Jtam.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 4, 39, 2001 FATIGUE CRACK GROWTH PECULIARITIES AND MODIFICATIONS OF THE WHEELER RETARDATION MODEL (PART 2) Sylwester Kłysz Materials Engineering Division, Air Force Institute of Technology, Warsaw e-mail: sklysz@itwl.waw.pl Modifications of the Wheeler retardation model have been presented in the paper. The modifications improve description of data resulting from investigations on fatigue crack growth phenomena, in particular, for higher values of overloads. Key words: crack growth, overloads, retardation,Wheeler model 1. Introduction Fatigue crack growth under a variable amplitude loading is usually accom- panied by the load interaction phenomenon, due to which the crack growth rate in a given cycle can differ from the growth rate observed for the same cycle in constant amplitude tests. The nature of the fatigue crack growth for various materials or conditions of loading – especially amplitude and frequen- cy of overload occurrences, or for randomly variable loads, is different. Crack growth prediction methods can be grouped by various criteria. An obvious requirement for prediction models is their capability to estimate variable am- plitude fatigue test results with a sufficient accuracy. It is also most desirable that the effect of any change in the load, material or geometry parameters on fatigue crack growth behaviour should be quantitatively predicted. The clas- sical Wheeler model has been modified in terms of extending the description of the fatigue crack propagation phenomena, among other things, those di- scussed in part 1 of the paper and by Kłysz (1998), where some peculiarities concerning the problem of the fatigue crack propagation have been presen- ted. TheWheeler model (Wheeler, 1972; Fuchs and Stephens, 1980; Kocańda 844 S.Kłysz and Szala, 1985) defines the retardation coefficient Cp in the following way (Fig.1a) Cp = ( rp,i aov+rp,ov−ai )n (1.1) where rp,i,rp,ov – radii of plastic zones of current and overload cycles, re- spectively ai,aov – crack lengths of the current and overload cycles, respec- tively n – exponent in theWheeler model. the range of using it is determined with the following condition: ai + rp,i ¬ aov+rp,ov. According to this model, the retardation phase exists until the plastic zone rp,i related with the propagating crack (i.e. in the current cycle of the load) is contained within the plastic zone rp,ov originated from the overload previous to the given cycle. With the constant-amplitude loading applied in cycles after the overloads had been imposed, the retardation coefficient Cp changes monotonically and increases up to unity at the moment of reaching the overload plastic zone by the current plastic zone of the crack. 2. Modification I The first modification has been based on the assumption that the crack growth retardation due to an overload is present up to the moment when the crack tip (and not the plastic zone spreading in front of it) reaches the boundary of the plastic zone produced by this overload. Hence, the condition of using the retardation model takes the form ai ¬ aov + rp,ov, whereas the equation that defines the retardation coefficient assumes the following form Cp = ( rp,i aov +rp,ov+rp,i−ai )n (2.1) Treating the boundary of the overload plastic zone as a physical barrier to be overcome by the front of the propagating crack seems to bemore justified than the approach presented in the initial form of themodel. Not earlier than after the crack itself (not the plastic zone of the crack) has overcome this pla- stic zone, the retardation coefficient reaches unity and the crack grows again at the rate as if no overload has occurred.Thismodification does not generate Fatigue crack growth peculiarities... – part 2 845 Fig. 1. Dependence of the retardation coefficient on the crack length (a) and plastic zones during overload crack propagation (b); curve No.1 – initialWheeler model, curve No.2-No.5 –Wheeler model after modification I-IV any significant change in the nature of the dependence of the retardation co- efficient upon the crack length. The only effect it produces is some delay (i.e. some elongation vs. crack length) of the range of retardation, i.e. up to re- aching unity by the retardation coefficient. It is evident that in the case of the constant-amplitude loading after the overload has occurred it is the extension by a distance equal to the size of the current plastic zone. Fig.1b shows some exemplary plots that illustrate these dependences for a selected computational case. Curve No.1 corresponds to the initial form of the Wheeler model after modification. A significant difference between these two cases consists