toe can be divided into a bending component: σ b = 0. whereas those falling on the lower curve had mainly tension/compression stress. Now the stresses used in the fatigue calculation are the absolute maximum principal stresses from the top and bottom surface of the weld toe elements. Figure 34 S-N data for a variety of different weld geometries. Loadings are constant amplitude. Another way of looking at this is to say that the results correlated to the upper curve were for joints that were “flexible” in character and those correlated to the lower curve were “stiff”.5(σ top − σ btm ) and a direct component: σ n = 0. 5 Theory Page 59 S-N Curves and Fatigue analysis method 1st Page Back Main Menu The structural stress at the weld toe may be used to correlate stress-life data from a variety of welded specimen configurations and loading directions.FE-Fatigue Rel. and all data is corrected to R = -1 using the Haig diagram (constant life diagram) in figure 34. The results illustrated below are taken from Reference 6. at the weld toe nodes.

M1 for R ≤ 0 and M2 for R ≤ 0. Failure is deemed to have occurred when there is a visible crack. by setting a threshold value of the flex ratio r. Clearly. preferably at load ratio R = -1. and the user must define the ratio M1/M2 to be used during the analysis. a Haig diagram is used.5. two mean stress sensitivity factors are defined. This ratio will therefore normally be expected to have a value of 2. Ideally.1 respectively. as illustrated below. In order to account for mean stresses. typical values for M1 and M2 are 0. a number of different geometries would be tested. Figure 35 Haig (constant life) diagram for welded joints. which may be in the range 0 ≤ r ≤ 1.FE-Fatigue Rel. The value of M1 may be defined in the material database.25 and 0. it is necessary also to make FE models of the specimens used. nCode n 1st Page Back Main Menu .5 for steels. S-N curves for use with this method should be determined by testing welded steel specimens. 5 Theory Page 60 We can now define a flexibility ratio r 1st Page Back Main Menu r= σb σb + σn This can be used to distinguish between the “flexible” and “stiff” cases. In order to obtain the structural stress values required to plot the results on a S-N diagram. In this diagram. The default value for this threshold in the software is 0. For steels. following the guidelines described above.

in this version. nCode n 1st Page Back Main Menu .FE-Fatigue Rel. otherwise use “flexible” curve Report worst (shortest life) result at each node Rainflow count top and bottom surface stresses and calculate lives Figure 36 Analysis process summary for Volvo seam weld approach. more research is needed to clarify this point. However. use “stiff” curve. 5 Theory Page 61 1st Page Back Main Menu The analysis process as implemented in FE-Fatigue is summarised in the following diagram: STATIC FE ANALYSIS MODAL TRANSIENT FE ANALYSIS Top and bottom surface stresses for each contributing mode (largest principal) DIRECT TRANSIENT FE ANALYSIS Top and bottom surface stresses for each static load case (largest principal) Load-time histories Linear superposition Linear superposition Modal participation factors Time histories of top and bottom surface stresses σtop (t) and σbtm(t) Determine minimum value of flexibility ratio r. the stresses that are superposed are the absolute maximum principal stresses for each of the static modes or loadcases. Note that the analysis may be based on a static superposition. There are some indications that it might be wise to use a Miner’s sum of less than 1 to calculate the life when using this method with variable amplitude loadings. Figure 35 illustrates the fatigue analysis process using the “Volvo” method as implemented in FE-Fatigue. and not the full stress tensors. when the static or modal superposition approach is used. modal superposition or transient approach. Note that. If rmin < threshold.

Koibuchi. “Method of Fatigue Life Estimation for Arc-Welded Structures”. Fermér. Nov 1997. Maruo. Frodin. Andréasson and B. No. Kasahara. 2001.7. BS5400 Pt 10:1980 “Code of practice for fatigue design of steel concrete and composite bridges”. 5. “Fatigue Life Prediction of MAG-Welded Thin Sheet Structures”. T. Svensson. IBEC ’98. Magara and K. Y. British Standards Institution 3. “Industrial Experiences of FE-Based Fatigue Life Predictions of Welded Automotive Structures”. M. British Standard BS7608:1993 “Code of practice for fatigue design and assessment of steel structures”. 1998. Chalmers University of Technology and Volvo Car Corporation. nCode n 1st Page Back Main Menu . 7. T. 6. SAE Technical Paper 982311. MSc Thesis. M. Fatigue and Fracture of Engineering Materials and Structures. British Standards Institution. Proc. 2. S. M. SAE Technical Paper 2001-01-0781. BS8118:1991 “Structural use of aluminium. Fermér and H. Nakamaru. Part 1:Code of practice for design”. M. Andréasson and B. “Fatigue Life Prediction of MAG-Welded Thin Sheet Structures – Theory and Experiments”. 24. Vol. 5 Theory Page 62 1st Page Back Main Menu 3.FE-Fatigue Rel.5 References 1. Frodin. 7. British Standards Institution 4.

It than uses the mean value to interpolate between the curves. life = 239768 cycles. then interpolation can be used using the four curve values surrounding the cycle value ( in this case.FE-Fatigue Rel. the values on the adjacent curves are as follows: Mean curve at 25 MPa. If the amplitude/mean pair lies inside the set of curves. which may or may not lie on the curve. The program first calculates the value on the adjacent curves at the required amplitude using a log-log interpolation. life = 617931 cycles. nCode n 1st Page Back Main Menu . where each curve represents the stress amplitude and life for a given mean stress or R-ratio. interpolation must be used. 5 Theory Page 63 1st Page Back Main Menu 3. for example in the case above.1 Interpolating life from a set of multiple mean stress life curves FE-Fatigue supports the input of and calculation from materials data in the form of multiple S-N curves. stress = 365 MPa. In the example shown. an amplitude of 400 MPa and a mean of 50 MPa. The cycle consists of an amplitude and a mean value.8.8 Multiple mean stress curve analysis 3. stress = 435 MPa. An example is shown below: Figure 37 Multiple S-N curves In order to use these curves instead of a mathematical mean stress correction such as Goodman. 2 values on the curve mean=25 and 2 values on the curve mean=75. Mean curve at 25 MPa.

((log10(435)-log10(400)) x (log10(239768)-log10(617931) / (log10(435)-log10(365))) L25 = 377016 cycles At 75 MPa: log10(L) = log10(126392) .((log10(435)-log10(400)) x (log10(126302)-log10(361833) / (log10(435)-log10(365))) L75 = 208905 cycles The life for a mean of 50 MPa is then interpolated as follows: log10(L50) = log10(377016) + ((log10(208905) . At 25 MPa: log10(L) = log10(239768) . In the plot below an interpolated curve at 50 MPa has been added: Figure 38 Includes interpolated curve at 50 MPa nCode n 1st Page Back Main Menu . Mean curve at 75 MPa.FE-Fatigue Rel. stress = 435 MPa. 5 Theory Page 64 1st Page Back Main Menu Mean curve at 75 MPa. life = 126302 cycles. The program first calculates the life at 400 MPa (L) on each of the curves.log10(377016)) x (50-25) / (75-25)) L50 = 280643 cycles Note that the mean stress is not logged. as the value can be zero or negative. life = 361833 cycles. stress = 365 MPa.

the program would extrapolate values for 100 MPa on the curves at 25 and 75 and then interpolate between as normal.((log10(435)-log10(400)) x (log10(30497)-log10(66306) / (log10(435)-log10(365))) L225 = 44210 cycles The life for a mean of 250 MPa is then extrapolated as follows: log10(L250) = log10(71274) + ((log10(44210) .log10(71274)) x (250-175) / (225-175)) L250 = 34819 cycles NOTE: this extrapolation may also be "turned off" by using the environment keyword MEANCEIL (set to "ON"). For example. 377016 cycles.FE-Fatigue Rel. which will then use the non-conservative value of the highest mean stress curve in the set. if the cycle has an amplitude of 400 and a mean of zero. 5 Theory Page 65 1st Page Back Main Menu If the data lies outside of the nest of curves. There is an upper limit of 1E30 on the value of life that can be extrapolated in this way. i. the program gives the value on the lowest curve as a conservative result.g. At 175 MPa: log10(L) = log10(48127) . ( e.e. If the mean is greater than 225 MPa. 250 MPa) the same equations are used with the two nearest curves and this extrapolates the value. nCode n 1st Page Back Main Menu .((log10(435)-log10(400)) x (log10(48127)-log10(109419) / (log10(435)-log10(365))) L175 = 71274 cycles At 225 MPa: log10(L) = log10(30497) . for example in this case if the amplitude was 100 MPa and the mean 50 MPa. For R-ratio curves.e. the data is converted to the equivalent mean values so that the same interpolation rules can apply. Extrapolation will also be performed if the amplitude value is less than the lowest amplitudes described in the set of curves. i. extrapolation is used only when the mean is higher than the highest mean curve.

FE-Fatigue Rel. 5 Theory Page 66

1st Page

Back

Main Menu

3.9 Vibration fatigue

Fatigue damage is traditionally determined from time signals of loading, usually in the form of stress or strain. This approach is satisfactory for periodic loading but requires very large time records to accurately describe random loading processes. These may prove prohibitive for many finite element analyses especially when modelling dynamic resonance. Alternatively, a compact frequency domain fatigue calculation can be utilised where the random loading and response are categorised using Power spectral density (PSD) functions and the dynamic structure is modelled as a linear transfer function. This chapter will review the available methods for performing fatigue analysis from PSDs and shows that the Dirlik method gives the best comparable results with the traditional time domain approaches. It also describes how these techniques are implemented in the Finite Element environment to rapidly identify critical areas in the structure.

