Strong is not the same as tough
A freshly drawn glass fibre a few micrometres across can carry a stress of several gigapascals, more than most steels. A window pane made of the same glass breaks at a few tens of megapascals. Scratch a glass rod with a diamond and it snaps cleanly along the scratch at the lightest bend; scratch a steel rod and nothing happens. The difference is not strength but toughness: the resistance of a material to the growth of a crack that is already there.
Toughness is the property that most often decides whether a failure is a nuisance or a disaster. A tough part dents, bends or leaks before it breaks, and gives warning. A brittle one fails suddenly, at a stress that may be far below its strength in a tensile test, because a flaw too small to see has concentrated the load. Lumenary ME's failure and fatigue introduced the engineer's tool for this, the stress intensity factor and the critical crack length. This lesson explains what happens at the tip of the crack: why toughness differs ten-thousand-fold between materials, why some steels lose it in the cold, and how ceramics, composites, bone and shell are made tough.
Two ways to break
Look at a broken part through a magnifying glass, or better a scanning electron microscope, and the fracture surface tells you how it broke.
Ductile fracture is a slow tearing. Inside a stretched metal, hard particles (inclusions of oxides and sulfides, carbides) crack or come loose from the metal around them, leaving tiny voids. The voids grow as the metal flows plastically around them and finally join up. Under the microscope the surface is covered with dimples, the halves of the burst voids, and a tensile bar that broke this way shows the classic cup and cone: a neck, a fibrous centre and a sheared rim. Ductile fracture absorbs a great deal of energy, because a large volume of metal deforms plastically before it separates.
Brittle fracture happens with little or no plastic deformation. In crystals it is usually cleavage: the crack runs along particular crystal planes, such as in iron, breaking bonds one row at a time. Each grain leaves a flat, shiny facet, and steps on the facets form "river patterns" that point back towards the origin. Glass and amorphous polymers have no planes to follow, and break with smooth, mirror-like surfaces. If grain boundaries are weakened, by segregated impurities such as phosphorus in a temper-embrittled steel, or by hydrogen, the crack instead follows the boundaries (intergranular fracture) and the surface looks like a pile of rock candy.
Which mechanism wins depends on a race at the crack tip. If the material can flow plastically before the bonds there reach their breaking stress, the tip blunts, energy is soaked up and the material is tough. If the bonds break first, the crack runs. Metals with easy slip (most FCC metals) almost always win the race for plasticity; covalent ceramics, with their enormous Peierls stresses (Elasticity, plasticity and strengthening), almost always lose it; and BCC steels can do either, depending on the temperature.
Griffith's energy balance
The previous lesson showed that a perfect crystal should be a tenth to a thirtieth as strong as it is stiff. For glass, with GPa, that is several gigapascals, yet windows break at a hundredth of that. In 1921 A. A. Griffith, working on the strength of aircraft materials at the Royal Aircraft Establishment in Britain, explained why with an argument about energy rather than stress.
Theorem— Griffith's criterion
A through crack of length in a large, ideally brittle plate under a remote tensile stress (plane stress) grows when
where is the surface energy per unit area of each new crack face.
Proof
Work per unit thickness of the plate. The crack has two faces, each of length , which cost of surface energy. It also lets the material around the crack relax: Inglis's elastic solution for an elliptical hole shows that the strain energy stored in the plate falls by when a crack of length is introduced at fixed remote stress. The energy of the system, relative to the uncracked plate, is therefore
The first term grows linearly and the second quadratically, so rises to a maximum and then falls. A crack shorter than the maximum would raise the energy by growing, so it is stable; a longer one lowers the energy by growing, so it runs. The maximum is where , that is, .
∎
The criterion says that a crack's danger grows with the square root of its length, and that a brittle material's strength is not a property of the material alone but of the material and its largest flaw. Griffith tested it by breaking glass bulbs and tubes with cracks of known length, and showed that his freshly drawn thin fibres, with fewer and smaller flaws, were many times stronger than ordinary glass rods.
Definition— Energy release rate and crack resistance
The energy release rate is the elastic energy made available per unit area of crack growth; for the Griffith crack, . The crack grows when reaches the material's crack resistance , the energy it absorbs per unit area of crack. For an ideally brittle solid, .