in the fact that the retardation model has not been based on the interdependence 846 S.Kłysz between the overload plastic zone and the current one only, but between the overload plastic zone and the front of the propagating crack. Thus, it allows some instances to be described, namely those when the crack growth retarda- tion takes place just as after thefirst overload cycle in the course of executing a number of subsequent cycles of applying the overload (e.g. while replacing the load block of the L-H type with a block of a higher maximum load). In such situations, with the initial model incorparted, the size of the current plastic zone always exceeds the size of the plastic zone reached in the previous cycle. As the condition that implies validity of the retardation model had not been satisfied, no account was taken of the retardation phenomenon in computa- tional analyses. In other words, the initial Wheeler model does not provide for any retardation in this case, because each subsequent overload cycle (i.e. a cycle from the block of a higher maximum load) generates a plastic zone beyond the zone originated in the course of the previously applied cycle (due to changes in both the crack lengths and the stress intensity factors in those cycles). The modified model relates in such situations the occurrence of the retardation phase to the sizes of crack increments in these overload cycles. If the crack increments are lower than the size of the plastic zone (e.g. at the initial stage of the crack growth), the retardation phase is to occur (before the crack front reaches this zone). On the other hand, if the increments are higher than the size of the plastic zone (like, for example, at the near-critical stage of the crack growth), the retardation phase will not occur at all. What can be referred to in this case, is the dependence of crack growth occurrence, or the crack growth not being retarded, on both: the size of the plastic zone in a given cycle – the retardation phase is to appear at large enough plastic zones (e.g. with plastic materials), and will not occur at small plastic zones (e.g. with brittle materials), and on purely fatigue properties of the material within which the crack propagates – the retardation phase is to occur with a slowly propagating crack, and will not occur with a quickly propagating one. In otherwords, themodel comprises also the cases when – as the experimental knowledge shows – different materials respond in different ways to identical overloads applied (e.g. at the same time intervals and of the same size). The modifiedmodel conditions the occurrence and the duration of the retardation effect not only with differences in values of the yield points of a given mate- rial but additionally with taking account of the crack-propagation properties of the materials (by dint of including the current crack length, i.e. indirectly, the C and m coefficients of the equation of propagation, and the form of the stress intensity factor). The very problem of thematerial response to overload sequences (decreasing/increasing blocks) and their modes (number of cycles, Fatigue crack growth peculiarities... – part 2 847 succession of applied overload blocks, mean values, stress ratios, etc.) is also of great significance to evaluation of the fatigue life – it will not be, however, analysed anymore in this paper. Let us then return to the problem under discussion, i.e. a single overload. It is evident that this modification does not ensure ability to describe the crack growth, like it did in figures included in part 1 of the paper and by Kłysz (1998). There is still the monotone dependence of the delay coefficient upon the crack length within the whole range of the retardation occurrence. Therefore, themodeldescribesonly thecases featuredwith the regular increase in the crack growth rate. 3. Modification II The secondmodification introduces the variable exponent n′ of theWhe- eler model, one that also depends (as the very retardation coefficient of the initial model) on the crack position and the current plastic zone previous to that originated in the overload cycle.Within the range, where the plastic zone of the propagating crack is totally included into the plastic zone originated in the overload cycle (ai+rp,i ¬ aov+rp,ov), assuming that the exponent n takes the form of n′ described with one of the following relationships is suggested n′ = (ai+rp,i−aov rp,ov ) n (3.1) or n′ = (ai+rp,i−aov rp,ov )n (3.2) Within the range, where the plastic zone in front of the propagating crack crosses theplastic zone originated in the overload cycle (ai+rp,i >aov+rp,ov), replacing the exponent nwith the n′ one describedwith one of the