3.9.1 Introduction

It is often easier to obtain a PSD of stress rather than a time history. Take, for instance, the dynamic analysis of complicated finite element models. Here it is often beneficial to carry out a rapid frequency response (transfer function) analysis instead of a computationally intensive transient dynamic analysis in the time domain. The offshore oil industry faced this problem in the early 1980's. An offshore oil platform is a hugely complex structure that is subjected to random wind and wave loading. A typical design analysis may have to consider over 70 load combinations on the structure. The analysis is further complicated because the imposed loads are random and dynamically excite the structure. A transient dynamic analysis in the time domain proves impossible to carry out in this case. A Finite Element analysis based in the frequency domain can simplify the problem considerably. The designer can now carry out a frequency response analysis on the FE model to determine the transfer function between wave height and stress in the structure. Using this, he simply multiplies the PSD of wave height by the transfer function to arrive at the PSD of stress. A review of this method is presented. To take advantage of the fast frequency domain techniques for fatigue analysis, we require a method of deducing damage from the PSD of stress. In this chapter we will briefly review the technology behind time domain S-N analysis and then draw a parallel approach in the frequency domain. A comparison study between the different fatigue analysis techniques is presented. The chapter concludes by describing the Finite Element approach for computing fatigue damage on vibrating components.

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. 5 Theory Page 67

1st Page

Back

Main Menu

**3.9.2 Review of S-N analysis in the time domain
**

The starting point for any fatigue analysis is the response of the structure or component. In the time domain this is usually expressed as a stress or strain time history. Fatigue occurs as a result of stress or strain reversals in the time history. These are known as cycles. The significant aspects of these are the ranges of stress in the cycle and also there mean stress. Today the range and mean information is usually extracted from the time history using a procedure known as 'Rainflow Cycle Counting'. Matsuishi and Endo first introduced the concept of rainflow ranges to the scientific community over twenty years ago. An example of the way rainflow ranges are extracted from a time signal is given by Downing [1]. The output from a rainflow cycle counting exercise is usually expressed as a range mean histogram such as that shown below. The stress range of each cycle is given along the x axis, its mean stress is shown on the y axis and the z axis gives the number of cycles contained in the time history for each particular range and mean. This data was taken from the Howden HWP330 wind turbine; Bishop [2] discusses the analysis.

4 0 0 stress MPa 3 0 0 2 0 0 1 0 0 0 -1 0 0 -2 0 0 0

5 0

1 0 0

1 0 5 T eS c . im e s

2 0 0

2 0 5

2 0 Cycles 1 5 1 0 5 0 0

2 2 4 6. 8 2. .7 17 58 .9 2 4 1 34 1. 3 43 7. R .6 an 51 3 8 ge 60 9. M 68 P a

-2 10 .3

an e M

a P M

Figure 39 Range-mean histogram derived from time history by rainflow cycle counting Each cycle will induce a certain amount of fatigue damage on the component. The total damage caused by the time history can therefore be obtained by summing the damage caused by each cycle shown in the stress range histogram. This approach is known as the Palmgren-Miner accumulated damage rule after the two independent people who proposed it.

nCode

n

1st Page

Back

Main Menu

FE-Fatigue Rel. 5 Theory Page 68

1st Page

Back

Main Menu

The damage caused by each cycle is calculated by reference to the material life curve, in this case the SN curve. The SN curve shows the number of cycles to failure, Nf, for a given stress range, S. The total damage caused by N number of cycles is therefore obtained as the ratio of cycles to the number of cycles to failure. The Palmgren-Minor rule can therefore be expressed as Equation (1).

Accumulated Damage =

åN

i

Ni

f

(1)

Where, Ni is the number of cycles with a particular stress range and mean; i is a ranging variable covering all the possible range and mean combinations; and Nf is the number of cycles to failure for a particular stress range and mean. The accumulated damage is expressed as a proportion of the damage required to fail the material. Therefore the fatigue life for the component can be determined from Equation (2).

Fatigue Life =

Length of time history Accumulated Damage

(2)

The fatigue life for the data shown in Figure 39 was found to be 0.2 years based on an aluminium alloy 2024_HV_T4.

nCode

n

1st Page

Back

Main Menu

30 Figure 40 Typical PSD of a random time history The mean square amplitude of a constituent sinusoidal wave can be determined by measuring the area under the PSD over the desired frequency range. frequency and phase. In practice we usually represent the frequency domain as a 'Power Spectral Density (PSD)' plot. however.9. The amplitude can be obtained from the modulus of the complex number while the phase is determined from the argument. If the time history is taken from an 'ergodic stationary Gaussian random process'. The PSD does not contain any phase data. nCode n 1st Page Back Main Menu . we are able to make assumptions about the original phase content and therefore regenerate a statistically equivalent time history. The frequency domain is simply another domain in which to view a time signal. This is a normalised density plot describing the mean square amplitude of each sinusoidal wave with respect to its frequency. 3E4 2E4 1E4 0 0 10 20 Frequency Hz. the x-axis now represents frequency instead of time. The amplitude can then be estimated using Equation 3. This section describes the theory of time history regeneration from PSDs and highlights some of the fundamental assumptions involved. To convert a time signal into the frequency domain we effectively split it up into a number of discrete sinusoidal waves of varying amplitude.3 Simple fatigue analysis using time history recreation The most obvious method for calculating fatigue from a PSD is to regenerate a characteristic time history and proceed as described above. When these are added together they form the original time signal.FE-Fatigue Rel. (3) Amplitude = 2 ⋅ Mean Square To regenerate a time signal from the frequency domain we would usually perform an 'Inverse Fourier Transformation' on the complex vector of frequency domain results. The algorithm used to split the time history into its constituent sinusoidal components is the 'Fourier Transform'. This would yield exactly the same time history as we started with. 5 Theory Page 69 1st Page Back Main Menu 3. This returns a complex vector of values where each value represents the amplitude and phase of the particular sinusoidal wave at a particular frequency. A typical PSD plot is shown below. When starting with a PSD this method is inappropriate because the PSD does not contain any of the original phase information. it is found that the phase is purely random between -π and +π radians. For certain time histories.

A stationary process is called an ergodic process if statistics taken from one sample are the same as those obtained for the ensemble. we can effectively take a single sampled time history from the process and safely assume that this contains all the required statistical properties of the parent process. For nonstationary processes the statistics obtained from a sampled time history would not be representative of those of the whole random process as these would be continuously changing. the vibrations occur as a result of many deterministic sources and the central limit theorem tends to make them behave in the desired fashion. 5 Theory Page 70 1st Page Back Main Menu A process is said to be stationary if its statistics are not affected by a shift in the time origin. therefore. wave height.FE-Fatigue Rel. the later examples are not random processes. however.) To test for stationarity we take a number of recordings of the random process at different times. Strictly speaking. If the ensemble probability density function is Gaussian then the process is known as a Gaussian random process. The process is stationary if the probability distributions of the ensemble are the same for all points in time. the statistics of a time history X(t) are the same as a time history X(t + τ) for all values of τ. nCode n 1st Page Back Main Menu . We can then generate random phase angles and add these to the amplitude data given in the PSD. With an ergodic stationary random process. Naturally occurring phenomena like wind speed. To regenerate a time signal from a PSD we therefore assume that the original process was 'ergodic stationary Gaussian and random'. In practice we see that many time signals obey this criterion. etc. (For example. At this point we use the Inverse Fourier Transformation to determine a statistically equivalent time history. usually obey this along with vibrations from generators and engines.

for a narrow banded time history Bendat assumed that all positive peaks in the time history would be followed by corresponding troughs of similar magnitude regardless of whether they actually formed stress cycles. The range mean histogram given by Bendat contains no cycle mean data. The term in brackets in Equation (4) is the Raleigh distribution. S ü ì S − ï 8⋅m0 ï ⋅e N (S ) = E [P ] ⋅ T ⋅ í ý 4 ⋅ m0 ï ï î þ 2 (4) Where N is the number of cycles of stress range S occurring in T seconds. directly from the PSD of stress as opposed to a time history. Bendat showed that the probability density function (pdf) of peaks for a narrow band signal tended towards a Rayleigh distribution as the bandwidth reduced.4 Fast fatigue analysis methods in the frequency domain This section describes a variety of approaches for computing fatigue life. For more background information the reader is referred to Bishop [3] and Halfpenny [4]. To complete his solution method. Furthermore. these methods are only applicable for PSDs describing an ergodic stationary Gaussian and random process.e. those that estimate fatigue life directly and those that compute range mean histograms as an intermediate stage. 5 Theory Page 71 1st Page Back Main Menu 3. or damage. the area beneath the curve). and E[P] is the expected number of peaks obtained by Equation (5). E[P] = m4 m2 (5) m4 and m2 are the 4th and 2nd moments of area of the PSD respectively where the nth moment of area is obtained as: mn = ò f n ⋅ G ( f )df and G(f) is the value of the single sided PSD at frequency f Hz. Bendat's narrow band solution for the range mean histogram is therefore expressed in Equation (4). In 1964 Bendat [5] proposed the first significant step towards a method of determining fatigue life from PSDs. The techniques fall into two broad categories. Bendat used a series of equations derived by Rice [7] to estimate the expected number of peaks using moments of area under the PSD. The figure below shows the rangemean histogram obtained from the PSD of the time history given in Figure 39.FE-Fatigue Rel. As before. m0 is the zeroth moment of area under the PSD (i. Using this assumption the pdf of stress range would also tend to a Rayleigh distribution. nCode n 1st Page Back Main Menu .9.

