What does safe mean?
In January 1954 a de Havilland Comet, the world's first jet airliner, broke up in the air near the island of Elba. In April another broke up near Naples. Both aircraft had passed every static strength test their designers had set; the stresses in their cabins were well below the strength of the aluminium alloy. They failed because a stress that is harmless once can be fatal when it is repeated a few thousand times, and because the stress at the corner of an opening was far higher than the average.
This lesson is about what it takes for a part to be safe: not to yield, not to break, and not to wear itself out with repetition. Almost every idea in it was learned the hard way, and the Comet's story closes it.
Three ways to fail
A metal part under load can fail in three broad ways (a fourth, buckling, was the subject of Deflection and buckling, and creep and corrosion belong to Lumenary Materials' fatigue, creep and corrosion).
- Yielding. A ductile part is loaded beyond its yield strength and deforms permanently. A bent bracket or a stretched bolt has failed even though it is in one piece.
- Fracture. A crack runs through the part. Brittle materials (glass, cast iron, ceramics, and some steels in the cold) fracture with little warning and little deformation.
- Fatigue. A load far below the yield strength, repeated thousands or millions of times, starts a tiny crack at the most stressed spot, grows it a little with each cycle, and finally leaves too little material to carry the load. Most mechanical failures in service are fatigue failures.
Analogy, not a formal construction
Bend a paper clip a little and it springs back: elastic. Bend it sharply and it stays bent: yielding. Now bend it back and forth through a modest angle, again and again. After some tens of bends it snaps, at a bend no bigger than the first one. Each bend did a little irreversible damage, and the damage added up. That is fatigue, in a form you can safely try at a desk.
The first two depend on how large the stress is; the third depends on how large the stress changes are and how many times they happen. A design must answer all three.
Yield under combined stress: Tresca and von Mises
A tensile test gives one number, the yield strength , for one stress state: uniaxial tension. But the surface of a shaft carries bending and torsion together, and the wall of a pressure vessel is stretched in two directions at once. When does such a combination yield? We need a rule that turns the three principal stresses (from Stress and strain) into one equivalent stress to compare with .
Metals yield by slip of crystal planes, which is driven by shear stress, and they hardly respond to hydrostatic pressure: squeezing a block of steel equally from all sides does not make it yield. Both classical criteria build in those two facts.
Definition— Tresca (maximum shear stress) criterion
Yield begins when the largest shear stress reaches the largest shear stress in a tensile test at yield:
For a surface point with normal stress and shear stress (a shaft in bending and torsion), .
Definition— von Mises (distortion energy) criterion
Yield begins when the elastic energy stored in changing the material's shape (not its volume) reaches its value in a tensile test at yield:
For a surface point with and , .
Both depend only on differences of principal stresses, so adding the same pressure to all three changes nothing. They agree in uniaxial and equal biaxial tension and differ most in pure shear, where Tresca predicts yield at and von Mises at . Experiments on thin-walled tubes under tension and torsion, from the 1930s on, put ductile metals between the two, mostly closer to von Mises. Tresca is always the more conservative, by at most 15.5%; von Mises is what finite element programs plot by default as "equivalent stress".
The block compares the two criteria for stress states that appear throughout mechanical engineering, and reports the factor of safety against yield, , for a structural steel with MPa.
import numpy as np
def equivalent_stresses(S):
"""Principal stresses, Tresca (s1 - s3) and von Mises equivalent stresses of a 3 x 3 tensor."""
s3, s2, s1 = np.linalg.eigvalsh(np.asarray(S, dtype=float)) # ascending order
tresca = s1 - s3
mises = np.sqrt(((s1 - s2)**2 + (s2 - s3)**2 + (s3 - s1)**2) / 2)
return (s1, s2, s3), tresca, mises
def plane(sx, sy=0.0, txy=0.0):
"""A plane stress state as a 3 x 3 tensor (MPa)."""