following relationships is recommended n′ = (aov +rp,ov−ai rp,i,max ) n (3.3) or n′ = (aov+rp,ov−ai rp,i,max )n (3.4) where rp,i,max stands for the maximum current plastic zone generated in the course of subsequent cycles after the overload has been applied, i.e. the zone 848 S.Kłysz that has overcome the overload zone during the retardation effected with this overload. As in the initial Wheeler model, the term in brackets in equations (3.1)- (3.6) illustrates just a fraction of the path that a crack has to cover to generate the plastic zone in front of it reaching the overload plastic zone (stage I) or escaping the retarded growth (stage II), and its value be contained in the (0,1) interval. The changes in variability of the retardation coefficient Cp against the crack length ai (together with changes of curves for the initial (No.1) and modified (No.2) Wheeler models) have been shown in Fig.1b. For the sake of simplicity in illustrating the problem, some exemplary values of the individual parameters (rp,i, aov, rp,ov, and rp,i,max) have been assumed, as well as the constant-amplitude loads to be applied in subsequent cycles to eliminate the shown changes (examplary plots) in the retardation coefficient due to load variability. CurveNo.3 correspondswith equations (3.1) and (3.3) (stage I and II) with the coefficient n in the form of a multiplier, whereas curve No.4 corresponds with equations (3.2) and (3.4) with the coefficient n in the form of an exponent. In both the cases the retardation coefficient takes some value from the interval (0,1) and reaches the minimum at the instance of crossing the overload plastic zone by the plastic zone in front of thepropagating crack. Itmeans that themodified retardationmodel of fatigue crack growth permits the cases to be analysedwhen the crack growth does not show the uniformly monotonic nature after the overload has occurred (as in the initial Wheeler model). It provides for a period of gradual retardation of the crack after applying the overload (up to the moment of intersection of both plastic zones, i.e. the current and the overload ones), followed by anwith the accelerated (but still slower than the crack propagation with no overload applied) escape from the phase of the growth retardation. This has found practical confirmation during testing the specimens. 4. Modification III On the account of thementioned in part 1 of this paper and in thework by Kłysz (1998) differences in crack length increments of 1-2 orders ofmagnitude experimentally observed in overload cycles and between the overloads appli- cation, it seems that the retardation model predicts these rapid crack length increments as well. In the considered model it is possible, if the retardation coefficient Cp could accept valueswithin the range of 1-2 orders ofmagnitude. Fatigue crack growth peculiarities... – part 2 849 The range of variability of Cp like of curves No.3 and No.4 in Fig.1b does not render such a description possible. As equations (3.1) and (3.3) in shortly expanded forms can be written down as n′ = ( 1− aov+rp,ov−ai−rp,i rp,ov ) n (4.1) and n′ = ( 1− ai+rp,i−aov−rp,ov rp,i,max ) n (4.2) it is suggested using these equations in the following forms: — at stage I n′ =1− (aov+rp,ov−ai−rp,i rp,ov ) n (4.3) — at stage II n′ =1− (ai+rp,i−aov−rp,ov rp,i,max ) n (4.4) This makes the exponent n′ capable of accepting positive and negative values (for specific positive values of n), and thus, the coefficient Cp capable of accepting the values higher than unity. Extending the range of possible applications of the retardation model is of crucial importance. Curve No.5 illustrates variability of the Cp coefficient in these cases. It is evident that immediately after applying an overload (i.e. for crack lengths close to aov at stage I) aswell as before abandoning the retardation phase (i.e. for crack lengths close to aov + rp,ov at stage II) the retardation coefficient is greater than one. The case that takes place at stage II is of no interest to the model-based description (or at least difficult to interpret as far as the mechanism of the crack growth at this stage is concerned), therefore, it will not be taken into account any more. On the other hand, the changes in the retarded crack growth throughout stage I meet the requirement assumed at the beginning – they ensure variability of the Cp coefficient within the range of 1-2 orders of magnitude and exactly the same differences in the increments of the crack length. Hence the suggestion: let the following relationships be accepted as the retardation coefficient: — in stage I (the left portion of curve No.5) Cp = ( rp,i aov+rp,ov−ai )n′ = ( rp,i aov+rp,ov−ai )1− ( aov+rp,ov−ai−rp,i