Process AI is the first application on the architecture — the one for industrial environments. Underneath it sits the Neuronal Autonomous Architecture, which carries other fields of use as well; everything is operated through the platform. If you want to find out whether this pays off in your plant, a pilot is the bounded first step.
The Problem: Alarm Limits Only See the Breach
A fixed alarm limit knows exactly two states: below and above. It reports when a value crosses a threshold — which is to say, at the moment the damage has already happened. Everything before that is structurally invisible to it: the bearing that announces itself over weeks, the nozzle that slowly clogs, the coupling whose torque shifts degree by degree. Every individual reading stays in the green. Nobody sees the movement towards the edge.
Then there is the second problem: an alarm tells you that something is wrong, not what. A red light on the line sends someone out who then has to start looking. The time between the alert and the diagnosis is the most expensive part of a stoppage — and the part no alarm limit shortens.
What Process AI Does
Process AI connects directly to the machine controller, reads every sensor value at the moment it is produced, and continuously decides whether the machine is behaving normally. Several hundred signals from one controller are assessed every second. Instead of waiting for one large deviation, the system accumulates many small ones — and reports the movement, not just the outcome.
The Explained Alarm
An alarm never arrives on its own. It carries four pieces of information, and it is those four that turn a notification into a work instruction:
What that changes on the floor: the maintenance engineer does not walk to the machine to search. They walk to a named measuring point.
What Is Demonstrable Today
The points below are implemented and verified against real plant hardware. Anything planned beyond that is in the roadmap — as a roadmap, not as a feature.
Already Connected
The question that follows “what can it do” is always: will it fit our equipment? These controller and robot systems are connected — as integrations we have built, not as a compatibility list off a datasheet.
- Beckhoff TwinCAT 3
- Siemens TIA Portal
- OPC UA
- KUKA KRC
- ABB
- FANUC
If your controller is not on the list, that is not a no — it means we name the effort during scoping, before a quote exists. For Beckhoff/TwinCAT there is one addition: verified against real plant hardware.
Connection: What Is Read, What You Provide
The connection is read-only, through the interfaces your controller already ships with. No gateway hardware is added, and nothing in the control programme is changed.
Legacy controllers without these interfaces are possible but cost effort. We say so during scoping, not afterwards.
What We Do Not Promise
A vendor who quotes you an accuracy figure before looking at your data is guessing. We do not name accuracy up front, we do not extrapolate results from other plants, and we do not promise predictions before our predictions have been measured against reality. We also will not tell you which component fails next. What is demonstrable today is above — that is what you work with from week one.
Above It: Henri
Process AI works entirely on its own. Henri — the autonomous neuronal core — is the layer above it: context, relationships, and states across all data sources, not just across one machine. Henri can be connected to Process AI — but it does not have to be.
The relationship in one sentence: Process AI is one application on the architecture — the first. The architecture is what lasts; the application is what can be replaced.
Frequently Asked Questions
Where does Process AI sit in the SynapSync architecture?
Process AI is the first application on the architecture — the one for industrial environments. The same architecture carries other fields of use as well; Process AI is the only one running in production today. Everything is operated through the Secure AI Infrastructure Platform. In practice: you buy Process AI, you work in the platform — and Henri can sit above it, but does not have to.
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 and robot systems are connected?
Controllers: Beckhoff TwinCAT 3, Siemens TIA Portal and OPC UA. Robotics: KUKA KRC, ABB and FANUC. Plus customer web data interfaces. For Beckhoff/TwinCAT there is one addition: verified against real plant hardware. In every case with no gateway hardware and no changes to your control programme. If your controller is not on the list, we name the effort during scoping — before a quote exists.
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 the pilot you receive an accuracy report for your own equipment, measured on your data.