The soft zone that takes all the stretch

The strength the welder gives back measured what a weld costs an aluminium member in strength: beside the weld, for twenty-five to forty millimetres either side, the heat dissolves the precipitates a heat-treated alloy was strengthened by, and the metal there keeps half its proof stress. It closed on one favourable note. The zone’s ultimate strength falls much less than its proof stress, so “a heat-affected zone is weaker and considerably more ductile than the parent — which means a member fails there in a way that gives warning”.

The first half of that is true of the material. The second is a statement about the member, and it does not follow. A member welded across its section is a soft zone in series with a strong one, pulled by the same force, and the question of how far it stretches before it fails is decided by which of the two reaches its limit first — not by which of the two is more ductile.

A soft zone in series with a strong one #

The two curves in the figure are the same alloy. The dashed one is the member as extruded, rising from its proof stress of 260 N/mm² to its ultimate of 310 over seven per cent of strain, and stretching 140 mm before it necks. The solid one has a weld across its middle and stops almost at once: it reaches its maximum load at 0.62 per cent overall strain, 12.3 mm of stretch, and then necks in the heat-affected zone.

The reason is in the numbers the Eurocode for aluminium gives the zone. For 6082-T6 it keeps 0.48 of the parent’s proof stress and 0.60 of its ultimate, so the zone yields at 125 N/mm² and fails at 185. The zone’s ultimate strength is below the parent’s proof stress. The force that breaks the zone is too small to yield the parent, so the parent stays elastic throughout, and every millimetre of permanent stretch the member ever makes is made in the sixty millimetres of metal beside the weld.

That the zone is ductile does not help. It is: on the assumptions drawn here it stretches twelve per cent before it necks, against seven for the parent. But twelve per cent of sixty millimetres is seven millimetres, and seven per cent of two metres is 140. Ductility is a strain, and a strain has to be multiplied by a length before it is a stretch.

All of the stretch in one short zone #

Drawn along the member, the strain at maximum load is a spike. The heat-affected zone is at twelve per cent, the limit of its uniform elongation; the parent on either side is at 0.26 per cent, which is 185 N/mm² over the modulus of 70,000 — an elastic strain that goes away when the load is removed. Of the 12.3 mm the member stretches, 7.2 are permanent and in the zone, and 5.1 are elastic and everywhere else.

This is the ductility that depends on the ruler turned from a measurement problem into a design one. A tensile test reports an elongation over a gauge length, and a short gauge across a neck reports a large number because the neck is most of what it measures. A welded member is a specimen whose gauge length is two metres and whose neck was put there deliberately by the welder, at a known position, before any load arrived. Its elongation is the zone’s strain times the zone’s length, spread over the member’s length, and the result is small in exact proportion to how short the zone is.

It is also the curve that was rising the whole time seen from the other side. The parent’s own curve keeps rising to 310 N/mm², and none of that rise is ever used: the member’s force is capped at 185 by the zone, and the parent spends its whole life on the straight elastic part of a curve whose useful part lies beyond a load it is never given.

Once welded, a longer member is less ductile #

For an unwelded member, ductility as a strain is a property of the alloy: seven per cent at any length. For a welded one it is not, because the plastic part of the stretch is fixed at the zone’s seven millimetres whatever the member’s length, and the rest is elastic. A metre-long welded member reaches 0.97 per cent overall; a ten-metre one 0.33 per cent, barely above the parent’s elastic strain at the zone’s failure load. The longer the member, the more completely its ductility is the zone’s, and the less the zone’s is worth.

That inverts the usual expectation of size. A longer tie is normally the more forgiving one — it has more metal to yield and stretches further before it breaks — and a structure that relies on a tie stretching to redistribute load, or to show distress before it fails, counts on that. A long welded aluminium tie gives almost none of it. Its warning is the few millimetres of a sixty-millimetre zone, and those millimetres arrive at a load well below the one the parent would have carried.

