What Process AI Is

Process AI sits between the machines and the platform: it connects directly to the machine controller, reads every sensor value as it happens, and continuously decides whether the machine is behaving normally. An alarm never arrives alone — it names the responsible sensor, the reading, the expected value, and the direction of the deviation. On the reference installation, Process AI watches 360 signals of an industrial line, assessed every second.

What It Provably Does Today

Explains every alarmNot a score — the named sensor, its reading, the expected value, and the direction it is wrong in
Catches slow problemsGradual drift over hours that fixed alarm limits are structurally blind to
Starts without training dataBaseline from the first readings — value in week one, no data-collection project
Nothing leaves the plantDetection, history, and the AI assistant on customer hardware — air-gap capable
Reads the equipment as it isBeckhoff/TwinCAT and OPC-UA, no new hardware, no changes to the control programme
Configured, not programmedA new machine is a configuration file, not a software project
Trained on real dataReal production data from a real line — not a simulation, not a public dataset

The Way In: Fixed-Price Modules

Each module is a service with a fixed scope, a fixed duration, and a named result — not a software subscription. They end with a report about your equipment, measured on your data.

What We Do Not Promise

A vendor who quotes an accuracy figure before ever seeing your data is guessing. We quote no accuracy up front, extrapolate no results from other plants, and promise no predictions before our predictions have been measured against reality. What is provable today is listed above — and that is what you work with from week one.

And Above It: Henri

Process AI works fully standalone. Henri — the Neuronal Core — is the intelligence layer above it: context, patterns, memory across all data sources. Henri can be connected to Process AI — but it doesn't have to be.

Frequently Asked Questions

Does Process AI need historical data before it starts?

No. The universal drift watchdog learns each signal’s normal range from the first readings and monitors continuously from then on — no training data, no data-collection project. Machine-specific trained models are added once your data supports them.

How is this different from the alarms we already have?

Two things. First, Process AI catches slow drift that never crosses a fixed alarm limit — it accumulates small deviations instead of waiting for one big one. Second, every alarm explains itself: the named sensor, its reading, the expected value, and the direction it is wrong in — a work instruction instead of a red light.

Which controllers are supported?

Beckhoff/TwinCAT and OPC-UA — verified against real plant hardware. No gateway hardware, no changes to your control programme. Customer web data interfaces as well.

Does our production data leave the site?

No. Detection, the failure history, and the AI assistant all run on your own hardware. No cloud connection is required — the system works air-gapped.

What accuracy can I expect?

We will not tell you in advance — and distrust anyone who does. Accuracy depends on your data, machines, and sensors. What we do commit to: within a module engagement you receive an accuracy report for your own equipment, measured on your data.

Related