Four hundred aluminium terrace chairs passed an inspection that sampled fifty of them, looked at every weld by eye and loaded each one to 150 kilograms. Every chair passed. Sixteen of them had a root that was never fused to the frame at all.
Illustrative example This case is written from the patterns we see in furniture export orders, and the figures are representative rather than from a specific client file. It shows how a check works and what it finds. We will replace it with a real, anonymised case as soon as we can publish one with the buyer's permission.
The order, and what nobody asked the factory to prove
A European hotel group ordered 400 aluminium terrace chairs for a coastal resort. Same model as the year before, same frame drawing, same 6063-T5 tube. Two hundred in each of two colours, one consignment, unit price $52, order value $20,800, thirty per cent deposit and seventy per cent balance against shipping documents.
The specification covered the things a chair specification usually covers. Aluminium grade, tube wall thickness, coating system, overall dimensions, seat height, static load, carton count, packing method. It described the frame as welded, and it described the coating as a powder coat to AAMA 2603 with an ISO 9227 salt spray requirement of 480 hours. It did not ask how the weld was made, who made it, or how it would be checked.
The pre-shipment inspection was routine and it was done properly. Fifty chairs were pulled for the AQL sample. Each one was measured against the drawing, each weld was looked at, each chair was loaded to 150 kilograms and held. The report recorded dimensions conforming, weld appearance conforming, static load conforming, coating conforming, carton count conforming. The lot was accepted and the balance of $14,560 was released.
This is an illustrative example built from the patterns behind furniture export disputes: the goods are representative of the category, the figures are representative of the trade, and no particular buyer, factory or hotel is described. It shows what a check finds when the scope agreed before production does not contain the failure the goods are going to have.
Where to go next
→ outdoor furniture verification in Foshan and Guangdong
→ the three levels of check we sell, and what each one covers
What a sectioned joint looks like when the weld was never made
Five months after the chairs were on the terrace, one cracked. Not a bent leg and not a broken slat. The back leg had come away from the seat rail, at the joint where the leg tube meets the frame under the seat, on a chair that had been leaned on and dragged around on tile.
The buyer's engineer cut the leg off the failed chair and had it sectioned and polished. On the visible face of the joint there was a clean, continuous, slightly convex bead, the kind of bead that wins a visual inspection every time. Underneath it, where the bead should have melted into the wall of the tube, there was a narrow band of unbonded metal: a cold lap running for part of the length of the joint, an interface that had never been joined to anything.
Aluminium welds that are not fused are a well-known failure and they have a well-known cause. The oxide layer on aluminium has to be cleaned off immediately before welding and it reforms within minutes. Where it is not cleaned, or where the shielding gas flow is too low and the weld cools before it wets the parent metal, the bead sits on the surface and the root is left unbonded. The joint looks correct from every angle a visual inspection can reach.
Nothing about the outside of that chair told anyone. The bead was continuous, the coating was intact, the frame was the right weight and the right colour. This is the whole difficulty of the defect: it lived inside the joint, and every check the inspection had performed was blind to it by construction rather than by mistake.
| Check performed on the delivered chair | What it would have seen |
|---|---|
| Visual inspection of the bead | A continuous, convex bead. Nothing. |
| Dye penetrant on the finished joint | No surface-breaking defect. The interface is closed and buried under the bead. |
| Static load to 150 kg | The joint held. Static load puts the bonded metal in shear, not the root in tension. |
| Sectioning and polishing the joint | An unbonded band along part of the root. The only method that found it. |
Where to go next
Why 150 kilograms did not find it
Static load testing on furniture is a shear and a deflection test. A chair loaded to 150 kilograms in the middle of the seat and held for a minute is being asked how much the frame bends and whether anything lets go. A joint that is unbonded along part of its length still has sound metal either side of the gap, and under a static load that sound metal carries the load without complaint.
The load case that breaks this joint is a cyclic one, and it is the one a terrace chair gets every single day. A person sits down and stands up. A child swings on the arm. The chair is dragged a few centimetres across the tile and rocks on one corner. Each of those cycles puts the root of the joint into tension, and an unbonded interface has almost no tensile capacity. It behaves as a pre-existing crack, and a pre-existing crack under cyclic loading grows.
The failure is therefore not a strength failure that a bigger load would have exposed. It is a fatigue failure at a stress level the chair sees every day, and a static test never reaches that stress level even once.
- Static load, held: shear through the bonded metal. The lap carries it.
- Sitting down and standing up: the leg joint rotates, and the root goes into tension.
- Rocking on one corner, dragged across tile: repeated rotation at the same point.
- Wind on an open terrace: cyclic side load on the back rail and the leg joint.