For metals Griffith's formula with the true surface energy, about 1–2 J/m², predicts absurdly low strengths. Egon Orowan and George Irwin explained why in the late 1940s: in a metal the crack tip is surrounded by a zone of plastic flow, and the plastic work done there, per unit area of crack, is hundreds or thousands of times the surface energy. Griffith's balance still holds, but with , where the plastic term dominates. Toughness is mainly the energy a material can dissipate in front of a crack.
Computing it: how strong is a window?
Fracture toughness tests on soda-lime glass give MPa·m, and the relation (next section) converts that into a crack resistance. The program then applies Griffith's criterion to flaws from a few nanometres to a millimetre, and checks the energy balance numerically at a stress of 50 MPa.
import numpy as np
# Griffith's criterion for window glass
E = 70e9 # Pa, Young's modulus of soda-lime glass
K_Ic = 0.75e6 # Pa m^0.5, a typical measured fracture toughness
G_c = K_Ic**2 / E # J/m^2 needed to make one square metre of crack (two new faces)
gamma = G_c / 2 # J/m^2 per face
print(f"crack resistance G_c = {G_c:.1f} J/m^2, so gamma = {gamma:.1f} J/m^2 per face")
print("\nhalf-length of crack strength sqrt(2 E gamma / (pi a))")
for a in [3e-9, 1e-8, 1e-7, 1e-6, 1e-5, 1e-4, 1e-3]:
s = np.sqrt(2 * E * gamma / (np.pi * a))
print(f" {a*1e6:10.3f} um {s/1e6:8.0f} MPa")
# The energy balance itself, per metre of plate thickness, at sigma = 50 MPa
sigma = 50e6
a = np.linspace(1e-6, 3e-4, 300_000)
W = 4 * gamma * a - np.pi * sigma**2 * a**2 / E # surface energy in, elastic energy out
i = np.argmax(W)
print(f"\nat {sigma/1e6:.0f} MPa the energy W(a) peaks at a = {a[i]*1e6:.1f} um "
f"(formula 2 E gamma / (pi sigma^2) = {2*E*gamma/(np.pi*sigma**2)*1e6:.1f} um)")
print(f"barrier height {W[i]*1e6:.2f} uJ per metre of crack front")
for frac in [0.25, 0.5, 1.0, 1.5, 2.2]:
j = np.argmin(abs(a - frac * a[i]))
bar = int(40 * W[j] / W[i]) if W[j] > 0 else 0
print(f" a = {a[j]*1e6:6.1f} um W = {W[j]*1e6:6.2f} uJ/m " + "#" * bar)A crack resistance of 8 J/m² (4 J/m² per face) is close to what bond-breaking alone would cost: glass has almost no way to dissipate energy. A crack with a half-length of one micrometre already limits the strength to about 420 MPa, and at the 50 MPa a window might see in a storm, a crack with a half-length of about 72 µm, too small to notice, is enough to run. Only flaws a few nanometres long, the size of the atomic structure itself, would let glass approach its ideal strength of 7–8 GPa: that is where Griffith's continuum argument and the bond-breaking estimate meet. The energy barrier at the critical length is tiny, under a millijoule per metre of crack front, which is why a crack, once critical, accelerates to a large fraction of the speed of sound.
Stress intensity, plastic zones and fracture toughness
Irwin's insight, at the US Naval Research Laboratory in the 1950s, was that the elastic stresses near any crack tip have the same shape,
where is the distance from the tip, are fixed functions of angle, and all the information about the load and geometry is in one number, the stress intensity factor . Two cracks with the same have identical surroundings, whatever the part. Energy and stress views agree: under plane stress , and under plane strain . Loads that open the crack faces are mode I; in-plane sliding is mode II and tearing out of plane is mode III. Mode I is the most dangerous and most often tested, which is why the toughness in tables is .
The singularity cannot be literally true: no material carries an infinite stress. In a metal, the stress near the tip is capped at around the yield stress, inside a plastic zone of radius roughly
As long as the zone is small compared with the crack and the part, the field surrounds it and controls it, and linear-elastic fracture mechanics (LEFM) works. The plastic zone is where toughness is earned: everything the material does inside it, slipping, voiding, transforming, sets .
