IMD SolutionComputing LibraryShop main store

Troubleshooting

Historian trend empty: PLC tag collection troubleshooting

Edge computing and IIoT Published 10 min read

Short answer

A live PLC value proves nothing about what the historian stored. Work one known, changing tag through four boundaries — controller to data server, server to collector, collector to point, point to chart — and with a 2 s update rate, a 1% deadband on a 200 in span and a 24 in/h fill, a 20 min window must show about 4 stored events. Zero events means the collector; four events with a flat chart means the display.

On this page
  1. Why does a live PLC value not guarantee a stored trend?
  2. Which tag do I test with first?
  3. How do I prove the data server actually reads the tag?
  4. How do I prove the collector is writing to the historian point?
  5. How much data should one working point archive?
  6. Why did every point go flat at the same minute?
  7. When is the empty trend only a display problem?
  8. How do I recover data lost while collection was down?
  9. Which boundary is failing, from the symptom alone?
  10. Questions asked

An empty trend is not a Studio 5000 problem to solve in Studio 5000: the engineering software shows live controller values, and it neither configures the historian nor writes anything to the archive. Collection breaks at one of four boundaries — controller to data server, data server to collector, collector to historian point, point to trend chart — and the fast route to a fix is proving each boundary with one known, naturally changing tag before touching anything else. Reinstalling software, rebuilding engineering trends and editing PLC logic cannot repair a broken link; finding the failing boundary can. This page gives the pass test for each boundary, the event count a healthy point must produce, and the rule that turns your symptom into the fix.

Why does a live PLC value not guarantee a stored trend? #

A live value on the controller or an HMI proves the tag exists, not that the historian stored it: the archive only holds what survived the data server, the collector and the point's own settings, so a blank trend means one of those links, or the chart itself, failed.

The archive sits at the end of a chain with five distinct parts, and each joint can fail alone:

  • Controller tag — the live value in the processor, visible in the programming software.
  • Data server — the OPC server or native interface that exposes that value to clients.
  • Collector or collection interface — the service that reads updates and writes them to the historian.
  • Historian point — the stored record, bound to a source address, not to a similar-looking name.
  • Trend client — requests stored values for display; it cannot collect and cannot repair.

Two consequences follow. A fault in the chart can never stop collection, so no display change fixes a blank archive. And a live read at one joint proves nothing about the next joint downstream. That is why the method below tests one boundary at a time, the same discipline behind why each troubleshooting test should prove a single cause: a passing test eliminates a cause, and a failing one names it.

Which tag do I test with first? #

Test one existing, noncritical tag that changes by itself at a roughly known rate — a loop process variable works — and record its scope, data type and current value before you configure anything; never force a running output just to make a chart move.

Before configuring anything, record three things: its exact name and scope as the controller holds it, its data type, and its current live value read online. Check its external-access or read setting at the same time — a tag an engineer sees online can still be refused to outside clients, and that refusal is cheaper to find now than after a silent collector.

One safety note: never force a running output or override a live loop to make a chart move — the plant acts on what you force. If nothing natural changes fast enough, pick a livelier signal or plan around a wide deadband instead.

Commission one tag, not a hundred: a single pilot point makes an addressing mistake visible, while a bulk import hides one bad mapping among forty working tags. When the pilot passes every boundary below, copy its pattern to the rest.

How do I prove the data server actually reads the tag? #

Read that tag through the configured data server and compare value and quality against the controller: a GOOD quality that matches the live value passes the first boundary, while any BAD quality is a failed read that will archive nothing usable.

Browse to the tag through the configured server — the one the collector is actually pointed at, not the one you are used to opening. Compare two things against the controller: the value and the quality.

A GOOD quality with a matching value passes the boundary; stop working on the controller link, because nothing on this side can cause the empty trend. A BAD quality is a failed read, and the OPC UA specification requires the value to be NULL when quality is BAD, so there is nothing to archive — the fault sits upstream of the collector and the chart. Read the substatus the server reports and fix what it names: controller route, connectivity, an access restriction on the tag, or a data type the interface handles poorly, such as arrays or structures. A source left off-scan or in a failed state also delivers nothing, so confirm the source itself is running.

How do I prove the collector is writing to the historian point? #

Map one historian point to that verified tag through that verified server, enable collection, and watch the point's current value, timestamp and quality in the historian's own tools; a point that never receives a write is a mapping, service or permissions fault, not a trend problem.