return [[sx, txy, 0.0], [txy, sy, 0.0], [0.0, 0.0, 0.0]]
Sy = 275.0 # MPa, yield strength of a structural steel
cases = {
"uniaxial tension 200": plane(200.0),
"pure shear 120 (shaft in torsion)": plane(0.0, 0.0, 120.0),
"biaxial 150, 150 (sphere)": plane(150.0, 150.0),
"hoop 160, axial 80 (cylinder)": plane(160.0, 80.0),
"bending 120 + torsion 70 (shaft)": plane(120.0, 0.0, 70.0),
"hydrostatic -100 (deep sea)": -100.0 * np.eye(3),
}
print(f"{'stress state (MPa)':34} {'s1':>6} {'s2':>6} {'s3':>6} {'Tresca':>7} {'Mises':>7}"
f" {'n Tresca':>9} {'n Mises':>8}")
for name, S in cases.items():
(s1, s2, s3), tr, vm = equivalent_stresses(S)
s1, s2, s3 = (round(v, 9) + 0.0 for v in (s1, s2, s3)) # print 0, never -0
nt = f"{Sy / tr:9.2f}" if tr > 1e-9 else " none"
nm = f"{Sy / vm:8.2f}" if vm > 1e-9 else " none"
print(f"{name:34} {s1:6.0f} {s2:6.0f} {s3:6.0f} {tr:7.1f} {vm:7.1f} {nt} {nm}")In pure shear, and in the cylinder whose middle principal stress lies halfway between the others, Tresca is 15.5% higher, so its factor of safety is lower (1.15 against 1.32 for the shaft in torsion). The shaft in bending and torsion has a von Mises stress of 171 MPa, larger than either applied stress alone. And 100 MPa of hydrostatic pressure gives zero equivalent stress: yield criteria for metals ignore pressure.
Stress concentrations
The formulas of the previous lessons give smooth, average stresses. Real parts have holes, grooves, keyways, shoulders and threads, and near each one the stress rises sharply. The ratio of the peak stress to the nominal stress the simple formula gives is the stress concentration factor, , which depends only on the geometry.
Two classical results set the scale. A small circular hole in a wide plate under tension has : the stress at the edge of the hole, across the load, is three times the remote stress (Ernst Kirsch, 1898). An elliptical hole with half-width across the load and half-height along it has (Charles Inglis, 1913). As shrinks, the ellipse becomes a crack, and grows without limit.
import numpy as np
# 1. A small round hole (radius a) in a wide plate under remote tension s: Kirsch's solution.
# Stress along the line through the hole's centre, across the load (distance r from the centre):
print(" r / a stress / remote stress")
for r in [1.0, 1.1, 1.25, 1.5, 2.0, 3.0, 5.0]:
k = 1 + 0.5 / r**2 + 1.5 / r**4
print(f" {r:4.2f} {k:5.3f} " + "#" * int(round(k * 10)))
# 2. An elliptical hole with half-axis a across the load and b along it: Kt = 1 + 2 a / b.
print("\n a / b Kt")
for ratio in [1, 2, 5, 10, 100, 1000]:
print(f" {ratio:5d} {1 + 2 * ratio:5d}")
# 3. Fracture mechanics: a through crack of length 2a in a wide plate fails when
# K = s * sqrt(pi * a) reaches the fracture toughness K_Ic. Critical crack length 2a_c:
materials = [ # (name, K_Ic in MPa m^0.5, working stress in MPa) -- rounded, illustrative
("window glass", 0.75, 20.0),
("alumina ceramic", 4.0, 100.0),
("high-strength aluminium alloy", 25.0, 200.0),
("ultra-high-strength steel", 50.0, 800.0),
("tough structural steel", 150.0, 200.0),
]
print(f"\n{'material':32} {'K_Ic':>6} {'stress':>7} {'critical crack 2a_c':>21}")
for name, K, s in materials:
a_c = (K / s)**2 / np.pi # metres
print(f"{name:32} {K:6.2f} {s:7.0f} {2 * a_c * 1e3:14.2f} mm")The first table shows that a concentration is local: the stress is three times the remote value at the edge of the hole, but only 22% above it one radius away and 2% above it four radii away. The second shows why sharp features are dangerous: a slot ten times wider than it is tall has .
How much a concentration matters depends on the material and the loading. In a ductile part under a steady load, the tiny region at the peak yields and passes load to its neighbours, so designers of steel structures under static load usually ignore . In a brittle part there is no such relief, and the peak stress starts a crack. Under repeated loads, even ductile metals feel nearly the full peak (measured by a notch sensitivity), because fatigue cracks start wherever the stress cycles hardest. Handbooks such as Peterson's Stress Concentration Factors give for thousands of geometries, and a finite element model can compute it for any other.