In comparison. Hz^-1 3 4 E 2 4 E 1 4 E 0 0 1 0 F q e c H .2 1 1 . As Bendat assumes that all positive peaks are matched with corresponding valleys of similar magnitude. 5 Theory Page 72 1st Page Back Main Menu RMS Power MPa^2.9 42 5 27 5 3. (We now observe a fatigue life of only 280. the damage is grossly exaggerated for wide band histories as shown. .9 an e M M a P Figure 41 Range-mean histogram derived from PSD using Bendat's method The problem with Bendat's narrow band solution is that it is extremely conservative when wider band time histories are used. re u n y z 2 0 3 0 Counts Cycles 8 0 6 0 4 0 2 0 0 0 .7 21 3 16 3 2. This effect is illustrated in Figure 42. a wide band time history is characterised by smaller waves riding on a low frequency carrier.) The reason for this lies in the assumption that peaks are matched with corresponding troughs of similar magnitude. A narrow band time history is characterised by each peak having a corresponding valley of similar magnitude.FE-Fatigue Rel.6 44 M 8 P a -4 28 . nCode n 1st Page Back Main Menu .1 hours. R 9 an 63 73 ge .6 05 .

these being m0. the notable ones being Wirsching. The dynamic wave and wind load data was already provided in the frequency domain and it therefore seemed sensible to make use of the speed advantages inherent in a frequency domain analysis. The Dirlik formulation is given in Equation (6). The transient dynamic analysis proved too intensive for the time domain software because of the large structural models and high number of possible load combinations. The latter two methods are both in the form of an equivalent stress parameter and neither tends to work particularly well when used outside the offshore platform industry. Kam & Dover and Hancock. Large jacket platforms were being designed and fatigue failures had to be avoided. Steinberg and Tunna both worked on the problem for the electronics and rail industries respectively. Several methods were developed to address this problem. m2 and m4. 5 Theory Page 73 1st Page Back Main Menu Narrow band Stress time Wide band Stress time Cycle counting using Bendat’s method Stress time Figure 42 Why Bendat's method is conservative During the 1980's the need for a rapid fatigue analysis method based in the frequency domain became apparent to the offshore oil industry.FE-Fatigue Rel. Again neither of these methods tend to work well outside their respective industries. Wirsching's approach was developed for the offshore industry but has been found to be applicable to a wider class of industrial problems. For more information on these methods see Bishop [3]. In 1985 Dirlik proposed an empirical closed form solution to the problem following extensive computer simulations using the Monte Carlo technique. The problem was how to calculate a reasonably accurate fatigue life using the resultant PSDs from the frequency domain analyses. This method has been found to be widely applicable and constantly outperforms all of the other available methods. Sea state spectra are relatively wide banded and this effectively rules out the use of Bendat's narrow band fatigue analysis because the results prove too conservative. Although apparently more complicated than some alternative methods it is still only a function of four moments of area of the PSD. In other industries advances were also being made. m1. nCode n 1st Page Back Main Menu . These are semi-empirical approaches based on the narrow band solution.

D1 Q D2 ⋅ Z 2⋅R 2 ⋅e + ⋅e + D3 ⋅ Z ⋅ e Q R2 p (S ) = 2 ⋅ m0 −Z −Z 2 −Z 2 2 2 ⋅ xm − γ 2 D1 = 1+ γ 2 ( ) D2 = 1 − γ − D1 + D12 1− R D3 = 1 − D1 − D2 Z= S 2 ⋅ m0 Q= 1. 5 Theory Page 74 1st Page Back (6) Main Menu N (S ) = E [P ]⋅ T ⋅ p (S ) Where: N(S) is the number of stress cycles of range S N/mm2 expected in time T sec. E[P] is the expected number of peaks obtained by Equation (5).25 ⋅ (γ − D3 − D2 ⋅ R ) D1 γ − x m − D12 R= 1 − γ − D1 + D12 γ = m2 m0 ⋅ m4 nCode n 1st Page Back Main Menu .FE-Fatigue Rel.

0 0 95 1 19 5. The theoretical approach devised by Bishop is computationally intensive and shows little improvement on accuracy over Dirlik's empirical approach.1 28 0 1 38 5. R 1 an 57 5. the need for certification of the technique before its use meant that theoretical verification was required. RMS Power MPa^2.FE-Fatigue Rel. re u n y z 2 0 3 0 Counts Cycles 1 0 2 1 0 0 8 0 6 0 4 0 2 0 2 . nCode n 1st Page Back Main Menu . Again the method ignores the cycle mean stresses but now gives an improved range mean histogram that is comparable with that obtained using a time domain approach. The fatigue life is correctly calculated as 0. 1 47 0.2 years. 5 Theory Page 75 1st Page Back Main Menu xm = m1 m2 ⋅ m0 m4 Figure 43 shows the range mean histogram obtained using Dirlik's method for the wind turbine data used in Figure 39. 1 ge 66 0. Dirlik's empirical formula for rainflow ranges has been shown to be far superior to the previously available correction factors. Bishop [6] achieved this when a theoretical solution for predicting rainflow ranges from the moments of area of the PSD was produced. . M 76 P a 0 0 -3 86 n ea M a P M Figure 43 Range-mean histogram derived from PSD by Dirlik's method In terms of accuracy. Hz^-1 3 E4 2 E4 1 E4 0 0 10 F q e c H . Bishop's method gives credence to the Dirlik method but is seldom used for analysis purposes. However. For this reason.

66 1.32 Figure 44 Comparison between different frequency domain fatigue analyses The table clearly shows that the Dirlik approach is remarkably robust.76 0.86 0.14 5. 5 Theory Page 76 1st Page Back Main Menu 3.08 2.15 1.15 2.86 3.84 0.14 1.07 1. It shows an average discrepancy of only 4% from the fatigue life calculated in the time domain.98 1.81 1.79 13.FE-Fatigue Rel.04 1.48 0.71 2.84 0.22 1.59 7.57 1.92 10.34 81.98 Dirlik Wirsching Bishop Chaudhury Hancock 1.81 3.01 1.47 5.98 1.98 1.34 0.98 1.36 2.31 2.95 2.74 1. computed fatigue lives in the frequency domain and then compared these with the life obtained from the time domain approach.77 0.80 3.25 1.50 2.9.22 5.03 1.75 2.98 3.87 2.65 1.29 1.43 1.53 1. These studies took data from the Howden HWP330 wind turbine.13 1.01 1.02 1.87 1.59 1.42 1.44 1.03 1.67 6.70 2.03 2.00 1.80 3.23 2.67 18.43 1.45 1.91 1.33 1. The studies made by Bishop [2] are of notable significance.33 2. nCode n 1st Page Back Main Menu .66 1.91 3.62 3.12 30.77 5.54 2.93 3.59 2.33 2.04 3.00 1.13 2.34 1.64 1.12 2.51 1.31 1.83 1.16 4.78 1.18 2. A summary of the results is given in the table below.15 1.27 1.50 1.74 0.97 3.23 1.83 25.50 2.86 0.09 2.08 1.61 1.23 1.72 0.15 14.99 1.20 1.03 3.13 2.14 5.11 0.95 1.08 0.22 2.50 8.52 1.54 1. Load Case Narrow Band y12a y19a y27a y35a y12b y19b y27b y35b y12c y19c y27c y35c y12d y19d y27d y35d y12e y19e y27e y35e y12f y19f y27f y35f avrg 5.91 62.15 1.29 1.76 2.17 1.75 1.31 1.87 1.01 0.31 1.10 1.17 3.25 1.46 1.99 1.92 6.76 0.92 7.96 5.54 2.02 3.92 0.03 1.65 3.08 1.25 6.06 3.12 0.95 0.01 4.95 1.68 0.5 Comparison between fatigue analysis techniques Many comparative studies have been carried out to ascertain the accuracy of all the various methods.73 0.

This section considers how these PSDs are obtained from finite element analysis. Transfer function [MPa/N] 2000 1000 0 50 Frequency [Hz] 100 Figure 45 Transfer function relating input force to output stress at a node nCode n 1st Page Back Main Menu . An example transfer function plot is shown below. The peaks represent resonant modes in the component.FE-Fatigue Rel. In the frequency domain it uses a very fast 'transfer function' technique. The transfer function relates the amplitude of the input force or moment to the amplitude of the output stress for each frequency of sinusoidal wave. In the time domain this involves a complicated and often lengthy transient analysis. 5 Theory Page 77 1st Page Back Main Menu 3. This relates input force (N) to output stress (MPa) for a certain node on a FE model.9. Essentially the frequency domain breaks down a signal into its constituent sinusoidal waves following Fourier's theory.6 FE based vibration analysis in the frequency domain The previous section discussed methods of determining fatigue life from PSDs of stress. The dynamic behaviour of components can be determined both in the time and frequency domains.

