Durability Analysis from FEA Models Incorporating High Temperature and Creep Effects

John Draper, Gary Malton Safe Technology Limited

Safe Technology Limited 1 The South West Centre Archer Road Sheffield S8 0JR UK Tel: +44 (0)114 255 5919 Fax: +44 (0)114 255 5910 Email: enquiries@safetechnology.com

2002 ABAQUS Users' Conference

1.0 INTRODUCTION This paper discusses a methodology for including the effects of temperature in the fatigue analysis of engine components. The methods are incorporated in the fe-safe software, for fatigue analysis from finite element models.

2.0 FATIGUE ANALYSIS FROM FEA MODELS. Fatigue analysis from finite element models is becoming an accepted design method in industry. Two basic techniques have been adopted. 2.1 Fatigue analysis from elastic FEA results. (i) In this method, a 'reference load' is analysed in the FEA, and the stress results are written to a results file, together with the model geometry. A linear elastic FEA is used. The fatigue software takes the stress tensor at a node and scales the stresses by a time history of the applied load, producing a time history of the elastically-calculated stress tensor. This stress history is converted into elastic-plastic stresses and strains using a cyclic plasticity model, and the fatigue life is calculated. The process is repeated for each node in the model, and the results are plotted as fatigue life contours. (ii) The method can be extended to analyse components which are subjected to many applied loads, applied at different points on the model. In this case, each applied load is analysed separately in the FEA, and the results written to a stress output file as separate stress solutions. The fatigue software scales each stress solution by the appropriate loading history, and superimposes the results. This 'scale and combine' procedure produces a time history of the nodal stresses for all the loading histories applied together. The fatigue life can then be calculated as before, using cyclic plasticity modelling and an appropriate fatigue algorithm. (iii) For engine components, it is often useful to model a sequence of events in the FEA. For example, a crank shaft may be analysed by calculating the stresses at each 5o of rotation of the crank shaft, through two or three complete revolutions. This need not be a linear FEA solution. The sequence of stresses (the 'data set sequence') may then be analysed in the fatigue software. If a linear elastic FEA is used, other types of loading may be superimposed, perhaps using the 'reference loads' methods described above. This method of analysis can also be used to analyse transient events. (iv) Other loading descriptions can also be used - for example loads described in the frequency domain.

2.2 Fatigue analysis of elastic-plastic FEA results. The methods described in 2.1 use cyclic plasticity corrections developed for yielding which is localised to notches. The methods have been shown to work to an acceptable level of accuracy for more general yielding, for example to smooth sections loaded in bending, or smooth uniformly stressed specimens. The methods do not take into account the stress redistribution in a more complex component which results from gross yielding. If this is important then an elastic-plastic FEA is required. Clearly no scaling of the stress results is then advisable in the fatigue analysis software, and the sequence of stresses and associated strains are analysed with no further plasticity correction required. Care must be taken with elastic-plastic FEA in order to ensure that an appropriate hardening model is used, with cyclic (not monotonic) materials properties.

2002 ABAQUS Users' Conference

3.0

THEORETICAL BACKGROUND

3.1 Fatigue analysis for uniaxial stresses. The theories for fatigue analysis based on local surface strains have been developed from the 1950s. A major feature of local strain fatigue is that fatigue lives to crack initiation can be calculated. This is of increasing importance as more emphasis is being placed on product safety and product liability legislation, and as a result the traditional approach to total life fatigue based on stress-life curves is becoming much less relevant. Local strain fatigue techniques were first applied to the analysis of simple strain gauges in the late 1970s [1]. Simple cyclic plasticity models were developed to estimate elastic-plastic stresses from the measured strains, so that the effects of mean stress could be included. The original theories were developed for uniaxial stress conditions. However, during the 1980s it became increasingly apparent that biaxial stresses were produced at critical locations in real engineering components, and that fatigue life estimates based on an assumption of uniaxial stress would produce significant errors. Devlukia and Davies [2] and Tipton and Fash [3] were among the many authors who documented such errors.

