Duromex closes the loop from perception to line-and-machine control. It runs at the factory edge, integrates deeply with the systems you already own, and earns autonomy one measured step at a time.
Fused vision, thermal, acoustic, pressure-decay, torque-tool, PLC and functional-test streams describe the panel, the surface, the assembly and the leak path on every unit — not a sample.
The orchestrator plans the form-and-finish, foam, assemble-and-wire and leak/functional-test sequence for the product actually on the line — including model changeovers.
Adaptive control writes back into the press, the enamel line, the foaming machine, the assembly stations and the testers — within the authority the plant grants.
Surface, weld and assembly defects, foaming voids, refrigerant and water leaks and functional performance are predicted before the tester confirms them.
First-pass yield, scrap, rework, energy and line balance are optimised continuously across the factory, not tuned once per quarter.
Every action lands in an immutable, assurance-grade quality and safety audit log. Every engineer correction trains the models.
Every layer is replaceable individually and useless alone. Together they are the reason a press, a foamer and a tester finally behave like one machine.
Fused vision, thermal, acoustic, pressure-decay, torque-tool, PLC and functional-test streams, serialised to the unit.
Fine-tuned defect, foam, leak, performance and yield models plus frontier reasoning for process and reliability questions.
Adaptive write-back into presses, enamel lines, foamers, stations, testers and robots — within granted authority.
Grounding, citations, immutable audit logging and twin validation before any change reaches the line.
Duromex does not stop at inspection. It owns the sequence — form, weld, enamel, foam, wire, test, rate and pack — and optimises across it, because the cause of a test failure almost never lives at the test station.
Takt time is not negotiable. If a decision arrives after the cabinet has moved on, it is a report, not a control action. That is why Duromex runs inference on the line, not in a distant region.
The edge runtime is written in Rust and C++, deployed on K3s on site, and updated over the air with canary rollout and rollback. The cloud does training, fleet management and cross-site learning — never the time-critical decision.
A read-only integration can only tell you that you already lost the unit. Duromex reads and writes: forming press, paint and enamel line, foaming machine, assembly stations, leak and functional testers, and the appliance MES.
Connectors speak the languages your plant already speaks — REST, OPC UA, MQTT, PLC tags and file drops — and every write is scoped to the authority you granted for that line.
No plant hands control of a press to a vendor on day one. Duromex is designed for the way trust is actually built on a factory floor.
The agent watches and predicts while your engineers and testers run the line. Nothing is written. We measure accuracy against your baseline on your product, and publish the gap.
The agent recommends forming, foaming and assembly moves and predicts defects, leaks and functional performance. A human approves each action; every correction trains the model.
Low-risk lines and test workflows run under adaptive control with automatic defect and leak triage. Exceptions escalate to a human. Authority is scoped per line and revocable.
Recipe, layout and changeover changes are validated in the line-and-product twin before they touch production.
The most expensive way to learn that a new model runs badly is to build it. The Omniverse-based line-and-product twin simulates forming, foaming, assembly and functional performance first.
Layouts, robot cells, foaming chambers and test stations simulated before a single fixture is moved.
New product variants and platforms validated for quality, yield and takt before release to production.
High-mix model changes sequenced and rehearsed so the line does not learn them the hard way.
Cosmos and Omniverse Replicator generate dent, seam, paint, wiring, foam, leak and robot-cell scenarios you cannot safely collect.
cuOpt optimises line balancing, changeover sequencing, rework routing and tester scheduling against your constraints.
The twin produces a predicted first-pass yield and takt before the build — the number the plant is actually judged on.
Every dimension below is the difference between a machine that measures and a system that acts.
| Dimension | Incumbents & copilots | Duromex |
|---|---|---|
| Scope | A single task or tool | The full autonomous loop — a system of action |
| Loop | Fixed motion or measurement | Closed perceive–decide–act at the factory edge |
| Data | Limited to one machine | Compounding cross-factory appliance quality and yield network |
| Integration | Read-only or shallow | Deep write-back into forming, assembly and test |
| Change control | Try it on the line | Validated in the line-and-product twin first |
| Outcome | Assist a human | Perform the work; humans supervise exceptions |
Reliability, latency and compliance targets that a plant can hold us to.
uptime target
station decision latency where takt demands it
synchronised sensor streams per line Aspirational
agent actions written to the audit log
“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.”
Design-partner deployments target shadow mode on one line within weeks, not quarters, because the wedge is deliberately narrow: one workflow, one line, one metric. Depth comes after the baseline is proven.
The edge keeps running. Inference, control and the audit log are local; the cloud is for training, fleet management and cross-site learning. Buffered telemetry syncs when the link returns.
Yes. An on-prem option exists for sensitive producers, with per-tenant data and model scoping and strict recipe IP protection either way.
Changes are gated: evaluated in CI against golden datasets, validated in the twin, then released under graduated autonomy with a human checkpoint for anything touching safety or warranty.
Bring us your worst first-pass-yield number. We will instrument it in shadow mode and show you what it is actually made of.
Duromex is pre-launch. Figures shown are design-partner targets and modelled economics, not audited results. Ask us for the methodology.