The flaw that sets the strength

A structural calculation compares a stress with a strength and passes or fails. That comparison contains an assumption so basic it is never stated: that the member is made of material, everywhere, with no gaps in it.

Nothing is. Rolled plate contains inclusions and laminations, welds contain lack-of-fusion and slag, flame cuts contain notches, and anything that has been in service contains fatigue cracks. Most of those are small and harmless. Some are not, and the boundary between the two is a length that can be computed.

Why a crack cannot be handled with a stress concentration factor #

The natural first move is to treat a crack as a very sharp notch and look up its stress concentration factor. A circular hole multiplies the stress by three; an ellipse three times as long as it is wide multiplies it by seven; the factor for an ellipse is $1 + 2\sqrt{a/\rho}$, where $\rho$ is the radius at the tip.

Send $\rho$ to zero, as a crack does, and the factor goes to infinity. That is not a large number; it is the wrong kind of answer. The stress at a crack tip is unbounded in the elastic theory whatever the applied stress, so the comparison “stress against strength” cannot be made at all — every crack under any load would fail.

What is finite is the rate at which the stress becomes infinite. Near a crack tip the elastic stress field goes as $\sigma \sim K/\sqrt{2\pi r}$, and the constant $K$ in front of the singularity is well-defined, has dimensions of stress times root length, and depends on the applied stress and the crack size in the combination

$$K = Y\sigma\sqrt{\pi a}$$

with $Y$ a geometry factor near 1. That is the whole of linear elastic fracture mechanics. Fracture happens when $K$ reaches a critical value $K_c$ — the fracture toughness — which is a property of the material measured on a specimen with a deliberate crack in it.

The move is the one physics makes repeatedly: when a quantity diverges, find the coefficient of the divergence and make that the thing with a critical value.

The transition length, and why it moves the wrong way #

Setting the fracture stress equal to the yield stress gives the crack length at which the two failure modes cost the same:

$$a^* = \frac{1}{\pi}\left(\frac{K_c}{Yf_y}\right)^2$$

It goes as the inverse square of the yield stress. Computed for one toughness across four grades: 33.5 mm at 275 N/mm², 20.1 at 355, 12.0 at 460, and 5.3 at 690.

So the stronger the steel, the shorter the crack that takes it away — and it is worse than the arithmetic alone, because toughness and strength are broadly in tension with one another. The metallurgical routes to high strength generally reduce toughness, so a stronger grade usually has a smaller $K_c$ as well as a larger $f_y$, and both terms move the transition length down.

Griffith’s route, which is a different derivation of the same expression #

The stress-intensity argument is a local one about the field near the tip. There is an entirely independent global one, and their agreeing is the reason to believe either.

Consider a crack of length $2a$ in a large plate under stress $\sigma$. Extending the crack releases strain energy from the material either side of it, at a rate proportional to $\sigma^2 a/E$ per unit of extension. It also costs energy: two new surfaces are being made, at a cost of $2\gamma$ per unit area. The crack grows when the release exceeds the cost, which gives

$$\sigma = \sqrt{\frac{2E\gamma}{\pi a}}$$

That is Griffith’s 1921 result, arrived at from thermodynamics with no mention of a stress field at all, and it has the same $\sigma\sqrt{a}$ structure as the stress-intensity criterion. The two are related by $G_c = K_c^2/E$, where $G_c$ is the energy release rate at fracture — for the steel in the figures, $K_c = 100$ MPa√m gives $G_c = 47.6$ N/mm.

The reason the energy route matters is that it explains the strangest feature of the subject: fracture toughness is enormously larger than surface energy. The surface energy of iron is around $2 \times 10^{-3}$ N/mm; the measured $G_c$ is 47.6, which is twenty thousand times greater. Almost none of the energy is going into making surfaces. It is going into plastically deforming a small volume of material at the crack tip, and toughness is essentially a measure of how much material yields before the crack advances.

Which is why toughness and ductility are so closely related, and why anything that suppresses local yielding — low temperature, high strain rate, triaxial restraint in a thick section — reduces toughness sharply while leaving the yield stress alone.

What the toughness is worth, against what the strength is worth #

The two properties can be put on the same axes and compared directly, and the comparison settles an argument that intuition gets wrong.

The two figures together give the practical rule. A member is safe from fracture if the largest flaw that could plausibly be present is well short of the critical length at the working stress, and the three ways of achieving that are: keep the stress down, keep the toughness up, and know what the largest flaw is. The third is inspection, and it is the only one of the three that is not a design decision.

Which free body produced the number #

The free body for a fracture calculation is unlike any other on this site, and naming it properly is the point of this section.

It is not a cut through the member with forces on the face. It is the region ahead of the crack tip, and what crosses its boundary is not a force but a flow of energy. The equilibrium statement — the sums cancel — is replaced by a balance statement: the energy released by the crack advancing equals the energy absorbed by advancing it.

That is a genuine change of method rather than a variation on one. Every other number on this site comes from cutting something open and insisting that forces and moments sum to nothing. This one comes from letting something move a little and insisting that the energy books balance, which is the same machinery as the virtual-work argument for a deflection and the same machinery as the work equation for a collapse mechanism — a family of methods in which nothing is cut and something is allowed to move.

The numbers quoted follow from it in two steps that can be checked against each other. $a^*$ from setting $Y\sigma\sqrt{\pi a} = K_c$ at $\sigma = f_y$, and $G_c$ from $K_c^2/E$; the first is a stress argument and the second an energy one, and they describe the same event.