3.2 Fatigue analysis for biaxial stresses. McDiarmid [4] proposed that for high cycle fatigue successful life estimates for biaxial stress conditions could be made using combinations of axial and shear stresses. Brown and Miller [5] extended this work into low cycle fatigue by using combinations of axial and shear strains. These methods have been very successful in practical fatigue analysis. Mroz [6] proposed a multisurface kinematic hardening model for variable amplitude cyclic loading which allowed stresses to be calculated from the measured strains. Lamba [7] later proposed a two surface model and this has been the basis for several general cyclic plasticity models. Bannantine and Socie [8] synthesised these methods into a unified multiaxial fatigue life estimation system for strain gauge rosettes, using critical plane procedures. For many ductile metals, fatigue cracks initiate on planes which experience the maximum shear strain amplitude. The normal strain and the associated stresses are modifiers. It has been shown that for cracks which originate from the surface, three basic planes are required (Figure 1), one perpendicular to the surface and two at 45o to the surface. For proportional stresses one of these planes will be the critical plane. For non-proportional stresses, where the principal stresses change direction as well as change their magnitude, each of these planes is rotated through 180o in small steps, typically 5o or 10o . The time history of shear and normal strain and the associated stresses are calculated for each plane and for each step through 180o.

Figure 1. The three planes used in critical plane analysis. The surface of the component is shown shaded.

The fatigue damage on the most damaged plane is determined, using an appropriate fatigue damage model. Applying the Brown-Miller combined shear and normal strain parameter in critical plane analysis has proved very successful. Various mean stress corrections have been proposed for the Brown-Miller parameter, one of the most recent being that of Chu, Conle and Hubner [9]. 2002 ABAQUS Users' Conference 3

For very high strength steels and some cast irons, fatigue lives are calculated using axial strains, again with a critical plane procedure for non-proportional stresses. 3.3 Fatigue analysis from finite element models In 1991 Devlukia, Draper and McDiarmid [10] put forward a proposal to apply these techniques to the analysis of finite element models. A 650,000 ($1million) research and development program was submitted to the UK Government, who offered substantial funding for an integrated program of research, software development and validation testing. The validation testing was carried out at the Rover Gaydon Technical Centre in England. The project was completed in 1996, and the software was subsequently developed into a commercial suite of software fe-safe by Safe Technology Limited [11]. The ability to analyse high temperature fatigue, creep fatigue, cast iron components and probability of failure were added subsequently. Current research is adding the capability to include the effects of press-forming sheet components [12]. Considerations of cost and processing speed mean that many designers wish to use an elastic finite element model as a basis for fatigue design. Glinka [13], Barkey [14] and others have proposed methods of calculating cyclic elastic-plastic stress-strains from the elastic FEA stresses. 3.4 Effects of stress gradient and notch sensitivity. There is much experimental evidence from fatigue testing carried out in the middle of the last century showing that stress gradients have an important effect on the total fatigue life of a component. This is particularly evident on smooth test specimens, but the effect seems to be less significant at geometric notches in real components. Local strain fatigue analysis from strains measured in notches has shown good correlation even though the effects of stress gradient are ignored. Stress gradients have also been used in an attempt to explain the effect of notch sensitivity. Traditional methods used the elastic stress concentration factor for a notch, combined with an algorithm to allow for the effects of stress gradient [15,16,17]. In general these methods have had mixed success. Finite element analysis provides surface strains on the model, but for real engineering components it is very difficult to determine the stress concentration factor at a notch. Taylor [18] has proposed a very successful method of estimating the effects of notch sensitivity on the fatigue limit stress, using the stress gradient from the finite element model and the threshold crack tip stress intensity factor. Taylors method is based on the observation that for a notch of specified depth, the fatigue limit stress reduces as the notch gets sharper, i.e. as the stress concentration factor increases, and that for many ductile metals a minimum value of fatigue limit stress occurs. Further increasing the stress concentration factor by sharpening the notch produces no further reduction in fatigue strength (Figure 2a). FEA surface stresses or strains calculated in the notch can therefore be used for blunt notches, but will give conservative life estimates for sharper notches. Taylor used the life to a 2mm crack as the fatigue life. Frost and Dugdale [19] and Frost [20], using test data on flat plate and round bar specimens in aluminium alloy and steel materials, showed that if fatigue life to first crack initiation is considered, then the fatigue strength reduces with increasing stress concentration with no limiting value (Figure 2b). It seems therefore that crack initiation may be determined by surface strains and stresses with no significant stress gradient effect, at least for the geometric features usually present in engineering design. However, in the high stress gradients which are present at sharper notches the crack may propagate into a region of low stress where the stress intensity factor at the crack tip is less than the non-propagating value Sth, and as a result the crack ceases to grow.

