Fatigue Failure

section epub:type=”chapter”>



Fatigue Failure



18.1 Introduction


In Chapters 14 and 15, we examined the conditions under which a crack was stable, and would not grow, and the condition


K=Kc



si1_e


under which it would propagate catastrophically by fast fracture. If we know the maximum size of crack in the structure we can then choose a working load at which fast fracture will not occur.


But cracks can form, and grow slowly, at loads lower than this, if the load is cycled. The process of slow crack growth—fatigue—is the subject of this chapter. When the clip of your pen breaks, when the pedals fall off your bicycle, when the handle of the refrigerator comes away in your hand, it is usually fatigue which is responsible.


18.2 Fatigue of Uncracked Components


Fatigue tests are done by subjecting specimens of the material to a cyclically varying load or displacement (Figure 18.1). For simplicity, the time variation of stress and strain is shown as a sine wave, but modern testing machines are capable of applying almost any chosen load or displacement spectrum. The specimens are subjected to increasing numbers of fatigue cycles N, until they finally crack (when N = Nf—the number of cycles to failure).


Figure 18.1
Figure 18.1 Fatigue testing.

Figure 18.2 shows how the cyclic stress and cyclic strain are coupled in a linear elastic specimen. The stress and strain are linearly related through Young’s modulus, and the stress range Δσ produces (or is produced by) an elastic strain range Δɛel (equal to Δσ/E).


Figure 18.2
Figure 18.2 Coupling of cyclic stress and cyclic strain for a linear elastic specimen.

Figure 18.3 shows how the cyclic stress and cyclic strain are coupled in a specimen that is cycled outside the elastic limit. There is no longer a simple relationship between stress and strain. Instead, they are related by a stress–strain loop, which must be determined from tests on the material. In addition, the shape of the loop changes with the number of cycles, and it can take several hundred (or even thousand) cycles before the shape of the loop stabilizes. The stress range Δσ produces (or is produced by) a total strain range Δɛtot—which is the sum of the elastic strain range Δɛel and the plastic strain range Δɛpl.


Figure 18.3
Figure 18.3 Coupling of cyclic stress and cyclic strain for a linear elastic/yielding specimen.

The best way to correlate fatigue test data is on a log-log plot of the total strain amplitude Δɛtot/2 versus the number of reversals to failure 2Nf (there are two reversals of load or displacement in each complete cycle). Figure 18.4 shows the shape of the curve on which the data points typically fall (although there is usually a lot of experimental scatter on either side of the curve). It is useful to know that this curve is the sum of two linear relationships on the log–log plot: (a) between the elastic strain amplitude and 2Nf, and (b) between the plastic strain amplitude and 2Nf. It can be approximated mathematically as:


Δɛtot2σf(2Nf)bE+ɛf(2Nf)c



si2_e  (18.1)


Figure 18.4
Figure 18.4 Relation between total strain amplitude and fatigue life.

b and c are constants determined by fitting the test data—they are generally in the range –0.05 to –0.12 for b, and –0.5 to –0.7 for c. σfsi3_e and ɛfsi4_e are the true fracture stress and true fracture strain (derived from a standard tensile test on the material).


The data in Figure 18.4 can be divided into two régimes:



  •  Low-cycle fatigue (less than about 104 cycles; plastic strain > elastic strain)
  •  High-cycle fatigue (more than about 104 cycles; elastic strain > plastic strain)

Until the 1950s, most fatigue studies were concerned with high-cycle fatigue (HCF), since engineering components subjected to cyclic loadings (e.g., railway axles, engine crankshafts, bicycle frames) were designed to keep the maximum stress below the elastic limit. Because of this, it is still common practice to plot HCF data on a log–log plot of the stress amplitude Δσ/2 versus 2Nf—see Figure 18.5. The test data can then be approximated as:


Δσ2σf(2Nf)b



si5_e  (18.2)


Figure 18.5
Figure 18.5 Relation between stress amplitude and fatigue life in high-cycle fatigue.