Thickness matters. In a thin sheet the material near the tip can contract freely through the thickness (plane stress), the plastic zone is large and the apparent toughness high. In a thick plate the surrounding material prevents that contraction (plane strain), the stress state at the tip becomes strongly triaxial, plasticity is suppressed and the toughness falls to a lower limit, . Standard tests (ASTM E399, ISO 12135) therefore require the specimen thickness to exceed . For tough materials where that is impossibly thick, the test measures instead the J-integral, a generalisation of to elastic–plastic material, or the crack-tip opening displacement (ASTM E1820). The cheaper Charpy test (ISO 148-1, ASTM E23), in which a swinging hammer breaks a notched bar 10 mm square and the energy it absorbs is recorded, does not give , but it is quick, and it is the standard way to check a steel's behaviour across a range of temperatures.
The program uses typical, rounded values for four structural metals to compute what these formulas mean in millimetres.
import numpy as np
# Typical, rounded room-temperature values: K_Ic in MPa m^0.5, yield strength in MPa, E in GPa.
# Real values depend on grade, condition, orientation and temperature.
materials = [
("high-strength aluminium alloy", 27, 500, 71),
("titanium alloy Ti-6Al-4V", 60, 900, 114),
("quenched and tempered steel", 50, 1500, 205),
("structural steel, upper shelf", 150, 355, 205),
]
print(f"{'material':31} {'r_p plane':>10} {'B for E399':>11} {'a at yield':>11} {'G_c':>8}")
print(f"{'':31} {'strain mm':>10} {'mm':>11} {'mm':>11} {'kJ/m^2':>8}")
for name, K, sy, E in materials:
ratio2 = (K / sy) ** 2 # m, since (MPa m^0.5 / MPa)^2 = m
r_p = ratio2 / (6 * np.pi) # plane-strain plastic zone radius (Irwin)
B = 2.5 * ratio2 # ASTM E399 minimum thickness for a valid K_Ic
a_y = ratio2 / np.pi # crack at which fracture and yield coincide (Y = 1)
nu = 0.3
G_c = (K * 1e6) ** 2 * (1 - nu**2) / (E * 1e9) # plane strain, J/m^2
print(f"{name:31} {r_p*1e3:10.2f} {B*1e3:11.1f} {a_y*1e3:11.1f} {G_c/1e3:8.1f}")
print(f"\nfor comparison, glass: G_c = {(0.75e6)**2 / 70e9:.0f} J/m^2")The column "a at yield" is the crack length at which the fracture stress equals the yield stress. With shorter cracks the part yields before it fractures, which gives warning; with longer ones it fractures first, without warning. The high-strength steel is the most flaw-sensitive: a crack of 0.4 mm makes it brittle in effect, though its toughness is respectable. The structural steel tolerates cracks of almost 6 cm, and a valid test on it would need a specimen 45 cm thick, which is why its toughness is measured with or CTOD instead. Its crack resistance, about 100 kJ/m², is more than ten thousand times that of glass, nearly all of it plastic work.
The ductile-to-brittle transition and the Liberty ships
Many steels are tough at room temperature and brittle in the cold. The Charpy energy of a ferritic steel follows an S-shaped curve: a high upper shelf where it tears in a ductile way, a low lower shelf where it cleaves, and a transition between them, often only a few tens of degrees wide.
The cause is the Peierls stress of BCC metals. Their yield strength rises steeply as the temperature falls, because thermal vibrations no longer help screw dislocations over the lattice's hills, while the stress needed to propagate a cleavage crack barely changes. At the root of a notch, constraint raises the local stress to two or three times the yield stress. When that local stress reaches the cleavage stress before the metal can yield and blunt the notch, the steel breaks by cleavage. Anything that raises the yield stress relative to the cleavage stress, such as cold, a high loading rate, a sharp notch or a thick section, moves the transition to higher temperatures. Anything that makes cleavage harder moves it down: fine grains (each boundary forces the crack to restart), nickel, and low levels of carbon, phosphorus and sulfur. FCC metals such as aluminium, copper and austenitic stainless steel have no steep rise in yield stress and no transition, which is why tanks for liquefied natural gas at −162 °C are built from 9% nickel steel, austenitic stainless steel or aluminium alloys, never ordinary carbon steel.