2002 ABAQUS Users' Conference

(a)

(b)

Figure 2. Relationship between endurance limit stress and the stress concentration factor Kt (e is the smooth specimen endurance limit stress, Sth is the threshold stress for non-propagating cracks) ) Use of stress gradients may therefore imply that the designer is no longer designing to prevent crack initiation, but instead is relying on a calculation that the cracks will initiate but not propagate. This may be an unsafe assumption in some complex components where crack growth may be accelerated by load redistribution.

3.5 Miners rule and the endurance limit. Conle and Topper [21], and Topper and co-authors [22] have shown that the constant amplitude endurance limit does not apply to the analysis of real service loading if some cycles in the loading exceed the constant amplitude endurance limit stress amplitude. For finite life design the larger cycles in the loading cause the endurance limit stress to be reduced significantly, with the result that small cycles contribute to the fatigue damage process. This effect has been included in many of the generic test signals derived by German industry (see for example the CARLOS car loading sequences), and must also be included in computer-based analysis.

3.6 Modelling cast iron Cast iron (and particularly grey iron) produces non-symmetrical hysteresis loops. These can be constructed by considering the separate contributions made by the bulk material, the graphite morphology and the crack opening and closure (Figure 3). In addition, fatigue damage accumulates non-linearly in cast irons, and this requires a multi-step damage summation procedure, changing the materials properties as damage accumulates. Downing [23] has proposed a successful method of modelling these processes. For multiaxial fatigue of cast irons, an axial strain parameter including mean stress effects, and critical plane analysis, are used.

2002 ABAQUS Users' Conference

200

100

Stress:MPa

Graphite Effect

Crack Closure Effect

-100

Bulk Response -200 Full Response -300 -4000

-3000

-2000

-1000

0 Strain:uE

1000

2000

3000

4000

Figure 3. The stress-strain response of grey iron

4.0 HIGH TEMPERATURE FATIGUE ANALYSIS A simple approach to high temperature fatigue analysis is to adjust the materials properties for the effects of temperature, using a separate adjustment for each node in the model. The materials data base contains either (i) a definition of the S-N curve at several (constant) temperatures, or (ii) a definition of the strain-life curve and cyclic stress-strain curve at several different constant temperatures. An example of strain-life data is shown in Figure 4.

Figure 4. Example of temperature-dependant strain-life data The FEA results file must contain the temperatures at each node (or the sequence of temperatures defining the time history of temperature at each node) and the thermal stresses. Additional stresses, from for example the dynamic forces and gas pressure on a piston, may be superimposed on the thermal stresses, if a linear elastic FE solution has been used. However, if this 'scale and combine' procedure is used it is not possible to allow for variations in contact pressure.

2002 ABAQUS Users' Conference

To analyse a node, the fatigue software reads the time history of the nodal temperature, and finds the maximum temperature. This temperature is used to determine an appropriate set of materials data. Using the materials data base, the software interpolates (using linear or logarithmic interpolations as appropriate) to produce a stress-life curve, or a strain-life curve and cyclic stress-strain curves, at this maximum temperature. This data is used for the analysis at the node. The procedure is repeated for each node in the model, adjusting the materials data for each nodal maximum temperature. An example of this type of analysis is shown in Figures 5 and 6, from Reference 24. The fatigue analysis of a diesel engine piston was carried out using the method described. For comparison, the analysis was repeated using constant room temperature materials data. The loading duty was taken from an accelerated fatigue test, simulating cycles produced when the engine is run at full load for a period of time then at zero load for a period of time. The thermal analysis of the loading was performed in ABAQUS.