The design response is simple to state: avoid sharp corners. Generous fillet radii at shoulders, rounded ends on keyways and slots, holes away from edges and from each other, gradual changes of section, and smooth surfaces.
Factors of safety
Definition— factor of safety
The factor of safety is the ratio of the strength of a part (the load or stress at which it fails in the relevant way) to the load or stress it is expected to carry: . A design is acceptable when is at least a required value , which covers the uncertainty in loads, material properties, dimensions, analysis and the consequences of failure.
Two styles are in use. In allowable stress design, a single factor divides the strength to give a permissible working stress. In limit state design, used by most modern structural codes, loads are multiplied by partial factors (larger for variable loads such as wind or traffic than for well-known self-weight), strengths are divided by other partial factors, and the factored load effect must not exceed the factored resistance. The second puts the margin where the uncertainty is.
Typical values reflect judgement and history. Transport aircraft must carry 1.5 times the largest load expected in service without failing, a deliberately small factor made acceptable by meticulous analysis, testing and inspection. Lifting equipment and wire ropes use much larger factors, often five or more on breaking strength, because loads are uncertain, people work close by, and failure releases a great deal of energy.
Fracture in brief
Inglis's ellipse predicts an infinite stress at the tip of a sharp crack, yet cracked parts carry load. A. A. Griffith resolved the puzzle in 1921 with an energy argument: a crack grows when the elastic energy it releases by growing exceeds the energy needed to make new surface. George Irwin, at the US Naval Research Laboratory in the 1950s, recast this as the stress intensity factor
where is the remote stress, the crack length (half-length for a crack in the middle of a plate) and a factor of order 1 for the geometry. A crack runs when reaches the material's fracture toughness , measured on cracked specimens by standard methods. The formula holds the key engineering insight: the strength of a cracked part falls with the square root of the crack size, and the critical crack length grows with the square of toughness over stress.
The third part of the block above computes that critical length for some representative, rounded values. A window pane at 20 MPa can tolerate a crack of only about 0.9 mm, and a ceramic at 100 MPa about 1 mm, which is why both are designed to keep tension low and surfaces flawless. A high-strength aluminium alloy at 200 MPa tolerates about 10 mm: big enough to find by inspection, which is the basis of aircraft damage tolerance. An ultra-high-strength steel at 800 MPa tolerates only 2.5 mm, the price of very high working stress. A tough structural steel would need a crack of over 350 mm.
That last number is beyond the formula's reach: a crack that large in a tough steel would make the remaining section yield long before reached , so linear-elastic fracture mechanics no longer applies and elastic–plastic methods take over. Toughness also depends on temperature. Below their ductile-to-brittle transition, many steels lose most of it, which is how welded ships broke in two in cold water in the 1940s.
Fracture mechanics also gives pressure equipment the principle of leak before break: choose material and wall thickness so that a crack growing through the wall becomes a detectable leak before it reaches the critical length for a sudden rupture. Why toughness differs so much between materials, and how it is measured with Charpy and fracture-toughness tests, is the subject of fracture and toughness in Lumenary Materials.
Fatigue: S–N curves and endurance limits
A fatigue failure has three stages. A crack initiates, usually at the surface where the cyclic stress is highest: a notch, a scratch, a corrosion pit, the toe of a weld. It grows a tiny distance each cycle, often leaving curved beach marks on the fracture surface that record where it stopped and started. Finally the remaining section is too small and it fractures suddenly. Investigators read a fatigue fracture surface like a map: smooth and marked where the crack crept, rough where it finally tore.
Definition— cyclic stress
A stress that cycles between and has amplitude , mean , range , and stress ratio . A rotating shaft under steady bending sees fully reversed stress (, ).
Definition— S–N curve and endurance limit
An S–N curve (or Wöhler curve) plots the stress amplitude against the number of cycles to failure , usually on logarithmic axes, from tests on many specimens at constant amplitude. Over much of its length it is close to a straight line on log–log axes, Basquin's law, with typically between and . Steels and titanium alloys tested in air show a knee, often around to cycles, below which specimens survive indefinitely; the stress there is the endurance limit . Aluminium and copper alloys show no true knee, so a fatigue strength is quoted at a stated life, such as cycles.
