Updated July 2026 · 8 min read · Adente Vision Engineering Team
How much integration does your inspection cell actually need?
The right amount of integration is set by one question: does the result have to carry a bit or data? A pass or fail decision that drives a reject is one bit, and one bit does not need a fieldbus. A coded result, such as which variant a part is or a numeric confidence, is more than a bit and wants a bus that can carry a value. A structured record, per-part millimetre measurements with a timestamp and a mode, headed for a quality database, wants a protocol that describes its own data. Answer that question first and the integration tier follows.
Cells go wrong by starting from the protocol instead of the result. A team standardises on a full fieldbus for a check that only ever hands over a pass or fail, then spends commissioning time configuring a bus the cell did not need. Adente Vision is an edge-AI visual inspection unit built by ADENTE Advanced Engineering Technologies, part of the Aden Group, sold through automation system integrators, and it exposes four 24V inputs and outputs plus five industrial protocols, so the same unit fits all three tiers. That means the tier is a wiring decision made per cell, not a product you have to pick at purchase time.
When is discrete 24V I/O all you need?
Four 24V discrete outputs are enough whenever the result is a single decision the controller acts on. Wire one output to pass, one to fail, one to ready and one to fault, and the controller has everything it needs to drive a reject and to know the inspection is running. There is no bus to configure, no address map to agree, and no gateway in the path, which is why this tier commissions fastest and has the fewest things to break. The unit carries 4 inputs and 4 outputs at 24V for exactly this handshake.
The limit of discrete I/O is how much it can say. Four outputs are four bits, so they carry a pass, a fail and a couple of status states, and no more. The moment you need to tell the controller which of several variants a part is, or pass a confidence value, or report a measurement, four bits run out and you step up a tier. Bit-coding a small number of variants across the outputs stretches this a little, but if the result is genuinely data rather than a decision, that is the signal to move to a fieldbus rather than to overload the I/O.
Discrete I/O, fieldbus or OPC UA: an integration decision at a glance
| Tier | Carries | Best when |
|---|---|---|
| Discrete 24V I/O (4-in / 4-out) | A few bits: pass, fail, ready, fault | The result is one pass or fail decision |
| Fieldbus (PROFINET, EtherNet/IP, Modbus TCP, EtherCAT) | A coded result: variant, confidence, part-count | The controller needs more than a bit |
| OPC UA | Structured per-part data: mm measurements, mode, timestamps | Results go to MES or SCADA for traceability |
The three tiers are not exclusive. A common pattern uses discrete I/O for the real-time reject and OPC UA in parallel for reporting, so the fast decision and the structured record each ride the layer that suits them.
When should you step up to a fieldbus?
Step up to a fieldbus when the result is a coded value rather than a single decision. Recognising which of several part variants is in frame, passing a per-part confidence the controller can threshold, or carrying a running part-count all need more than four bits, and a fieldbus carries them as a value the PLC reads on its scan. This is the tier where a recognition result becomes a program-select code for a robot, or where counts feed a line dashboard directly instead of being inferred from a pulsed output.
Which fieldbus is a match-what-the-controller-speaks decision, not a ranking. The unit exposes PROFINET, EtherNet/IP, Modbus TCP and EtherCAT, so it can speak the native bus of a Siemens or Rockwell line, the neutral Modbus TCP of a mixed cell, or the deterministic EtherCAT of a fast synchronised one. The commissioning cost is higher than plain I/O because there is a bus to configure and an address map to agree, but it is bounded and predictable, and it buys the ability to carry data the wires cannot. For the neutral, lowest-friction option on a mixed-controller line, the Modbus TCP path is covered in its own note.
When is OPC UA worth the extra layer?
OPC UA is worth its extra layer when the result is a structured record that has to reach the MES or SCADA layer with meaning attached. Where discrete I/O sends a bit and a classic fieldbus sends a value, OPC UA sends named, typed, browsable data: a per-part millimetre measurement, the active inspection mode, a timestamp, all self-describing so a quality system can consume them without a hand-maintained mapping. When the unit is measuring feature positions and dimensions per part in millimetres and those numbers have to be traceable per unit produced, that is the tier that carries them cleanly.
The cost is that OPC UA is the heaviest of the three to stand up, so it is overkill for a cell that only needs a reject. The honest default is to use the lowest tier that carries the result: discrete I/O for a pass or fail, a fieldbus for a coded result, OPC UA for structured data to MES, and a combination only when the cell genuinely has both a real-time decision and a reporting need. Because the unit installs in about 30 minutes and exposes all four tiers, the integration tier, not the hardware, is what sets the commissioning effort, and choosing the smallest one that does the job is the fastest way to a running cell. For the four-step Mount, Aim, Configure, Wire install this connects into, see the 30-minute install guide, match the tiers to the connectivity on the system page, and for the full method see the pillar guide.