How Science Works | Why Do Some Solids Slowly Change Shape Under a Constant Load?

Imagine observing a test specimen through a camera while a controlled tensile load is applied. Its length changes quickly at first. Then, while the load remains, it keeps changing more slowly. To separate these observations, record both the loading interval and the subsequent holding interval. Calling the whole extension “creep” without explaining the convention can conceal the initial response.

An ideal elastic model gives strain determined by the current stress, with no continuing time-dependent deformation once a constant stress is established. Real materials may add delayed, irreversible or otherwise history-dependent responses. Creep measurements investigate that time dependence. They do not require the material to have become liquid.

Engineering strain is the change in gauge length divided by the initial gauge length. If a marked length starts at 100 millimetres and becomes 100.20 millimetres just after loading, its engineering strain is 0.20/100 = 0.002, or 0.20%. If it later reaches 100.50 millimetres, the total strain is 0.50% and the additional strain since the first reading is 0.30 percentage points.

Those are invented measurements. They establish the arithmetic, but not a mechanism. Grip movement, temperature drift or instrument drift could also change an apparent length reading. A good experiment verifies that it is measuring deformation of the intended gauge section under the stated conditions.

Contents · Next chapter

Load is force. Stress expresses force relative to an area. For a simple tensile specimen, engineering stress uses the original cross-sectional area, while true stress uses the current area. Keeping a force constant therefore keeps engineering stress constant, but can allow true stress to change as the specimen changes shape.

Take a hypothetical load of 100 newtons on a specimen whose original area is 10 square millimetres. The engineering stress is 10 newtons per square millimetre, or 10 megapascals. If the current area falls to 8 square millimetres while the force stays at 100 newtons, the average true stress at that section becomes 12.5 megapascals. The force has not increased, yet the stress based on current area has risen by 25%.

To maintain 10 megapascals at that new area, the force would instead need to be 80 newtons. Constant-force and constant-true-stress tests are different experiments. The distinction becomes especially important when deformation is large or localised.

Our calculation assumes a simple, uniformly loaded section. A necked specimen can have a more complicated local stress state, so force divided by area is not a complete description of every internal stress component. It is enough here to show why the control variable must be named.

Whenever you read a creep graph, look for the material condition, temperature, force or stress convention, and strain definition. A smooth curve with missing test conditions is not a transferable material property.

Contents · Previous chapter · Next chapter

Draw time horizontally and strain vertically. Mark the initial loading response separately. A common subsequent pattern has a decreasing slope, an approximately steady slope, and finally an increasing slope. These regions are conventionally called primary, secondary and tertiary creep.

Slope is creep rate: the amount of additional strain per unit time. A decreasing slope means deformation is continuing more slowly, not reversing. A straight rising segment means strain continues accumulating at approximately constant rate. An increasing slope means acceleration. These distinctions are easy to lose when the eye concentrates on the height of the curve rather than its local steepness.

NIST’s materials-data guide uses such curves to explain creep modelling and discusses the assumptions behind stress-, temperature- and microstructure-dependent rate laws. The three-stage sketch is a useful organising picture, but it should not be forced onto every material or test record. Some experiments end before later stages, and some responses do not show three clearly separated regions. NIST materials-data evaluation guide.

For an original numerical illustration, suppose additional strain rises from 0.0030 to 0.0034 over 20 hours. The interval-average rate is 0.0004/20 = 0.00002 per hour. In percentage terms, that is 0.002 percentage points per hour. It is not 0.002% of the specimen’s current strain each hour.

One interval-average slope is not automatically the minimum rate or a true steady-state rate. More closely spaced measurements may reveal curvature. And a curve alone rarely identifies which atomic or molecular process produced it; different mechanisms can generate similar-looking records.

Contents · Previous chapter · Next chapter

A spring is an idealised elastic element: a force produces an extension, and removing the force allows recovery. A dashpot is an idealised viscous element: motion under force takes time and dissipates energy. Combining these mathematical elements gives simple ways to think about delayed response without claiming that literal springs and pistons exist inside a polymer.

Place one spring and one dashpot in parallel. They must undergo the same strain, while sharing the stress. Under a suddenly imposed constant stress, the dashpot initially resists rapid change. Over time, the spring carries more of the load and the strain approaches a finite value. This arrangement is the Kelvin–Voigt model.

