Specimen geometry and size effects

Define which dimension is changing

A thin strip and a thick block made from the same material can give different fracture results. That observation has several possible causes. The crack’s driving field may differ; the surrounding material may restrict deformation differently; or the tests may sample different defects. Begin with a drawing so that these explanations do not become one vague size effect.

For an edge through-crack, define a as its depth from the cracked edge, W as the total width in that direction, and b = W − a as the remaining uncracked ligament. B is the specimen thickness, perpendicular to the drawing plane. The ligament is the connected material ahead of the crack that still transfers load. For a central through-crack, a commonly denotes half the total length 2a. The edge-crack expression b = W − a must not be silently transferred to that central convention.

Crack orientation matters too. A rolled, layered, textured or otherwise directional microstructure can resist one crack path differently from another. Record the loading direction and the crack plane and advance direction relative to the material’s processing axes. The same composition and nominal dimensions do not establish the same material condition or crack path.

Shape ratios and absolute crack size enter differently

In the admitted elastic opening-mode model, K_I = Yσ√(πa). The remote stress σ and crack parameter a need the definitions for that geometry. The dimensionless factor Y accounts for such features as crack location, relative crack size, boundaries and loading arrangement. It can depend on a/W, and also on the convention used for nominal stress. A factor supplied for one geometry is not an all-purpose multiplier.

If a crack grows while the width stays fixed, a/W changes. Both √a and Y may then change, so doubling a does not necessarily multiply K_I by exactly √2. To isolate the square-root effect, either use an effectively infinite-plate idealization or explicitly stipulate a fixed Y. For a real finite specimen use its matched expression or analysis.

Now scale every relevant length of a geometrically similar cracked body by s while preserving the nominal stress, loading arrangement and dimensionless shape ratios. In the same elastic model Y stays unchanged and K_I scales as √s. This differs from a dimensionless concentration factor K_t for a geometrically similar finite-radius hole, which can remain unchanged. Absolute crack size enters the dimensional intensity even when the shape ratios match. The fracture criterion and the model’s validity must still be examined.

Compare specimen dimensions with a local yielding scale

A real crack tip has a process zone where elastic response is replaced by plastic flow or other damage. A linear-elastic intensity is useful only if there is an elastic near-tip region outside that zone and inside the larger crack/specimen geometry. A specimen with almost no remaining ligament cannot supply that separation merely because its total length is large.

A simple estimate illustrates the scale. Use an isotropic ductile-metal model with the von Mises yield criterion, which combines the stress components into one yielding measure. For the leading mode-I field directly ahead of the tip under plane stress, that equivalent elastic stress is K_I/√(2πr), where r is distance from the tip. Set that expression equal to the uniaxial yield stress σ_Y: σ_Y = K_I/√(2πr_e). Squaring and rearranging gives r_e = K_I²/(2πσ_Y²). The units are length: (MPa√m)²/MPa² = m.

Call r_e a first elastic-field yield-distance estimate. The real yielded region redistributes stress and is not obtained simply by cutting off every stress component at σ_Y. Its shape and size depend on the yield criterion, hardening, constraint and loading. The estimate is not the full actual plastic-zone radius and is not a universal test-validity rule.

At K_I = 20 MPa√m and σ_Y = 400 MPa, r_e ≈ 0.000398 m = 0.398 mm. A 1 mm thickness is only about 2.5 times this scale, whereas 20 mm is about 50 times it. That comparison helps explain why identical K values can be easier to characterize elastically in some dimensions than others. It does not certify either specimen as a valid plane-strain toughness test.

Thickness changes crack-tip constraint

At an unloaded outer face, the traction normal to that face must vanish. For a sufficiently thin region, the out-of-plane normal stress σ_z can remain small across much of the thickness. This is the plane-stress approximation. Out-of-plane strain need not be zero: material is comparatively free to contract or expand in that direction as the in-plane loading changes.

Far inside a sufficiently thick specimen, surrounding material can restrict that transverse deformation near the crack. The out-of-plane strain ε_z can then be approximately zero over a relevant region: plane strain. A nonzero σ_z develops to enforce the restriction. Zero out-of-plane strain therefore does not mean zero out-of-plane stress. Even a thick specimen still has free-surface regions, so the state varies through the thickness rather than switching everywhere at one boundary.

This extra constraint changes the combination of stresses driving plastic flow, separation and cavity growth. It can reduce the deformation available before crack advance, and measured critical resistance can change. Do not assume that all thin specimens are tougher, that thickness always acts monotonically, or that every critical K is a material constant. Plane-strain mode-I toughness K_IC is a restricted test result, established only when the applicable test’s preparation, loading, dimensional and other validity conditions are met.

A thick specimen can still be too small relative to its yielded zone or remaining ligament; a large specimen can still have the wrong crack preparation, orientation or test record. When extensive yielding is unavoidable, an appropriate elastic-plastic resistance description may be needed. Increasing B alone does not turn an invalid calculation into a valid K_IC measurement.