Expressing this in exponential form yields the expressions for load and displacement shown in Equation (8). nCode n 1st Page Back Main Menu . Consider the simple bracket component shown below subjected to a single input load. PSDs of stress are required at each node on the model in order to determine the fatigue life of the component. [M ] ⋅ (t ) + [C ] ⋅ x(t ) + [K ] ⋅ x(t ) = f (t ) x Where: [M] is the global mass matrix [C] is the global damping matrix [K] is the global stiffness matrix x(t) is the vector of nodal displacements f(t) is the vector of nodal forces (7) In the case of a single input system the vector f(t) has only one non-zero component. Figure 46 Simple FE model of a mounting bracket The equation of motion for the component is given in Equation (7).FE-Fatigue Rel.9. 5 Theory Page 78 3.6.1 Calculating the linear transfer function 1st Page Back Main Menu This section describes the calculation performed by the Finite Element solver when calculating the linear transfer function. If the input load is sinusoidally varying with respect to time then the displacement vector v(t) will also be sinusoidally varying.

for the two dimensional plate elements shown in Figure 7. H xz (ω ) N. Axial H xx (ω ). H yy (ω ). nCode n 1st Page Back Main Menu . These are given in Equation (9).B. Hyz(ω) and Hxz(ω) are all zero. H yz (ω ). 5 Theory Page 79 1st Page Back Main Menu x(t ) = X ⋅ e i ⋅ω ⋅t f (t ) = F ⋅ e i ⋅ω ⋅t Where: (8) F is the amplitude of the sinusoidally varying force of frequency ω rad/sec and X is the amplitude of the induced displacement. Classical FE analysis results in a tensor of 6 component stresses at each node. These consist of 3 axial and 3 shear stress components and are expressed mathematically in Equation (11). the values of Hzz(ω). Differentiating displacement with respect to time yields expressions for the velocity and acceleration of the component. This is carried out in the FE program following the back substitution stage.FE-Fatigue Rel. Frequency can also be expressed in terms of Hertz after using the appropriate conversion. x(t ) = i ⋅ ω ⋅ X ⋅ e i⋅ω ⋅t (t ) = −ω 2 ⋅ X ⋅ e i⋅ω ⋅t x (9) Substituting these into Equation (7) and simplifying yields the following formula for the amplitude of displacement with respect to frequency: X (ω ) = H (ω ) ⋅ F (ω ) Where: H(ω ) is the linear transfer function given by: (10) H (ω ) = − [M ] ⋅ ω 2 + [C ] ⋅ i ⋅ ω + [K ] X(ω) is the vector of displacement amplitude F(ω) is the vector of force amplitude [ ] −1 It is usual to compute the transfer function between input force and output stress instead of displacement shown in Equation (10). H zz (ω ) (11) Shear H xy (ω ).

A typical application would be to determine the bending stress at the base of the flagpole as a result of the two random wind speeds seen by the flags. PSDs of wind speed can be calculated from these. To calculate the stress at the base of the flagpole it is insufficient to simply sum the reactions from the two input PSDs. If the two flags are far enough apart then the wind speed witnessed by one will be completely independent of that on the other.9. The two time histories are correlated because the random wind turbulence incident on one flag has a sufficiently large range of influence to also affect the response at the other.2 Single force input 1st Page Back Main Menu Having found the relationship between component stress and applied load. As they are moved closer together then a correlation between the two time histories will be seen. 5 Theory Page 80 3.6. instead we must sum the reactions from the input and cross-power spectra.FE-Fatigue Rel. however. nCode n 1st Page Back Main Menu . The time history of wind speed at the location of the flags can be recorded using anemometers. The cross-power spectra contain information on the joint statistics of the two processes.3 Multiple force input This section introduces the analysis of multiple random inputs. If the two processes are correlated then the sequencing effects may act to increase or decrease the base bending stress depending on whether the forces are in or out-of-phase.9.6. For a mathematical explanation of this see Newland [8]. the PSDs alone do not provide information on the phase relationships between the two measured time histories. the component stress PSD 'G(f)' for a single input load PSD 'W(f)' can be found very rapidly by multiplying the input PSD by the transfer function as in Equation (12). Wind Speed x(t) Wind Speed y(t) Bending stress required at the base of the flagpole Figure 47 Example of two random process inputs With multiple random processes we also require the sequential relationship between the two time histories. G( f ) = H ( f ) ⋅ W ( f ) 2 (12) 3. Figure 63 shows a simple example of a flagpole with two flags flying at different heights.

In reality each node or element has 6 component stresses pertaining to 3 axial and 3 shear components. nCode n 1st Page Back Main Menu . if we assume the stresses are uniaxial then all component responses will be in-phase. respectively. In general.and cross-spectral densities) has been included in a random vibration finite element analysis. Gzz(f). The figure below shows the fatigue log damage plot arising on the bracket due to a vertical vibration at the bolt hole. We must now consider how to obtain the response PSD for the maximum principal stress. An alternative method (not currently implemented in FE-Fatigue) for evaluating principal stresses is also described by Halfpenny [9]. This method does not require a uniaxial stress state but requires a rather more complicated analysis procedure. For a general loading with n simultaneous forcing functions the PSD of stress can be obtained from Equation (14). 5 Theory Page 81 1st Page Back Main Menu The single-sided PSD function of component stress at the base of the flagpole. G ( f ) = åå H a ( f ) ⋅ H b ( f ) ⋅ Wab ( f ) * a =1 b =1 n n (14) 3. respectively.9. éGσ 1 ( f )ù éG xx ( f ) G xy ( f ) Gxz ( f )ù ê ú ê ú êGσ 2 ( f )ú = eigen values êGxy ( f ) G yy ( f ) G yz ( f )ú êGσ ( f )ú êG xz ( f ) G yz ( f ) G zz ( f )ú ë û ë 3 û (15) This is the method that is currently implemented in FE-Fatigue.6. The FE results will be component stress PSDs (with units 'stress^2/Hz') and these are translated by the FE2FES module into a FES file of principal stress PSDs. The critical fatigue life is determined as 206 hours. It requires that all the loading (both auto. The absolute maximum principal stress is used for fatigue calculations. is therefore determined by Equation (13). we cannot calculate the principal stresses using component stress PSDs as they do not have information on the relative phase between each component. The asterisk indicates the complex conjugate. W11(f) and W22(f) are the PSD functions of wind speed at flags x and y. G ( f ) = åå H a ( f ) ⋅ H b ( f ) ⋅ Wab ( f ) * a =1 b =1 2 2 (13) H1(f) and H2(f) are the transfer functions relating stress at the base of the flagpole to wind load incident on flags x and y. and W12(f) and W21(f) are the cross-power spectral density functions. However. The response PSDs for the three principal stresses can therefore be obtained using an Eigenvalue routine as shown in equation (15). This is valid where there is a single loadcase or a uniaxial stress condition.4 Finding the response PSD for the maximum principal stress To this point we have considered calculating the response PSD for a component of stress. This was discussed in Equation (11).FE-Fatigue Rel.

Dover. Bendat JS." Ph. Bishop NWM. (1999). 7. and Sherratt F (1989). 4." Int. "The use of frequency domain parameters to predict structural fatigue. thesis.9.7 Conclusions The concept of frequency domain fatigue analysis has been presented." British Wind Energy Conference. "A frequency domain approach for fatigue life estimation from Finite Element Analysis." nCode International Ltd. 9.FE-Fatigue Rel. (1993).” International Conference on Damage Assessment of Structures (DAMAS 99) Dublin. and Socie DF. J Fatigue. 6. New York.8 References 1. UK. (1982). 2. Rice SO. and Bishop NWM. Longman Inc. (1984). Bishop NWM. York. (1997). Hu Z. "Mathematical analysis of random noise. Downing SD. Halfpenny A. "Simple rainflow counting algorithms. "Fatigue life prediction from power spectral density data. 2. "Vibration Fatigue. 5 Theory Page 82 1st Page Back Main Menu Critical region Figure 48 Fatigue contor plot of Log Damage 3. "Probability functions for random responses. New York. UK. 8.. Wang R. "Methods for rapid evaluation of fatigue damage on the Howden HWP330 wind turbine. Quarton D.9." Selected papers on noise and stochastic processes. 5. S9 3LQ. nCode n 1st Page Back Main Menu . where the random loading and response are categorised using Power spectral density (PSD) functions. "An introduction to random vibrations and spectral analysis” (2nd edition). 3. 31-40." Environmental Engineering. Newland DE. Bishop NWM (1988). (1964). University of Warwick. All the current methods are briefly reviewed and conclusions are drawn showing that the Dirlik method is recommended for general use." NASA report on contract NAS-5-4590. The Dirlik formulation is presented along with an explanation of the FE analysis involved. January 1982. Halfpenny A.D. 230 Woodbourn Road. 3. (1954). Sheffield.

2 to 4.5 2 3 4 -1 -1. nCode n 1st Page Back Main Menu .5 0 1 -0. the load must be cyclic. Constant amplitude loading assumes that the same cyclic loading sequence will occur throughout the lifetime of the component.5 1 Load scale factor 0. There are two ways in which FE data can be subjected to cyclic loading. • by running a transient or time-step analysis in the FE itself • by applying time histories of load or modal participation to static or modal FE stress analysis results Sections 4.5 Figure 49 This example shows a “peak to peak” or “fully reversed” history which will subject the FE model to a peak positive loading equivalent to the load applied in the FE model followed by a peak negative loading equivalent to -1 times the load applied in the FE model. the loading can be defined as either a constant amplitude or variable amplitude. 5 Theory Page 83 1st Page Back Main Menu 4. Variable amplitude loading allows a sequence of load values to be defined which will typically result in a number of differently sized cycles being experienced by the component. Time Histories 4. This defines 1 cycle and the damage for this cycle is then used to calculate the expected lifetime of the component in cycles. To induce fatigue failure at loads below this limit. section 4.5 are concerned with the use of external time histories.2 Constant amplitude versus variable amplitude If only a single static load case has been applied to the FE model.6 with the use of transient or time step analyses. 4.FE-Fatigue Rel.1 Why time histories are required Components under static loading at normal ambient temperatures will typically not fail unless that loading exceeds the static fracture limit of the material. A constant amplitude loading time history typically looks like this: Peak to peak Peak to peak 1.

the stress data in the FE model. as follows : Sc aled data Sc aled data 2000 1500 Load scale factor 1000 500 0 1 -500 2 3 4 Figure 51 In this case.8 0. This defines a stress cycle that can be used to calculate nCode n 1st Page Back Main Menu . which was the result of a unit load being applied. where the user can specify the maximum and minimum values to be applied.4 0. from zero to the maximum load applied in the FE model and back to zero. Finally FE-Fatigue recognises a third type of constant amplitude definition. as shown below: Zero to peak Zero to peak 1.2 0 1 2 3 Figure 50 This example shows only tensile loading. is multiplied by 1776 to give the stress as a result of the peak positive loading and then by -300 to give the result of the negative loading.FE-Fatigue Rel.6 0. This is particularly necessary when a unit loading has been applied in the FE model and therefore the time history is describing the maximum and minimum load. 5 Theory Page 84 1st Page Back Main Menu Another example of a constant amplitude history is the zero-peak example.2 1 Load scale factor 0.