For a hypothetical model with final strain 0.010 and a time constant of 100 seconds, the predicted strain is 0.010 × [1 − exp(−t/100)]. At 100 seconds it is about 0.00632, or 0.632%. At 300 seconds it is about 0.00950, or 0.950%. The strain is still approaching its limit; it has not suddenly completed a process at one time constant.

If the load is removed, this ideal model recovers gradually toward zero strain. It has delayed deformation but no permanent set after complete recovery. A spring and dashpot in series instead allow an accumulating viscous contribution under constant stress. More elaborate combinations can include an immediate elastic response, delayed recovery and a residual deformation.

These models make a precise lesson visible: time-dependent deformation and permanent deformation are not identical categories. A recovery record helps separate them. Real specimens may also undergo damage or structural change that these small linear models do not describe.

Contents · Previous chapter · Next chapter

A polymer contains long molecular chains whose movements are constrained by their neighbours, entanglements, crystalline regions or crosslinks. Loading can alter chain conformations and local arrangements. Some changes occur rapidly, while others require cooperative motion and take longer. The observed response depends on which motions are accessible during the experiment.

Temperature can change those motions dramatically. Moisture can also matter for polymers that absorb it, and processing or ageing can alter the starting structure. It is therefore possible for a polymer to show appreciable creep at ordinary temperatures even when a familiar structural metal shows very little change on the same time scale.

This does not mean that every polymer behaves like the same dashpot. A crosslinked network, a semicrystalline plastic and an amorphous glassy polymer can have very different long-time responses. A model chosen because it fits a short section of one curve may fail when the loading history changes.

An original NIST study by Rouiller and McKenna tested a constitutive model for polyurethane rubber using stress-relaxation and creep histories. Model parameters obtained from one kind of measurement were used to evaluate responses under other histories. That is a stronger test than merely drawing a curve through the data used for fitting, although the reported agreement remained specific to the studied material and conditions. NIST polyurethane study.

A useful molecular diagram would show chains constrained at several points, with a few segments changing conformation between two times. Label it as a schematic. It should not imply that all chains straighten completely, that every constraint is permanent, or that the specimen’s extension comes only from stretching chemical bonds.

Contents · Previous chapter · Next chapter

In a crystal, atoms occupy an ordered structure, but the structure contains defects. A dislocation is a line defect that allows plastic deformation through a sequence of local rearrangements. It lets a crystal change shape without requiring an entire plane of atoms to move past another plane simultaneously.

Obstacles can impede dislocation motion. At suitable temperatures, diffusion-assisted processes can help dislocations change position and bypass some obstacles. Dislocation climb involves exchanging point defects with the surrounding lattice; it is distinct from glide along a slip plane. NIST’s structural-materials modelling work explicitly connects precipitate geometry and dislocation movement with superalloy creep behaviour. NIST structural-materials research.

Creep can also involve redistribution of matter through grains or along grain boundaries, with accompanying shape change. Grain-boundary sliding may contribute when other processes accommodate the movement so that neighbouring grains remain compatible. These routes can act together rather than occupying perfectly separate boxes.

A theoretical study of coupled diffusional creep, for example, links grain-boundary dislocations, vacancy transport and deformation, and examines how the controlling response changes with stress, temperature and boundary mobility. It is a model of specified mechanisms, not evidence that one equation describes every crystalline solid. Original coupled-creep model.

The important physical contrast with the polymer picture is the nature of the moving structures. Both produce time-dependent strain, but one may involve chain rearrangement while another involves crystal defects and mass transport. Naming the macroscopic response is the beginning of the explanation, not the end.

Contents · Previous chapter · Next chapter

Many microscopic rearrangements must cross energy barriers. Temperature changes the rate at which thermally activated events can occur. Stress can bias those events, while the internal structure changes the available paths and obstacles. A process too slow to matter during one observation can become important during a hotter or longer exposure.

For an invented activated-rate comparison, let rate be proportional to exp[−Q/(RT)], with activation energy Q = 80,000 joules per mole and gas constant R approximately 8.314 joules per mole per kelvin. Compare 300 kelvin with 330 kelvin while keeping the same mechanism and every other factor fixed.

The warmer-to-cooler rate ratio is exp[(80,000/8.314) × (1/300 − 1/330)], approximately 18.5. A 10% increase in absolute temperature can therefore produce a much larger rate change in this hypothetical model. The number is not a measured acceleration factor for a named plastic, metal or household object.

The assumptions are crucial. A change in temperature may change the material’s structure or dominant mechanism. Then a single activation energy and prefactor may no longer apply. A fit over a narrow range does not automatically justify extrapolating to another range.