rp,ov ) n (4.5) — in stage II (the right portion of curve No.4) Cp = ( rp,i aov+rp,ov−ai )n′ = ( rp,i aov +rp,ov−ai ) ( aov+rp,ov−ai rp,i,max )n (4.6) 850 S.Kłysz The exponent of retardation n′ can be then written down with only one equation of the following form n′ = sgn(ai ¬ aov+rp,ov−rp,i) [ 1− (aov+rp,ov−ai−rp,i rp,ov ) n ] + (4.7) + sgn(ai >aov+rp,ov−rp,i) (aov+rp,ov−ai rp,i,max )n The range of the retardation effect and changes in variability of the Cp coefficient in the cases under discussion depend, of course, on specific values of the individual parameters: rp,i, aov, rp,ov, and rp,i,max, i.e. on the crack length at the instance of the overload occurrence, on the overload size kov, and plasticity of the material. Themodel still retains its essential, alreadyknown–owing to experimental work – property: when the overload increases, the crack length corresponding to stage I increases aswell, and the crack over someconsiderable length escapes the retardation phase. Moreover, all equations for n = 1 are equivalents, which seems to confirm the theoretical roots common for all of them. For values of the exponent n from within the (0,1) interval, the variability of the retardation coefficient Cp increases monotonically – the model enables obtaining the properties like in the initialWheelermodel over thewhole range of its applications; in other words, the initial Wheeler model is a special case of the modifiedmodel. Fig.2 shows amore complete display of the changes in the retardation co- efficient Cp against the crack length for the suggested formsof the relationship defining this coefficient (equations (4.5)-(4.7)), for awider interval of values of the coefficient n, i.e. from 0 up to 6. The conditions of the simulation were: – application of constant-amplitude loads in subsequent cycles – to eliminate changes in the retardation coefficient due to variability of the load level, hence, rp,i = rp,i,max = const (although in practice the coefficient chan- ges as the crack length increases – a constant value has been assumed for the sake of clarity of the figure) – the same length of the crack at themoment of the overload aov occurrence – values of kov =1.75, 1.4, and 1.2 – values of n=0.5, 2, 4. So the extensive range of variability of the properties of themodifiedmodel ensures high flexibility in describing the phenomenon of fatigue crack growth retardation. Fig.3 shows exemplary relationships of a = f(N) determined Fatigue crack growth peculiarities... – part 2 851 Fig. 2. Variations of the retardation coefficient as a result of modifications I-III (described in detail in the text); aov =10mm, rov =1.695mm, ri =0.551mm according to the model described by Kłysz (1991), and the corresponding retardation coefficient Cp against the number of cycles, for simulation-based analyses of the crack propagation in a standard flat specimen with a single- edge notch (SEN). The calculations have been carried out for the following conditions: – constant-amplitude load of the specimen,with overloads kov =1.75 applied every 10000 cycles – pre-determined values of the coefficients of the Paris equation: m = 2, C = 1 ·10−10 – materials with various yield points, i.e. 360MPa – variability of the coefficient nwithin the range (0,9). It has become evident that themodifiedmodel enables description of a ve- rywide spectrumofmaterial behaviour under the above-mentioned conditions of loads application. For plastic materials (large plastic zones) no escape from the retardation phase occurred in amajor part of the specimen life. The crack length increments between the overloads did not surmount the large overload plastic zone up to themoment just before the specimen failure, when the rate of propagation was the highest. The crack did not escape the retarded phase of growth. In the case of a material with a high yield point the situation is different.While between the overloads, the retardationmanifested itself in its full range – the retardation coefficient was changingwhile passing through the 852 S.Kłysz Fig. 3. Simulation-based plots of crack propagation.Modified retardationmodel, the yield point 360MPa Fatigue crack growth peculiarities... – part 2 853 minimum to reach unity before the subsequent overload occurred. An inter- mediate situation takes place in materials with mid-range values of the yield point. Therefore, the points of inflexion (or lack of them) and the directions of convexities on the a = f(N) curves and, for the cases with high values of n, also considerable crack length increments in the overload cycles, are characteristic for those variants. All these features are easily observed in the course of fatigue tests. Tomake the comparisons