Figure 5. Fatigue life contours

Figure 6. Fatigue life contours, room temperature materials data

Figure 5 shows that the fatigue crack initiation site is identified correctly in the crown on the piston, if the materials data is adjusted for the effects of temperature as described. If room temperature data is used for the analysis, this crack site is not identified in the fatigue analysis. An example of crack initiation, and the calculated life contours, is shown in Figure 7.

Crack initiation site

Figure 7. Correlation between test data and fatigue analysis 2002 ABAQUS Users' Conference 7

This demonstrates that it is essential to adjust the materials data for the effects of temperature if fatigue crack sites are to be correctly identified. It also demonstrates that this relatively simple method of high temperature fatigue analysis can give excellent correlation with test results.

5.0 TIME-DEPENDANT TEMPERATURE EFFECTS For some materials, and at higher temperatures, a simple adjustment of materials data is not sufficient, and time-dependant effects must be included. For the analysis of engine components, four effects must be taken into account. These are described below. The strains on the critical plane are cycle counted to extract the fatigue cycles. This is carried out on the time history, not on the peak-valley sequence, and all the strain/time points forming the cycle are stored, not just the peak and valley.

5.1 Calculation of stresses incorporating temperature changes and changes in strain rate. For each cycle the strain rate information is calculated. This is a multi-stage procedure. In the first stage, the strain rate between each data point is calculated. An average strain rate is determined. For the second stage, periods where the strains are held constant are identified. These hold periods can be formed by smaller cycles within the cycle, and by periods of constant strain.

Figure 8 Strain-time history

For the third stage, these holds are excluded, and a new strain rate for each time step in the cycle is calculated, and an average calculated. Any new holds are identified, using a strain rate of 5% of the new average. New strain rates are calculated with these new holds excluded. This process is repeated until no new holds are being identified and the average strain rate attains a stable value.

2002 ABAQUS Users' Conference

The result of this process is a fatigue cycle, complete with all its strain-time points, and no holds. The stresses are now calculated from these strains. The stresses are dependant on strain rate and temperature between each pair of points.

800 600 Stress (MPa) 400 200 0 -6000 -4000 -2000 -200 -400 -600 Strain () 0 2000 4000 6000

Figure 9 Fatigue stress-strain cycle constructed using strain-rate and temperature effects

This procedure produces a set of strains and associated stresses for each cycle, incorporating the effects of temperature and strain rate. Because the stress between each pair of data points is used, changes in strain rate and temperature during the cycle are allowed for. The diagrams show the calculation for a uni-axial stress state. For biaxial stresses a similar procedure is used with a biaxial cyclic plasticity model. Masings hypothesis (that the room-temperature hysteresis loop is symmetrical in tension and compression) is used as a basis for stress-strain relationships for conventional metallic materials. For cast irons, and particularly for grey iron, more complex methods of constructing roomtemperature hysteresis loops are used. In both cases the temperature and strain-rate effects are incorporated as described above.

5.2 Calculation of the effect of phase relationship between strain and temperature. Fatigue damage is dependant on the phase relationship between temperature and strain. The materials data base requires additional parameters calculated from the plastic strain component of the strain life curve, for two conditions: (i) temperature and strain cycled in-phase, and (ii) temperature and strain cycled 180o out-of-phase.

2002 ABAQUS Users' Conference

(i) In-phase

(ii) Out-of--phase

Figure 10. Material testing for in-phase and out-of-phase stress and temperature

Figure 11 Effect of phase relationship on fatigue strength

For the cycle, the phase relationship between strain and temperature is calculated, and an adjustment is made to the strain-life curve. This fatigue damage curve is used as a basis for all subsequent adjustments to the damage curve for this cycle.

5.3 Calculation of bulk stress relaxation Bulk stress relaxation is a time-dependant and temperature-dependant phenomenon (Figure 12). These effects are incorporated into the incremental damage procedure described in section 5.5. The calculated bulk stress relaxation curves are one of the plots that can be generated in fe-safe (Figure 13).

