Leak-before-break

Why do engineers deliberately design a pressure vessel so that a growing crack will dribble rather than hold?

Here is a design objective that sounds like an admission of defeat: make sure the vessel leaks. Not "make sure it never cracks" — that goal was abandoned deliberately — but arrange the steel, the wall thickness and the working stress so that when a crack does grow, the first thing it does is weep into the room rather than tear the vessel apart. Why would anyone accept a leak as an outcome, and what is the alternative that makes leaking the better one?

Begin by discarding the goal that seems obvious. A defect-free vessel is not available: welds contain flaws below the resolution of any inspection you can afford, and vessels are cycled, heated and corroded for decades. The honest question is never whether a flaw exists but what it will do.

So what governs whether a crack grows quietly or runs? A sibling explanation here on stress concentration follows a notch as far as its tip radius allows, then stops: as the radius goes to zero the elastic stress at the tip goes to infinity and becomes useless. Fracture mechanics replaces it with a quantity that stays finite: the stress intensity factor, K, combining applied stress and crack size into one number for the severity of the whole crack-tip field. For a crack of half-length a under a stress sigma, K is sigma times the square root of pi times a, adjusted by a geometry factor near one in simple cases. The material answers with one property, its fracture toughness K-one-C, and the crack runs when K reaches it.

Set those equal and rearrange, because the rearrangement is the entire subject. The critical half-length is one over pi, times the square of the toughness-to-stress ratio — so halve the working stress and the tolerable crack quadruples.

Now compare two lengths. A buried flaw in a wall of thickness t grows outward until it breaks through, and the through-wall crack it makes at that moment has a surface length of the order of twice the wall thickness. Against that, the critical length above. If the critical length is comfortably the larger, the crack penetrates while still stable, and pressure escapes through a crack not yet long enough to run. If it is the smaller, the flaw reaches instability before it ever reaches the far surface, and the vessel opens with no warning at all.

Put illustrative numbers on it, chosen to show the argument's shape rather than any real component. Take a hoop stress of 150 megapascals and a toughness of 200 megapascals root-metre. The critical half-length is one over pi times the square of 200 over 150, which is 0.57 metres — a stable crack over a metre long. A 40-millimetre wall breaks through at a crack of roughly 80 millimetres. Not close: the vessel leaks with enormous margin.

Now drop the toughness to 50 and change nothing else. Because the relationship is quadratic, a fourfold fall in toughness cuts the critical length sixteenfold, to 35 millimetres of half-length, or 71 overall — now shorter than the crack the wall produces on penetration. Same vessel, same stress, same wall, and it has silently stopped being a leak-before-break vessel.

That is the binding constraint, and it is not the steel's strength. It is the ratio of toughness to working stress, squared — so the argument is destroyed by anything lowering toughness: operating below the ductile-to-brittle transition, neutron embrittlement, thermal ageing of cast stainless, or the extra constraint a very thick section imposes on metal at the crack tip. A thicker wall, which intuition says is safer, makes the case harder on both counts at once.

The second half of the argument is not fracture mechanics at all. A leak only warns if it is heard, so the case must also show that flow through a crack of that size beats the plant's leak-detection sensitivity by a real margin, with time to depressurise before the crack grows on to critical length. The leak is a sensor for a defect nobody can see.

Separate this from the fail-safe designs described elsewhere in this collection. A train's air brake makes the loss of a signal mean "stop", so any interruption becomes the safe instruction. Leak-before-break makes no claim about signals: it compares two lengths — the crack that penetrates a wall, and the crack that runs — and is only as good as that arithmetic.

The load-bearing claim is that the critical crack length scales as the square of toughness over stress, and leak-before-break holds only while that length exceeds the through-wall crack the vessel's own thickness produces. The direct test is qualification practice: machine through-wall slits of increasing length into test vessels, pressurise until one runs, and compare that length with the prediction. The sharper prediction is temperature — the same vessel should be leak-before-break when warm and lose the property when cold, as toughness collapses through the transition. The refuting observation would be a vessel rupturing from a crack well below its predicted critical length at the assumed toughness and stress.

Two honest limits. The geometry factor I glossed as "near one" is not one for a through-wall crack in a pressurised cylinder, where the shell bulges at the crack and raises K; real assessments carry that correction and I have left it out. And leak-before-break is partly a regulatory instrument: formalised in nuclear licensing practice, where demonstrating it permits pipe-whip restraints and impingement shields to be omitted — a large saving. That commercial motive is why the codes exclude degradation mechanisms it cannot cover.

• How stress-corrosion cracking produces long shallow flaws that defeat the leak-before-break comparison outright.

Every term the collection defines is gathered in the glossary.