How-to
Trace a PLC tag from controller to historian display
Short answer
Trace one existing, noncritical measurement end to end: controller source, collector mapping, historian point, optional calculation, then display binding. Check the value and the timestamp at each boundary before you change any configuration. A value that matches and a timestamp that agrees at every step means the path is sound; the first mismatch names the layer to fix.
On this page
A value on a historian display is the end of a data path, not a direct view of the controller. Between a PLC tag and the number an operator sees sit the collector or interface, the historian point, an optional calculation, and the display's binding and time selection. The fastest way to learn that path is to trace one existing, noncritical measurement end to end, checking the value and the timestamp at each boundary before changing any configuration. This article walks the trace in the order data flows, works one scaled count through every layer with the arithmetic shown, and ends with the checks that name the broken layer when a value goes stale or wrong.
What sits between the controller tag and the display #
The layers are the same on most PI systems even though product names differ between interface and connector generations:
| Layer | What it holds | What you record there |
|---|---|---|
| Controller | The tag or address in the PLC program | Online value, data type, raw or scaled |
| Collector or interface | The mapping from controller item to point | Source reference, poll rate or scan class, scaling |
| Historian point | One named stream of time-series values | Value type, archiving and compression settings |
| Calculation (optional) | Analyses that derive new values from points | Input references, units, trigger conditions |
| Display | The binding and time selection on screen | Data reference, time range, retrieval mode |
Two naming facts keep the trace straight. A PI tag and a PI point are the same object: one named stream of time-series values coming from an instrument, device or sensor. And the same measurement usually carries a different name in the controller, the collector, the point and the display, so matching names never proves a connection. Values and timestamps prove connections; names only label them.
Choose the tag you trace #
Pick a measurement that is already collected, barely matters to production, has units you understand and moves normally during operation — a tank level or a motor winding temperature qualifies; a point feeding a safety function or an interlock does not. Trace it read-only: the exercise needs no controller edit, no forced value and no configuration change, and forcing a value on a running process is how a learning exercise stops being one.
Trace the tag end to end #
- Start at the screen. Open the display object and record its data reference, engineering units, selected time range, current value and timestamp. Note whether the object shows a stored measurement or a calculated result.
- Query the point directly. With a historian client or query tool, read the current value and timestamp for the same point, then retrieve a historical interval containing a change you recognise — a startup, a fill, a batch step — and confirm the stored events follow what the process actually did.
- Open the collector mapping. Find the configured source connection and the item reference behind the point, and record how values are acquired: an OPC DA interface, for example, can poll tags on a schedule or collect them through advise scan classes, and the configured method sets when values arrive. Record any scaling or conversion applied at this hop.
- Check the controller source. Read the online value, its data type and its engineering meaning, and decide whether the tag holds a raw count or an already scaled engineering value. An unscaled integer must be converted somewhere downstream, and the collector mapping is the usual place — if the displayed units make no sense against the controller's raw span, that is the hop to inspect.
- Compare each boundary. Screen against point, point against collector, collector against controller: values must agree after exactly one conversion per hop, and timestamps must agree within the collection interval. Know which timestamp your collector stores — the source's or its own receipt time — because that decides what a gap means. Check clock alignment between clients and server before you read any timestamp difference as collection delay.
- Write the path down. One line per layer with the exact reference at each hop, including any intermediate data server or calculation. That line is the map you will use for every future fault on this measurement.
Worked example: one level count through every layer #
The tag names and numbers below are illustrative; read the equivalents from your own system.
Suppose controller tag LI101_RAW holds an unsigned 16-bit integer, scaled in the program so that 0–32000 counts span 0–100.0 % of tank level. The collector maps LI101_RAW to historian point LI101_PCT, converts counts to percent on the way through, and polls the controller once per second. The display binds LI101_PCT over the last eight hours.
