Twelve years ago I stood in a shop in Sharjah and watched a crew scrap forty-eight elbows in one afternoon. The cheeks were square. The gores were sheared clean. Everything looked right on the bench, and nothing would close up on the floor.
The fault was on the drawing. Someone had dimensioned the elbow to the outside of the heel instead of the centreline, and every part had been cut to a radius 300 mm too small. Forty-eight fittings went into the scrap bin, and the line sat idle for a day and a half while new blanks were nested.
Rectangular elbows are where a duct line stops behaving like a straight-line machine. A line will cut and notch a straight section to a tenth of a millimetre all day long. An elbow asks you to do geometry first, cutting second, forming third and assembly fourth, and a mistake in step one only surfaces in step four. This is the sequence I use now, with the numbers that actually matter.
I have seen more elbows ruined by a radius argument than by bad welding. Decide three numbers with the customer before the first blank is cut: duct width, centreline radius, and gore count. Everything else is arithmetic.
A centreline radius of 1.5 times the duct width is the normal choice for HVAC work. On a 600 mm duct that puts the throat radius at 600 mm and the heel at 1,800 mm. Air turns through a throat as wide as the duct itself, the pressure loss stays close to a long-radius fitting, and the parts are still small enough to handle with two men.
Short-radius elbows at 1.0D get specified when a ceiling void is tight. The throat radius drops to half the duct width, 300 mm on the same 600 mm duct. That throat is where the trouble starts. The airstream has to turn 90 degrees through a passage narrower than the duct, flow separates on the inside of the turn, and the loss coefficient roughly doubles compared with 1.5D. On a supply run moving 3,000 m³/h at 8 m/s, two or three of those short elbows can cost you 60 to 80 Pa that the fan has to make up for the next fifteen years. If a consultant asks for 1.0D, I quote it, but I also put the pressure penalty in writing.
There is a hard floor as well. Below a 1.0D centreline radius the throat gets so tight that a standard Pittsburgh lock will not close on the cheek-to-gore joint, and you end up welding or riveting. That is a different price.
Every joint in a 90 degree elbow turns the airstream by a fixed angle. A five-piece elbow has four joints, so each joint turns 22.5 degrees, and each gore end is cut at half of that: 11.25 degrees. A three-piece elbow turns 45 degrees per joint and needs a 22.5 degree cut. Fewer pieces means less assembly labour and a coarser-looking fitting; more pieces means a smoother turn and more seams to seal.
For shop work I stay with five pieces up to about 1,000 mm wide and go to seven pieces above that, mainly because a wider gore becomes awkward to feed through a lock former and the seam starts to wander.
The gore width falls straight out of the half-angle:
Outside width = 2 × (R + W/2) × tan(11.25°)
Inside width = 2 × (R − W/2) × tan(11.25°)
tan(11.25°) is 0.1989. That single constant is worth writing on the wall next to the shear. Here is what it gives for common sizes at 1.5D with a five-piece elbow:
| Duct width W (mm) | Centreline R (mm) | Throat radius (mm) | Gore width, outside (mm) | Gore width, inside (mm) | Cheek blank, one side (m²) |
|---|---|---|---|---|---|
| 300 | 450 | 300 | 239 | 119 | 0.212 |
| 500 | 750 | 500 | 398 | 199 | 0.589 |
| 600 | 900 | 600 | 478 | 239 | 0.848 |
| 800 | 1,200 | 800 | 637 | 318 | 1.508 |
| 1,000 | 1,500 | 1,000 | 796 | 398 | 2.356 |
| 1,250 | 1,875 | 1,250 | 995 | 497 | 3.682 |
Those cheek figures come from the quarter-annulus area, π/4 × (Rout² − Rin²), and they are the number I use for coil planning. A 600 mm elbow eats 1.7 m² of sheet for the two cheeks before a single gore is cut. That is why a quote built only on duct weight comes out short on fittings.
An auto duct line is built for rectangles. Coil feeds in, the shear cuts to length, the notcher takes the corners, the beader puts in the stiffening ribs, and TDF or Pittsburgh rolls form the seams. None of that helps much with a quarter annulus. So I split the work.
There are two practical routes for the cheek. The first is to cut a rectangle that bounds the quarter annulus and then trim the arc on a plasma table or with a template and hand shear. You lose material, up to 21 percent on a 1.5D cheek, but you keep the line's speed and the edge quality of a sheared blank. The second is to nest the annulus directly on a CNC plasma table with a shared common-line cut, which recovers most of that 21 percent but loads the plasma table and leaves a slightly rougher edge.
On Line 5 with a 1,500 mm coil, four 600 mm elbows come off one 2,500 mm advance if the cheeks are cut as bounding rectangles with the arcs nested inside. The line produces the blank in about 26 seconds including notching. The plasma table would take four times as long per part but use 15 percent less coil. Below about 40 elbows a month I take the line and the waste. Above that the plasma table earns its keep.
Gores are trapezoids with two angled ends and two square ends. They are simple shapes, and a shear handles them well if you build a stop. I set an angle gauge on the backstop at 11.25 degrees and cut gores in stacks of six to eight sheets, which is fine on 0.8 mm galvanised and gets ugly above 1.2 mm because the stack creeps. For anything over 1.5 mm, or for stainless, cut them singly on the plasma.
The detail that catches people is the seam allowance. If the gore is going to lock into the cheek, each of the two long edges needs 8 to 10 mm of allowance for the Pittsburgh lock, and the lock is formed after cutting. Cut the gore to finished width and you will be 16 mm short on every joint, which shows up as a gap that no amount of sealant will close.