Example— The Liberty ships
During the Second World War, American shipyards built about 2,700 Liberty ships, together with many T2 tankers, at unprecedented speed. They were among the first ships with fully welded hulls, which saved time and steel. Hundreds suffered serious cracks, and some broke completely in two. On 16 January 1943 the new T2 tanker Schenectady split almost in half while moored at her fitting-out dock in Portland, Oregon, in calm, cold weather.
The investigations, in which Constance Tipper at the University of Cambridge played a leading part, found three causes acting together. The ship steel of the time had a transition temperature near the temperature of winter seas, so it was brittle in service. The welded hull was one continuous piece of metal, so a crack that started had nothing to stop it, whereas in a riveted ship it would have stopped at the edge of a plate. And the cracks started at stress concentrations: square hatch corners and weld defects. Tipper's notched tests showed that the same steel broke in a ductile manner above a fairly sharp temperature and by cleavage below it, and that the transition temperature depended on the steel.
The remedies are still the basis of practice: steels specified by Charpy energy at a low temperature, rounded corners, better welding and inspection, and tough "crack arrestor" strakes in the hull.
The transition has not gone away; it is managed. Steel specifications such as EN 10025 encode it in their grade names (S355J2 must absorb at least 27 J at −20 °C), and in nuclear reactors, neutron irradiation slowly raises the transition temperature of the pressure vessel steel, so capsules of Charpy specimens are placed inside the vessel and tested at intervals throughout its life.
Toughening ceramics, composites and natural materials
A ceramic cannot be toughened by making it more plastic, because its dislocations will not move. Robert Ritchie and others classify toughening mechanisms into two kinds. Intrinsic mechanisms act ahead of the crack tip and make the material itself harder to break; plasticity is the main one, available to metals. Extrinsic mechanisms act behind the tip and shield it from part of the applied load, and they are the ceramic designer's tools:
- Transformation toughening. Zirconia () stabilised with a little yttria can be kept in a tetragonal form that wants to transform to a monoclinic form with a volume increase of about 4%. In the high stress ahead of a crack, it does, and the expanding particles squeeze the crack shut. Garvie, Hannink and Pascoe, in Australia, called their 1975 paper "Ceramic steel?". Zirconia's toughness reaches 5–10 MPa·m or more, several times that of alumina, which is why it is used in hip-joint heads, dental crowns and cutting blades.
- Crack deflection and bridging. Silicon nitride grown with elongated, interlocking grains forces a crack to zigzag, and unbroken grains bridge the crack behind its tip, pulling its faces together.
- Fibre bridging and pull-out. In a ceramic-matrix composite, fibres with a weak coating (pyrolytic carbon or boron nitride) let the crack pass around them; the fibres stay intact behind the tip and are then pulled out against friction, absorbing energy. Silicon carbide fibres in silicon carbide have been used since 2016 in the hottest parts of some jet engines (Composites).
Extrinsic toughening produces a rising R-curve: the resistance to cracking grows as the crack extends, because the shielding zone behind the tip builds up. Such materials tolerate damage far better than their initial toughness suggests.
Nature is the master of extrinsic toughening. Nacre, the mother-of-pearl lining of shells, is about 95% brittle aragonite, a form of calcium carbonate, in microscopic tablets glued by a few per cent of protein; tablets slide and interlock, cracks deflect, and nacre is far tougher than the mineral it is made of. Bone combines collagen fibrils with mineral crystals in a hierarchy of structures, so a crack in bone twists, is bridged by fibrils and is deflected by the boundaries of osteons. Materials scientists now copy these designs in "bioinspired" ceramics and freeze-cast composites (Biomaterials).
Glass, finally, is made safer not by raising its toughness but by putting its surface into compression. In toughened (tempered) glass the hot surface is cooled quickly with air jets; the surface freezes first, and when the inside contracts later it pulls the surface into compression of the order of 100 MPa. A surface flaw must overcome that compression before it can open. Chemically strengthened glass for phone screens is treated in a bath of molten potassium salt, where larger potassium ions replace sodium ions in the surface and wedge it into compression. Laminated glass, with a polymer interlayer, holds the pieces together when it does break.
