A weld can pass inspection and fail anyway.
Not immediately. Two days later, sometimes three, long after the welder has packed up and the joint has been signed off.
The crack was not there when anybody looked.
It opened while the steel sat there cooling.
Which is a strange enough failure on a workshop floor, and a serious one on a span over open water.
So the question is what the joint was breathing.
The gas that gets in and cannot get out
Moisture is the problem, and an estuary supplies it endlessly.
Water reaching the arc splits apart, and the hydrogen goes straight into the molten pool, which absorbs it happily because liquid steel will hold a great deal of it.
Then the pool freezes, and the solid steel that forms holds far less.
The surplus has to go somewhere. It diffuses through the metal toward the places where stress is concentrated, the toe of the weld and its root, and gathers in microscopic voids there.
Enough of it in one void and the pressure pries the metal apart from the inside.
Three things have to be present at once: the hydrogen, tensile stress locked into a joint that shrank as it cooled, and a hard brittle structure formed when hot steel chilled too fast.
Cold wet air delivers the first and causes the third, which is why the counter move is not a better welder but a different atmosphere.
The bridge and the water under it
The crossing stands at the mouth of the Tamsui River in northern Taiwan, joining Bali on one bank to Tamsui on the other, both districts of New Taipei City.
It runs 3,018 feet in total, with a main span of 1,476 feet hung from a single tower 656 feet high, leaning away from the load rather than standing between two equal halves.
The deck is 233 feet across, which is wide even by the standards of a major crossing, and it carries a highway, footways, a cycle route and reserved space for a light rail line.
Wind comes off the Taiwan Strait with nothing in front of it.
The humidity barely drops, the salt never stops arriving, and both of them land on every joint that has to be made in the open.
That is the working environment, and it does not take a day off.
What is on the record about it
The bridge opened on May 12 of 2026, after a build that ran years past its first target.
It was designed by a British architecture practice with a German bridge engineering firm and a Taiwanese consultancy, and it is described as the longest single tower asymmetric cable stayed bridge anywhere.
Seismic protection is heavy, as it has to be here: hydraulic dampers, friction pendulum bearings and synthetic rubber pads, sized against a magnitude 7 or above.
The practical payoff is unromantic. The trip between the two banks drops by about 25 minutes.
A single tower also means one foundation in the riverbed instead of two, which is the environmental argument for the shape.
Cost is quoted at anywhere from 400 million to 766 million dollars depending on the source, and the construction start is given as either 2014 or 2019, which are different claims about different things nobody separates.
Where this account runs out of evidence
The enclosures themselves are the part to be careful with.
Welding habitats with dried air and controlled preheat are ordinary practice for site welding of steel bridge decks in marine conditions, and the physics above is settled. What is missing is any published account confirming that arrangement on this particular crossing.
It is very likely true and it is not documented, and those are different states.
The same goes for the segment weights in circulation, which carry no source, in the way that build details for a major crossing or a record tower often reach the public as architecture rather than as method.
The design gets photographed. The procedure does not.
Why the last joints are the hard ones
Most of a steel deck is welded in a shop, indoors, with the plate flat, the humidity known and the joint turned to whatever position suits the welder.
None of that survives the trip to site.
The closing welds have to be made where the bridge is, in the position the structure dictates, with both halves already carrying load and the weather doing whatever it is doing that week.
Every one of those conditions makes the joint worse, and only one of them can be bought back, as the opening coverage of the bridge notes about the wider difficulty of the site.
You cannot move the span indoors, so you build a small indoors around the joint, as the project record sets out for the structure itself.
The sealed enclosures are not weather protection for the crew.
They are protection for the steel.