10

2002 ABAQUS Users' Conference

Figure 12 Bulk stress relaxation

200

-200 Str es s -400

-600

-800 0 5000 10000 Hours 15000 20000

Figure 13 Example of calculated bulk stress relaxation curves

5.4 Calculation of pre-soak/strain ageing. This is most pronounced early in the life of the component (Figure 14), and so an incremental damage procedure must be used to incorporate its effect. fe-safe generates calculated pre-soak curves (Figure 15).

2002 ABAQUS Users' Conference

11

Figure 14
1

Effect of pre-soak on strength

0.9

0.8
Factor

0.7

0.6

5000

10000 Hours

15000

20000

Figure 15 Example of calculated pre-soak strength reduction curve.

5.5 Incremental damage procedure Bulk stress relaxation and pre-soak effects vary with time. Using strain-life analysis, the fatigue damage is calculated for one repeat of the load history. A critical plane procedure is used to determine the most damaged plane. Once the critical plane has been determined it is then used for the subsequent damage calculations at this node. The bulk stress and pre-soak effects are then recalculated for the time taken for this one repeat of the loading history, and a new damage curve is constructed. These effects are then presumed to apply for the next two repeats of the loading, and the fatigue damage is calculated for two repeats of the loading. The bulk stress and pre-soak effects are then recalculated for the time taken for the total of the three repeats analysed so far. These effects are presumed to apply for the next four repeats of the loading, and the fatigue damage is calculated for four repeats of the loading. This procedure is repeated, for 8, 16, 32, etc. repeats of the loading, until the fatigue damage exceeds 1. The length of the final block of repeats is then reduced until the fatigue damage equals 1. The summation of all the blocks of repeats is the life at this node. 12 2002 ABAQUS Users' Conference

5.6 Superimposition of higher frequency cycles. The thermo-mechanical cycles described above would normally be obtained from a finite element analysis which produced the time history of the temperatures and elastically-calculated stresses. The time steps between the results do not need to be equal times. fe-safe allows other stress histories (from another FE analyses, or from an FE unit load analysis and a specified time history of loading) to be superimposed. In this way, centrifugal stresses can be superimposed on to the results of thermo-mechanical analysis, and the combined stresses used as input to the whole creep-fatigue analysis procedure. The superimposed loading may be at a much higher frequency, and fe-safe uses interpolation routines to apply the superimposed loading as the underlying temperature changes.

6.0 SUMMARY Fatigue analysis methodology is evolving continuously, particularly in the area of multiaxial fatigue (see for example [25]). However, existing algorithms carefully implemented allow complex multiaxial stress states to be analysed successfully. Analysis of elastic FE results, with the elastic-plastic conversion built in to the fatigue software, provides a computationally-efficient method of processing fatigue loading sequences. The methods have been extended to the analysis of cast irons using an incremental damage procedure. For high temperature analysis, it has been demonstrated that some components can be analysed successfully using a relatively simple temperature correction, with a separate correction applied to each node on the model. Examples of the fatigue analysis of diesel pistons have demonstrated that this method can predict crack initiation sites and fatigue lives to a very acceptable level of accuracy. At higher temperatures time-dependant phenomena must be taken into account. Four such phenomena - strain rate and instantaneous temperature, the phase relationship between stress and temperature, bulk stress relaxation, and pre-soak/strain ageing - have been included in fatigue analysis software.

7.0 REFERENCES [1] Musiol C, Draper J, Sykes N, Morton K Advances In Computer Aided Design Against Fatigue. Institution of Mechanical Engineers, London. Paper C234/81, 1981 [2] Devlukia J, Davies J Fatigue Analysis Of A Vehicle Structural Component Under Biaxial Loading. Biaxial Fatigue Conference, Sheffield University, Dec 1985 [3] Tipton S M, Fash J W Multiaxial Fatigue Life Predictions For The SAE Specimen Using Strain Based Approaches. Multiaxial Fatigue: Analysis and Experiments, SAE AE-14, 1989 [4] McDiarmid D L A General Criterion For High Cycle Multiaxial Fatigue. Fatigue Fract Engng Mater Struct Vol 14, No 4, pp429-453, 1991 [5] Brown M W, Miller K J A Theory Of Fatigue Under Multiaxial Strain Conditions. Proc Inst Mech Eng 187, pp745-755 ,1973 2002 ABAQUS Users' Conference 13

