Skip to content
LumenaryAcademy
Families 75 min

Ceramics and glasses

Why ceramics are hard, stiff and brittle; alumina, silicon carbide and zirconia; how glass is made and toughened; the chemistry and carbon footprint of cement and concrete; and the Weibull statistics engineers use to design with brittle materials.

Code tested · 6AI-draftedAwaiting review by a materials scientist

Your guides: Chloe · The Materials Scientist, Margaret · The Tester, Gideon · The Failure Analyst

MotivationH

The oldest engineered materials, and some of the newest

People were firing clay long before they smelted metal. Pottery fragments from Xianrendong Cave in China are about 20,000 years old (Wu et al.), and fired clay figurines from Moravia are older still. Glass followed in Mesopotamia more than four thousand years ago. Today the same family of materials includes the spark-plug insulator, the hip-joint ball, the phone's cover glass, the coating on a turbine blade, the silicon carbide chip in an electric car's inverter — and concrete, the most used material on Earth after water.

Ceramics and glasses share a character: hard, stiff, heat-resistant, chemically stable, and brittle. That last word dominates their engineering. A steel beam that is overloaded bends; a ceramic one breaks without warning, at a stress that varies from one piece to the next. This lesson explains where that character comes from, tours the main ceramics and glasses, follows the chemistry of cement and its carbon footprint, and ends with the statistics engineers use to design safely with materials whose strength is a matter of chance.

IntuitionH

Structure and bonding: why ceramics are brittle

A ceramic is an inorganic, non-metallic solid, usually a compound of a metal or semi-metal with oxygen, nitrogen or carbon, held together by ionic and covalent bonds (bonding in solids). Alumina (Al₂O₃) and zirconia (ZrO₂) are mostly ionic; silicon carbide (SiC) and silicon nitride (Si₃N₄) are mostly covalent. A glass is a solid without long-range order (glasses and nanomaterials); the commercial ones are silicates.

Strong bonds give high melting points (alumina melts at 2072 °C), high stiffness (alumina's Young's modulus is around 380 GPa, nearly twice that of steel), great hardness, and resistance to heat and chemicals. The same bonds make dislocations nearly immobile at room temperature. In an ionic crystal, sliding one plane of ions over the next by a part of a lattice spacing brings like charges face to face; in a covalent crystal, slip means breaking strong, directional bonds. With dislocations stuck, a ceramic cannot blunt a crack tip by yielding, and fracture mechanics from fracture and toughness takes over: a crack of depth aa grows when YσπaY\sigma\sqrt{\pi a} reaches the fracture toughness KIcK_{Ic}.

MaterialKIcK_{Ic} (MPa√m, typical)
soda-lime glass0.7–0.8
alumina3–4.5
silicon carbide3–4
silicon nitride5–7
yttria-stabilised zirconia (Y-TZP)5–10
structural steel50–150

Two consequences follow. Ceramics are far stronger in compression, which closes cracks, than in tension, which opens them; concrete, brick and stone have always been used where they are squeezed. And their tensile strength scatters from specimen to specimen, because flaw sizes scatter. The last section of this lesson turns that scatter into a design method.

DefinitionsH

Traditional and advanced ceramics

Traditional ceramics are made from clay, the platy aluminosilicate mineral kaolinite, Al₂Si₂O₅(OH)₄, with quartz and feldspar. Wet clay is plastic and can be shaped; drying removes free water; firing drives out the chemically bound water near 500 °C, forms needles of mullite above about 1000 °C, and melts the feldspar into a glass that binds everything together (vitrification). Fired cooler and shorter, the body stays porous (earthenware, brick); hotter and longer, it becomes dense and translucent (stoneware, then porcelain, perfected in China over many centuries and first reproduced in Europe, at Meissen, around 1710). Refractories, the bricks that line furnaces and kilns, are traditional ceramics chosen for heat resistance.

Advanced (technical) ceramics are made from purified or synthetic powders, shaped by pressing, slip casting, injection moulding, tape casting or 3D printing, and then sintered: heated, usually to 1400–1800 °C, until diffusion fills the pores between particles, driven by the reduction of surface area (processing and heat treatment). A few stand out.