Computing it: brittle strength is a statistic
If a brittle part breaks at its worst flaw, its strength depends on how large the largest flaw happens to be, and nominally identical specimens scatter. Waloddi Weibull, in Sweden in 1939, described this with a "weakest link" distribution: the probability that a part of volume fails at a stress or below is
where the Weibull modulus measures the scatter: about 5–20 for engineering ceramics, the higher the better, and far higher for ductile metals. The program builds bars of an alumina-like ceramic out of random flaws, breaks thirty of them as a laboratory would, fits from the classic Weibull plot of against , and then tests the prediction that bigger parts are weaker.
import numpy as np
# Weakest-link strength of a ceramic: each bar contains many flaws; the largest one breaks it.
rng = np.random.default_rng(7)
K_Ic = 4.0e6 # Pa m^0.5, an alumina-like toughness
a0 = 5e-6 # m, smallest flaw size considered
k = 5.0 # flaw-size tail: P(flaw > a) = (a0 / a)^k
def bar_strengths(n_bars, flaws_per_bar):
a = a0 * (1 + rng.pareto(k, size=(n_bars, flaws_per_bar)))
return (K_Ic / (1.12 * np.sqrt(np.pi * a))).min(axis=1) # surface flaws, Y = 1.12
s = np.sort(bar_strengths(30, 200)) # 30 test bars, as in a standard test series
F = (np.arange(1, 31) - 0.5) / 30 # failure probability assigned to each rank
x, y = np.log(s), np.log(-np.log(1 - F))
m, c = np.polyfit(x, y, 1) # Weibull plot: slope = Weibull modulus
s0 = np.exp(-c / m)
print(f"30 bars: weakest {s[0]/1e6:.0f} MPa, median {np.median(s)/1e6:.0f} MPa, strongest {s[-1]/1e6:.0f} MPa")
print(f"Weibull modulus m = {m:.1f} (theory for this flaw population: 2k = {2*k:.0f}); "
f"characteristic strength {s0/1e6:.0f} MPa")
# Size effect: a part with 10 times the stressed volume has 10 times as many flaws
small = bar_strengths(500, 200)
big = bar_strengths(500, 10 * 200)
print(f"\nmedian strength: test bar {np.median(small)/1e6:.0f} MPa, part 10x larger {np.median(big)/1e6:.0f} MPa")
print(f"ratio {np.median(big)/np.median(small):.3f}; Weibull predicts 10^(-1/m) = {10**(-1/(2*k)):.3f}")
for p in [0.5, 0.01, 1e-4]:
print(f"large part, failure probability {p:g}: stress {s0 * 10**(-1/m) * (-np.log(1 - p))**(1/m) / 1e6:.0f} MPa")Thirty bars range from about 355 to 590 MPa, and the fitted modulus, 11.4, is close to the value of 10 that this flaw population should give: thirty specimens pin down only to within roughly 20–30%, which is why standards such as ASTM C1239 ask for at least thirty and report confidence bounds. A part with ten times the stressed volume has a median strength 21% lower, exactly as Weibull predicts. And the design stress for a failure probability of one in ten thousand, 190 MPa, is less than half the median strength. Brittle design is design for the weakest of very many flaws, which is why ceramic parts are often proof-tested like optical fibres.
Designing against fracture
The tools of this lesson add up to a discipline, damage-tolerant design, which assumes that cracks exist and asks how large they may grow:
- Choose materials by toughness, not just strength. For a given working stress, the critical crack length scales as . A design that must yield visibly before it breaks needs a high ; a part that must survive impacts needs a high , the energy it can absorb per unit area of crack.
- Keep the temperature in mind. Check the toughness at the lowest service temperature, and for ferritic steels specify Charpy energy at or below it.
- Inspect, and prove what inspection can find. Ultrasonic, radiographic, eddy-current and dye-penetrant inspection find cracks above some size. If the critical crack length is larger than that size, with margin, and the time a crack takes to grow between the two (calculated in Fatigue, creep and corrosion) exceeds the inspection interval, the structure is safe. Aircraft, pressure equipment and pipelines are managed this way, and ME's failure and fatigue describes leak-before-break.
- Assess flaws that are found. Fitness-for-service procedures such as BS 7910, API 579-1/ASME FFS-1 and the R6 procedure use a failure assessment diagram, which combines the risk of fracture (as a ratio of to toughness) with the risk of plastic collapse, to decide whether a cracked component may stay in service.