For steels a rough starting estimate of the endurance limit of a polished laboratory specimen is half the tensile strength (up to about 1400 MPa). A real part has a rougher surface, a larger size, a different kind of loading and must survive with high reliability, so engineers reduce the specimen value with modifying factors from data and design codes, often to half of it or less.
Mean stress: Goodman and friends
Tension held steady underneath the cycling opens cracks and shortens life; compression helps. The Goodman line (after the British engineer John Goodman, 1899) is the most widely used correction: a part survives indefinitely if
a straight line on a plot of amplitude against mean from at zero mean to the tensile strength at zero amplitude. Heinrich Gerber's parabola (1874) fits test data for ductile steels more closely; Richard Soderberg's line (1930) runs to the yield strength instead and is more conservative. A separate check makes sure that does not exceed the yield strength on the first cycle.
The block builds the S–N curve for an illustrative steel part, MPa and MPa, with a corrected endurance limit of 180 MPa, using a common textbook construction: a straight log–log line from at cycles to at . Then it checks a part loaded with an amplitude of 130 MPa about a mean of 200 MPa, like a pre-tensioned bolt or a spring.
import numpy as np
# An illustrative steel part: tensile strength 600 MPa, yield 400 MPa. Endurance limit of a
# polished specimen about half the tensile strength, reduced by an overall factor of 0.6 for
# a machined surface, size and reliability (real factors come from data and design codes).
Sut, Sy = 600.0, 400.0
Se = 0.6 * 0.5 * Sut # 180 MPa, corrected endurance limit
S1000 = 0.9 * Sut # strength at 1000 cycles
# A straight line on log-log axes (Basquin's law) from (1e3, 0.9 Sut) to (1e6, Se):
# S = a N^b
b = -np.log10(S1000 / Se) / 3
a = S1000 / 1e3**b
def cycles_to_failure(Sa):
"""Fully reversed stress amplitude -> cycles to failure (inf below the endurance limit)."""
return np.inf if Sa <= Se else (Sa / a) ** (1 / b)
print(f"S-N curve: S = {a:.0f} N^({b:.4f}); endurance limit {Se:.0f} MPa")
print(" amplitude (MPa) cycles to failure")
for Sa in [500, 400, 300, 250, 200, 180, 150]:
N = cycles_to_failure(Sa)
print(f" {Sa:5.0f} {'infinite' if np.isinf(N) else f'{N:12,.0f}'}")
# A part sees an alternating stress of 130 MPa about a mean of 200 MPa (a bolt, a spring).
Sa, Sm = 130.0, 200.0
print(f"\nalternating {Sa:.0f} MPa about a mean of {Sm:.0f} MPa:")
print(f" ignoring the mean: n = Se / Sa = {Se / Sa:.2f} (looks safe)")
for name, n in [("Goodman", 1 / (Sa / Se + Sm / Sut)),
("Gerber", 0.5 * (Sut / Sm)**2 * (Sa / Se) * (np.sqrt(1 + (2 * Sm * Se / (Sut * Sa))**2) - 1)),
("Soderberg", 1 / (Sa / Se + Sm / Sy))]:
print(f" {name:9} n = {n:.2f}")
Sar = Sa / (1 - Sm / Sut) # equivalent fully reversed amplitude (Goodman)
print(f" Goodman equivalent fully reversed amplitude = {Sar:.0f} MPa "
f"-> life {cycles_to_failure(Sar):,.0f} cycles")
print(f" first-cycle yield check: (Sa + Sm) / Sy = {(Sa + Sm) / Sy:.2f}")Life is extremely sensitive to stress: cutting the amplitude from 400 to 300 MPa multiplies the life by six, and from 250 to 200 MPa by four. Ignoring the mean stress, the part looks safe with . With the Goodman correction, its 130 MPa amplitude on a 200 MPa mean is as damaging as a fully reversed 195 MPa, above the endurance limit, and the predicted life is about 600 000 cycles; Soderberg is harsher still, while Gerber's parabola, closer to the test data for ductile steels, gives . When criteria disagree like this, a designer treats the part as marginal: reduce the stress range, improve the surface, or test it.