Where the model stops #

Linear elastic fracture mechanics requires the plastic zone to be small, and it stops being small exactly where it matters most. The zone size is roughly $(1/2\pi)(K/f_y)^2$, which for the steel here at yield-level stress is about 4 mm — comparable with the transition crack length itself. The standard validity requirement is that the crack and the specimen thickness both exceed $2.5(K_c/f_y)^2$, which for these numbers is 198 mm. A specimen thinner than that cannot measure the toughness being used.

That is not a footnote. It means the quantity in every calculation on this page is properly defined only for sections much thicker than most structural members, and the values quoted for thinner ones are transitional. For genuinely ductile behaviour the theory is replaced by elastic-plastic fracture mechanics, with the crack-tip opening displacement or the J-integral as the parameter.

Toughness is not a number, it is a curve against temperature. Mild steel’s toughness falls by an order of magnitude over a range of perhaps 50°C, and the transition sits uncomfortably close to ordinary service temperatures for thick sections. This is the property that makes a steel meeting every strength requirement structurally useless in the cold, and it is the reason toughness is specified by a subgrade tied to thickness and minimum service temperature rather than by a design calculation. It is also the sharpest instance of the general point that a material property is a summary of a test: the Charpy test that fixes a subgrade measures energy absorbed by a small notched bar struck by a pendulum, and it correlates with $K_c$ without being convertible to it.

Nor is a member’s toughness the toughness of its plate. A weld is a different material, and the heat-affected zone beside it is a third — coarse-grained, sometimes hardened, and usually the least tough part of the assembly. The joint is where the ductility runs out, and it is where the flaws are too.

And a crack in service is not a crack of fixed length. Fatigue grows it, and the interesting question then is not whether the present crack is critical but how long it takes to become so. That is what makes inspection intervals a calculable quantity rather than a habit.

What the picture cannot show #

The crack is drawn as a length and it is a shape. A crack in a real member is a surface flaw of some depth and some aspect ratio, embedded in a plate of some thickness, at some distance from a free edge — and all of that is folded into the single factor $Y$, which is 1.12 for one particular idealisation and is a handbook entry for anything else. The horizontal axis says “crack length” and means “the crack length of the configuration $Y$ was chosen for”.

And the plot has no time in it. Every point on it is a crack that exists now, at a stress applied now. A member’s actual history is a small flaw that grows under repeated load, and the useful question is not whether the present crack is critical but how many cycles remain before it becomes so — which is a question the fatigue essay asks with the same $K$ on its axis.

Nor is there a temperature on it. The single toughness value that fixes the falling curve is the property that varies most and most abruptly of anything in this field. The same steel at 20°C and at −20°C would be two curves a factor of two apart in stress at every crack length, and the plot draws whichever one it was given.

The generalisation #

The pattern is that a strength is a property of a specimen and a structure is not a specimen.

Every material property in this field has turned out to be a summary of a test, and this is the sharpest case: the tensile test measures a bar with no crack in it, and reports a number that a cracked member cannot achieve. The two quantities are not related — knowing the yield stress tells nothing about the toughness, and steels of identical strength differ in toughness by a factor of ten.

The deeper generalisation is about which kind of quantity governs. This site’s other failure modes are governed by a stress compared with a stress, or a load with a load. Fracture is governed by a stress times a root length compared with a material constant of the same odd dimensions, and the appearance of a length in it is the whole difference. A quantity with a length in it does not scale — two geometrically similar structures do not have the same fracture behaviour, because the crack in the larger one is longer in absolute terms while the stress is the same.

A surprising place this turns up #

The scaling consequence has a name and it is one of the reasons large structures are not small ones enlarged.

Take a structure and double every dimension, including the flaws that scale with the process — weld defects, laminations, the notch left by a flame cut. The stresses are unchanged if the loads scale with the areas. But $K = Y\sigma\sqrt{\pi a}$ has gone up by $\sqrt{2}$, because $a$ doubled and $\sigma$ did not. The larger structure is 41% closer to fracture with every stress in it identical.

This is why the thickness of a plate appears in toughness specifications at all — a thicker plate is closer to plane strain, which suppresses yielding at the tip, and is likely to contain larger flaws, and has more restraint against the local yielding that toughness depends on. Three separate mechanisms, all pushing the same way, none of them visible in a stress calculation.

It is also why the failures that established this subject were large structures. Ships, pressure vessels, bridges and rocket casings: the members that fractured in service were the thick ones, and the same detail in a thinner section had been performing without complaint for years.

Where the ladder goes next #

Later rungs on this anchor: the geometry factor $Y$ for real configurations — edge cracks, surface cracks, cracks at holes — and how a handbook of them is compiled. The plastic zone and its correction to the effective crack length. Elastic-plastic fracture mechanics, with CTOD and the J-integral, which is what applies to structural steel at service temperatures. The ductile-brittle transition and the Charpy test that measures it, and why a Charpy value is correlated with toughness rather than convertible to it. Constraint and thickness, and the plane-stress to plane-strain shift. Leak-before-break as a design philosophy for pressure vessels. And fracture in concrete, where the process zone is large, the material is quasi-brittle, and the size effect on nominal strength is a measurable rather than a theoretical curiosity.

Historically the subject arrived from failures. Griffith’s 1921 paper was about glass fibres and was ignored by engineers for twenty-five years. What ended the neglect was the Liberty ships: over two hundred serious hull fractures and a dozen vessels breaking completely in two, in cold water, at stresses well inside the design values, on steel that passed every specification then in existence. The investigation established that a material could be strong and useless, and Irwin’s reformulation of Griffith’s energy argument into the stress-intensity factor in the 1950s gave the profession something it could compute with. Nearly every advance in this field has followed the same sequence.