easier, the Cp = f(N) curves have been plotted up to the same scale; unfortunately, to the effect that the values of the Cp coefficient exceeding 2 (for n=5 and n=6) have been cut off. The occurrence of something like ”double retardation phase” at n = 0 (up to 150000 cycles), and n= 0.5 (up to 30000 cycles), for a material with a low yield point, is an interesting detail of this simulation. The retardation coefficient Cp was not subjected to cyclic changes every pre-determined num- ber of the cycles consistent with the frequency of overload occurrences, but every twice as large number of cycles. The crack length increment between the overload applications was small enough (as referred to the size of the overload plastic zone) not to escape the retarded phase of growth (the crack tip did not reach this zone) even at themoment of applying the subsequent overload. The result was that the previous retardation phase seemed to be continued. The retardation effect was repeated in cycles, every 20000 cycles instead of every 10000 cycles. It could be equivalent to a situation met in research practice, when the response of a given material to some overloads is evidently different from that to other ones (e.g. different nature of the relationships a= f(N) or COD = f(F), the crack opening displacement against the loading force, i.e. the specimen flexibility, or the relationship da/dN = f(∆K), the crack pro- pagation rate vs. range of stress intensity factor). It takes place particularly in the cases of high overloads, when the classical response to an overload does not occur after every overload application, but every second, third, or further one, and thematerial behaves between the overloads either in a differentmanner or ”as if nothing had been happening”.According to the terminology assumed in the paper, it may correspond to the above-mentioned: as if ”double”, ”triple”, etc. retardation phase. Fig.4 shows, according to a slightly different approach, exactly the same results for a pre-determined value of the coefficient n. The mid-range value has been selected fromamong thepreviously assumedvalues, i.e. n=2.Thus, oneof the essential factors that shape the retardationmodel (the coefficient n) has been eliminated for the sake of this analysis. From the above-shown plots it follows that the mentioned factor has originated a 300 per cent change in the specimen life and is responsible for physically different interpretation of 854 S.Kłysz Fig. 4. Simulation-based plots of crack propagation.Modified retardationmodel, the coefficient n=2 Fatigue crack growth peculiarities... – part 2 855 the changes in the retardation process of the individual curves a = f(N). Therefore, the influence of the two other factors (i.e. overload quantity, and yield point) on description of the changes in the retardation process with the model under discussion has been presented. The plots confirm the above discussed relationships between the changes of the retardation coefficient in individual computational modes. It is evident that even for such a limited range of the analysed mode (n = 2) the model is highly sensitive to the magnitude of the applied overload and the sort of material used. It proves suitability of themodifiedmodel to the analysis of fatigue crack growth under random loadings. 5. Modification IV In all cases the minima of the curves Cp = f(a) (Fig.2) occurring for the crack length ai = aov + rp,ov − rp,i are either 0.5 (for curves No.4 and No.5) or (1/2)n (for curve No.3) – the exponent n′ takes – according to Eqs (4.5)-(4.7) – values 1 or n. For the ascending overload coefficients kov the model responses with a prolongation of stage I, like the initial Wheeler model does. The initial model reduces, however, the value of the retardation coefficient Cp – down to the minimum immediately after the overload cycle. In improving the description of the experimental data it seems reasonable tomake theminimumvalue of the Cp coefficient of themodifiedmodel change within a wider range, e.g. to make it decrease down to zero, or at least down to the range (0.5-0). It is suggested that the formula for Cp should take then the following form Cp = ( 1 αkov rp,i aov+rp,ov−ai )n′ (5.1) where α is to be found experimentally. Fig.5 shows some exemplary dependences of the retardation coefficient upon the crack lengths (based on Eqs (4.7) and (5.1) for some selected values of α and kov). Each curve models different ”dynamics” of the changes in the retardation coefficient. Such behaviour can happen in reality for various materials (brittle, plastic, cyclic-hardening and cyclic-softening ones, at va- rious stages of treatment), or different loading conditions (different values