Variable loads: rainflow counting and Miner's rule
A crane jib, a car suspension arm or a wind-turbine blade sees an irregular history of peaks and valleys, not neat constant-amplitude cycles. Two tools turn that history into a life estimate.
Rainflow counting, devised by Tatsuo Endo and Masanori Matsuishi in Japan in 1968, breaks the history into closed stress–strain loops, each a cycle of definite range and mean; the name pictures rain running down a pagoda roof drawn from the history turned on its side. The algorithm keeps a stack of turning points and removes a cycle whenever the newest range is at least as large as the one before it.
Miner's rule (Arvid Palmgren, 1924, for ball bearings; Milton Miner, 1945, for aircraft) assumes that each cycle of a given range uses up a fraction of the life, where is the life at that range from the S–N curve, and that the fractions add:
For welded joints, codes give S–N curves in terms of stress range alone, because the residual stresses left by welding are already near the yield strength, so the applied mean makes little difference. The block counts the cycles in one lift of a crane, as measured at a weld on its jib, and adds up the damage on an illustrative welded-joint curve.
import numpy as np
def turning_points(history):
"""Keep only the peaks and valleys of a load history."""
h = [history[0]]
for v in history[1:]:
if v == h[-1]:
continue
if len(h) >= 2 and (h[-1] - h[-2]) * (v - h[-1]) > 0:
h[-1] = v # still rising (or falling): extend
else:
h.append(v)
return h
def rainflow(history):
"""Rainflow cycle counting (three-point method). Returns (range, mean, count) with count
1.0 for a full cycle and 0.5 for a half cycle left over at the end."""
stack, cycles = [], []
for p in turning_points(history):
stack.append(p)
while len(stack) >= 3:
X = abs(stack[-1] - stack[-2]) # the newest range
Y = abs(stack[-2] - stack[-3]) # the range before it
if X < Y:
break
if len(stack) == 3: # Y starts at the beginning: half cycle
cycles.append((Y, (stack[0] + stack[1]) / 2, 0.5))
stack.pop(0)
else: # Y is closed by X: a full cycle
cycles.append((Y, (stack[-2] + stack[-3]) / 2, 1.0))
del stack[-3:-1]
for s0, s1 in zip(stack, stack[1:]): # what is left: half cycles
cycles.append((abs(s1 - s0), (s0 + s1) / 2, 0.5))
return cycles
# A welded joint: S-N curve in terms of stress RANGE only, a straight line of slope 3 on
# log-log axes through an illustrative 80 MPa at 2 million cycles (codes give such curves,
# with refinements, for each kind of welded detail).
def cycles_to_failure(stress_range, S_ref=80.0, N_ref=2e6, m=3):
return N_ref * (S_ref / stress_range) ** m
# Stress (MPa) at a weld on a crane jib during one lift: pick up, swing, gusts, set down
lift = [0, 60, 35, 95, 50, 110, 65, 90, 40, 120, 70, 95, 20, 50, 0]
print(" range mean count cycles to failure damage")
D = D_big = 0.0
for r, m, n in sorted(rainflow(lift), reverse=True):
N = cycles_to_failure(r)
D += n / N
D_big += n / N if r == max(lift) - min(lift) else 0.0
print(f" {r:5.0f} {m:5.0f} {n:3.1f} {N:14,.0f} {n / N:.2e}")
print(f"\nMiner's sum for one lift D = {D:.3e}, so about {round(1 / D, -3):,.0f} lifts to failure")
print(f"the largest cycle alone gives {D_big / D:.0%} of the damage")
print(f"at 60 lifts a day, 250 days a year: {1 / D / (60 * 250):.1f} years")The fifteen readings of one lift contain one big cycle, from zero to the peak of 120 MPa and back, counted as two half cycles, and six smaller cycles of 25 to 70 MPa riding on it. With the slope of 3 typical of welds, damage goes as the cube of the range, so the big cycle does 77% of the damage and the three 25 MPa cycles together barely 2%. The joint would last about 458 000 lifts, some 30 years at 60 lifts a day, before any design factor is applied. Doubling the stress range would cut that by a factor of eight.