[6] Mroz Z On The Description Of Anisotropic Work-hardening. Journal of the Mechanics and Physics of Solids 15:163-175, 1967 [7] Lamba H S Non-proportional Cyclic Plasticity. PhD Thesis, Theoretical and Applied Mechanics Report No 413, University of Illinois at UrbanaChampaign, 1976 [8] Bannantine J A, Socie D F A Variable Amplitude Multiaxial Fatigue Life Prediction Method. Fatigue under Biaxial and Multiaxial Loading, Proc. Third International Conference on Biaxial/Multiaxial Fatigue, Stuttgart, 1989. EISI Publication 10, MEP, London, 1989 [9] Chu C, Conle F A, Hubner A An Integrated Uniaxial and Multiaxial Fatigue Life Prediction Method VDI Berichte NR 1283, 1996 [10] Devlukia J, Draper J, McDiarmid D L Integrated Software For Durability Design On Metallic Materials Report to the Department of Trade and Industry LINK Enhanced Materials Committee, 1997. [11] Software Manual for fe-safe Fatigue Analysis From Finite Element Models. Safe Technology Limited, 1997-2001 [12] Hughes A, Draper J, Kemp M Fatigue Assessment Of An Oilpan Incorporating Manufacturing Effects. Engineering Integrity, The Journal of the Engineering Integrity Society, January 2002. [13] Glinka G Calculation Of Inelastic Notch-tip Strain-stress Histories Under Cyclic Loading. Engineering Fracture Mechanics, 1985, 839-854 [14] Barkey M E and Socie D F Calculation Of Notch Strains For Nonproportional Cyclic Loading Using A Structural Yield Surface. Fourth International Conference on Biaxial/Multiaxial Fatigue, Paris, France, 1994 [15] Peterson R E Notch Sensitivity. Metal Fatigue Chapter 13, Sines and Waisman, Editors, McGraw-Hill Book Co., In., 1959. [16] Kuhn P, Hardrath H F An Engineering Method For Estimating Notch-Size Effect In Fatigue Tests On Steel. National Advisory Committee for Aeronautics, Technical Note 2805, October 1952. [17] Kuguel R The Highly Stressed Volume Of Material As A Fundamental Parameter In The Fatigue Strength Of Metal Members. University of Illinois, T&AM Report No. 169, June 1960. [18] Taylor D and ODonnell M Notch Geometry Effects In Fatigue: A Conservative Design Approach. Engineering Failure Analysis, Vol 1, No 4, pp275-287, 1994

14

2002 ABAQUS Users' Conference

[19] Frost N E and Dugdale D S Fatigue Tests On Notched Mild Steel Plates With Measurements Of Fatigue Cracks. Journal of the Mechanics and Physics of Solids 5:182-192, 1957 [20] Frost N E Notch Effects And The Critical Alternating Stress Required To Propagate A Crack In An Aluminium Alloy Subject To Fatigue Loading. Journal of Mechanical Engineering Science 2, 109-119, 1960 [21] Conle F A and Topper T.H Overstrain Effects During Variable Amplitude Service History Testing. International Journal of Fatigue, Vol 2, No.3, pp130-136, 1980 [22] DuQuesnay D L, Pompetzki M A, Topper T H Fatigue Life Prediction for Variable Amplitude Strain Histories. SAE Paper 930400, Society of Automotive Engineers
[23] Downing S D

Modelling Cyclic Deformation and Fatigue Behaviour of Cast Iron Under Uniaxial Loading Report No 101, Ref UILU 84-3601, (University of Illinois, 1984). [24] Tharratt D, Malton G and Draper J Fatigue Analysis of a Diesel Piston from a Finite Element Model Proceedings of the ABAQUS International User's Conference, England 1999. [25] Varvani-Farahani A and Topper T H A New Energy-Based Multiaxial Fatigue Parameter. Conference Proceedings, Fatigue 2000, Cambridge, England. Editors Bache M R, Blackmore P A, Draper J, Edwards J H, Roberts P and Yates J R. Engineering Integrity Society, 2000

2002 ABAQUS Users' Conference

15