For FactoryTalk Historian SE, tags are imported and their attributes set in the PI SMT add-in; the installation and configuration guide for your installed release carries the exact workflow. Map the point to the exact source address through the verified server. A point whose name looks right but whose address is wrong either stays silent or quietly records a different signal — always verify the address, never the spelling.

Check the collection service is running, and test its connection under the account it actually runs with. Read access for people viewing trends and write access for the collector are separate grants at separate boundaries; a successful manual test from an administrator login proves nothing about the background service, which may hold neither.

If the configuration tool offers no PLC connection options at all — common right after a server or host migration — the source interface never registered, which is a fault of its own; the fix is the same one behind why a migrated historian shows no PLC connection options.

Decision: the point's current value updating with an advancing timestamp and clean quality passes this boundary. A silent point behind a GOOD server read is a mapping, service or permissions fault — fix it here, before touching the chart.

How much data should one working point archive? #

Expect stored events set by the exception deadband, not one per scan: with a 2 s update rate, a 200 in level span, a 1 percent deadband and a tank filling at 24 in/h, a 20 min window should hold about 4 events, not 600.

Those inputs are illustrative — substitute your own. A level tag spans 0–200 in, the collector updates every 2 s, the exception deadband is 1% of span, and the tank fills steadily at 24 in/h. Collection enables at the start of a 20 min window:

  • Deadband per stored event: 0.01 × 200 in = 2 in
  • Updates offered: 20 min × 60 s/min = 1200 s, and 1200 s ÷ 2 s = 600
  • Rise inside the window: 24 in/h × 20 min ÷ 60 min/h = 8 in
  • Stored events expected: 8 in ÷ 2 in = 4, plus the initial snapshot written when collection enables
  • Same rate over a day: 1440 min ÷ 20 min = 72 windows, so 4 × 72 = 288 events/day with the deadband, or 600 × 72 = 43,200 events/day with it off

Read the result against this scale. About 4 GOOD events in the window is a healthy point — exception reporting is thinning a quiet signal exactly as designed. Zero events with no current value is a collection failure; go back to the previous boundary. Six hundred events means the deadband is disabled or zero: it works, but 43,200 events/day from a single tag is the storage bill for that choice.

The count moves with rate of change and deadband: events ≈ rate of change × window ÷ deadband, so doubling the fill rate or halving the 2 in deadband doubles the events, and a signal slower than the deadband archives almost nothing — a flat trend can mean a quiet process, not a dead collector. Exception and compression settings act after a working write, thinning stored events; they never substitute for one.

Why did every point go flat at the same minute? #

A simultaneous stall across every point means the shared path failed — collector service stopped, OPC session dropped or a controller route lost — not dozens of separate point faults; fix the shared link once and the whole trend returns together.

Read the timestamps before you read the points. If every point's last good sample carries the same stamp, one shared thing died at that minute: the collector service, the OPC session, the network route or the controller connection. Fix that one link and the whole trend returns together — editing individual point configurations while the service is down just tunes points that were never at fault.

A single stale point among healthy neighbors points the other way: that point's own source mapping, a disabled state, or a deadband widened past the signal. Several outputs failing together behind one shared input is a pattern that reaches well beyond historians — the same diagnosis explains one shared trigger leaving four log entries missing — so hunt the upstream common cause first.

When is the empty trend only a display problem? #

If the point updates now and a direct archive query returns events inside your window, the chart is empty by selection: the wrong, similarly named point, a time range that ends before collection started, or a vertical scale far wider than the signal.

Prove the archive first, away from any chart: query the point directly for the exact window and check the returned timestamps, values and quality. If the query returns the expected events, the archive is fine and the chart is making a selection mistake. Four candidates, in order:

  • Wrong point selected — a controller tag with a near-identical name instead of the historian point.
  • Time range ending before collection was enabled, asking for a window that was never stored.
  • Vertical scale far wider than the signal: the worked example's 8 in rise covers 4% of a 0–200 in chart and looks flat, while the same four events on a 0–10 in scale show clear 2 in steps.
  • The client itself failing to reach the historian — a connection fault with its own path, the same one described for historian trend client connection failures.

Decision: events in the query but none on the chart is display-only. Fix point, range and scale; collection needs nothing.

How do I recover data lost while collection was down? #

You can only backfill what another store still holds: a controller is not an archive, so treat the outage as a gap unless a documented buffer or an upstream historian provably covers the window.