The Comet and other lessons
The de Havilland DH.106 Comet entered service with BOAC in 1952, flying higher and faster than any airliner before it, with a cabin pressurised to keep passengers comfortable at nearly 12 km. Each flight inflated the fuselage like a balloon and let it down again: one large stress cycle per flight. After the two break-ups of 1954, all Comets were grounded, and the Royal Aircraft Establishment at Farnborough did something new: it put a complete Comet fuselage in a water tank and pressurised it again and again, while jacks flexed the wings, to simulate flight after flight. After the equivalent of about three thousand pressurised flights, the fuselage split open from a fatigue crack at the corner of a window. Wreckage recovered from the sea near Elba showed that the first accident had started at the corner of a cut-out in the roof for a navigation aerial.
The inquiry, chaired by Lord Cohen, found that the stresses near the corners of the cabin's openings were much higher than the designers had believed, and that the fatigue tests done during development had misled them: a test section had first been loaded to a high proof pressure, which left beneficial residual stresses at the corners and made its fatigue life look far longer than a production aircraft's. The redesigned Comet 4 served airlines into the 1980s without a repeat of those failures, and the lessons outlived it:
- Full-scale fatigue testing of a whole airframe, representative of production and loaded as in service, became standard.
- Fail-safe and damage-tolerant design: structures with several load paths and crack stoppers, designed so that a crack big enough to find by inspection cannot cause a catastrophe before the next inspection.
- Detail design: generous radii and local reinforcement at every cut-out.
- Inspection programmes planned from crack-growth calculations, now required by aviation regulators.
Other failures taught related lessons. Welded ships of the Second World War, Liberty ships and T2 tankers among them, cracked in two by brittle fracture in cold water, from square hatch corners and weld defects. In 1980 the accommodation platform Alexander L. Kielland capsized in the North Sea after a bracing failed from a fatigue crack at a fillet weld around an instrument fitting; 123 people died. In 1988 an ageing Boeing 737 of Aloha Airlines lost part of its cabin roof in flight when many small fatigue cracks along riveted joints linked up, which led to new rules for ageing aircraft. In 1998 an ICE train derailed at Eschede in Germany after a fatigue crack broke a wheel tyre. Each investigation changed rules, inspections and designs.
Look at where all of these cracks started: a corner, a weld toe, a rivet hole, a fitting. Fatigue design is mostly detail design. The overall stress in the Comet's cabin was modest; the trouble was the few square centimetres around each opening, where the shape decided everything.
In professional practice
Codes and data. Pressure equipment follows the ASME Boiler and Pressure Vessel Code or the European Pressure Equipment Directive with its harmonised standards; steel structures follow the fatigue parts of Eurocode 3 or national equivalents, which classify each welded detail and give its design S–N curve; the International Institute of Welding publishes widely used fatigue recommendations; aircraft follow the certification rules of regulators such as the FAA and EASA. Material data come from tests to standard methods published by ASTM International and ISO, and from databases such as the MMPDS for aerospace metals.
Analysis. Finite element programs compute stresses at every point, including concentrations; fatigue post-processors, commercial and open source (Python packages such as rainflow and fatpack), then do the rainflow counting and damage summation over measured or simulated load histories. Crack-growth analysis uses the Paris law, which relates the crack's growth per cycle to the range of the stress intensity factor, to set inspection intervals.
Testing and inspection. Components that matter are tested, from coupons on servo-hydraulic machines to whole aircraft for several lifetimes, and in service non-destructive testing (dye penetrant, magnetic particle, ultrasound, eddy currents, X-ray) finds cracks before they become critical. Test laboratories guard their fatigue rigs and pressure tests, because parts fail suddenly and release stored energy; nobody stands near a pressure test or under a suspended test load.
Improving fatigue life. Shot peening and rolling put the surface into compression, polishing removes scratches, weld toes can be ground or peened, and a few millimetres more fillet radius often does more than a stronger material.
Common misconceptions
Misconception
"A part stressed below its yield strength will last for ever." Under repeated loading, cracks can start and grow at stresses well below yield. Only below the endurance limit, for materials that have one, in a benign environment and at constant amplitude, is life practically unlimited.