The ratio that decides where it stretches #

One ratio decides the whole behaviour: the zone’s ultimate strength over the parent’s. The figure sweeps it and holds everything else. Below the parent’s own ratio of proof stress to ultimate, 0.84 for 6082-T6, the zone fails before the parent can yield, and the member keeps a tenth of its ductility. Between 0.84 and one, the parent yields before the zone necks, but only just, and the member’s stretch climbs slowly. Only as the ratio approaches one does the neck move out of the zone and into the parent, and then the member stretches like an unwelded one.

The shape is a cliff rather than a slope, and it is the shape of every race between two parts in series. The tear that goes diagonally found the same race in steel, between gross yielding over the member and net-section rupture across a bolt hole — $f_y A_g$ against $f_u A_n$ — and the same conclusion: a member whose net section breaks before its gross section yields has lost its ductility, however ductile the steel. Steel usually wins that race because its ultimate is well above its yield. A heat-treated aluminium alloy welded across loses it by a wide margin, because the weld has taken four tenths of the ultimate away from exactly the section that decides it.

The thicker land that gives it back #

The remedy an extrusion designer has is geometry. An aluminium section is extruded to whatever shape the die makes, and a section that will be welded across can carry a thickened land at the weld position, so that the zone has more area than the parent. The question is how much, and the figure gives two answers that differ by more than one might expect.

Thickening the zone to 1.41 times the parent’s area, which is 260 over 185, makes the zone’s failure load equal to the parent’s proof load. By a strength criterion that is enough: the member can now carry the load at which the parent yields, and nothing about its capacity is lost to the weld. But the member keeps 13 per cent of its ductility, because the zone still fails the moment the parent begins to yield. Restoring the strength does not restore the stretch. For that the zone must outlast the parent’s whole hardening curve, and the land must be 1.68 times as thick, 310 over 185. Then the parent reaches its ultimate first, necks there, and the member stretches 142 mm — slightly more than the unwelded member, because the thick, soft zone adds a little of its own.

The gap between those two numbers is the gap between the two questions a designer can ask of a joint. A strength check asks whether the weakest section can carry the design load; a ductility check asks whether the member yields before its weakest section breaks. For a steel member the two usually give the same answer, and the property that appears in none of the equations is taken on trust. For a welded aluminium member they give different answers, and only the first is routinely asked.

An alloy chosen for welding #

The alloy that avoids all of this is the one that looks worse on paper. 5083 in the annealed H111 temper has a proof stress of 125 N/mm², half that of 6082-T6, and the weld’s heat has nothing to take away: the metal was annealed before the arc arrived, and the zone beside the weld is the same metal as the rest. A member of it welded across stretches exactly as an unwelded one does, 320 mm on two metres, because there is no soft zone for the stretch to collect in.

The first essay on this found that the welded strengths of the two alloys land within a few per cent of each other, so the heat-treated alloy’s advantage is spent on its first weld. Ductility completes that comparison in the annealed alloy’s favour: at the same welded strength, one member stretches 320 mm before it necks and the other 12. It is part of why the 5000-series alloys are the usual choice for welded plate structures such as hulls and tanks, and it appears nowhere on the alloys’ data sheets, where the heat-treated one wins on every line.

Steel, where it starts to matter #

Structural steel of ordinary strength has no soft zone worth the name — the metal beside a weld is usually harder than the plate — and its ultimate strength is well above its yield, 490 against 355 N/mm² for S355. Both margins protect it. The case that erodes them is the high-strength quenched-and-tempered plate, whose strength was bought by a heat treatment much as 6082’s was, and whose ratio of yield to ultimate is near 0.9 — 690 against 770 N/mm² for S690.

The same series model makes the consequence plain. If the zone beside the weld loses five per cent of its strength, as such steels can, its ultimate is 731 N/mm²: above the parent’s yield, so the parent does yield, but below the parent’s ultimate, so the neck still forms in the zone. On the same two-metre member, with an assumed uniform elongation of eight per cent for both, the welded member keeps 24 per cent of its stretch. Lose ten per cent and it keeps 11. The margin a high-strength steel has against this is its own ratio of yield to ultimate, and it is ten per cent wide. It is one reason the weld procedures for these steels limit the heat input so tightly: the heat input decides the softening, and the softening decides where the member stretches.