- Avoid stress concentrations and residual tension: rounded corners, smooth welds, and post-weld heat treatment where codes require it.
Common misconceptions
Misconception
"Strong materials are tough." Strength and toughness are often in conflict: the mechanisms that stop dislocations also stop the crack tip from blunting. Quenching a steel to full hardness raises its strength and can lower its toughness sharply, which is why it is tempered afterwards.
Misconception
"Brittle means weak." Glass fibres, silicon wafers and alumina can be very strong. Brittle means flaw-sensitive: the strength is set by the largest flaw and scatters accordingly, and there is no warning before failure.
Misconception
"Toughness is the area under the stress–strain curve." That area, the energy per unit volume to break an unflawed tensile specimen, is sometimes called tensile toughness. Fracture toughness, or , measures resistance to the growth of an existing crack. The two can rank materials differently.
Misconception
"Charpy energy is a design value." The Charpy test uses a blunt notch, a small specimen and an impact, so its energy is a comparative measure for specifying and checking steels. Correlations between Charpy energy and exist, but they are empirical and approximate, and safety-critical assessments use measured fracture toughness.
Exercises
1. Fracture and toughness
Warm-upHumanA failure investigator examines a broken steel bolt in a scanning electron microscope. The surface is covered with small cup-shaped hollows, many with a tiny particle at the bottom. How did the bolt break?
2. Fracture and toughness
CoreHumanA glass panel ( GPa, surface energy J/m² per face) carries a tensile stress of 30 MPa. Using Griffith's criterion for a through crack of length , what is the critical half-length , in micrometres, rounded to the nearest whole number?
3. Fracture and toughness
CoreHumanA steel has a plane-strain fracture toughness MPa·m, GPa and . What is its crack resistance in kJ/m², rounded to the nearest whole number?
4. Fracture and toughness
CoreHumanA laboratory wants to measure a valid plane-strain (ASTM E399) on a steel with yield strength 600 MPa, expecting MPa·m. What is the minimum specimen thickness, in millimetres, that the standard's size requirement demands?
5. Fracture and toughness
CoreHumanWhich of these materials are suitable for the inner tank of a store for liquefied natural gas at −162 °C? Select all that apply.
6. Fracture and toughness
CoreComputationalWrite weibull_fit(strengths), which estimates the Weibull modulus and characteristic strength from a list of measured fracture strengths (in any order) by the classic Weibull plot:
- sort the strengths and give the -th smallest () the failure probability ;
- fit a straight line with and ;
- return
(m, sigma0)with .
Try it on ten flexure tests of a ceramic (made-up practice data): 512, 488, 530, 455, 501, 547, 476, 519, 495 and 463 MPa.
7. Fracture and toughness
StretchComputationalCharpy results across the transition are often fitted with , where are the upper and lower shelves, the mid-transition temperature and its half-width. Write transition_temperature(A, B, T0, C, target=27.0) that returns the temperature at which the fitted curve equals target joules, and raises ValueError if the curve never reaches that energy. For J, J, °C, °C (the curve in the lesson's figure) the answer is about °C.
Further reading and resources
- The phenomena of rupture and flow in solids · A. A. Griffith · 1921The founding paper of fracture mechanics, with Griffith's experiments on glass.
- Analysis of stresses and strains near the end of a crack traversing a plate · G. R. Irwin · 1957Journal of Applied Mechanics, volume 24; introduced the stress intensity factor.
- The Brittle Fracture Story · Constance Tipper · 1962Cambridge University Press. Tipper's own account of the welded-ship failures and the ductile-to-brittle transition.
- Ceramic steel? · R. C. Garvie, R. H. Hannink and R. T. Pascoe · 1975The short Nature paper that announced transformation toughening in zirconia.
- The conflicts between strength and toughness · R. O. Ritchie · 2011A review of intrinsic and extrinsic toughening, from metals to bone and nacre.
- A fracture-resistant high-entropy alloy for cryogenic applications · B. Gludovatz and others · 2014An FCC alloy that becomes tougher as it gets colder.
- ASTM InternationalTest standards for fracture toughness (E399, E1820), Charpy impact (E23) and the strength statistics of ceramics (C1239).
- TWIA research organisation for welding and structural integrity, with accessible technical knowledge articles on fracture and Charpy testing.