Fit-up is the weld decision nobody logs
Root gap and alignment cause more lack of fusion than any parameter choice, yet almost no shop records them per joint. What changes when you do.
Journal
Field notes fromthe fab shop andthe control loop.
In brief
Engineering writing about the parts of welding autonomy that are genuinely hard: measuring fit-up, planning heat, reading indications, and agreeing what a pilot actually proved.
Latest
Recent writing
Three pieces on the upstream decisions that determine weld quality.
Agreeing a pilot scorecard you can both live with
Five lines, two arguments, and how to settle attribution before integration week rather than after the review.
The welder shortage is a rate problem
Modelling what happens to takt when the constraint station cannot be staffed at all — and why overtime stops working sooner than expected.
Editor's note
Why a fabrication company keeps a journal
Industrial AI is full of claims that cannot be checked. Accuracy figures with no baseline, yield gains with no attribution, autonomy demonstrations on joints chosen for the camera.
Writing is the cheapest available correction. If we describe our method in public — how we baseline, how we attribute a rework hour, how we handle a low-confidence indication — then a welding engineer can tell us where we are wrong before a customer pays for the lesson.
So the journal is technical, specific and occasionally unflattering to us. That is the point of it.
Method
On measurement
How Weldeon decides whether something worked.
What a good baseline period looks like
Mix, shift pattern and seasonality — and the four weeks that decide whether your pilot means anything at all.
Attribution: the rework hour and the station that caused it
Rework is booked to the station that repaired it, which systematically hides the station that caused it.
Confidence is a control input, not a report
Why model confidence should stop a cell rather than colour a dashboard.
Process
On welding
The physics and craft the platform has to respect.
Heat input is a promise, not a setting
Why holding kJ/mm across a real joint requires continuous control rather than a number entered at shift start.
Sequencing a panel: distortion as a design variable
Treating weld order as something to optimise rather than inherit, and what it saves at block erection.
Reading an indication with an inspector
Sitting with qualified inspectors adjudicating machine calls, and what disagreement taught us about our own thresholds.
Systems
On building the platform
Architecture arguments, in public.
Why the envelope belongs in the adapter
Enforcement below the intelligence: where to clamp a qualified range, and why every other location eventually fails.
Twenty milliseconds at the edge
What a welding control loop's latency budget is actually spent on, line by line.
Genealogy as a by-product of control
Traceability compiled afterwards is always incomplete. Traceability produced by the controller cannot be.
“If we describe our method in public, a welding engineer can tell us where we are wrong before a customer pays for the lesson.”
Series
Running series
Longer arcs we return to.
- 01
The upstream series
Cutting, forming and fit-up — the decisions made before the arc that determine whether the weld can succeed.
- 02
Reading metal
Radiography, ultrasonics and vision, and what machine interpretation changes about the inspection bay.
- 03
Rate
Takt, line balance and the constraint station, in shops that cannot hire their way out.
- 04
Evidence
Baselines, attribution and scorecards — how to prove an industrial AI claim honestly.
Journal
What we publish
4
running series across process, systems, rate and evidence
12
pieces written and queued for individual publication
0
vendor-sponsored posts, now or later
100%
of methodology claims accompanied by their measurement definition
Contribute
Writing with us
We publish guest pieces from welding and quality engineers, and we pay for them.
- What we want
- Specific, technical accounts of real fabrication problems — including ones we cannot solve.
- What we will not do
- Sponsored placements, vendor comparisons written by vendors, or anonymous customer results.
- Editing
- Light and technical. We fact-check process claims with our welding engineers before publication.
- Rights
- You keep your work. We ask for first publication and a canonical link.
Subscribe
Get the journal by email
One email when something substantial is published. No campaign sequences, no drip.
Begin
Bring autonomy to thearc, the fit-up andthe inspection bay.
Book a pilot on one cell or line. We agree the scorecard before we start — first-pass yield, rework hours, defect escapes, takt — and we report against it in the open.