How to Number Your Tags
A naming convention is not bureaucracy. It is the difference between a historian you can query and forty thousand points nobody can find, and it is nearly impossible to fix later.
Tag naming looks like an administrative question and is actually an architectural one. The names assigned during commissioning become the permanent addresses of every measurement the plant will ever make. They end up in the historian, in reports, in alarm lists, in the analytics project three years later, and in whatever system replaces the current one. They are also the single thing in an industrial system that is hardest to change afterwards, because every consumer of the data has hard-coded them.
The failure mode is familiar. A plant reaches forty thousand points and nobody can find anything. Three tags measure the same physical quantity under different names because three contractors were on site in different years. A tag named TT-101 exists in two units. The letters mean different things on different lines. Nobody knows whether FIC_203_PV is a flow controller's process value or a totaliser, and the only person who did has retired. At that point the data exists and is unusable, which is the expensive version of not having it.
A convention that works has a small number of properties. It is systematic, so a name can be constructed and parsed rather than remembered. It encodes location, equipment and measurement type in fixed positions. It is unique plant-wide rather than unique per unit, because the moment two systems are joined, per-unit uniqueness becomes a collision. It is stable under change — meaning it does not encode anything likely to be reorganised, such as a department, a cost centre, a vendor name or the controller's rack and slot. And it is mechanically checkable, so a script can reject a malformed name at the point of entry rather than an auditor finding it later.
The standards worth knowing are ISA-5.1 for instrument identification, which gives the familiar letter codes — first letter the measured variable, following letters the function, so TIC is a temperature indicating controller and PDT a differential pressure transmitter — and KKS or its successor RDS-PP in power generation, which is more thorough and more hierarchical. Neither is mandatory outside particular industries, but adopting one rather than inventing a private scheme has a durable advantage: the next contractor already knows it, and so does the next software package.
There is a deeper decision underneath the string, which is whether the name carries the meaning at all. The traditional approach encodes semantics in the identifier, which means the identifier has to be long and every consumer has to parse it. The modern alternative is to keep the identifier short and stable — ideally meaningless, like a UUID or a sequence — and hold the meaning in an asset model alongside it: this point belongs to that pump, measures temperature, has these engineering units, this range, this criticality. Standards like OPC UA's information model, and the various asset frameworks in historians, exist to hold exactly this. The advantage is that meaning can be corrected without renaming anything, and a point can have several meanings for several audiences. The cost is that you now have a model to maintain and a migration to perform, and the model has to be populated by someone who knows the plant.
The pragmatic answer for most sites is both: a structured, parseable name for human use and continuity with existing drawings, plus an asset model layered over it that carries units, ranges, equipment relationships and criticality. The name gets you compatibility; the model gets you queries like "every bearing temperature on every compressor in this area" without pattern-matching on strings.
Whatever is chosen, three rules pay off disproportionately. Write the convention down with examples and counter-examples, and put it in the contract, because the contractor will otherwise use whatever their last job used. Enforce it with a script rather than a review, at the moment tags are created. And record engineering units and range as attributes rather than in the name, because units change, names should not, and a tag called Flow_GPM in a plant that moved to cubic metres per hour is a permanent small lie.
The test of a convention is not whether it looks tidy. It is whether someone who has never seen this plant can be given a name and say what it measures, where it is, and what it is attached to — and whether a script can find all of them.