Many industrial AI offerings rely on cloud platforms and generic models. That rarely fits machine operation: production knowledge is confidential, signals are machine-specific, and a system that must understand “everything” quickly becomes large, opaque, and dependent on external infrastructure.
elvnode therefore builds its own local learning architecture — the Copass cognitive engine — for industrial machines. Not a chatbot, not a cloud language model. Copass places operating signals in context, learns how each machine runs, detects change, and informs people with traceable guidance. Everything runs locally at the machine or in the plant network.
To keep the engine practical on the shop floor, we split processing into layers. Each layer handles less for the next — keeping the overall architecture compact.
Layer 1 — Signals: Copass connects to machine signals in read-only mode — today via OPC UA, prospectively via manufacturer-specific interfaces. Only signals relevant to this machine are processed. No central data pool, no recording of the entire plant in one shared structure.
Layer 2 — Context: Raw signals are not stored as a permanent data stream. They become observations with context: which state is active, what is happening in operation. Material change, maintenance, normal run — signals gain meaning instead of just delivering values. That reduces what the engine actually needs to process.
Layer 3 — Experiences: Recurring situations are stored as experiences — not as an endless log of every measurement point. Copass compares today with what this machine has already shown in similar context. Learning means patterns from experience, not big-data collection.
Layer 4 — Patterns: From experiences, machine-specific patterns emerge — the foundation for understanding this exact asset. Each machine builds its own knowledge. The engine does not need one large structure for every machine type and every plant. That keeps the learning load per machine manageable.
Layer 5 — Local execution: Learning and evaluation logic runs on hardware in the plant — without permanent cloud connectivity for the core. A focused cognitive engine is not a drawback here but a prerequisite: fast enough for live operation, explainable enough for operators and technicians, sovereign enough for confidential production data.
These layers are not a marketing diagram. They are the architecture behind Copass: read signals, build context, store experiences, recognize patterns, classify change — and inform people before the classic alarm fires.
In practice, Copass can run without an internet connection. Production knowledge stays in the company. Every piece of guidance can be traced back — not as an opaque score, but as classification in machine context.
We believe industrial AI must be decentralized: one cognitive engine per machine, small enough for the shop floor, large enough for real understanding. That is what we are building in Copass — with the first partners we seek.
More on industrial AI for machines — how Copass contextualizes machine data locally.