Three routes are common in export shops, and the choice is mostly about how many elbows leave the door each week.
The cheek is bent on a press brake or a hand folder, the gores are formed on a hydraulic gorelocker, and the whole thing is assembled on the bench. This is the cheapest entry point and the most flexible. A hydraulic gorelocker will handle 0.5 to 1.5 mm galvanised and will put a proper double-seam on a gore in about 8 seconds. Two men can build a 600 mm five-piece elbow in roughly 22 minutes, of which 14 minutes is handling and sealing.
The weak point is consistency. Two men making the same elbow on Monday and Thursday will not produce the same throat radius, because the cheek bend line is set by eye against a template. On a job where 120 elbows have to stack in a riser, that variation becomes a site problem.
An elbow making machine rolls the gore to a set radius and closes the seam in one pass. You feed a flat blank, the rolls curve it, the lock closes, and the gore comes out with the curvature already in it. Cycle time drops to about 12 seconds a gore and the radius is repeatable because it comes from the roll set, not from a man's eye.
The trade-off is tooling. Each radius band needs its own roll set, and a set for 1.5D covers roughly one duct width either side before the fit gets loose. A shop running 300 to 1,250 mm duct needs four or five roll sets, and each change is a 20 to 30 minute job. If your order book is mostly one or two sizes, this route wins on labour. If you are quoting twenty different sizes a month, you will spend your life changing rolls.
For odd sizes, stainless, or anything under ten pieces, buying from a fitting house is rational. You pay 2.5 to 3.5 times the material cost, but you avoid setup, scrap and the risk of a wrong radius. I keep a bought-in allowance in every quote for the sizes I know we will not run economically.
| Route | Labour per 600 mm 5-piece elbow | Scrap rate | Best fit |
|---|---|---|---|
| Folder + gorelocker | 22 min, two men | 4–6% | Under 40 elbows a month, mixed sizes |
| Dedicated elbow machine | 9 min, two men | 1–2% | Over 60 elbows a month, repeat sizes |
| Bought in | 0.5 min handling | 0 | Under 10 pieces, stainless, odd radii |
Most leaks on a rectangular elbow are not on the seam. They are at the four corners where the cheek meets the first gore, because three edges converge there and the lock cannot close all of them.
My sequence: lock the two cheeks to the first and last gore on the bench, then close the intermediate joints from the inside out. Before closing, run a bead of sealant along the inside of each lock, not on the outside after assembly. Sealant applied outside a Pittsburgh lock sits on top of a gap and peels off within a season. Applied inside before closing, the lock crimps it into the joint and it stays.
For medium-pressure work I add a 3 mm closed-cell gasket at the access panel and at any bolted connection, and I rivet rather than self-tap the corners. Self-tapping screws into 0.8 mm galvanised strip their thread before they pull the joint tight, and every stripped screw is a leak path with a screw head sitting in it.
If the elbow carries a TDF flange at each end, the flange goes on last, after the body is closed and squared. Flanging a body that is still twisted locks the twist in permanently. I have pulled apart elbows where the flange was perfectly square and the body was 6 mm out of wind, and the only fix was to cut the flange off and start again.
Four checks, five minutes, first article and then one in twenty.
Face-to-face dimension. Tape from the outside of one flange face to the other. Tolerance plus or minus 3 mm on anything up to 1,000 mm. Beyond that, plus or minus 5 mm.
Squareness of the end faces. Stand the elbow on a flat bench and check the gap under the flange with a feeler gauge. More than 2 mm across a 600 mm face means the body is out of wind.
Throat radius. A plywood template cut to the nominal throat radius, offered up to the inside of the turn. A 5 mm gap is acceptable. A 10 mm gap means the gores were formed to the wrong radius and the fitting will not sit in the riser.
Joint gap. Run a finger along each lock. Any gap you can feel with a fingernail is a leak. On a 600 mm elbow at 500 Pa, a 0.5 mm gap along 200 mm of seam will pass enough air to fail a leakage test on its own.
On 0.8 mm galvanised at our coil price, the material in a 600 mm five-piece elbow is about 2.4 m², roughly 15 kg, and lands around USD 19. Add 22 minutes of two-man labour at USD 9 an hour and you have another USD 6.60, plus sealant, rivets and 5 percent scrap. Call it USD 28 in-house. A bought-in equivalent of the same size quoted to me last quarter was USD 71 delivered.
That gap is the whole argument for making fittings yourself, but it only holds if the line is not waiting. If producing elbows takes the shear operator off the line for two hours, the real cost is the straight duct that did not get made. That is why I schedule fittings in a block on Friday afternoon, when the line is doing maintenance anyway.
Usually not on 0.8 mm and thicker. The gore curvature needed at 1.0D is tight enough that the lock former cannot reach the seam without marking the part, and the cheek-to-gore lock will not close cleanly. Expect to weld or rivet, and price it accordingly.
No. One roll set covers a band of roughly one duct width either side of its nominal size before the seam fit gets loose and the gore starts to spring. In practice, four sets cover 300 to 1,250 mm.
Almost always the seam allowance. If the gore was cut to finished width and the lock was formed after cutting, the extra material goes into the joint and pushes the faces apart. Cut gores 8 to 10 mm oversized per locked edge, or allow for it in the development length.
Only for stainless, for high-pressure systems above 1,500 Pa, or where the specification calls for it. Welding a galvanised elbow burns the zinc and leaves you chasing porosity. Lock plus internal sealant passes a normal leakage test and takes a fifth of the time.