At 10:42:15 the controller reports 20800 counts. The collector converts:
20800 counts × 100.0 % ÷ 32000 counts = 65.0 %
The point stores 65.0 %, stamped with the acquisition time 10:42:15. A minute later the display shows 65.0 % at 10:42:15. Both boundary checks pass: the value matches after exactly one conversion, and the timestamps agree.
The one-second poll bounds the lag. A step change that happened at 10:42:14.2 cannot become a stored event before the next poll, so the point can trail the controller by up to the one-second scan interval plus transmission time. That is expected behaviour, not a fault. Separately, the archive's exception and compression settings decide which polled values become stored events — the deviation limits live in the point's archiving configuration — so read them before treating a missing intermediate value as a communications failure.
When the numbers disagree #
A stale value or an empty trend does not name the broken layer. Work backward from the screen and let each boundary tell you where the path stops:
| Symptom | Deciding check | Layer it names |
|---|---|---|
| Display frozen, point current | Compare display binding and time range with a direct point query | Display binding |
| Controller changes, point stays old | Collector connection status, mapping, diagnostics | Collection path |
| Raw value right, calculated value wrong | Calculation inputs, units, trigger conditions | Calculation |
| Repeatable scale or offset | Conversions at controller, collector, calculation, display | Duplicated conversion |
| Value right, trend shape odd | Stored events against trend time range and retrieval mode | Presentation, not data |
For faults at the collector boundary — timestamps that jump, items that fail to browse, bad quality codes — the guide on resolving OPC collector error tags for timestamp, browse and quality faults covers those three failure families in detail. An empty trend with a healthy-looking collector is usually a collection problem rather than a display one, and troubleshooting an empty historian trend with no PLC tag collection walks that diagnosis step by step. A steady scale or offset means the conversion is applied twice somewhere; scaling an unscaled PLC integer for the historian shows how to keep it in exactly one place.
After the first trace #
Repeat the trace on a measurement of a different kind — a temperature, a counter, a digital state — until the layers are routine. Then rebuild the path in a training or lab environment: define the expected signal from a simulated or spare source, configure collection per the installed component's documentation, create the point, confirm stored events against a known change, and bind the display last. Keep the exercise small enough that every stored value has an explanation.
The verified point becomes the building block for everything downstream of the archive: exposing historian latest values over OPC UA or MQTT shows how to serve it to dashboards and message brokers without building new collection. And when the plant later argues about historian products, read the candidates layer by layer the same way — a comparison of PI Historian with InSQL shows how another historian splits the same controller-to-display path.
The decision rule the whole exercise teaches: agreement on value and timestamp at all three boundaries means a sound path; the first disagreeing boundary names the layer to fix. Correct that one layer, re-verify the value and the timestamp across it, and only then move on.
Questions asked
Does the historian store every PLC scan?
No. The collection cadence and the point's exception and compression filters decide which polled values become stored events, so the archive holds a filtered subset, not every scan. Between stored events, most clients draw interpolated values, which is why a smooth trend is not proof that every scan arrived. Retrieve archived values in recorded mode when you need the actual stored events.
What access do I need for a trace?
Read-only access to four things: the display configuration, the historian point data, the collector mapping and one live controller value. PI security can be set per point or per database, so your read rights may differ between layers; ask for the least access that still shows values and timestamps. None of it needs to include write access to controller logic.
Is there structured training after the first trace?
Yes. The vendor's overview material explains the PI System architecture, and its Configuring a Simple PI System course is written for the IT professional who maintains the system. Take it after your first trace rather than before: every tool the course introduces then maps to a value, a timestamp and a mapping you have already followed, which is what makes the administration stick.
Why do two displays show different values for the same point?
Usually they are not showing the same thing: one displays the latest archived event, the other an interpolated or averaged value over its own time range. Compare timestamps first, then the time ranges, then the retrieval mode each display uses. Check clock alignment between the clients too; two correct systems with unsynchronised clocks disagree by exactly the clock offset.