- A child on the arm: an off-design moment, delivered at the worst joint in the frame.
What the salt and the season added
These chairs were for a coastal resort. Sea salt is hygroscopic, it holds moisture against the surface and it turns a passive film into an electrolyte. The powder coat over a lap joint is thinner at the toe than anywhere else, and salt solution finds exactly that place.
Crack growth in a corrosive environment is not linear. A fatigue crack in dry air advances at a steady rate. The same crack in salt-laden humidity advances faster, because every cycle of opening the crack exposes fresh metal and the environment does part of the work of breaking it. A crack that would have taken a year to become visible in a dry inland winter can become visible in one season at the coast.
This is why the failure appeared in month five and not month one, and it is the part of the story a pre-shipment inspection cannot reproduce. The inspection sees the chair as it leaves the factory. This chair fails three years into a service life, in weather the factory never saw.
The salt spray requirement in the specification was ISO 9227 at 480 hours, on the coating. That test tells the buyer the coating resists corrosion. It says nothing about the fatigue life of a lap joint under the coating, and it is run on flat test panels rather than on the assembled frame.
Where to go next
The number that explains the whole miss
Here is the arithmetic that turns this from a bad-luck story into a specification story. The lot was 400 pieces, so ISO 2859-1 at general inspection level II puts the sample size at 50. At AQL 2.5 the acceptance number is 3 and the rejection number is 4: three defective chairs found in the sample and the lot still passes.
The lot contained sixteen defective chairs. Put those two facts together and the number that matters is this one. Even with an inspector who never misses a defective chair, and a chair that always fails when it should, this lot still passes 87.4 per cent of the time.
The expected number of defective chairs in a fifty-piece sample is 2.0, which is below the acceptance number of 3. AQL 2.5 is not a search plan. It is a rule for deciding when a lot is good enough to ship, and at this lot size and this defect rate it is calibrated to pass a lot containing four per cent defective chairs. If the real requirement is that not one chair in a thousand has an unfused root, then AQL 2.5 is the wrong instrument, and no amount of care in the field turns it into the right one.
| Sampling plan for a 400-piece lot | Sample | Accept / reject | Chance the lot passes with 16 defective chairs |
|---|---|---|---|
| Level II, AQL 2.5, as ordered | 50 | 3 / 4 | 87.4% |
| Level II, AQL 0, zero acceptance | 50 | 0 / 1 | 11.3% |
| Level III, AQL 2.5 | 80 | 3 / 4 | 59.8% |
Two chairs sectioned at random find this 8 per cent of the time
Widening the sample does not rescue a clustered defect. Sectioning is the only method that reliably finds a lap joint, and sectioning destroys the chair, so nobody sections chairs at random on a four-hundred-piece order. If two chairs are sectioned at random out of the lot, the chance that at least one of them comes from the affected window is 7.8 per cent. Three chairs: 11.6 per cent.
The number changes completely when the sample is drawn from the process rather than from the product. Inside the two-hour window on the third production day, sixteen of the twenty chairs welded had the defect, which is 80 per cent of that window. Section two chairs taken from that shift and there is a 96.8 per cent chance of catching it. Section two chairs from that day and it is 36.2 per cent.
So the useful sampling question is not how many chairs to pull from the shipment. It is how many welds to section per production shift, per machine, per welder. The defect rate was eighty per cent inside one window and zero everywhere else in the lot. Against a distribution like that, a shift-stratified sample is enormously more powerful than a product-random sample of the same size, and it costs no more, because the frames come off the line one shift at a time anyway.
| How two sectioned chairs are chosen | Chance of catching the defect |
|---|---|
| Two chairs at random from the 400-piece lot | 7.8% |
| Two chairs at random from the affected production day | 36.2% |
| Two chairs from the affected two-hour shift window | 96.8% |
| Five test coupons from the same welder, machine and wire in that shift | 99.97% |
Where to go next
The gas log that turned 400 chairs into 20
The number of chairs actually at risk turned out to be twenty, not four hundred, and the only reason anyone could establish that was a record the factory had kept for its own reasons. The welder's shift card carries an hourly check of the argon flow on each machine. On the third afternoon the readings on that machine were 4.5 and 5.0 litres a minute, against the 10 to 14 the welding set-up sheet calls for.
Two hours of low flow. The twenty chairs welded in that window were coated, packed and closed up before anyone compared the readings with the set-up sheet.
This is the part worth arguing for in a purchase order, because it costs the buyer nothing. Require the factory to keep a production trace for every chair, down to the shift and the machine, and require the inspection to read it. With that record the buyer can isolate twenty chairs and test those twenty. Without it the only defensible assumption is that the defect rate in the lot is the defect rate in the affected window, which is four per cent, and the only safe action is to treat all four hundred as suspect.