  • Alumina (Al₂O₃): cheap, hard, an excellent electrical insulator. Spark-plug insulators, circuit substrates, wear parts, cutting-tool inserts and the balls of hip replacements.
  • Silicon carbide (SiC): very hard, conducts heat well, keeps its strength above 1400 °C. Abrasives since 1891, kiln furniture and brake discs. As a single crystal it is a wide-band-gap semiconductor, and SiC transistors now switch the power in many electric-car inverters and solar converters (see power electronics in Lumenary EE).
  • Silicon nitride (Si₃N₄): tough for a ceramic, with interlocking elongated grains; bearing balls, cutting tools and engine parts.
  • Zirconia (ZrO₂): the toughest of the oxide ceramics, thanks to a phase transformation.

Pure zirconia is monoclinic at room temperature and tetragonal above about 1170 °C, and the tetragonal-to-monoclinic transformation on cooling expands it by about 4 %, enough to shatter a sintered piece. Adding about 3 mol % of yttria keeps fine grains tetragonal at room temperature, in a metastable state. In 1975, Ron Garvie and colleagues in Australia reported that such zirconia was surprisingly tough (Garvie, Hannink and Pascoe). The reason is shown below.

□ metastable tetragonal grain■ transformed to monoclinic (+4 % volume)dashed line: transformed zone; arrows: the zone squeezes the crack shutcrack
Transformation toughening in yttria-stabilised zirconia, schematic. The high tensile stress near the crack tip triggers metastable tetragonal grains to transform to the monoclinic phase, which takes up about 4 % more volume. The transformed zone, which the crack drags along behind it, pushes on the crack faces and squeezes them shut, so a larger applied load is needed to make the crack grow.

Transformation toughening raises zirconia's toughness to two or three times alumina's. Y-TZP makes dental crowns, knife blades and ferrule sleeves for optical connectors, and zirconia-toughened alumina makes many ceramic hip balls (biomaterials). With more yttria (about 8 mol %), zirconia becomes fully cubic and full of oxygen vacancies, a solid that conducts oxide ions: the electrolyte of car oxygen sensors and solid-oxide fuel cells. And as a porous coating it is the thermal barrier on turbine blades you met in metals and alloys.

Worked examplesH

Making glass, and making it tougher

Window glass is soda-lime-silica glass: about 72 % silica (SiO₂), 13–14 % soda (Na₂O), 9 % lime (CaO) and a few per cent of magnesia and alumina. Silica alone would need about 2000 °C to melt and work; the soda breaks up the silicate network and lowers the working temperature to around 1000 °C, and the lime makes the glass resistant to water. Recycled glass, cullet, melts at a lower temperature than the raw materials, so every tonne used saves energy and the CO₂ from decomposing carbonates.

Almost all flat glass is made by the float process, developed by Alastair Pilkington and Kenneth Bickerstaff in Britain in the 1950s. Molten glass pours from the furnace onto a bath of molten tin, spreads into a ribbon with perfectly flat, parallel faces, cools until it is stiff enough to touch rollers, and passes through a long annealing lehr that cools it slowly through the range where stresses could freeze in. The industry describes that range by viscosity: about 10310^3 Pa s at the working point, 106.610^{6.6} at the softening point, 101210^{12} at the annealing point, where stresses relax in minutes, and 1013.510^{13.5} at the strain point, below which they cannot relax at all.

Other compositions serve other needs. Borosilicate glass (about 80 % silica and 13 % boron oxide) expands only a third as much as soda-lime glass on heating, so it survives sudden temperature changes in laboratories and kitchens. Aluminosilicate glasses make phone covers. Glass-ceramics, glasses crystallised under control, can be made with almost zero thermal expansion for cooktops and telescope mirrors.

Thermal tempering

Since glass fails from surface flaws opened by tension, the way to strengthen it is to put its surface into compression. In thermal tempering, a glass sheet is heated to about 620–650 °C, where it is just soft, and its surfaces are blasted with cold air. The surfaces cool and stiffen first while the core is still soft. When the core cools in turn, it contracts and pulls the rigid surfaces into compression, leaving itself in tension. The stress through the thickness is close to a parabola, and force balance fixes its shape.

python
import numpy as np

# Residual stress through a thermally tempered glass plate (tension positive), modelled as a
# parabola: sigma(u) = sigma_mid - (sigma_mid + sigma_surf) * u**2, with u = 2z/t from -1 to 1.
t = 6.0             # plate thickness, mm
sigma_surf = 100.0  # surface compression, MPa (fully tempered glass is typically 80-150 MPa)

# Force balance: the stress must integrate to zero through the thickness.
# The integral of sigma over u in [-1, 1] is 2*sigma_mid - (2/3)*(sigma_mid + sigma_surf) = 0.
sigma_mid = sigma_surf / 2
u = np.linspace(-1, 1, 4001)
sigma = sigma_mid - (sigma_mid + sigma_surf) * u**2
net = np.sum((sigma[1:] + sigma[:-1]) / 2 * np.diff(u))     # trapezium rule
u0 = np.sqrt(sigma_mid / (sigma_mid + sigma_surf))           # where sigma = 0
depth = (1 - u0) / 2 * t

print(f"mid-plane tension {sigma_mid:.0f} MPa, net force {abs(net):.3f} (balanced)")
print(f"compressive layer depth: {depth:.2f} mm = {depth / t:.1%} of the thickness")
print()
for z in np.linspace(-t / 2, t / 2, 13):
    s = sigma_mid - (sigma_mid + sigma_surf) * (2 * z / t) ** 2
    bar = int(round(abs(s) / 5))
    left = ("#" * bar).rjust(20) if s < 0 else " " * 20
    right = "#" * bar if s > 0 else ""
    print(f"z = {z:5.2f} mm {s:7.1f} MPa  {left}|{right}")

# Strength in bending: surface flaws open only once the applied stress overcomes the
# built-in compression. Annealed float glass has a characteristic strength of about 45 MPa.
annealed = 45.0
print(f"\nbending stress to break: annealed {annealed:.0f} MPa, tempered about {annealed + sigma_surf:.0f} MPa")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

With 100 MPa of surface compression, the core carries 50 MPa of tension and the compressive layer reaches 21 % of the thickness from each face — the same fraction, 12(1−1/3)\tfrac12(1 - 1/\sqrt3), for any parabolic profile. An applied bending stress must first cancel the 100 MPa before any surface flaw feels tension, so the glass can carry roughly three times the load. (European product standards use characteristic bending strengths of 45 MPa for annealed float glass and 120 MPa for toughened glass; the latter is less than our simple sum because a characteristic value is a 5 % fractile that also covers weaker edges and variable tempering.)

+50 MPa−100 MPa−100 MPatension (core)compressioncompressionsurfacesurfaceposition through the 6 mm thickness
The tempering stress profile computed above, through a 6 mm plate. The surfaces (left and right) are in compression, down to −100 MPa; the core is in tension, up to +50 MPa. The stress crosses zero 1.27 mm inside each face, and the areas above and below the axis balance.

The stored energy has a second effect. When a tempered pane does break, the core tension drives cracks through the whole sheet, which shatters into small, blunt fragments instead of long, sharp shards: that is why it is called safety glass and is required in doors, shower screens and car side windows. For the same reason it cannot be cut or drilled after tempering. Laminated glass, two sheets bonded by a polymer interlayer, holds its fragments together when broken and is used for windscreens and overhead glazing.

Chemical strengthening reaches stresses no air jet can. An aluminosilicate glass is immersed in molten potassium nitrate at around 400 °C; potassium ions diffuse into the surface in exchange for smaller sodium ions, and, crowded into sites made for smaller ions, wedge the surface into compression of several hundred MPa over a layer tens of micrometres deep. Phone cover glasses from makers in the United States, Japan, Germany and China are made this way, and their composition and ion-exchange schedules are among the most closely studied in the industry.

ComputationC

Cement and concrete

Concrete is about 10–15 % cement, with water, sand and gravel; the cement paste glues the stones together. Portland cement (named in an 1824 patent for its resemblance to Portland stone) is made by heating limestone and clay to about 1450 °C in a rotary kiln. The limestone decomposes, CaCO₃ → CaO + CO₂, and the lime reacts with silica, alumina and iron oxide to form clinker, nodules that are ground with about 5 % gypsum. Cement chemists write the clinker phases in shorthand (C = CaO, S = SiO₂, A = Al₂O₃, F = Fe₂O₃, H = H₂O):

PhaseFormulaMass fractionRole
alite, C₃SCa₃SiO₅50–70 %fast hydration, early strength
belite, C₂SCa₂SiO₄15–30 %slow hydration, later strength
aluminate, C₃ACa₃Al₂O₆5–10 %very fast reaction, controlled by the gypsum
ferrite, C₄AFCa₄Al₂Fe₂O₁₀5–15 %gives grey cement its colour

Concrete does not set by drying. It sets by hydration: the silicates react with water to form a nanometre-scale, poorly crystalline calcium silicate hydrate, C–S–H, which binds everything, plus crystals of calcium hydroxide. The reaction gives out heat and continues for months, as long as water is present, so fresh concrete is kept moist while it cures. Strength falls as the water-to-cement ratio rises (Duff Abrams's rule of 1918), because the excess water leaves pores. Typical structural concrete reaches 25–50 MPa in compression but only about a tenth of that in tension, so it is reinforced with steel bars, which expand with temperature almost exactly as concrete does and are protected from rust by the high alkalinity of the pore water. Most deterioration of concrete structures comes from losing that protection, when carbon dioxide from the air neutralises the surface layer or chlorides from de-icing salt or seawater reach the steel.

Roman builders used lime and volcanic ash, a pozzolan that reacts with lime to form binding hydrates; some of their harbour and dome concretes have lasted two thousand years, and recent work suggests that lumps of lime left by hot mixing help heal cracks (Seymour et al.).

The carbon footprint

Cement production is responsible for roughly 7 to 8 per cent of the world's CO₂ emissions (IEA). Most of it is not from fuel. It is the CO₂ released by the limestone itself, fixed by stoichiometry.

python
# CO2 from making Portland cement clinker, cement and concrete: a back-of-the-envelope model.
M_CO2, M_CaO = 44.01, 56.08

# Assumptions (typical values; real plants vary):
cao_in_clinker = 0.65        # mass fraction CaO in clinker, almost all of it from limestone, CaCO3
kiln_fuel_GJ = 3.5           # fuel energy per tonne of clinker, GJ (modern dry kilns: about 3-3.5)
fuel_factor = 0.095          # t CO2 per GJ, for coal (natural gas would be about 0.056)
power_kWh = 110              # electricity per tonne of cement, mostly grinding
grid_factor = 0.00045        # t CO2 per kWh

process = cao_in_clinker * M_CO2 / M_CaO      # CaCO3 -> CaO + CO2
fuel = kiln_fuel_GJ * fuel_factor
per_clinker = process + fuel
print(f"per tonne of clinker: calcination {process:.2f} t CO2 + kiln fuel {fuel:.2f} t = {per_clinker:.2f} t")
print(f"share from the chemistry itself: {process / per_clinker:.0%}")

binder_per_m3 = 0.32          # tonnes of cement in a cubic metre of ordinary structural concrete
print("\ncement type              clinker  t CO2 per t cement  kg CO2 per m3 concrete")
for name, clinker_factor in [("ordinary Portland (CEM I)", 0.95), ("world-average blend", 0.72),
                             ("limestone calcined clay", 0.50)]:
    per_cement = clinker_factor * per_clinker + power_kWh * grid_factor
    print(f"{name:26} {clinker_factor:5.2f} {per_cement:15.2f} {1000 * per_cement * binder_per_m3:20.0f}")

world_cement_Gt = 4.0
print(f"\nworld: {world_cement_Gt} Gt of cement at the average blend -> about "
      f"{world_cement_Gt * (0.72 * per_clinker + power_kWh * grid_factor):.1f} Gt CO2 a year")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Each tonne of clinker releases about 0.51 t of CO₂ from the limestone and 0.33 t from coal, so about 60 % of the emissions are process emissions that no switch of fuel can remove. A cubic metre of ordinary concrete carries about 210–270 kg of CO₂, and four billion tonnes of cement a year add up to about 2.6 Gt, consistent with the published share. The levers follow from the model:

  • Less clinker per tonne of cement. Replace part of it with fly ash, blast-furnace slag, or calcined clay with limestone (LC3, developed by an international team led from EPFL in Switzerland, with partners in India and Cuba). Fly ash and slag supplies shrink as coal plants close and steel plants change; clay is abundant. Calcining clay at about 800 °C releases water rather than CO₂, though its fuel use brings the net saving of LC3 to about 30–40 % (LC3).
  • Less cement per cubic metre, by better mix design and by not over-specifying strength.
  • Cleaner kiln heat: biomass and waste fuels, and electric kilns under development.
  • Capturing the CO₂. Since the process emissions cannot be avoided by chemistry alone, the industry's roadmaps (GCCA) rely on carbon capture for the rest; Heidelberg Materials' plant at Brevik in Norway began capturing CO₂ for permanent storage under the North Sea in 2025, the first at full scale in the sector.
DefinitionsH

Designing with scatter: Weibull statistics

Break thirty identical alumina bars and you get thirty different strengths, perhaps from 280 to 460 MPa. Steel bars would agree to within a few per cent. The difference is the flaws: each ceramic bar fails from its single worst flaw, and the worst flaw varies. In 1939 the Swedish engineer Waloddi Weibull turned this into a distribution.

Theorem— Weakest-link failure and the Weibull distribution

Suppose a body of volume VV under uniform tensile stress σ\sigma fails if any part of it fails, and that disjoint parts fail independently. Then its probability of survival has the form Ps(V,σ)=exp⁡[−(V/V0) ϕ(σ)]P_s(V, \sigma) = \exp[-(V/V_0)\,\phi(\sigma)] for some increasing function ϕ\phi. Weibull's choice ϕ(σ)=(σ/σ0)m\phi(\sigma) = (\sigma/\sigma_0)^m gives the Weibull distribution

Pf=1−exp⁡[−VV0(σσ0)m],P_f = 1 - \exp\left[-\frac{V}{V_0}\left(\frac{\sigma}{\sigma_0}\right)^m\right],

where mm is the Weibull modulus and σ0\sigma_0 the characteristic strength of the reference volume V0V_0.

Proof

Split the body into two parts of volumes V1V_1 and V2V_2. It survives only if both survive, and they are independent, so Ps(V1+V2)=Ps(V1) Ps(V2)P_s(V_1 + V_2) = P_s(V_1)\,P_s(V_2). Writing g(V)=−ln⁡Ps(V)g(V) = -\ln P_s(V) turns this into g(V1+V2)=g(V1)+g(V2)g(V_1 + V_2) = g(V_1) + g(V_2); since g≥0g \ge 0, it is non-decreasing, and the only monotone solutions of this equation (Cauchy's) are linear, g(V)=V g(V0)/V0g(V) = V\,g(V_0)/V_0. Calling g(V0)=ϕ(σ)g(V_0) = \phi(\sigma) gives Ps=exp⁡[−(V/V0)ϕ(σ)]P_s = \exp[-(V/V_0)\phi(\sigma)]. The power law for ϕ\phi is not derived: it is an empirical choice, justified when the flaw-size distribution has a power-law tail, and checked against data.

∎

The modulus measures scatter: a large mm means consistent strength. Engineering ceramics have mm of about 5 to 20; metals, if described the same way, would be near 100. Taking logarithms twice turns the distribution into a straight line, ln⁡ln⁡[1/(1−Pf)]=mln⁡σ−mln⁡σ0\ln\ln[1/(1 - P_f)] = m\ln\sigma - m\ln\sigma_0, so a set of strengths is ranked, each given an estimated failure probability, and fitted. Here thirty strengths are simulated from a known distribution and the parameters recovered, first by least squares on that line, then by maximum likelihood, the method testing standards for advanced ceramics (ASTM C1239, ISO 20501) prescribe.

python
import numpy as np

rng = np.random.default_rng(2026)
m_true, s0_true = 10.0, 400.0          # Weibull modulus and characteristic strength (MPa)
n = 30                                 # specimens broken in a bend test
# Inverse-transform sampling: if U is uniform, s0 * (-ln U)^(1/m) is Weibull distributed.
strengths = np.sort(s0_true * (-np.log(rng.random(n))) ** (1 / m_true))

# 1. Linear regression on a Weibull plot, with median-rank estimates of P_f.
i = np.arange(1, n + 1)
F = (i - 0.3) / (n + 0.4)
x, y = np.log(strengths), np.log(-np.log(1 - F))
m_lsq, c = np.polyfit(x, y, 1)
s0_lsq = np.exp(-c / m_lsq)

# 2. Maximum likelihood: solve the score equation for m by bisection, then s0 follows.
def score(m, s):
    w = s ** m
    return np.sum(w * np.log(s)) / np.sum(w) - 1 / m - np.mean(np.log(s))
lo, hi = 0.5, 100.0
for _ in range(80):
    mid = (lo + hi) / 2
    lo, hi = (mid, hi) if score(mid, strengths) < 0 else (lo, mid)
m_mle = (lo + hi) / 2
s0_mle = np.mean(strengths ** m_mle) ** (1 / m_mle)

print(f"weakest {strengths[0]:.0f} MPa, median {np.median(strengths):.0f} MPa, strongest {strengths[-1]:.0f} MPa")
print(f"least squares:      m = {m_lsq:5.2f}, s0 = {s0_lsq:5.1f} MPa")
print(f"maximum likelihood: m = {m_mle:5.2f}, s0 = {s0_mle:5.1f} MPa   (true: {m_true}, {s0_true})")

# Use the fit: a component with 50 times the effective volume of a test bar, and the stress
# at which only one part in ten thousand may fail.
m, s0 = m_mle, s0_mle
for Pf in (0.5, 1e-2, 1e-4):
    bar = s0 * (-np.log(1 - Pf)) ** (1 / m)
    part = bar * (1 / 50) ** (1 / m)
    print(f"P_f = {Pf:<6g}: test bar {bar:5.0f} MPa, component (50 x volume) {part:5.0f} MPa")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Both fits find σ0\sigma_0 within 1 MPa of the truth, but both overestimate the modulus (11.7 and 12.3, against 10). The last lines show why this matters. The median bar breaks at 387 MPa, but if only one part in ten thousand may fail, the allowed stress in the bar falls to 189 MPa, and in a component with fifty times the stressed volume — more volume, more chances of a bad flaw — to 137 MPa, about a third of the median bar strength. This size effect, σ2/σ1=(V1/V2)1/m\sigma_2/\sigma_1 = (V_1/V_2)^{1/m}, is why strengths measured on small bars cannot be used directly for large parts, and why fine fibres and thin wires of brittle materials are so much stronger than bulk pieces (an idea you will meet again in composites).

How much should we trust mm from a test programme? Simulate the programme many times.

python
import numpy as np

rng = np.random.default_rng(7)
m_true, s0_true, trials = 10.0, 400.0, 4000

def mle_modulus(s):
    """Maximum-likelihood Weibull modulus for each row of s (bisection on the score equation)."""
    ls = np.log(s)
    lo, hi = np.full(len(s), 0.5), np.full(len(s), 100.0)
    for _ in range(60):
        m = (lo + hi) / 2
        w = s ** m[:, None]
        score = (w * ls).sum(1) / w.sum(1) - 1 / m - ls.mean(1)
        lo, hi = np.where(score < 0, m, lo), np.where(score < 0, hi, m)
    return (lo + hi) / 2

print(f"{trials} simulated test programmes, true m = {m_true:.0f}")
print(" n   mean m   5th pct  95th pct")
for n in (5, 10, 30, 100):
    s = s0_true * (-np.log(rng.random((trials, n)))) ** (1 / m_true)
    est = mle_modulus(s)
    lo5, hi95 = np.percentile(est, [5, 95])
    print(f"{n:3d} {est.mean():8.2f} {lo5:9.2f} {hi95:9.2f}")
Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Two further cautions. Ceramics in moist air suffer slow crack growth (water helps break the strained Si–O and Al–O bonds at a crack tip), so their strength falls with time under load; and a part can contain more than one population of flaws — machining damage at the surface, pores inside — each with its own modulus. Design programmes for ceramic components, such as NASA's CARES code, handle both.

Further reading

Ceramics and glass in industry and research

Testing and design. Flexural strength is measured in three- or four-point bending on small bars under standards such as ASTM C1161 and ISO 14704, fracture toughness on pre-cracked bars, and Weibull parameters as above. Critical parts — implant heads, bearing balls, turbine components — are made under tight process control, inspected by ultrasound and X-ray computed tomography, and often proof tested.

Frontiers in 2026. Ceramic-matrix composites of silicon carbide fibres in silicon carbide now fly in the hot sections of jet engines (composites). Ultra-high-temperature ceramics such as hafnium carbide and zirconium diboride, which melt above 3000 °C, are studied for hypersonic vehicles. Silicon carbide wafers for power electronics have moved to 200 mm diameter. Oxide, sulfide and halide solid electrolytes are being developed for solid-state batteries (battery materials). And 3D printing of ceramics, by binding powder layers or curing ceramic-loaded resins, makes shapes that cannot be pressed.

Decarbonisation. Glass furnaces are moving to more cullet, oxygen firing and electric melting, and brick and tile kilns to electric and hydrogen firing. Low-clinker cements, carbon capture, and accounting for the CO₂ that concrete reabsorbs slowly by carbonation over its life are all active, and argued over, in standards bodies and research groups worldwide. The broader picture is in sustainable materials.

Common misconceptionsH

Common misconceptions

Misconception

"Ceramics are weak." Alumina's compressive strength is several gigapascals, and its bending strength can exceed the yield strength of structural steel. Ceramics are brittle and flaw-sensitive: their tensile strength depends on their worst flaw, and they give no warning before failing.

Misconception

"Old windows are thicker at the bottom because glass flows." At room temperature, glass is so viscous that sagging measurably would take far longer than the age of the universe (Zanotto). Medieval panes were uneven because they were made by spinning or blowing, and glaziers often set the thicker edge at the bottom.

Misconception

"Concrete hardens by drying out." It hardens by hydration, a chemical reaction with water. Concrete that dries out too early stops gaining strength, which is why it is kept moist while it cures.

Misconception

"Cement and concrete are the same thing." Cement is the grey powder, about a seventh of the mix; concrete is cement, water, sand and gravel. Mortar is cement, water and sand, without the gravel.

Misconception

"The strength of a ceramic is a single number." It is a distribution, and it depends on the volume and surface under stress. A design strength is meaningful only with a failure probability and a size attached.

Exercises

Exercises

1. Ceramics and glasses

Warm-upHuman

Why does a dense alumina bar break in a brittle way at room temperature, while an aluminium bar of the same shape bends?

Choose one

2. Ceramics and glasses

CoreHuman

An alumina component with KIc=4K_{Ic} = 4 MPa√m must carry a tensile stress of 350 MPa. Taking the geometry factor Y=1Y = 1, what is the largest crack depth it can tolerate, in micrometres? Give the nearest whole number.

3. Ceramics and glasses

CoreHuman

A 10 mm thick sheet of thermally tempered glass has a parabolic residual stress profile with 90 MPa of surface compression. What is the tensile stress at its mid-plane, in MPa?

4. Ceramics and glasses

Warm-upHuman

A clinker contains 64 % CaO by mass, all of it from limestone (CaCO₃). How many kilograms of CO₂ does the decomposition of that limestone release per tonne of clinker? Use molar masses CO₂ = 44.01 g/mol and CaO = 56.08 g/mol, and give the nearest whole number.

5. Ceramics and glasses

CoreHuman

Which of these would genuinely reduce the CO₂ emitted per cubic metre of concrete of a given strength? Select all that apply.

Choose all that apply

6. Ceramics and glasses

CoreHuman

Small silicon nitride bend bars have a characteristic strength of 400 MPa and a Weibull modulus of 12. A component has 1000 times the effective volume under stress. At the same probability of failure, what strength should we expect for the component, in MPa? Give the nearest whole number.

7. Ceramics and glasses

CoreComputational

Write weibull_fit(strengths), which takes a list or array of fracture strengths (in any order) and returns a tuple (m, s0) estimated by linear regression on a Weibull plot:

  1. sort the strengths;
  2. give the ii-th smallest (i=1,…,ni = 1, \ldots, n) the median-rank failure probability Fi=(i−0.3)/(n+0.4)F_i = (i - 0.3)/(n + 0.4);
  3. fit a straight line y=mx+cy = m x + c to x=ln⁡σix = \ln\sigma_i, y=ln⁡[−ln⁡(1−Fi)]y = \ln[-\ln(1 - F_i)];
  4. return the slope mm and s0=exp⁡(−c/m)s_0 = \exp(-c/m).

You may use numpy (for example np.polyfit).

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

8. Ceramics and glasses

StretchComputational

Write allowable_stress(s0, m, volume_ratio, pf), which returns the uniform tensile stress at which a component fails with probability pf, given Weibull parameters s0 and m measured on test bars, and volume_ratio = (component's effective volume) / (test bar's effective volume). Raise ValueError unless 0 < pf < 1, and unless m, s0 and volume_ratio are positive.

Runs entirely in your browser, in an isolated worker. Nothing is sent to a server.

Further reading and resources