Larger specimens can sample different flaws

There is another size effect even when the stress state is otherwise comparable. A larger stressed region may contain more opportunities for a sufficiently severe flaw. This is a population argument, distinct from changing Y or deformation constraint. Its form depends on where the relevant flaws are: a surface-controlled population should not automatically be modeled by bulk volume.

Use a deliberately simple weakest-link model. Suppose one region has probability p of surviving the specified stress, and the whole specimen fails if any region fails. If n regions have identical flaw statistics, are loaded equally, and survive independently, whole-specimen survival is pⁿ. For two regions with p = 0.90, survival is 0.90 × 0.90 = 0.81; failure probability is 1 − 0.81 = 0.19. Doubling the number of regions does not simply double a strength value or give a universal percentage strength reduction.

Actual gradients, correlated manufacturing defects, residual stresses and changing microstructure can violate those assumptions. A probability model needs data for its intended population and scale. A thin-versus-thick resistance comparison alone cannot tell us whether geometry, constraint, sampled flaws or material processing caused the difference. Keep those hypotheses separate and seek discriminating evidence.

Use dimensions to explain a result, then judge its label

A useful report supplies more than failure force. It defines the crack and remaining ligament, thickness, loading geometry, nominal-stress convention, material state, orientation, temperature and rate. A fracture test also records how its sharp crack was prepared and how the load/deformation and crack extension were interpreted. These details determine which driving-force and resistance model the result can support.

A plateau in measured critical K with increasing thickness may be consistent with an approach to a constrained regime, but it is not a substitute for the method’s other validity checks. Similarly, a specimen many times larger than a rough r_e estimate has only passed a scale comparison, not an entire test. Current numerical requirements belong to the applicable test method and edition; the physical explanations here do not supply a compliance procedure.

When two reports disagree, first identify which question changed. Was Y recalculated for a different a/W? Was the out-of-plane state changed? Did the microstructure or flaw population change? Were the resistance labels and conditions actually comparable? This order turns a list of dimensions into a causal interpretation.

Worked examples

Try the reasoning

1. An edge-cracked specimen has W = 25 mm, a = 10 mm and B = 4 mm. Find its remaining ligament. A central crack in a different specimen has full length 10 mm. Which a belongs in the central-crack formula?

   Show the answer and reasoning

   The edge ligament is b = 15 mm. The central-crack parameter is a = 5 mm.

   The edge definition gives W − a. The central drawing uses total length 2a. Equal printed numbers called crack length can refer to different parameters; read each geometry definition.

2. A central through-crack of total length 10 mm is in the ideal very wide plate at 100 MPa. Find K_I and explain the result obtained if a student uses the full length as a.

   Show the answer and reasoning

   The correct value is about 12.53 MPa√m; using a = 0.010 m instead gives 17.72 MPa√m, too large by √2.

   Use a = 0.005 m and Y = 1. The error doubles the parameter inside the square root. It does not double K_I and does not change the supplied geometry factor.

3. A crack doubles in a finite-width specimen while W remains fixed. A student predicts exactly √2 times the intensity at the same remote stress. What additional fact is required?

   Show the answer and reasoning

   Whether the geometry factor Y remains unchanged over that crack growth.

   The ratio is K_new/K_old = (Y_new/Y_old)√2. Unlike a full geometric scale-up, growth at fixed width changes a/W and may also challenge the elastic regime.

4. A report describes a constrained interior as plane strain, then sets σ_z = 0 because ε_z ≈ 0. Identify the error and explain what the outer free face requires.

   Show the answer and reasoning

   Plane strain restricts strain, not stress. The constraint can require nonzero σ_z; an unloaded outer face instead requires zero normal traction there.

   The same specimen can have a constrained interior and near-surface regions closer to plane stress. A uniform through-thickness state is an approximation, not implied by the word thick.

5. At the same K_I, doubling σ_Y changes the first-yield estimate by what factor? Does that prove a fourfold increase in fracture resistance?

   Show the answer and reasoning

   r_e decreases to one quarter. No resistance increase follows.

   The estimate is proportional to 1/σ_Y², so doubling σ_Y divides it by four. This is a local yielding-scale calculation at fixed applied K_I; it does not measure the material’s critical resistance.

6. A brittle specimen’s important flaws occur mainly on its loaded surface, and all dimensions are doubled. Why is multiplying a reference volume’s survival by itself eight times not automatically the right model?

   Show the answer and reasoning

   The controlling flaw population may scale with surface area and nonuniform surface stress rather than volume.

   The factor eight comes from a volume scale-up. A suitable sampling model must identify the relevant population, stress distribution and dependence; a geometric volume ratio alone supplies none of them.

What to carry forward

• Define a, 2a, W, ligament and thickness for the actual geometry.
• Shape similarity can preserve Y while absolute crack size changes K_I.
• First-yield scale, through-thickness constraint and complete test validity are distinct questions.
• Weakest-link size effects require an explicitly justified flaw population and probability assumptions.