So far, we have only considered test data obtained with zero mean stress (σm = 0). However, in many design situations, there will be a tensile mean stress (σm > 0). Intuitively, we would expect that the component would be more prone to fatigue if it were subjected to a large mean stress in addition to having to cope with repeated cycles of stress. The test data confirm this—to keep the fatigue life the same when the mean stress is increased from 0 to some large tensile value, the stress (or strain) cycles must be reduced in amplitude to compensate.


In terms of the strain approach to fatigue, the test data can be approximated as follows:


Δɛtot2(σfσm)(2Nf)bE+ɛf(2Nf)c



The equation clearly shows that, to keep the fatigue life the same, the strain amplitude must be decreased to compensate for a mean stress.


In terms of the stress approach to fatigue, the test data can be approximated as:


Δσσm2Δσ(σfσm)2σf



Δσ/2 is the stress amplitude for failure after a given number of cycles with zero mean stress, and Δσσm/2 is the stress amplitude for failure after the same number of cycles but with a mean stress. So if, for example, the mean stress is half the fracture stress, then the applied stress amplitude must be halved to keep the fatigue life the same.


18.3 Fatigue of Cracked Components


Large structures—particularly welded structures such as bridges, ships, oil rigs, and nuclear pressure vessels—always contain cracks. All we can be sure of is that the initial length of these cracks is less than a given length—the length we can reasonably detect when we check or examine the structure. To assess the safe life of the structure we need to know how long (for how many cycles) the structure can last before one of these cracks grows to a length at which it propagates catastrophically.


Data on fatigue crack propagation are gathered by cyclically loading specimens containing a sharp crack like that shown in Figure 18.6. We define


ΔK=KmaxKmin=YΔσπa



si8_e


Figure 18.6
Figure 18.6 Fatigue-crack growth in precracked components.

The cyclic stress intensity ΔK increases with time (at constant load) because the crack grows in tension. It is found that the crack growth per cycle, da/dN, increases with ΔK in the way shown in Figure 18.7.


Figure 18.7
Figure 18.7 Fatigue crack-growth rates for precracked material.

In the steady-state régime, the crack-growth rate is described by


dadN=A(ΔK)m



si9_e  (18.5)


where A and m are material constants. Obviously, if a0 (the initial crack length) is given, and the final crack length (af) at which the crack becomes unstable and runs rapidly is known or can be calculated, then the safe number of cycles can be estimated by integrating the equation


Nf=0NfdN=a0afdaAΔKm



remembering that ΔK=YΔσπa.si11_e


18.4 Fatigue Mechanisms


Cracks grow in the way shown in Figure 18.8. In a pure metal or polymer (left diagram), the tensile stress produces a plastic zone (Chapter 15) which makes the crack tip stretch open by the amount δ, creating new surface there. As the stress is removed the crack closes and the new surface folds forward extending the crack (roughly by δ). On the next cycle the same thing happens again, and the crack inches forward, roughly at da/dNδ. Note that the crack cannot grow when the stress is compressive because the crack faces come into contact and carry the load (crack closure).


Figure 18.8
Figure 18.8 How fatigue cracks grow.

We mentioned in Chapter 15 that real engineering alloys always have little inclusions in them. Then (see right diagram of Figure 18.8), within the plastic zone, holes form and link with each other, and with the crack tip. The crack now advances a little faster than before, aided by the holes.


In precracked structures these processes determine the fatigue life. In uncracked components subject to low-cycle fatigue, the general plasticity quickly roughens the surface, and a crack forms there, propagating first along a slip path (“Stage 1” crack) and then, by the mechanism we have described, normal to the tensile axis (Figure 18.9).


Figure 18.9
Figure 18.9 How cracks form in low-cycle fatigue. Once formed, they grow as shown in Figures 18.10 through 12.

High-cycle fatigue is different. When the stress is below general yield, almost all of the life is taken up in initiating a crack. Although there is no general plasticity, there is local plasticity wherever a notch or scratch or change of section concentrates stress. A crack ultimately initiates in the zone of one of these stress concentrations (Figure 18.10) and propagates, slowly at first, and then faster, until the component fails. For this reason, sudden changes of section or scratches are very dangerous in high-cycle fatigue, often reducing the fatigue life by orders of magnitude.


