Perception under real conditions
Lighting drift, reflective enamel, occlusion, and product variants with no labelled data.
Illustrative · publishing begins with cohort one
What we are learning building physical AI for appliance manufacturing — including the parts that did not work. Written by the engineers doing the work.
Outlines are real; the posts publish as the underlying work completes. Marked Illustrative until then.
Why direct observation is impossible after the jig closes, and what shot weight, mix ratio, jig temperature and cure timing actually tell you. Illustrative
EdgeHow multi-camera fusion, model serving and control writes are scheduled so a station decision never misses takt. Illustrative
QualityPressure-decay curves contain more information than a pass/fail verdict. What the shape of the curve predicts. Illustrative
SafetyDesigning bounded authority envelopes so autonomy is scoped, observable and instantly revocable. Illustrative
FieldWhat a baseline actually looks like, why it is usually worse than the plant believes, and how to publish it honestly. Illustrative
ProductThe case for vertical physical AI, and the specific things a horizontal platform structurally cannot learn. Illustrative
Field notes exist to be useful to the process engineer who reads them, not to generate leads. That means the failure modes get published alongside the wins.
Where a note discusses design-partner work, the partner reviews it and identifying detail is removed. Where a number appears, so does its baseline and its window.
Forming, welding, enamel, foaming, assembly, test and packing — each stage has its own physics, its own failure modes and its own notes.
Each theme maps to part of the platform.
Cabinets are formed, foamed and charged — then judged by a leak test and an energy rating. Duromex links the foam shot and the seam to the number the tester prints.
Drums, cabinets and wiring harnesses at high volume. Balance, seal and electrical integrity decide whether a machine survives its warranty period.
Tub forming, insulation, pump and heater assembly, then a wet functional test that is expensive to repeat.
Enamel is the product. Coverage, film build, cure and chipping decide whether a range ships as first quality or a downgrade.
Brazed joints, refrigerant charge and performance test under tightening F-gas rules — a leak here is a compliance event, not just a rework.
Slow and specific beats fast and generic.
What we learn instrumenting real lines and measuring honest baselines.
What engineer corrections teach the models, and where the models were confidently wrong.
How scoped write authority behaved in production, including rollbacks.
Full studies with partner approval, method and caveats.
The things we keep coming back to.
Lighting drift, reflective enamel, occlusion, and product variants with no labelled data.
Scheduling, batching and serving decisions that keep station decisions inside takt.
Tracing an end-of-line failure back to the upstream stage that caused it.
Bounded envelopes, fail-safe defaults, and what happens when the model is uncertain.
How good a twin has to be before it is allowed to gate a production change.
Building assurance records that hold up in a regulatory or warranty dispute.
“Our end-of-line functional testers told us what failed. They never told us which press stroke or foam shot caused it. That link is the whole product.”
“Foam voids are invisible until the energy rating comes back wrong. Predicting density distribution before the cabinet cures is the part I could not buy anywhere else.”
“If it writes to the press, it needs an audit trail a quality auditor accepts. Duromex started there instead of bolting it on.”
topic per note
notes with a stated method
notes without partner approval
AI-generated filler posts
Segments we write about
Design-partner programme, 2026. Duromex is pre-launch — partner identities are withheld under NDA and shown by segment, not by logo.
Because they are not written yet, and pretending otherwise would contradict the point of the page. The outlines are real.
Yes. If you are a process, quality or OT engineer in appliance manufacturing and there is something you wish someone would write about properly, tell us.
From design partners and practitioners, yes. From content agencies, no.
No newsletter noise — one message when a note goes live.
Duromex is pre-launch. Figures shown are design-partner targets and modelled economics, not audited results. Ask us for the methodology.