A buyer who asked for the gas log and was given it turned a four-hundred-chair problem into a twenty-chair problem before a single chair was cut open.
Where to go next
The line in the report that decided the balance
The report the buyer received was not wrong. It recorded what was checked and found all of it conforming: fifty chairs measured, welds visually inspected, static load taken to 150 kilograms, coating checked against the specification, carton count and carton marks verified. Everything inside the agreed scope was done and everything inside it passed.
What the report could not say was that weld integrity had not been checked, because nothing in the scope was capable of checking it. And because the scope was agreed before the goods were made, there was no argument to have once the chair failed: the report had been accepted, the lot had shipped, and $14,560 of balance money had gone out with it.
An outdoor furniture inspection that stops at the surface of the frame cannot see an unfused root, and a report that reads like a clean bill of health invites the buyer to treat it as one. The honest version of the same report carries a line that reads something like this: weld integrity on structural frame joints was not verified by sectioning or by any method capable of detecting an unfused root, and visual inspection of the bead does not detect this condition. That line changes what the buyer can do with the report, and it costs nothing to add.
That is the difference between a report that describes what was checked and a report that describes what the goods actually are.
Where to go next
What it cost, and what one more man-day would have cost
The buyer's engineer loaded and sectioned chairs from the third production day and confirmed the defect in sixteen of the twenty from that window. Four had already cracked in service. All sixteen were replaced, because the factory could not show which of the remaining chairs were sound, and because the join is structural.
Adding a shift-coupon sectioning check and a cyclic load test to the pre-shipment scope would have cost one extra man-day. The coupon comes off the line with the shift it belongs to, and the cyclic load test runs on a chair that was already going to be sacrificed for load testing. Neither line requires the factory to buy anything.
| After the failure | Amount |
|---|---|
| 16 chairs replaced, at the contract unit price | $832 |
| Second inspection of the 400 chairs on site, sectioning and load testing, L3 at 3 man-days | $897 |
| Air freight and import entry for the replacement consignment | $640 |
| Total spent after one chair cracked | $2,369 |
Where to go next
The arithmetic of a pre-shipment check against a failure
Two thousand three hundred and sixty-nine dollars against two hundred and ninety-nine for the man-day that would have looked at the coupons. The other half of the arithmetic is the balance payment. The lot passed, so the buyer released $14,560. A report that said weld integrity on structural joints was not verified would not have been signed off, or would have been signed off with the additional check attached, and the money would still have been in the buyer's account on the day the first chair cracked.
The arithmetic of a pre-shipment check is always cheaper than the arithmetic of a failure. The only reason it feels expensive is that the invoice arrives before the problem does.
Where to go next
→ the three levels of check we sell, and what each one covers
Seven lines that would have caught this
None of the seven lines below are unusual, and none of them require the factory to own equipment it does not already own. They require the question to be asked before the goods are made, which is the only time the answer is cheap.
- Weld integrity on structural frame joints shall be verified by macro-section of test coupons, not by visual inspection of the bead.
- One sectioned coupon per production shift, per machine, welded from the same wire by the same welder at the same point on the frame.
- Acceptance for a coupon: continuous root fusion along the full length of the section. A well-formed bead is not evidence of fusion.
- Shielding gas flow shall be recorded on the shift card for every machine, with a stated minimum and maximum, and the record shall be available to the inspector.
- Frame joints shall be classified critical with zero acceptance, or the inspection shall specify the number of frames to be sectioned per shift.
- Structural frame joints shall be verified by cyclic loading to a stated number of cycles, in addition to static load, with the outdoor seating standard named on the order.
- Every chair shall be traceable to production date, shift and machine, and that trace shall be supplied with the shipping documents.
Where to go next
The three questions to put in your next enquiry
Before asking what a check will cost, ask what it is going to find. The chair that failed had been inspected and passed, so the inspector was not the problem and the scope was. Three questions put that scope to the supplier in writing, before anything is cut:
How do you verify that an aluminium weld is fused to the tube, and what evidence appears in the report? Which joints in this frame are structural, and what load case is each one actually verified to? And if I ask you to section two frames from one shift, how long does that take, and does it move the booking date?
The defect in this order was never going to be found by looking harder at a chair. It was going to be found by asking a question the agreed scope had no room to answer.
Where to go next
→ send us your PI and factory address
→ the three levels of check we sell, and what each one covers
Have an order you need verified? Send us your PI and factory address. Three levels, published prices, $149 per man-day from.