of DN, kov, R, but also randomness, multiaxiality, changes in temperature or frequency), which proves the universal nature of the suggested model. Same example of good fitting the above described theoretical model to experimental 856 S.Kłysz Fig. 5. Variability of the retardation coefficient after modification IV; aov =10mm, rp,ov =1.695mm, rp,i =0.551mm Fatigue crack growth peculiarities... – part 2 857 data shows Fig.6. It cannot be accomplished using the basic Wheeler retar- dation model. Fig. 6. Description of experimental data with the modifiedmodel 6. Conclusion Crack growth prediction in structured elements in service can be based upon various techniques with different degrees of sophistication and varying life prediction accuracy. In rising order of complexity these are: rough estima- tes, analyticalmethods combinedwith theFEM, sophisticated crackmodelling in 3D using automatic mesh-genetarion software and fracture mechanics ana- 858 S.Kłysz lyses. The fracture behaviour of a given structure or material will depend on the stress level, material properties and themechanisms bywhich the fracture progresses. The most successful approach in prediction and prevention of the fracture ismodelling the crack growth rate, especially whenunder a service lo- ading. The different features of a= f(N) and da/dN = f(∆K) relationships (presented in part 1) make a uniform theoretical description rather difficult, and sometimes preclude simple mathematical models from being applied. A number ofmodels have been developed to describe the observed variabi- lity in the crack growth. Themodifications of theWheeler retardation model have been proposed in the paper. They widen the range of possibilities of de- scribing the experimental data using this model. They also enable application of it to any of the dependences a = f(N) or da/dN = f(∆K) found in the research practice. This model has shown the ability to characterise both constant and variable amplitude fatigue crack growth. The above-introduced model guarantees good representations of the com- plex, experimentally determined relationships a= f(N) and, which is proba- bly of fundamental importance, very precise estimations of the coefficients of the equations of fatigue crack growth rate propagation, as well as the resul- tant fatigue life. The possibility of matching the theoretical description with experimental data is well confirmed. Hence, with such precise representations of the experimental data the model makes the estimations of dispersion of these parameters possible (e.g. while analysing some specific kinds of tests with a suitable number of the specimens). All mentioned features are very useful in the twomain approaches to the design of critical structural elements with respect to fatigue safe-life design and damage-tolerance design. The first approach aims at withdrawal of the elements from serwice before the crack is detectable.The secondapproach tolerates agrowing crack andaims at removal of the part when the crack is detectable or has the critical size. All the problems discussed above give grounds for correct analysis of fati- gue life of specimens and structural elements. It is a very important problem especially for all kinds of devices responsible for peoples’ life. It is hard to find tasks in engineering practice beingmore critical than the prediction of service lives of critical structural elements and prediction of their fatigue strength. References 1. Fuchs H.O., Stephens R.I., 1980, Metal Fatigue in Engineering, A Wiley- Interscience Publication Fatigue crack growth peculiarities... – part 2 859 2. Kłysz S., 1991, Modelling of Fatigue Crack Growth Within the Stress- Concentration Regions (in Polish), The PhD Dissertation, Military Academy of Technology,Warsaw 3. Kłysz S., 1998,Peculiarities of FatigueCrackGrowth and of theModel-Based Description of the Phenomenon (in Polish), Proc. of XVll Symposium of Fa- tigue and Fracture Mechanic of Materials and Structures, 159-164, Pieczyska k/Bydgoszczy 4. Kocańda S., Szala J., 1985, Fundamentals of Fatigue Calculations (in Po- lish), PWN,Warsaw 5. Wheeler O.E., 1972, Spectrum Loading and Crack Growth, Trans. ASME, J. Basic Eng., 94 Osobliwości rozwoju pęknięć zmęczeniowych i modyfikacja modelu opóźnień Wheelera – część 2 Streszczenie Przedstawiono modyfikacje modelu opóźnień Wheelera rozwoju pęknięć zmęcze- niowych. Modyfikacje te poprawiają możliwości opisu danych doświadczalnych z ba- dańpropagacji pęknięć zmęczeniowych,w szczególnościdla dużychprzeciążeń.Przed- stawiono przykłady opisu danych doświadczalnychprzy użyciu zmodyfikowanegomo- delu opóźnieńWheelera. Manuscript received March 19, 1999; accepted for print May 9, 2001