A backfill needs a source that held the values while collection was down: a documented controller or interface buffer, or an upstream historian. If nothing documented covers the whole window, the gap is permanent — no chart setting, reinstall or recommissioning reconstructs samples that were never stored. When an upstream historian does hold the window, recover it by backfilling trend history from a plant historian.

Close the loop so the next failure costs minutes instead of weeks: alert on sample age, not on value. In the worked example's terms, a 2 s tag whose last sample is 10 min old has already missed 10 min × 60 s/min ÷ 2 s = 300 updates — far past any excuse, and worth a standing alarm on the point's timestamp.

Decision: no documented source covering the window means the gap stays; add the staleness alarm so a dead collector is noticed the minute it dies.

Which boundary is failing, from the symptom alone? #

Match the symptom to the boundary and fix there: a failed server read is a controller or access fault, a silent point is a collector fault, missing events are a deadband fault, and an empty chart with stored events is purely a display fault.

Run the checks in this order on one pilot tag and each symptom lands on exactly one boundary:

  • Data server read fails or returns BAD — controller route, access setting or source state; fix the read before configuring anything downstream.
  • Server reads GOOD, the point never updates — collector side: service state, the source address in the point mapping, or the service account's write permission.
  • Point updates now, but the archive query returns nothing for the window — deadband or enable time; recompute the expected event count before blaming collection.
  • The archive query returns events but the chart stays empty — display only: point selection, time range, vertical scale.
  • Every point stopped at one shared timestamp — collector service, OPC session or network route.
  • Values arrive but look wrong rather than missing — raw integers that were never scaled into engineering units, a different fix covered in historian tag scaling for an unscaled PLC integer.

Sign off when the pilot tag returns its predicted event count with GOOD quality over a defined window. Only then import the rest of the tag list, with one proven boundary already behind every point.

Questions asked

Does reinstalling the historian fix an empty trend?

Rarely. A reinstall replaces program files, not configuration: it cannot repair a point mapped to the wrong source address, a collector service account without write permission, or a controller that blocks external reads. Run the four boundary tests on one tag first, and reinstall only when the installation itself is damaged, then rebuild and remap the point.

Should I change the PLC program to feed the historian?

Not as a first step. If the value already exists in the controller and is externally readable, collection reads it with no logic change; editing the program before the boundary tests adds risk and fixes nothing. Change logic only when the computed value you need genuinely does not exist, and then through your normal change process.

Why do archived samples sit hours away from when they happened?

Clock skew or a time-zone mismatch between the collector host, the historian server and the trend client shifts every stored timestamp, so a chart range can miss data that exists. Keep all three on the same time source, keep offsets to seconds or minutes, and remember each sample keeps the stamp it received at collection, not the one the viewer shows now.

Which quality codes can you trust in a stored trend?

Count only GOOD-quality samples toward a sign-off. The OPC UA specification requires the value to be NULL when quality is BAD, so a BAD sample carries nothing to plot and marks a failed read upstream of the archive. If BAD samples dominate a window, fix the controller link, the access setting or the data server that produced them.

Keep reading

  • Troubleshooting

    Historian tag scaling for an unscaled PLC integer value

    Scale the tenths-encoded integer on the historian tag itself: linear scaling with gain 0.1 and offset 0 turns raw 144 into 14.4 while keeping the direct PLC acquisition. Check the tag's existing scaling first so a second divide-by-ten does not log 1.44, and repoint the tag at a SCADA-supplied value only if logging that stops with that runtime is acceptable.

    Edge computing and IIoT

  • Troubleshooting

    Troubleshooting historian trend client connection failures

    A trend client connection failure has two first checks. If trend pens or Historian Client still load data, compare the failing client's connection definition against a working one instead of restarting the server. If every client fails, verify the historian and SQL services, then test TCP reachability from the failing node to the configured endpoint, port 1433 only if that is the configured port.

    Edge computing and IIoT

  • Troubleshooting

    Fixing a historian device shortcut that will not appear

    A ping, an open web page and an online programming session prove IP reachability, not a working data path. When the historian drops a device shortcut, the RSLinx Enterprise tree usually holds the gateway interface instead of the controller behind it. Stop browsing and the related services, delete the interface entry, define the SLC 5/03 manually under ENI Devices at the same IP address, point a fresh shortcut at the controller icon, restart, and confirm with a live item read.

    Edge computing and IIoT

Show a historian tag average over three days on an HMI