Although time histories are typically measured with time as the X-axis. Section 4.4955 at 1743 Secs Mean = 0. 5 Theory Page 85 damage. variable amplitude time histories are converted into a series of loading cycles using peak-picking and rainflow cycle counting. The normal description is in cycles to failure.778 Figure 52 During the fatigue life calculation process. 1st Page Back Main Menu Note: There are no specified X-axis units.3 Obtaining and creating variable amplitude time histories FE-Fatigue can process nCode . Typically time history data comes in many other forms and so it is necessary to use the nSoft tools provided with FE-Fatigue to convert data from other formats. The following table shows which nSoft functions can be used to import. The format description for DAC files are detailed in the document called format. from pt 1 1 Load (No un its) Full file data: Max = 1 at 0 Secs Min = -0.5 describes how to pre-process the time histories using these techniques to speed up the calculation process. The following graph shows a typical variable amplitude load sequence from the SAE validation cases (Appendix C).FE-Fatigue Rel. or number of repeats of the history (in the constant amplitude case.mts.) The type of loading and the units of loading must correspond to the loading of the FE model. 4. = 0.3857 S. these are the same thing. The file contains 17081 points and has been normalised to unit maximum.DAC 17081 points. Warning: The software makes no checks with respect to the suitability of the units.6 0 Tim e (Secs) RMS = 0. create and view time nCode n 1st Page Back Main Menu .4517 1897. For details of these algorithms. refer to the Fatigue theory manual. DISPLAY OF SIGNAL: SAETRN.2352 -0.pdf and the MTS file format can be obtained from MTS via their website on www.D. this time is almost never used as a measure of life. 9 pts/Secs Displayed: 17081 points.DAC format.com. and MTS RPC format time histories.

Creates time histories from rainflow and Markov matrices Views single channel time history data Views multiple channel time history data Views and edits multiple channel DAC files and RPC files Many software packages for data acquisition and analysis create time histories directly into .5 for details on using pre-calculated rainflow data for single load case calculations. although this does not necessarily mean that their frequencies are in-phase. 4. nSoft utilities can be used to synchronise the time histories. with respect to each other. and to use this as an input to a single load case calculation in FE-Fatigue. Pre-calculated rainflow matrices cannot be used as input for multiple load case analysis as the all important phase information has been lost. for details of how linear superposition works.4 Multiple load cases When two or more independent load case results have been obtained from the FE analysis there are special conditions applied to the time histories that can be used in the analyser. Also. they must have the same number of points.FE-Fatigue Rel. start time and sampling rate.DAC or RPC III format. If this is not the case. Principles of Linear Superposition” on page 96. Program atd remdac coe waves regen qld mfd muledt Function Performed Converts data from a variety of ASCII formats Converts data from MTS RPC II and RPC III formats Used to type in values from the keyboard Creates sine. triangular and square waves. Consult the nSoft online manuals for details of the use of these functions. It is also possible to pre-calculate the rainflow matrix using the nSoft module cyc. See “6. Imposing these conditions on the time histories allows the principle of linear superposition to be applied to the loads and stresses at each point in the time history. or in-phase. Consult your software supplier for further information on third party . thereby creating a local stress or strain time history that combines the load cases together. The histories can then be said to be synchronised. See section 4.DAC support. swept sine. and that each point in each time history occurs at the same time as the same point number in the other histories. the analyser automatically runs a rainflow cycle count on the input time history and then calibrates the rainflow matrix for each node or element to be processed. 4. thereby preventing the time consuming cycle counting at each calculation point. 5 Theory Page 86 1st Page Back Main Menu history data.5 Using nSoft time history processing to increase calculation speed For single load case calculations. The results will be the same if the same limits nCode n 1st Page Back Main Menu . The analysis procedure requires that the time histories have the same number of points.

rainflow matrices cannot be used as they lose the synchronisation information between the channels. This presents a problem as shown in the data below: Time series 1 0 1 2 3 5 10 7 6 9 3 -5 -10 -5 0 3 0 Time series 2 0 -3 -5 3 7 9 6 1 -4 -1 3 7 -3 -7 -5 0 Figure 53 nCode n 1st Page Back Main Menu . each of which can be applied independently or they can be applied together. 5 Theory Page 87 1st Page Back Main Menu are used for the rainflow matrix. for the purposes of quick interactive analysis. • • The first is to reduce the number of nodes or elements to be calculated. This section focuses on the second method.FE-Fatigue Rel. However. multiple load case analysis is the most time consuming and it is therefore useful if some form of data reduction can be performed to speed up the calculations. Both methods involve some risk in that it is possible to create situations in which they will lose information. For multiple load case calculations. There are two ways to do this. and in the case of proportionally loaded components this means we can reasonably use the peaks and valleys thus extracted and discard the points in-between. both methods can be used with caution and engineering judgement. Alternatively if rainflow data is available from another source nSoft can be used to convert it into a rainflow file compatible with FE-Fatigue. Rainflow cycles are extracted from a time history using the turning points. where a peak-slicing technique is used to extract the turning points in each time history. and ultimately it is strongly advised that a complete superposition analysis be performed on the whole model if the correct results are to be obtained. The second is to remove data from the input time histories that is not contributing to the damage. Typically.

To compensate for this. the synchronisation has been lost and points which now correspond on the X-axis did not originally correspond before the peak-valley extraction. is not matched by a turning point in Series 1. the following reduced histories are produced. If only the peaks and valleys are extracted.FE-Fatigue Rel. 5 Theory Page 88 15 1st Page Back Main Menu 10 5 Series1 1 -5 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Series2 0 -10 -15 Figure 54 The first turning point in Series 2. which exists at point 3. peak-slicing keeps not only nCode n 1st Page Back Main Menu . 15 10 5 Series1 1 -5 2 3 4 5 6 7 Series2 0 -10 -15 Figure 55 Although there are the same number of peaks and valleys (and this is not generally the case).

PVX that can be used as input to the loading form of FE-Fatigue. This leads to the following reduced histories in our small example: 15 10 5 0 1 -5 -10 -15 2 3 4 5 6 7 8 9 Series 1 Series 2 Figure 56 By keeping 2 extra points in total the synchronisation is retained. it may be handled more efficiently by using a modal superposition approach. If the transient problem is one that can reasonably be solved using a modal formulation. nCode n 1st Page Back Main Menu .DAC files with extension . However extreme caution must be employed when using this option as data which contributes to damage may be lost.FE-Fatigue Rel. 4. In addition a peak picking or cycles gate can be applied to reduce the length of the data further. 5 Theory Page 89 1st Page Back Main Menu all the peaks and valleys but also the corresponding points in all other channels. .DAC format files. No external time history file is required and hence the FE-Fatigue loading form is not required. creating reduced . the time history data is embedded in the FE results and hence in the FES file. In full multiaxial loading situations the damage does not necessarily occur at the time when a turning point exists in any of the input loading channels.6 Transient or time-step analysis In a transient or time step analysis. When the loading is not proportional the risk in using this technique is increased significantly. This method is discussed in more detail in the worked example Transient Analysis and Modal Superposition using NASTRAN. The nSoft program pvxmul performs this task on .

It also has the ability to generate cyclic fatigue data from static properties. Data integrity and traceability issues dictate that the ability to change shared materials data should be restricted and it is strongly advised that in an organisation where a number of users have access to a shared database a materials administrator be appointed to care for the entry and editing of data in a shared database. 5 Theory Page 90 1st Page Back Main Menu 5. Consult the mdm on-line manual for details on setting up and using nCode materials databases. FE-Fatigue includes over 150 data sets for steel and aluminium alloys in its standard materials database. Access to the shared database is secured by a password known to your systems administrator. which is available from the nSoft menu system or from the group information form in FE-Fatigue (see ”5. Proprietary data can be merged into nCode’s standard database. 5.FE-Fatigue Rel. At present the only data format available for use in FE-Fatigue is a parametric form which describes the S-N relationship as two straight lines on a log-log scale. See ”5.5 Auto-generated data on page 92’). or a new database can be created which contains only company data or a mixture of company data and some of the data from the standard database. 5.2 S-N data S-N data can be characterised and stored in many forms. using the databases provided with the system. The mdm program allows the entry. Materials Data Good quality materials data is crucial to good fatigue life prediction. display and editing of materials data in either a shared or local database. If there is no appropriate data in the database nCode has a large library of data from tests conducted at its own laboratories which it can make available. or tests for a specific material can be made to determine the necessary properties. Materials Data on page 90’ re.1 Materials database nCode’s materials database is managed by an nSoft program called mdm (materials data manager). Stress SRI1 UTS Slope=b2 Slope=b1 NC1 Life Figure 57 The S-N relationship The required parameters are as follows : • SRI1 • b1 • NC1 • b2 Intercept of first slope at Life=1 cycle in range Slope of first line Intercept point of the two slopes Slope of second line nCode n 1st Page Back Main Menu .

