The Only Area Already Running in Production
Of the four solution areas, Plant Availability is the only one already used in production — on real machine signals in continuous operation, not a simulation and not lab numbers. We publish no details about individual installations; what the system does on your equipment ends up in your own report at the end of the pilot.
Parallel, Mutually Isolated Detection Skills
Henri's anomaly detection is structured as multiple independent skills, each targeting a distinct signal class. They run in parallel and share no state — a deliberate design decision so a false positive in one domain doesn't contaminate classification in another. Incoming sensor data passes through a three-stage pipeline: signal fingerprinting, cross-cycle correlation, then classification and action.
What the Architecture Is Built For
- Early detection of degradation patterns on real machine signals — making signatures visible before the deterioration becomes mechanically measurable.
- Planned instead of unplanned maintenance — interventions in windows the system names in advance.
- Sub-40ms classification latency per cycle, fully on embedded hardware, no cloud offload.
How much lead time that yields on your equipment depends on your machines, your sensors, and your data. We quote no figure for it in advance — you receive one at the end of the module, measured on your own data.
Fast Entry Point
Instead of starting straight into a full platform deployment, a bounded pilot validates a single use case first — fixed price, fixed timeline. You pick one of the three applications for it, or a combination:
- Process AI — understands what the plant is doing
- Vision AI — understands what the cameras see
- Production AI — patterns across orders, shifts and periods
The first step is deliberately small.
A bounded pilot examines, in four phases, on your real data, whether one of the three applications pays off in your production. Open-ended: there is no automatic continuation — you decide after the pilot whether and how it goes on.
Discuss a pilot →Frequently Asked Questions
Is Plant Availability already running in production, or is this still a concept?
Already running in production — Henri has been continuously monitoring real machine signals on customer hardware since going live, not a simulation.
What accuracy can I expect?
We do not quote a figure upfront. Accuracy depends on your machines, your sensors and your data, and a number measured in someone else’s plant says nothing about yours. What you do get is an accuracy report for your own equipment — including when it comes out unfavourably.
Do we have to commit to a full platform deployment to start?
No. The same approach is also available as a bounded pilot with a set price and timeline — with one of the three applications (Process AI, Vision AI, Production AI), before any full deployment decision.
How does detection stay reliable across different failure types?
Multiple independent, parallel detection skills, each specialized on its own signal class, sharing no state — so a false positive in one domain doesn't contaminate classification in another.