Misconception
"Von Mises is the safer criterion." Tresca is the more conservative of the two: its hexagon lies inside the von Mises ellipse. Von Mises is usually the more accurate for ductile metals, which is why it is the default.
Misconception
"A stronger steel always gives a longer fatigue life." Smooth specimens of stronger steels do have higher endurance limits, but notched and welded parts gain much less: high-strength steels are more notch-sensitive, and the fatigue strength of welded joints hardly depends on the steel's strength at all.
Misconception
"A factor of safety of 2 means the part can carry twice the load." Only if the failure mode scales linearly with load. Buckling, fatigue (where life falls steeply with stress), and loads that change the geometry do not, and a factor on load is not a factor on life.
Exercises
1. Failure and fatigue
Warm-upHumanA point on the surface of a rotating shaft carries a bending stress MPa and a torsional shear stress MPa. What is the Tresca equivalent stress there, in MPa?
2. Failure and fatigue
CoreHumanA spring steel part has a tensile strength of 600 MPa and a corrected endurance limit of 200 MPa. It carries a steady mean stress of 240 MPa. Using the Goodman line with a factor of safety of 1, what is the largest stress amplitude it can take for infinite life, in MPa?
3. Failure and fatigue
CoreHumanEach year a bracket sees three kinds of load cycle:
| cycles per year | cycles to failure at that level |
|---|---|
| 5 000 | 50 000 |
| 40 000 | 800 000 |
| 1 000 000 | below the endurance limit |
Using Miner's rule, after how many whole years of service is the damage sum still below 1 for the last time? (That is, the number of complete years the bracket survives.)
4. Failure and fatigue
CoreHumanA wide plate of an aluminium alloy with fracture toughness carries a tensile stress of 150 MPa. A through crack lies across the load in the middle of the plate. Taking , where is the crack's half-length, what is the total crack length at which the plate fractures, to the nearest mm?
5. Failure and fatigue
Warm-upHumanA wide plate in tension at a remote stress of 50 MPa has an elliptical hole 12 mm wide across the load and 4 mm high along the load. Using , what is the peak stress at the edge of the hole, in MPa?
6. Failure and fatigue
CoreHumanWhich of these statements about fatigue are true? Select all that apply.
7. Failure and fatigue
CoreComputationalWrite safety_factors(S, Sy) that takes a symmetric 3 × 3 stress tensor S (MPa; a NumPy array or nested list) and a yield strength Sy (MPa), and returns a tuple (n_tresca, n_mises): the factors of safety against yield by the Tresca and von Mises criteria, . If an equivalent stress is zero (for example, pure hydrostatic pressure), return math.inf for that factor.
Use all three principal stresses: the tests include a plane stress state where the largest shear acts out of the plane.
8. Failure and fatigue
StretchComputationalWrite fatigue_life(Sa, Sm, Sut, Se) returning the predicted number of cycles to failure (a float, or math.inf) for a steel part loaded with stress amplitude Sa about a mean Sm (MPa), using the lesson's model:
- Goodman correction: the equivalent fully reversed amplitude is
Sar = Sa / (1 - Sm / Sut)whenSm > 0; for a zero or compressive mean, useSar = Sa. - S–N curve: a straight line on log–log axes, , through and .
- If
Sar <= Se, the life is infinite (math.inf).
Further reading and resources
- ASTM InternationalPublishes the test methods for fatigue (for example E466), fracture toughness (E399) and cycle counting (E1049).
- ASMEPublishes the Boiler and Pressure Vessel Code, which sets allowable stresses and fatigue rules for pressure equipment.
- European Union Aviation Safety Agency (EASA)Publishes the certification specifications for large aeroplanes, including the rules on fatigue and damage tolerance.
- U.S. Federal Aviation Administration (FAA)Its airworthiness regulations and advisory circulars set the 1.5 ultimate factor of safety and the damage-tolerance requirements for transport aircraft.
- A history of fatigue · Walter Schütz · 1996A review in Engineering Fracture Mechanics, volume 54, tracing fatigue research from Wöhler's railway axles to modern damage tolerance.
- Report of the Court of Inquiry into the Accidents to Comet G-ALYP and Comet G-ALYY · 1955The UK inquiry chaired by Lord Cohen, published by Her Majesty's Stationery Office; widely summarised in histories of aviation safety.