and the other containing data measured for the nugget. To use the data.3 E-N data The strain life analysis method allows fatigue calculation on low cycle problems. All the multiaxial damage models use common uniaxial fatigue parameters. n and S respectively. where significant plasticity may occur. For multiaxial strain-life. The two parts of the equation come from regression analysis on the elastic and plastic life lines from strain controlled life tests. Spot weld data is stored in this format. Youngs Modulus and Yield Stress. If S-N data is available for the model to be analysed which is not in the form described above. However. Below 1000 cycles the S-N method is likely to be invalid. The standard errors for the three regression analyses may also be entered. In addition to these parameters an optional parameter SE can be entered which describes the standard error in log life. This is described by the Ramberg-Osgood equation: e = (s/E) + (s/K’)1/n’ where e is strain. s is stress.000 cycles. Yield Stress is used by the Fatemi-Socie model. two of the algorithms. all the regular strain life properties must be entered along with UTS.⋅ ( 2 Nf ) + ef ( 2 Nf ) E where e is strain and Nf is the number of cycles to failure.FE-Fatigue Rel. Both parameters can and should be obtained from results of pure uniaxial and pure torsion fatigue tests. and any stress range values greater than twice the UTS will cause a static fracture failure and to calculate a finite life for such values of stress would be incorrect. one for the data measured for the sheet.4 Multiaxial data In the materials data manager properties can be entered for either multiaxial strain life or multiaxiial safety factor or both. The life curve is described by the Coffin-Manson equation: Sf b ′ c e = -----. 5 Theory Page 91 1st Page Back Main Menu If b2=0. this means that at the value of stress indicated by NC1 there is in effect a fatigue limit. but it has been nCode n 1st Page Back Main Menu . where any cycle with a stress below this value will cause zero damage. If SRI1 is greater than the UTS of the material then the software adds an extra line between the UTS (multiplied by 2 to convert from amplitude to range) and the value on curve one at 1. Fatemi-Socie and Wang-Brown. The obtained parameters are the coefficients sf’ and ef’ and the exponents b and c. In addition to these parameters. Default values of 0. E is Young’s Modulus and n’ and K’ are parameters derived from regression analysis of test data. contact nCode support for advice on how to convert data to this form.6 and 1 can be used. require one additional parameter each. There are typically two separate entries in the database. the spot weld flag must be set to indicate that the data is for spot welds. which takes into account the variability in the physical tests. This is obtained as part of the regression analysis performed to derive the linear parameters from the life test results. This allows a confidence factor to be entered in the calculations. a material cut-off may be entered which describes the life beyond which the damage is assumed to be zero. 5. In addition to the damage curve the relationship between stress and strain is also required. ′ 5.

However it requires knowledge of the parameter RA (reduction in area) which is not always available.168 -0. and Seeger [2] have compiled an alternative approach based on the results of more nCode n 1st Page Back Main Menu .56 sf' /(ef') 0. When using Dang Van only. 5.76 ef0.53 -0.05.2 Modified Universal Slopes (Muralidharan) 0. E-N data The first method of approximating the strain life relationship from monotonic properties was proposed by Manson and later modified by Muralidharan[1].6 and the Wang-Brown parameter is set to 1 when data is generated.2 where : Rm = the ultimate tensile strength ef’= the true fracture strain calculated from ln (1 / (1 . Dang Van requires the tension fatigue limit (entered as Type A fatigue limit) and the hydrostatic stress sensitivity.12 0.2 0.6 sf ' /( ef ')0.RA)) and RA = the reduction in area.2 0.623 Rm 0. If this property is known. properties can be calculated using this method and entered into the mdm database.5 Auto-generated data Data for E-N analysis may be generated automatically in FE-Fatigue using the following rules: 5.5. 5 Theory Page 92 1st Page Back Main Menu observed that the Wang-Brown method can be quite sensitive to the value of S whereas FatemiSocie is less sensitive to the value of n. Baumel Jr. For multiaxial safety factor.3. The procedure is usually referred to as the method of universal slopes and can be applied to any metal. See the section 3. McDiarmid requires the type A and type B fatigue endurance limits. However this is not implemented directly in the FE-Fatigue analyser. set the two fatigue limits to the same value.0196 ef 0.9 Rm -0. set HSS to 0.823 E0. Multiaxial safety factor Generated properties are not available for multiaxial safety factor methods. 5.5.155 (Rm / E) -0.4.FE-Fatigue Rel.2.5. Dang Van Criterion for more information. Multiaxial E-N data The Fatemi-Socie parameter is set to 0.09 0.1.6 -0. three values are required. 5.5. When using McDiarmid only. Parameter (Manson) sf' b ef' c K' n' Universal Slopes 1.

000. fixing the stresses at 1000 cycles and the endurance limit according to the fractions of UTS detailed below: Cycles Ferrous Alloys.800 x UTS 0.5.000 Other Alloys.59 α -0.307 x UTS 1. 1 1. 1 cycle.000 100.095 0.11 The ductility factor α is calculated from: α = 1.000 500. The curves are constructed by fixing the stress axis intercept.000 1.15 Aluminium/Titanium alloys 1.000 1.000.5 Rm -0. 5.900 x UTS 0.000 Aluminium Alloys. 1 1.800 x UTS 0.087 0.61 Rm 0.35 -0. 5 Theory Page 93 1st Page Back Main Menu than 1500 fatigue tests. aluminium and titanium alloys. 1 1. UTS.000 Stress 1. 1 1.65 Rm 0.000.125 R m / E) for R m / E > 3 x 10 -3 where Rm is the UTS and E is Young’s modulus. at the value of the UTS.700 x UTS 0.67 Rm -0.375 .357 x UTS 1.000.58 1.258 x UTS 1. Since only Rm and E are required this is a simpler option and has been included directly in the analyser. Parameter sf' b ef’ c K' n' Steels 1.FE-Fatigue Rel. Currently the approach is limited to plain carbon and low to medium alloy steels.000 x UTS 0. S-N data Approximate S-N damage curves can be generated purely on the basis of ultimate tensile strength.69 1.0 for values of R m / E < 3 x 10 -3 α = (1.000 x UTS 0.000 x UTS 0.4.000 x UTS 0.000 Titanium Alloys.274 x UTS S-N data can be derived directly from the elastic portion of the strain-life curve by converting strain nCode n 1st Page Back Main Menu .

The nCode Fatigue Theory Book gives a full description of these methods and their effects.5. 5 Theory Page 94 1st Page Back Main Menu to stress and reversals to cycles. It is useful for approximate calculations and for assessing the likely increase or reduction in life as a result of changing to a stronger or a weaker material. Csur . particularly for high cycle environments. machining marks influence fatigue strength by providing additional stress raisers that aid the process of crack nucleation.FE-Fatigue Rel. The program then uses a series of look-up tables to calculate the surface effect factor which is then applied as a fatigue strength reduction factor (see 5. A very high proportion of all fatigue failures nucleate at the surface of components and so surface conditions become an extremely important factor influencing fatigue strength. for this reason the surface correction factor. Generated data is not suitable for spot weld analysis.5. Similarly. machined or forged.7) The actual calculated values for the surface factor can be viewed within the materials manager.6. nCode n 1st Page Back Main Menu . can be extended significantly by treating the surface using methods such as nitriding and shot-peening. In FE-Fatigue the surface finish and treatment are simply selected from a list. The surface finish correction factor is often presented on diagrams that categorize finish by means of qualitative terms such as polished. See the section on multiaxiality safety factor theory for suggestions on the estimation of the required properties for those algorithms. 5. The usual standard by which various surface conditions are judged is against the polished laboratory specimen. This data is only valid above the transition life (1000 to 10000 cycles) as it does not take any plasticity into account. Broadly speaking. fatigue lives. 5. mdm. high strength steels are more adversely affected by a rough surface finish than softer steels. Surface finish correction Note: This option is not available for multiaxial safety factor or spot weld. scratches. is strongly related to tensile strength. Validity of generated data The use of generated properties is only encouraged when no good quality measured data is available. Normally. pits.

Each group of nodes and elements within FE-Fatigue can have a fatigue strength reduction factor associated with it. This value affects the fatigue limit as shown in the following diagram : S tres s S1 S2 L ife Figure 58 If S1 is the original fatigue limit and k is the fatigue strength reduction factor then S2=S1*k and the slope is recalculated accordingly. 5 Theory Page 95 1st Page Back Main Menu 5. The default is 1. Fatigue strength reduction factor Note: This option is not available for multiaxial E-N or multiaxial safety factor. In the case of surface effect the value k is less than 1 for rougher surfaces ( as shown in the diagram) and greater then 1 for smoother surfaces.7. nCode n 1st Page Back Main Menu . 55-58. MANSON (1988) A Modified Universal Slopes Equation for Estimation of Fatigue Characteristics of Metals. Journal of Engineering Materials Technology.. T. or treated surfaces. 2] BAUMEL JR. pp.FE-Fatigue Rel. A and SEEGER. Vol. Materials Data for Cyclic Loading. loading or notch effects. Materials Science Monographs. Supplement 1. 1] MURALIDHARAN. The fatigue strength reduction factor can be used to address other factors such as size. 61 published by ELSEVIER ISBN: 0444 88603 6. 110.

Stresses must only be averaged at a node if they have the same co-ordinate system. Principles of Linear Superposition In FE-Fatigue. and the time histories corresponding to the load cases were as follows: Table 2: Load case 1 0 10 20 -5 25 -10 -15 -5 0 Load case 2 10 -5 10 -5 10 -5 10 -5 10 nCode n 1st Page Back Main Menu . So at each point in the time history we can calculate the total stress by dividing all stresses by the value of the input load to produce a unit load case stress. 5 Theory Page 96 1st Page Back Main Menu 6. Therefore. It is not safe to add together principal stresses as the direction of the principals is often different.FE-Fatigue Rel. multiplying by the value of the point in the time history and adding the values from each load case together. multiple input out-of-phase loading is dealt with by assuming that the stresses from the independent load input calculations can be added together to produce a net total stress. Component stresses in the same direction are typically used. 2 – The stresses are linear elastic. if we have two load cases with the following stress values: Table 1: LC 1 LC 2 X 20 -30 Y -50 50 Z 10 0 XY -25 -10 YZ -28 0 XZ 15 0 Where load case LC1 had 2 units of load applied and LC2 had 5 units of load applied. There are some important factors which allow this assumption to be made: 1 – The stresses must be in the same direction. Elastic-plastic stresses should not be superimposed in this way. Note: This also affects nodal averaging. 3 – That each point in the input load-time histories occur at the same time as each other and are independent of the points before and after.