Figure 18.10
Figure 18.10 How cracks form in high-cycle fatigue.

Figure 18.11 shows the crack surface of a steel plate (40 mm thick) which failed by fatigue. The shape of the fatigue crack at any given time is indicated by “beach marks”. Figure 18.12 shows a fatigue crack in a human tooth caused by the repeated loads of chewing food.


Figure 18.11
Figure 18.11 Fatigue crack surface in a steel plate. The arrows show the direction of crack growth.

Figure 18.12
Figure 18.12 Fatigue crack in a human tooth.

Worked Example 1


You have just been told that some copper water-cooling plates have begun failing in a furnace. The suspected cause is low-cycle fatigue, caused by thermal expansion movements between the cooling plates and the furnace shell. To make a preliminary assessment, you urgently need the strain–life plot for annealed deoxidized copper—the material of the plates. But where do you start?


Remember that Figure 18.4 shows the typical form of the strain–life plot. Looking at the left side of the plot, you can see that it should at least be possible to find standard tensile testing data for copper, which would give you values for the true fracture stress and strain. There are plenty of handbooks and/or websites that can give you tensile test data, but of course this is always listed as nominal stress and strain, so needs to be converted into true stress and strain. This is OK, because you already know about the equations that do this conversion (see Chapter 9). You also need a value for Young’s modulus, but again that is easily available (and will be the same whatever the grade of copper). You easily locate the necessary data for annealed copper: σTS = 216 MN m-2, ɛf = 48% → 0.48, E = 130 GN m-2. Then:


ɛf=ln(1+ɛf)=ln(1+0.48)=ln(1.48)=0.39



si12_e


σf=σTS(1+ɛf)=216(1+0.48)=320MN m–2,σfE=320MN m–2130 GN m–2=0.0025



So you already have two critical points to put on your plot—we have marked them on the diagram.


Unlabelled Image

We could now do with something at the right side of the plot, to define the elastic fatigue line. There are lots of data available for HCF—in fact, much more than there are for LCF. But almost always, handbooks and/or websites list just one value—the stress amplitude for failure after typically 108 cycles. In the case of annealed copper, a quick handbook search comes up with a stress of ±76 MN m-2 after 2 × 107 cycles (4 × 107) reversals. The strain amplitude corresponding to the stress amplitude of 76 MN m-2 is:


Δɛel2=Δσ2E=76MN m–2130 GN m–2=0.00059



Marking this point on the diagram fixes the elastic fatigue line.


Finally, we need some points in the middle of the plot, to define the plastic line. Then we can just add the plastic line and the elastic line to get the overall strain–life plot (but remember that the vertical axis is not a linear scale!). Fortunately, a quick trawl through some standard textbooks comes up with a drawing of the cyclic stress–strain loop for annealed copper under a fixed strain range of 0.0084 (Hertzberg). The stabilized stress range is 252 MN m-2, and the number of reversals to failure is 8060. The total strain amplitude is then 0.0042. The elastic part of the total strain amplitude is:


Δɛel2=Δσ2E=252MN m–22×130 GN m–2=0.001



The plastic part of the total strain amplitude is then 0.0042 − 0.001 = 0.0032. The three strain amplitudes are marked on the preceding diagram, and they allow the overall strain–life plot to be drawn. This is sufficient for your preliminary assessment. However, predictive design should always rely on actual fatigue test data—and plenty of it, so that the variability of the data (mean and standard deviation) can be established.