For crystalline materials, temperature relative to melting temperature is often a useful organising comparison, using kelvin for both temperatures. It helps explain why the same room temperature represents very different conditions for different materials. It does not create a universal sharp temperature below which creep is impossible or above which a component is unsafe.

Contents · Previous chapter · Next chapter

A material’s microstructure determines more than its short-test strength. Grain size, precipitates, interfaces, processing history and environmental reactions can change the routes available for time-dependent deformation. An obstacle effective against one process may be less effective against another.

A 2024 study of a cobalt-based superalloy with chromium and titanium additions provides a concrete example. The additions changed precipitates and oxidation behaviour. Grain-boundary precipitates inhibited diffusional creep and boundary sliding in the studied comparison, but overall creep resistance was lower. The authors connected that result with other microstructural changes. The lesson is that improving one measured feature does not guarantee improvement in the total long-time response. Original superalloy study.

This is why “more obstacles” is not a complete design principle. Their size, stability, spacing and relationship to the active deformation mechanism matter. Nor does a high room-temperature hardness value settle a high-temperature creep question. Different measurements probe different responses.

Accelerating late-stage strain can involve shrinking load-bearing area, cavities, cracks or microstructural degradation. Under constant force, an area reduction can increase true stress, which can further accelerate deformation. Our earlier 10-to-12.5-megapascal example isolated that geometrical feedback; it did not calculate a rupture time.

A damage mechanism must be established by evidence such as microscopy, interrupted tests and comparison across conditions. A rising curve alone does not tell us which feature is responsible, and the absence of visible surface damage does not establish that the interior is unchanged.

Contents · Previous chapter · Next chapter

In a creep experiment, sustained stress or load is the control, and strain is a response. In a stress-relaxation experiment, a deformation is imposed and held while the required stress is monitored. The same time-dependent material can exhibit both behaviours, but the plots answer different questions.

For an ideal Maxwell model, a spring and dashpot in series share stress. Holding their total strain fixed allows the dashpot to take more deformation while the spring unloads, so stress decays. This is a useful conceptual counterpart to the constant-load creep model. NIST’s account of the development of nonlinear viscoelastic modelling explains why complete loading histories matter when going beyond simple linear elements. NIST account of viscoelastic constitutive modelling.

Fatigue concerns damage associated with repeated loading. Creep can occur during a sustained hold. A real component can experience both, especially when load cycles include long holds at elevated temperature. Separating the definitions helps identify the experiment; it does not imply that their effects always remain independent in service.

For a safe paper-based investigation, compare fictional records with the same initial strain but different controls. Record A holds force and measures increasing extension. Record B holds extension and measures decreasing force. Record C repeatedly changes the load and records damage after many cycles. Identify what is controlled before naming the behaviour.

For terminology transfer between languages and engineering records, use the separate creep and stress-rupture translation guide. This physical explanation does not replace its terminology checks or professional assessment of service-life data.

Contents · Previous chapter · Next chapter

Begin with the specimen: composition, processing and initial structure. Then identify the control mode, dimensions, temperature, environment and measurement method. Ask whether the recorded change is specimen deformation or movement elsewhere in the apparatus. A measurement of time dependence needs a stable reference as much as it needs patience.

Next examine the curve’s slope and the observation window. A short nearly straight segment supports an approximately constant rate over that interval. It does not establish that the rate will remain constant indefinitely. If unloading data are available, use them to distinguish immediate recovery, delayed recovery and deformation remaining at the end of observation.

Then connect the curve with independent evidence. Does microscopy support the proposed defect or damage mechanism? Does changing temperature produce the predicted trend within the same structural regime? Can a model fitted to one loading history predict another? Such comparisons make a mechanism more persuasive than a single successful fit.

The worked examples have deliberately stopped short of predicting safe lives. A specimen can exceed a deformation limit before rupture, and a component can experience conditions unlike a laboratory test. Qualification requires appropriate data, validated models and the relevant engineering judgement; invented teaching values cannot supply them.

The answer to the title is that a constant external load does not freeze a solid’s internal structure in place. Molecular rearrangements, defect motion and transport processes can continue over time. Understanding creep means discovering which processes are active, measuring their rates under defined conditions, and recognising where an extrapolation would outrun the evidence.

Return to the Materials Science guide to connect structure and processing with other properties, or explore the How Science Works master guide.

Contents · Previous chapter · Continue to Materials Science