This method is discussed in more detail in the worked example Transient Analysis and Modal Superposition using NASTRAN.0) + (-30.0*0. so long as the problem can reasonably be assumed linear.5 -20 YZ 0 -140 -280 70 -350 140 210 70 0 XZ 0 75 150 -37. The principle of linear superposition works equally well for quasi-static cases (where the instantaneous structural response is simulated by the linear superposition of scaled static loading cases) and transient-dynamic cases.5 -75 -112.0/2.0/5.0) + (50.0) = 100. or more commonly be combined together to produce a maximum principal stress or Von Mises stress value.5 -332.5 -37.0) = -60. 5 Theory Page 97 1st Page Back Main Menu Then the combined stress time history for the X stress at point 1 is calculated as follows : Total X= (X1*(LC1 value)/(LC1 applied load)) + (X2*(LC2 value)/(LC2 applied load)) = (20.5 0 The 6 stresses can then be used directly.0/2.5 187.0*10.FE-Fatigue Rel.0 Total Y= (Y1*(LC1 value)/(LC1 applied load)) + (Y2*(LC2 value)/(LC2 applied load)) = (-50. nCode n 1st Page Back Main Menu . the instantaneous structural response may be simulated by the linear superposition of scaled mode shapes.0*0. giving the following results: Table 3: X -60.0 and so on.0*10. where.5 135 167 72.0 130 140 -20 190 -70 -210 -20 -60 Y 100 -300 -400 75 -525 200 475 75 100 Z 0 50 100 -25 125 -50 -75 -25 0 XY -20 -115 -270 72.0/5.

800. Each cycle can be described as follows: • Maximum strain value • Minimum strain value • Direction of loading These values can then be classified into a variety of histograms. For Example: Consider the following time series: 1000.FE-Fatigue Rel.1000 It is displayed below: nCode n 1st Page Back Main Menu . Calculation of fatigue life from rainflow matrices This appendix shows how fatigue life is calculated from a rainflow matrix.1. The only difference is in the interpretation of the loading environment.1000. In this document we assume that all the reversals are converted to closed cycles by adding an extra data point to the sequence which corresponds to the absolute largest value of the sequence.200. 5 Theory Page 98 1st Page Back Main Menu 7.800.800.500. Typically these forms are: • Maximum-Minimum-Counts (X-Y-Z) • Range-Mean-Counts (X-Y-Z) • Range-Counts (X-Y) • From-To-Counts (X-Y-Z) where Range = (max-min) Mean = (max+min)/2 and From-To uses the maximum and minimum values but positions the count in a different part of the histogram depending on whether the maximum or minimum occurs first.500. Therefore the result of the rainflow count will consist solely of closed cycles. Types of rainflow matrix The result of a rainflow cycle count of a strain history can be described as a list of closed cycles and unclosed reversals (See the technical background of rainflow cycle counting for details).0. Strain-life Fatigue Analysis from a Rainflow Matrix Technical Overview The theory of strain-life analysis from a rainflow matrix is based on the theory used for a time history. 7.500. the form of which is dependent on user preference.

FE-Fatigue Rel. 5 Theory Page 99 1st Page Back Main Menu 1200 1000 800 600 400 200 0 1 2 3 4 5 6 7 8 9 10 11 Series1 Figure 59 The rainflow procedure yields the following results : Table 4: Cycle 1 2 3 4 5 maximum 800 1000 800 800 1000 Minimum 500 200 500 500 0 direction down down up down down Figure 60 After a time history fatigue analysis. a typical hysteresis plot would look like this: Cross Plot of Data : CYC01 Material Factor Life : Generated : 1 : 1.83E5 Repeats UTS Kf : 100 : 1 60 40 S tress(M Pa) 20 0 -20 -40 0 200 400 Strain(uE) 600 800 1000 Hysteresis loops for file CYC01 Figure 61 nCode n 1st Page Back Main Menu .

this is a conservative approach as larger mean values give more damage. The example above can be classified into an 8x8 matrix as follows : Table 5: Range 0:150 0:150 150:300 300:450 Mean 450:600 600:750 750:900 900:1050 1050:1200 0 0 0 0 0 0 0 0 150: 300 0 0 0 0 0 0 0 0 300: 450 0 0 0 0 3 0 0 0 450: 600 0 0 0 0 0 0 0 0 600: 750 0 0 0 0 0 0 0 0 750: 900 0 0 0 0 1 0 0 0 900: 1050 0 0 0 1 0 0 0 0 1050: 1200 0 0 0 0 0 0 0 0 Figure 62 Note that the choice of bin size and limits and the policy of where to place a count when the value is between bins may have a significant effect on the life calculation. 2 .FE-Fatigue Rel. calculate the range and mean and insert the counts into a 2D histogram. It is also important to ensure that the outer limits of the histogram are sufficient to contain all potential cycles as information on cycles which fall outside the matrix is lost. 1 . The range is 800 which falls neatly into the bin -750:900 but the mean is 600 which lies on the border of two bins. Similarly the 3 cycles which have a range of 300 are put into the bin with the higher range. Consider the cycle whose maximum is 1000 and minimum is 200. It is possible to ascertain whether the cycles are “hanging” from the top limb of a loop or “standing” on the bottom limb by the direction of loading ( “down” or “up” ) but once the data is in a histogram the sequence is lost and it is not possible to identify from which loop the cycle is hanging from or standing on. In either case the approach of using a matrix leads to an approximation of the answer derived from using the time history directly.Range-Mean To classify the data into a range mean matrix.Range only This option ignores the mean entirely. This is only important if a mean stress correction is used and the method of estimating the mean stress in this case is described in the calculation procedure below. 5 Theory Page 100 1st Page Back Main Menu The three cycles which have the same strain maximum and minimum each have a different value of mean stress. In our simple case the histogram becomes single dimensional as follows : nCode n 1st Page Back Main Menu . In this case we have put it into the bin with the larger limits .

5 Theory Page 101 1st Page Back Main Menu Table 6: Range 0:150 0 150:300 300:450 450:600 600:750 750:900 900:1050 1050:1200 0 3 0 0 1 1 0 If the mean is to be ignored this approach makes sense as the data storage saved can be used to refine the resolution of the bins in the range axis. In the example above. the matrix becomes : Table 8: 0:150 0 0 0 0 0 0 0 1 150:300 300:450 450:600 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 1 0 0 To 600:750 750:900 900:1050 1050:1200 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0:150 150:300 300:450 From 450:600 600:750 750:900 900:1050 1050:120 0 nCode n 1st Page Back Main Menu . 3 . 4 . e. The lower diagonal must therefore contain zeroes. However it is normally recommended to save the mean information as this may be required later and once lost cannot be retrieved.Max-Min This option bins the maximum and minimum values directly.From-To This matrix utilises the lower area to include directional information. This area can be utilised in the from-to matrix. Table 7: Max 0:150 150:300 300:450 450:600 600:750 750:900 900:1050 1050:1200 0:150 150:300 300:450 Min 450:600 600:750 750:900 900:1050 1050:1200 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3 0 0 0 0 0 0 0 0 0 0 0 0 1 1 0 0 0 0 0 0 The values in this matrix must by definition populate only a triangular region above the diagonal where max>=min.FE-Fatigue Rel.g.

5 Theory Page 102 1st Page Back Main Menu Cycles with a “down” direction (“hanging”) populate the lower diagonal and the “up” (“standing”) cycles populate the upper diagonal. together with their average. The second problem can be resolved in one of two ways. can be reported to give a results envelope. e.g. e. The worst case occurs when the range and mean ( or max/min ) are at their largest corner. The outside loop. The range-mean and max-min matrices do not contain any directional information and so we do not know whether the loops hang or stand. in the case above the range would be 975 and the mean 525. Both methods are approximations and neither is “right”. This corresponds to the time histories identified below : 1200 1000 800 Actual 600 400 200 0 1 2 3 4 5 6 7 8 9 10 11 Worst Best nCode n 1st Page Back Main Menu . The actual value of the maximum and minimum is lost and so must be approximated.FE-Fatigue Rel. In this case the maximum has the absolute largest and so the entry is in the lower diagonal. The first problem is resolved according to whether the matrix contains the directional information. the best case a range of 900 and a mean of 450. Two values can be calculated which correspond to the worst and best cases and these. Calculation Procedure When calculating strain-life fatigue lives from a pre-counted rainflow matrix there are two problems : In a range-mean or single-sided max-min rainflow matrix the sequence of cycles is completely lost and it is therefore not possible to correctly define the hysteresis loops.g. for the outside loop in the range-mean matrix above the worst case would have a range of 1050 and a mean of 600. The second way would be simply to take the middle value from the bin. which goes from the maximum in the series to the minimum is placed according to which has the absolute largest value. where the worst case involves hanging all cycles from the top of the outside loop ( maximising the mean stress) and the best case involves standing all cycles on the bottom of the outside loop. In a low resolution matrix like this the difference is large but in a 128x128 matrix this difference would be much smaller and the impact on the life much less. This is not important unless mean stress is a consideration. Therefore we adopt a best-worst approach. the best case when the values are selected from the smallest corner.

nCode n 1st Page Back Main Menu . The loops for the best and worst cases look like this: : Cross Plot of Data : BEST Material Factor Life : Generated : 1 : 8.08E4 Repeats UTS Kf : 100 : 1 60 40 S tre s(M a s P ) 20 0 -20 -40 0 200 400 Strain(uE) 600 800 1000 Hysteresis loops for file BEST Figure 63 The best case Cross Plot of Data : W ORST Material Factor Life : Generated : 1 : 7. in that we know there are two hanging and one standing cycles rather than three whose status is unknown.FE-Fatigue Rel. When we have the From-To matrix we have slightly more information.72E4 Repeats UTS Kf : 100 : 1 60 40 Stress(M a) P 20 0 -20 -40 0 200 400 Strain(uE) 600 800 1000 Hysteresis loops for file WORST Figure 64 The worst case The average damage. can then be calculated. The sequence effect leads to different mean stress values for the smallest cycles. 5 Theory Page 103 1st Page Back Main Menu These time histories have exactly the same values but in a different order. and hence the average life.