Worked example 2


In Equation (18.5), we showed that the rate of fatigue crack growth was proportional to (ΔK)m, where ΔK is the range of the stress intensity for the crack, and m is a material constant. In Equation (18.6), we showed that we could calculate the number of fatigue cycles Nf required to grow the crack from an initial length to a final length by integrating up Equation (18.5). The integration is somewhat tricky, so we give an example of it here, using a value of m = 3 (which is typical for steels). Starting with an Equation of the kind (18.5), we get


dadN=AΔK3=AYΔσπa3=AYΔσπ3a3/2,a1a2daa3/2=AYΔσπ3dN,21aa1a2=21a11a2=AYΔσπ3N,N=2AYΔσπ31a11a2si16_e


Typically, A = 6 × 10–12 m (MN m–3/2)–3. We assume Y = 1, and Δσ = 100 MN m–2. We assume initial and final cracks of length 2 mm and 10 mm. To get the units right, these must be converted into m, i.e., to 0.002 m and 0.010 m. We then input numbers as follows


N=26×1012mMN3m3/23100MNm23π3/210.002m10.010mN=7.4×105si17_e


Putting the units in verifies that N is dimensionless, as it must be. Next time, there is no need to write out the units as long as they are all in MN and m.


Examples




  1. 18.1 The copper water-cooling plates in the Worked Example failed at welded joints which connected the plates to the inlet and outlet water pipes. During thermal cycling of the furnace, the pipes were bent back and forth, which resulted in cyclic straining of the welds. The strain amplitude was estimated to have been 0.01. Using the strain–life plot in the diagram, estimate the number of cycles required to cause the welded joints to fail. Your estimate of the strain amplitude could be in error by a factor of 2. What would be the failure life if the strain amplitude were as large as 0.02?
  2. 18.2 An electric iron for smoothing clothes was being used when there was a loud bang and a flash of flame from the electric flex next to the iron. An inspection showed that the failure had occurred at the point where the flex entered a polymer tube which projected about 70 mm from the body of the iron. There was a break in the live wire and the ends of the wire showed signs of fusion. The fuse in the electric plug was intact. Explain the failure. Relevant data are given in the table that follows.

  3. 18.3 A large component is made of a steel for which Kc = 54 MN m-3/2. Nondestructive testing by ultrasonic methods shows that the component contains long surface cracks up to a = 2 mm deep. Laboratory tests show that the crack-growth rate under cyclic loading is given by

    dadN=A(ΔK)3


    where A = 6 × 10-12 m (MN m-3/2)-3. The component is subjected to an alternating stress of range

    Δσ=180MNm2



    about a mean tensile stress of Δσ/2. Given that ΔK=YΔσπasi20_e, estimate the number of cycles to failure. Assume that Y = 1.12.


    u18-01-9780081020517

  4. 18.4 The preceding photograph shows the fracture surfaces of two broken tools from a pneumatic drill. The circular fracture surface is 35 mm in diameter, and the “rectangular” fracture surface measures 24 mm × 39 mm. The shape of the fatigue crack at any given time is indicated by the beach marks, which are clearly visible on the fatigue part of the fracture surface. Indicate the following features on each fracture surface:

    1. a. The point where the fatigue crack initiated
    2. b. The position of the crack just before the final fast fracture event
      The fatigue crack traverses much more of the cross section in the circular tool than in the rectangular tool. What does this tell you about the maximum stress in the fatigue cycle?

  5. 18.5 An aluminum alloy for an airframe component was tested in the laboratory under an applied stress which varied sinusoidally with time about a mean stress of zero. The alloy failed under a stress range, Δσ, of 280 MN m-2 after 105 cycles; under a range of 200 MN m-2, the alloy failed after 107 cycles. The fatigue behavior of the alloy can be represented by

    Δσ=C(Nf)b



    where b and C are constants. Find the number of cycles to failure, Nf, for a component subjected to a stress range of 150 MN m-2.
  6. 18.6 The following photograph shows a small part of a fatigue fracture surface taken in the scanning electron microscope (SEM). The material is stainless steel. The position of the crack at the end of each stress cycle is indicated by parallel lines, or “striations.” By taking measurements from the photograph, estimate the striation spacing.
    u18-02-9780081020517

  7. 18.7 After 3 years in service, a steel boiler started to leak steam from a position below the water line. The leak was traced to a fine crack 20 mm long in the wall of the boiler barrel alongside the entry position of the water feedpipe. An internal inspection showed severe cracking of the boiler plate. It was concluded that the cracks had initiated at the inner surface of the barrel and had then propagated through the wall. A baffle had been fitted in the water space a short distance in from the feed pipe in order to deflect the cold feed water sideways along the boiler barrel. Indicate the likely mechanism of failure and support it with a simple calculation. Supporting data is as follows: water temperature of boiler = 180oC; feed water temperature = 10oC; coefficient of thermal expansion of steel = 12 × 10-6 oC-1; yield strength of boiler plate = 250 MN m-2.
  8. 18.8 Referring to Example 18.7, what effect would removing the baffle have on the thermal fatigue life?
  9. 18.9 Referring to Example 18.7, why is the fatigue likely to be low-cycle fatigue?
  10. 18.10 Referring to Example 18.7, estimate the number of thermal cycles over the 3-year period of operation. How does this confirm low-cycle fatigue?

Answers




  1. 18.1 Approximately 1000 and 300 reversals (500 and 150 cycles). The answers can be read directly off the graph in Worked Example 1.
  2. 18.2 Each time the iron was moved backwards and forwards the flex would have experienced a cycle of bending where it emerged from the polymer sheath. The sheath is intended to be fairly flexible to avoid concentrating the bending in one place. Possibly the sheath was not sufficiently flexible, and the flex experienced a significant bending strain amplitude at the location of failure. The number of cycles is well into the range for high-cycle fatigue, so fatigue is the likely cause. The scenario is that the individual strands in the live conductor broke one by one until the current became too much for the remaining strands to carry. At this stage, the remaining strands would have acted like a fuse and melted, causing the fire. If 23 strands are rated to carry 13 A, then a single strand should carry about 0.57 A. The iron draws 4.8 A, which is 8.4 times the capacity of a single strand. It is not surprising when only a few strands were left intact that the flex was no longer able to take the current without overheating. Failures of this type have also occurred with appliances such as vacuum cleaners. However, these tend to have a smaller current rating, so failure does not always cause a fire.
  3. 18.3 First find a2 using the fast fracture equation
    Kc=YΔσπa2a2=1πKcYΔσ2si22_e
    Input Kc = 54, Y = 1.12 and Δσ = 180. This gives a2 = 0.0228 m. Then follow the method of Worked Example 2, integrating from a1 = 0.002 m to a2 = 0.0228 m. This gives 1.15 × 105 cycles.
  4. 18.4 

    1. (a) Circular = 0730 hrs position. Rectangular = 0600 hrs position.
    2. (b) Circular = 1300 hrs position. Rectangular = 1200 hrs position.

  5. 18.5 5.2 × 108 cycles. Rewrite the given equation as C = Δσ(Nf)a, then write it out twice and equate to give 280(105)a = 200(107)a. This gives a = 0.073. Putting this back into the equation for a (Δσ, N) pair (e.g., 280, 105) gives C = 649 MN m–2. The equation with these values of a and C then gives Nf for 150 MN m–2.
  6. 18.6 0.5 μm. Scale off the 1 μm “bar” at the bottom LHS of the photo (it is 3.0 mm long on the photo). To measure the striation spacing more accurately, take a length on the photo of, say, 20 mm, and count the number of striation spacings over this length. Repeat this on different parts of the photo to get an idea of the average value.
  7. 18.7 Low-cycle fatigue, produced by repeated cycles of thermal contraction and expansion (thermal fatigue). Show that the maximum strain due to thermal contraction αΔΤ = 2.04 × 10–3, which is 1.6 times greater than the yield strain σy/E = 1.25 × 10–3, so there is plenty of thermal strain to spare.
  8. 18.8 The baffle makes the cold water sweep sideways over the inner surface of the boiler plate, ensuring efficient cooling. The problem would have been much less serious if a baffle had not been fitted. The fatigue life would have been much longer, and in fact there might not have been any fatigue cracking at all.
  9. 18.9 Because the thermal strain is greater than the yield strain, the plate is experiencing a plastic strain amplitude—see Figure 18.4.
  10. 18.10 Assume 10 feeds per day at 365 days per year over 3 years = 104 cycles. This is consistent with low-cycle fatigue—see Figure 18.4.

Aug 9, 2021 | Posted by in General Engineer | Comments Off on Fatigue Failure
Premium Wordpress Themes by UFO Themes