What the strength check does not ask #

Aluminium design rules check a welded member at its heat-affected zone against the zone’s reduced strength, and that check is correct: the capacity it computes is the capacity the member has. What a capacity check does not do, by itself, is compare the zone’s failure load with the parent’s yield load, because it has no reason to. A designer who then assumes, as every plastic method does, that the member will yield along its length before anything breaks has made an assumption the check never tested.

Which free body produced the number #

Cut the member anywhere and the force on the cut is the same: one force, carried in series from end to end. That single fact is the whole of the argument. The force at every section is the force at the weakest section, so no section can be taken past the stress the weakest one fails at, and a strong section beside a weak one is a strong section that is never used.

The member’s extension is the sum over its length of each section’s strain, and each section’s strain is its own material’s response to the common force. At the zone’s failure load the zone is on the plastic part of its curve and the parent is on the elastic part of its own; the sum is a large strain over a short length plus a small strain over a long one, and on any member much longer than its zone the second term wins.

The 6082-T6 tie by hand #

At the zone’s failure load of 185 N/mm², the parent’s strain is $185/70{,}000 = 0.26$ per cent, which over the 1,940 mm of parent is 5.1 mm. The zone, at its uniform elongation of 12 per cent, adds $0.12 \times 60 = 7.2$ mm. The member stretches 12.3 mm, against $0.07 \times 2{,}000 = 140$ mm for an unwelded member. The land that moves the neck is $f_u/f_{u,haz} = 310/185 = 1.68$; the land that only restores the proof load is $f_o/f_{u,haz} = 260/185 = 1.41$.

Where the model stops #

The zone is uniform and sharp-edged. A real heat-affected zone is graded, softest near the fusion line and recovering over its width, and the weld metal is a third material with its own curve. The grading concentrates the strain further into the softest few millimetres, so the uniform zone here overstates the stretch the weld gives rather than understating it.

The uniform elongations are assumed. The strength reductions are the Eurocode’s; the seven and twelve per cent are representative rather than specified, and the argument does not depend on them. A zone with twice the assumed elongation would still stretch only fourteen millimetres, because its length, not its strain, is what limits it.

The member is followed to maximum load. After that the neck in the zone thins and the load falls; the neck adds some millimetres that depend on the thickness rather than the length, and the member’s final elongation at fracture is a little larger than the figures show. It is still a local number on a long member.

The zone is across the whole section. A weld along a member softens a strip rather than a section, and the rest of the section is parent metal in parallel with it, which yields at its own proof stress and carries the member to a stretch much nearer the unwelded one. That is the orientation the strength the welder gives back found cheap in strength, and it is cheap in ductility for the same reason.

What the pictures cannot show #

That the lost stretch is a lost warning. A structure designed to redistribute — a statically indeterminate frame whose moments are moved on purpose, a joint whose bolts must share load, a tie that is meant to yield visibly before anything breaks — assumes its members can be strained beyond yield without failing. A welded 6082-T6 member cannot, and it gives no sign of it in any strength check: the capacity computed at the zone is correct, and the assumption that the member will yield first is silently false.

The same localisation governs the end bolt of a long joint, whose deformation capacity decides whether its neighbours ever share the load. Wherever a weak part sits in series with a strong one and the design relies on the strong part yielding, the question is the same one the ratio answers here: does the weak part outlast the strong part’s whole curve, or only its first point?

Still open: two welds, and the length between them #

A member welded at both ends, or spliced twice along its length, has two soft zones, and the parent between them is the same as before. If the zones are identical, the neck forms in one of them and the other unloads with it, so the second weld adds nothing to the stretch and nothing to the strength. But no two welds are identical: one was made hotter, or slower, or on a thicker section. Whether a member with two welds is reliably no more ductile than a member with one — or whether the weaker weld is simply the one that fails, so that the scatter between welds, rather than their number, decides where and how the member breaks — is the question two zones in series leave, and it is the one a statistical view of weld quality would have to answer.