5 Theory Page 104 1st Page Cross Plot of Data : FROMTO Back Main Menu Material Factor Life : Generated : 1 UTS Kf : 100 : 1 : 7. i. In conclusion.84E4 Repeats 60 40 Stress(M a) P 20 0 -20 -40 0 200 400 Strain(uE) 600 800 1000 Hysteresis loops for file FROMTO Figure 65 Average damage However.2. this is still not totally accurate as one of the small cycles in reality hangs from an inner loop. Stress-life Fatigue Analysis from a Rainflow Matrix Types of rainflow matrix The basic types of matrix are the same as for the strain-life approach. nCode n 1st Page Back Main Menu . However the From-To option is not typically required as non-linear behaviour is not modelled in the S-N approach. but by selecting a high resolution matrix and storing directional information a reasonable answer can be obtained. particularly if the cycles are well distributed and a good sample size is captured. the only approximation in this case is related to using the bin corners and the same methods can be used to estimate the cycle range and mean.e. or use the mid-point of the bin. the use of histograms to store rainflow data is not ideal. 7. take the best and worst cases from the bin corners. Calculation Procedure Because the sequence effect is not considered.FE-Fatigue Rel.

Auto-elimination uses the magnitude of the stress due to each loadcase and the magnitude of the load to calculate an approximate worst case stress for each node or element. i n σvm. This method exploits the fact that for most structures the high stresses are concentrated in local areas and large proportions of the model experience little or no stress. For the stress-based option. For example.FE-Fatigue Rel. if a value of 10% is entered and the worst stress combination is 500 MPa.i)] where.i is the loadcase number is the total number of loadcases is the Von Mises stress for loadcase i (Lmax.Lmin.Lmin. Calculation method All auto-elimination methods use the combination of the Von Mises stress for each load case and the range of each time history to produce a worst case superposition value for each element. % to retain by group – To avoid the problem of only retaining elements from a single “hot-spot” on the model.0) is the 'divide by' factor for loadcase i (default 1. % of model to retain – Selects only the highest n percent of nodes or elements for analysis. Auto elimination A key influence on the length of time FE-Fatigue takes to process a model is the number of nodes or elements. (Lmax.i . this total value can be expressed as: σtotal = i=1 to nΣ[(αi/βi) . Auto elimination methods The following auto elimination methods are available: % of maximum stress – Eliminates the nodes or elements with a calculated stress less than n percent of the model's maximum. This value of stress is then used to filter out the lower stressed nodes/elements.0) This total value is used to form an approximate ranking of nodes or elements likely to be damaged.i . then all nCode n 1st Page Back Main Menu . the program also requires a percentage value to be entered which then removes all elements whose stress combination is less than that percentage of the maximum. auto-elimination on a group basis retains the top n percent nodes/elements from each material group defined in the FES file.i . σvm.i) is the range or largest cycle of the loading time history associated with loadcase i (maximum value subtract the minimum value) αi βi is the scale factor for loadcase i (default 1. Mathematically. 5 Theory Page 105 1st Page Back Main Menu 8. So care must be taken if contour plotting these results. FE-Fatigue includes an auto elimination option to enable the user to preselect nodes or elements on which to focus the analysis. Nodes that are eliminated will have no results.

If one element of a 3 sheet weld is selected for analysis.min for the maximum and minimum values of each associated time history. The ranking parameter for each spotweld is the larger of the values of ∆σtotal for sheet 1 and sheet 2 (the nugget is neglected for ranking purposes). the other will also be included.fes file scaled by +1 and –1. a ranking stress parameter is required for each spot weld. 5 Theory Page 106 1st Page Back Main Menu elements whose combination was less than 50 MPa would be omitted from the calculation. Spot weld auto elimination To perform auto elimination.000 element model. nCode n 1st Page Back Main Menu . θ max.000 elements with the largest “total Von Mises” stress will be used in the calculation. If the FATFE preferences are used to select ranking without consideration of the loading history.i is calculated for each loadcase using the forces and moments from the . the program also requires a percentage to be entered which is the % of nodes or elements to be analyzed.σmax(My1))) These angles are then substituted back into the structural stress equations and the ‘effective’ stress is calculated at θ max.i = i=1 to n Σ ∆σ max.3) can be differentiated to find the angles of maximum and minimum stress for each load case. if 10% is entered then the 1. The auto-elimination method is very useful but should be used with caution. For example for a 10.6.min is ∆σmax. the ranking value of the middle sheet is neglected. and where other factors such as mean stresses may increase damage. the extreme values of stress used to calculate ∆σmax. This ranking stress can be expressed as: ∆σtotal where: i n ∆σmax. When 3 sheet spot welds are considered.min (180 degrees apart) are calculated from: θ = arctan ((σmax (Fy1) + σmax(Mx1) )/( σmax(Fx1) . especially where a large proportion of the model is being discarded. These angles.FE-Fatigue Rel. The bigger of the two ranges calculated at θ max.i is the loadcase number is the total number of loadcases is the maximum range of stress associated with each loadcase and loading history ∆σmax. For the point-based options.fes file. together with the maximum and minimum values of the associated time history and any scaling factors.i will be based on the forces and moments in the .i. The spot weld structural stress equations (see section 3.

Element or node strains that are calculated in the FE solver are exported such that the strain output will emulate a strain gauge rosette at the same location on the actual part. the following equation should be used: E(T) = (Ex * cos2(T)) + (Ey * sin2(T)) + (Exy * sin(T) * cos(T)) Where Ex. This provides an improved facility for road to CAE correlation and calibration to occur and improves the confidence of FE models by a more direct comparison of strain results.3. At the moment the existing eigenvector method must be used. so stress to strain conversion must be used before and after as necessary. 3D to 2D conversion If a file contains 3D stresses or strains. where the elastic outputs are created first and then converted to elastic-plastic using the Mroz-Garud cyclic plasticity model described in section 3. 5 Theory Page 107 1st Page Back Main Menu 9. Ey and Exy are the X. The eigenvector method transforms stresses. The following types of gauge are supported: • Single • Rectangular (45 degrees) • Delta (60 degrees) • Tee (90 degrees) It is recommended that virtual strain gauge data is exported from shell elements rather than nodes.8. If an error is generated in the course of calculating the eigenvectors. these must be converted to surface resolved strains. Shell element data will typically be in local coordinate system which will help the user to compare orientation with actual strain gauge positioning. no output can be produced. nCode n 1st Page Back Main Menu . Elastic-plastic conversion This is a two-pass operation. Y and shear strains. although this is not 100% reliable. Note that the X-axis in the plane of the surface is effectively arbitrary after conversion from 3D data. Virtual strain gauges This feature allows the user to place a virtual strain gauge on the part being analysed.FE-Fatigue Rel. Angle calculation To calculate the strain at an angle T to the X axis.

- Students Reporting for Admission 1-4-2015 (1)
- New Doc.1
- Paramjeet Resume (2)
- Cover Letter
- Cover Letter
- syllabus
- Defence Day Flag Hosting
- InitiateSingleEntryPaymentSummary25-05-2016
- InitiateSingleEntryPaymentSummary18-04-2016
- Project Synopsis Solar Car New
- QP-FET-Dec-2014.74
- EX1-Model
- QP-FET-Dec-2014.75
- QP-FET-Dec-2014.74
- QP-FET-May-2015
- Gas Analyzer
- Mechanical Engineering Objective question paper
- Industrial Promotion Officer JKSSB
- Payment(1)
- 20091202
- 42V
- Bestpptforseminaronhybridelectricvehiclebyrahul 141017074707 Conversion Gate02
- Reference Edited
- SiSo-Aufg01-v2
- Transfer Certificate

- Fatigue Theory
- Fatigue Mechanics
- MSC.Fatigue User's Guide
- Akrache_04
- Fatigue Analysis
- Fatigue Thesis
- 3150_MSC.FatigueTut1_v1.3
- 12.1-ANSYS-ncode
- Fatigue Analysis of Welded Components Designers
- User
- MSC.Fatigue Quick Start Guide
- 1-6
- EN 1993-1-9 2005
- Aircraft Design Day6
- Fatigue Analysis - Automotive Components
- Aircraft Design Day2
- Fatigue Analysis Using Ansys Fatigue Module and Ansys Ncode
- Aircraft Design Day5
- Aircraft Design Day8
- IIW-1823-07-FatigueRecomm2
- Recommendations for Fatigue Design of Welded Joints and Components
- Calcul Fatigue Resistance Materiaux
- Fundamentals Metal Fatigue
- ANSYS Fatigue
- Rainflow Counting Methods
- Atlas of Fatigue Curves
- Altair Hyper Works Hypermesh 8.0 Tutorial Meshing
- MSC Nastran 2010 Release Guide
- FKM Analytical Strength Assessment 5th Edition
- Crack propagation in Ansys

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd