Creep

What creep actually is — and why time is the variable

Pull on a steel bar at room temperature with a stress below its yield strength and it stretches elastically by a fixed amount, then stops. Hold that stress forever and nothing more happens: strain is set by stress at the instant you apply it, and time does not appear in the equation. That intuition — strain follows stress — is correct at low temperature and is the basis of almost all everyday structural design.

Creep is what happens when that intuition breaks down. Heat the same bar above roughly 40 percent of its absolute melting temperature, apply a steady stress that is still below yield, and the bar keeps stretching — slowly, permanently, on its own — for as long as the load and the heat are present. The driving variable is no longer just stress; it is stress multiplied by time at temperature. The strain that accumulates is irreversible plastic flow, and it does not need a stress anywhere near yield to occur. A boiler tube loaded to a third of its yield strength will, over a decade at 560 °C, slowly bulge, thin, and finally burst.

The reason is atomic mobility. Above the threshold temperature, atoms have enough thermal energy to hop between lattice sites by diffusion, and dislocations — the line defects that carry plastic flow — can climb out of the slip planes where obstacles normally pin them. Both processes are thermally activated, which means their rate scales with exp(−Q/RT): an Arrhenius term that is brutally sensitive to temperature. That single exponential is why a component can be perfectly safe at 500 °C and creep-limited at 600 °C, and why creep is the defining life-limiter of every hot section in a power plant or jet engine.

The temperature threshold is expressed as the homologous temperature, the ratio of the operating temperature to the melting temperature, both in kelvin: T_h = T / T_m. Below about 0.3 creep is negligible for most engineering purposes; between 0.4 and 0.5 it must be checked; above 0.5 it usually dominates design. Because lead melts at 600 K, room temperature (293 K) is already T_h ≈ 0.49 for lead — which is exactly why old lead roof sheeting visibly sags over centuries, and why solder joints creep at their service temperature.

The three stages of a creep curve

Run a constant-load creep test — hang a fixed weight on a specimen inside a furnace and record elongation against time — and the strain-versus-time curve has a characteristic three-part shape. Every creep curve, from a 3 mm tensile coupon to a full turbine blade, follows it.

• Primary (transient) creep. Strain accumulates quickly at first, then the rate decreases. The material is work-hardening: dislocations multiply and tangle faster than thermal recovery can clear them, so each increment of flow is harder than the last. Primary creep is mostly recoverable strain and a small permanent set.
• Secondary (steady-state) creep. The strain rate flattens to a near-constant minimum value. Here work-hardening and thermal recovery (dislocation climb, annihilation, subgrain formation) reach a dynamic balance, so the rate stays roughly constant for most of the component's life. This minimum creep rate is the single most important number in creep design, because the steady-state stage occupies the great majority of the service life and is the regime the governing equations describe.
• Tertiary (accelerating) creep. The rate turns upward and runs away. Internal damage now dominates: cavities nucleate on grain boundaries, link into microcracks, and the load-bearing cross-section shrinks. As area drops the true stress on the remaining metal climbs, which accelerates creep further — a positive-feedback spiral. In ductile creep the specimen also necks geometrically. Tertiary creep ends in stress rupture: a final, often surprisingly brittle-looking fracture.

Engineers almost never design to rupture. They design to a fixed allowable strain reached deep inside the secondary stage — typically a total creep strain of 0.5 to 1 percent — because a turbine blade that has elongated 1 percent will already be rubbing its shroud, and a pipe that has crept 1 percent has lost dimensional integrity long before it bursts. The rupture point matters mainly as the absolute backstop in a stress-rupture (creep-rupture) test.

The governing equation — Norton's law and Arrhenius

The steady-state creep rate is captured by the Norton-Bailey power law with an Arrhenius temperature term:

ε̇_s = A · σ^n · exp(−Q / R·T)

ε̇_s = minimum (secondary) creep strain rate   [1/s]
A    = material constant
σ    = applied stress                          [MPa]
n    = stress exponent (dimensionless)
Q    = activation energy for creep             [J/mol]
R    = gas constant = 8.314 J/(mol·K)
T    = absolute temperature                    [K]

Two terms do all the work. The power law σ^n sets how steeply creep responds to stress, and the exponent n identifies the underlying mechanism — near 1 for diffusion creep, 3 to 8 for dislocation (power-law) creep. The exponential exp(−Q/RT) sets the temperature sensitivity. Because Q for self-diffusion in metals is often 250 to 350 kJ/mol, the exponential is savage: with Q = 300 kJ/mol, raising the temperature from 800 °C (1073 K) to 825 °C (1098 K) multiplies the creep rate by

exp[ −300000/8.314 · (1/1098 − 1/1073) ]
  = exp[ −36084 · (−2.12e−5) ]
  = exp(0.766)  ≈  2.15×

A 25 °C miscalculation in metal temperature — entirely plausible in a turbine where gas, film-cooling, and conduction all interact — roughly doubles the creep rate and halves the life. This is why turbine metal-temperature prediction is treated as a safety-critical discipline in its own right.

To compress life data, engineers fold temperature and rupture time into the Larson-Miller parameter:

LMP = T · (C + log₁₀ t_r)

t_r = time to rupture (hours)
C   = material constant, ≈ 20 for most steels and superalloys
T   = absolute temperature (K)

A single plot of stress versus LMP — the master rupture curve — then lets you trade temperature for time. The same LMP value can be reached by a high temperature for a short time or a lower temperature for a long time. Worked example: a 1Cr-1Mo-V rotor steel with C = 20 reaches a given LMP of about 20,300 either at 540 °C (813 K) for roughly 11 years, or at 565 °C (838 K) for only about two years. That is the quantitative statement of how a 25 °C rise can cut rupture life several-fold.

The mechanisms — what is actually moving

The value of the stress exponent n and the activation energy Q together fingerprint which microscopic mechanism is carrying the creep. A deformation mechanism map (Ashby map) plots normalized stress against homologous temperature and shades in which mechanism dominates each region.

• Dislocation (power-law) creep. At higher stresses, dislocations glide until pinned by precipitates or other dislocations, then climb over the obstacle by absorbing or emitting vacancies — a diffusion-controlled step. Stress exponent n ≈ 3–8; activation energy equals that of lattice self-diffusion. This is the dominant mechanism in turbine blades and most loaded high-temperature parts.
• Diffusion creep. At low stress and very high temperature, no dislocation motion is needed: atoms simply diffuse from grain faces in compression to faces in tension, elongating each grain. When the path is through the grain interior it is Nabarro-Herring creep; when it is along grain boundaries it is Coble creep. Both give n ≈ 1, and both speed up dramatically as grain size shrinks (rate ∝ 1/d² for Nabarro-Herring, 1/d³ for Coble) — which is why fine-grained metals creep faster, the opposite of room-temperature strength.
• Grain-boundary sliding. Grains slide past one another along their boundaries, accommodated by diffusion or local plasticity. It is a major contributor in the tertiary stage because the sliding opens up cavities at boundary triple points, and it is the reason creep resistance improves when grain boundaries are eliminated entirely.

That last point drives a counter-intuitive design rule. At room temperature, fine grains make a metal stronger (the Hall-Petch relationship) because grain boundaries block dislocations. At creep temperatures, fine grains make a metal weaker, because boundaries are now the fast paths for diffusion and the planes that slide. The optimal creep microstructure therefore has few, large grains aligned with the load — or, ideally, no grain boundaries at all.

The canonical case — single-crystal turbine blades

The high-pressure turbine blade of a modern jet engine is the most creep-critical component ever mass-produced. It spins at 10,000–15,000 rpm, so centrifugal loading puts the airfoil root under a steady tensile stress of roughly 100–200 MPa. It sits in a combustion gas stream above 1,500 °C, with the metal itself running near 1,000–1,100 °C — a homologous temperature above 0.7 for nickel. Constant high stress plus very high homologous temperature is the textbook creep condition, and a blade that elongates even a fraction of a millimeter will rub its shroud, overheat, and fail.

The entire evolution of superalloy turbine blades is a war against creep:

• Equiaxed castings (1950s–60s). Ordinary polycrystalline nickel superalloy. Grain boundaries transverse to the load slide and cavitate — the weakest link.
• Directionally solidified (DS) blades (1970s). Controlled cooling grows columnar grains aligned with the centrifugal axis, eliminating transverse grain boundaries. Creep life roughly tripled.
• Single-crystal (SX) blades (1980s onward). The casting is grown from one grain through a helical "pigtail" selector, so the finished blade has no grain boundaries at all. With grain-boundary sliding and diffusion removed, the blade can run roughly 30–50 °C hotter for the same life — and in a turbine, 30 °C is worth a meaningful chunk of fuel efficiency.
• Gamma-prime strengthening. The matrix is filled with 60–70 percent by volume of ordered Ni₃(Al,Ti) gamma-prime precipitates that pin dislocations. Uniquely, gamma-prime gets stronger with temperature up to about 800 °C.
• Heavy refractory alloying. Rhenium (3 percent in 2nd-gen, 6 percent in 3rd-gen SX alloys like CMSX-10) and ruthenium slow diffusion in the matrix, directly attacking the exp(−Q/RT) term.
• Internal cooling and thermal-barrier coatings. Serpentine passages flow compressor bleed air through the hollow blade, and a ~150 µm yttria-stabilized-zirconia ceramic coating drops the metal temperature 100–150 °C below the gas temperature — directly lowering the homologous temperature, the most effective single lever of all.

Stack those measures and a 3rd-generation single-crystal blade survives thousands of hours under conditions that would rupture 1960s cast nickel in minutes. Every measure maps onto one of two strategies in the Norton equation: lower T (cooling, coatings) or raise the resistance to dislocation and diffusion motion (single crystal, gamma-prime, rhenium).

Creep versus the other ways materials fail

Creep is one of several time- or cycle-dependent damage modes, and engineers must know which one is in play because the design rules are completely different. The table below contrasts creep with the most-confused alternatives.

Property	Creep	Fatigue	Yielding (plastic)	Stress relaxation
Driving variable	Time at constant stress	Number of load cycles	Instantaneous stress	Time at constant strain
Temperature needed	High (T/T_m ≳ 0.4)	Any, incl. room temp	Any	High (same as creep)
Load character	Steady	Fluctuating / cyclic	Monotonic, above yield	Fixed displacement
What changes with time	Strain grows	Crack grows per cycle	Nothing — set at load	Stress decays
Governing relation	ε̇ = Aσⁿexp(−Q/RT)	Paris / S-N (Basquin)	σ ≥ σ_yield	Norton law, σ falling
End state	Stress rupture	Fatigue fracture	Permanent set / collapse	Loss of clamp / preload
Classic example	Boiler tube, turbine blade	Rotating shaft, aircraft wing	Overloaded bracket	Bolted flange losing torque hot

Stress relaxation is creep's twin: it is the same physics under the complementary boundary condition. Fix the strain instead of the stress — a tightened bolt holds a flange at constant stretch — and creep flow lets the material slowly redistribute that strain into permanent set, so the stress (and the bolt preload) decays over time. Hot bolted joints in steam plants must therefore be re-torqued periodically, and pre-stressed components are sized for end-of-life preload, not initial preload.

Creep-fatigue interaction — the real-world combination

Pure creep and pure fatigue are textbook idealizations. A real turbine disc lives in creep-fatigue: each flight cycle (startup, climb, cruise, shutdown) imposes a thermal-mechanical fatigue loop, while the long cruise hours at temperature add steady creep. The two interact destructively — a creep cavity is a perfect initiation site for a fatigue crack, and a fatigue crack tip held open at temperature creeps and oxidizes.

Life is estimated with a linear damage summation, the time-fraction plus cycle-fraction rule:

Σ (t_i / t_r,i)  +  Σ (n_j / N_f,j)  ≤  D

t_i  = time spent at condition i
t_r,i = creep-rupture life at condition i
n_j  = cycles applied at condition j
N_f,j = fatigue life at condition j
D    = allowable damage (often 0.3–1.0, with margin)

The allowable D is pushed below 1.0 to account for the interaction, because the combined damage is empirically worse than the sum of the parts. This is why hot-section components are tracked not by calendar age but by accumulated time-at-temperature and start-stop cycles — an engine's "life used" is a running tally of both fractions.

Designing against creep — the practical levers

• Use a high-melting matrix. Higher T_m means a lower homologous temperature at the same service temperature. Nickel and cobalt superalloys for the hottest parts; ferritic and martensitic steels (9–12 % Cr, "P91/P92") for steam plant up to ~620 °C; refractory metals and ceramics beyond.
• Lower the metal temperature. The single most powerful lever, because of the exponential. Internal air cooling, film cooling, and thermal-barrier coatings buy 100–150 °C, often worth more than any alloy change.
• Engineer the grain structure. Coarse grains, directional solidification, or single crystals to suppress grain-boundary sliding and Coble creep. The opposite of room-temperature strengthening.
• Pin the dislocations. Stable precipitates (gamma-prime, carbides, oxide dispersions in ODS alloys) and slow-diffusing solutes (rhenium, molybdenum, tungsten) raise the effective activation energy.
• Size to a strain or rupture limit, not to yield. Hot components are designed against a creep-strain allowable (often 1 % over design life) and a stress-rupture margin from a Larson-Miller master curve, never against the room-temperature yield strength.

Common pitfalls in creep assessment

• Designing to yield strength at temperature. A material can be far below yield and still creep to failure. The relevant property is the stress for a given rupture life or creep strain at temperature, not yield.
• Trusting an uncalibrated metal temperature. Because of the exponential, a 25 °C error roughly doubles the creep rate. Hot-spot prediction must be conservative, not nominal.
• Extrapolating short-term tests too far. Larson-Miller lets you extrapolate, but only modestly — pushing a 10,000-hour test to a 200,000-hour prediction can cross into a different mechanism (e.g. from dislocation to diffusion creep) and falsify the curve.
• Forgetting that fine grains help at room temperature but hurt at temperature. A heat treatment chosen for tensile strength can ruin creep life.
• Ignoring stress relaxation in bolted hot joints. Preload that is correct cold will decay hot; re-torque schedules and end-of-life preload margins are mandatory.
• Neglecting environment. Oxidation, sulfidation, and coating spallation expose fresh metal and accelerate tertiary creep — creep life in clean argon overstates life in real combustion gas.