How to Set Historian Deadbands Without Hiding Events
Set a historian deadband by defining the smallest process change your historical record must preserve, then choosing a candidate setting in engineering units that keeps that change visible. Compare the stored trace with the incoming readings during steady operation, transitions and short excursions before accepting it. Review the maximum recording interval separately: regular stored points do not, by themselves, prove that brief events have been captured.
What should a historian deadband preserve?
Start with the question someone will ask of the record. A temperature trace used to review a batch needs different evidence from a pressure trace used to investigate a brief disturbance.
For each signal, write down:
- The decision the history must support.
- The smallest meaningful change in value.
- The shortest event that must remain visible.
- Whether peak value, duration, timing or all three matter.
- What should happen when readings stop or quality becomes uncertain.
These are acceptance criteria. “The trend looks smooth” is not one.
Do not confuse a historian deadband with an alarm deadband or a controller setting. Check which setting you are changing and where it acts before adjusting anything.
Which part of the data path are you changing?
Check collection and storage separately. A collection setting may limit which readings reach the historian; a storage setting may reduce which received readings are retained. Product terminology and behaviour vary, so confirm the actual comparison rule rather than relying on a field labelled “deadband” or “compression”.
Ask whether the rule compares each reading with the last retained value, checks departure from a reconstructed line, or uses another method. Establish whether the threshold is absolute or relative, and how it treats timestamps and quality changes.
A useful commissioning record identifies each stage: source update, collection, filtering, storage and display. Otherwise, a missing excursion may be blamed on storage when it never reached the historian.
How do you choose a deadband in engineering units?
Work in the units operators and engineers use: degrees, pressure, flow or level. Record the configured range alongside the chosen deadband so that a later range change prompts a review.
Use the following as an illustrative planning example, not recommended settings:
| Signal | Historical question | Smallest change to preserve | Initial trial |
|---|---|---|---|
| Vessel temperature | Did temperature leave the agreed operating band? | 0.5 °C movement near the boundary | Trial a 0.1 °C deadband |
| Header pressure | Was there a short pressure dip? | 0.2 bar dip lasting 2 seconds | Trial 0.05 bar and verify collection timing |
| Tank level | How did inventory change during transfer? | 10 mm movement | Trial 2 mm during filling and draining |
The trial values deliberately leave room below the stated meaningful changes. They still need testing. A small amplitude threshold cannot compensate for readings collected too slowly to represent the event.
How should you account for noise?
Observe the incoming signal during a genuinely steady period. Then repeat the observation under normal running conditions. Record the size and pattern of the fluctuations rather than assuming every small movement is noise.
Investigate whether the variation reflects the process, the instrument or the signal path. Avoid increasing the deadband simply to make an unexplained trace quieter.
If an illustrative steady signal varies by about 0.03 bar, while a 0.2 bar dip matters operationally, there may be room to reject small fluctuations while preserving the dip. Test that proposition against actual readings.
If normal variation overlaps the change you need to detect, record the limitation. Do not claim that one threshold can reliably distinguish the two without evidence.
What should the maximum recording interval do?
Treat the maximum interval as a separate requirement: how long may the record go without a retained observation?
Check what the setting actually guarantees. Does it retain a newly received reading after the interval, create a periodic record, or depend on another update arriving? What happens during a communication failure?
Choose the interval around the historical question. For a slowly changing vessel, periodic observations may support a useful operating record. For a short pressure dip, an interval chosen for routine trending is not sufficient evidence of event capture.
Test a constant signal and an interrupted connection separately. A flat process value and an unavailable measurement should be distinguishable in the records used for investigation.
When can compression hide a useful excursion?
Pay particular attention to brief peaks, dips, reversals and movement near an operating limit. Depending on the algorithm and settings, reduced records can omit useful intermediate values or change how an excursion appears.
Consider an illustrative pressure sequence: 5.00 bar, 4.85 bar, then 5.00 bar. The historical question is whether the pressure dipped and recovered. A retained record showing only the two outer readings would not answer it.
Test ramps too. Retaining the start and finish of a rise may be enough for one review, while another needs to establish when a boundary was crossed.
For alarm investigations, retain the relevant event records alongside the analogue trace. Agree which record answers each question; a visually quiet trend should not be treated as proof that no alarm event occurred. Our guide to why operators ignore alarms explains why alarm handling needs its own review.
How do you verify the resulting historical trace?
Run a controlled comparison with a reference record that preserves the incoming readings for the test period. Apply the candidate settings through the intended collection and storage path.
Include these cases:
- Steady operation: assess normal variation.
- Slow movement: compare ramps and boundary crossings.
- Short excursions: test the smallest and shortest events in the acceptance criteria.
- Operating transitions: include starts, stops and recipe changes where relevant.
- Data interruption: check gaps, quality and recovery.
Compare retained timestamps and values, peak and minimum values, event duration and the order of changes.
Inspect the stored records as well as the chart. A display may interpolate or aggregate data, so its appearance alone may not reveal what was retained.
Keep the reference, candidate settings and comparison results together. Put the agreed behaviour in the functional design specification so future changes have a clear acceptance basis.
What common mistakes should you avoid?
Applying one setting to every analogue signal. Group signals by purpose, units and event requirements before considering shared settings.
Choosing settings only to reduce storage. First establish what information must survive. Measure storage effects after the trace passes that test.
Testing only steady production. Include transitions and the excursions that motivated the recording requirement.
Assuming the chart shows recorded points. Confirm how the display constructs its line and examine retained values.
Changing several settings together. Adjust one stage at a time so the comparison can show which change affected the record.
When does this approach need a different recording strategy?
Use separate acceptance criteria for discrete states, counters, batch events and quality changes. Do not apply an analogue deadband rule to them by habit.
Where a signal supports a safety-related decision, define its recording requirements with a qualified assessor. For contractual, quality or other formal records, agree the retention and evidence requirements with the responsible owner before applying compression.
If the required event cannot be preserved reliably by the proposed path, change the recording strategy rather than weakening the acceptance criteria.
Related reading: diagnosing gaps in historian data and standardising engineering units in plant data.
What else should you know about historian deadbands?
Is a smaller deadband always better?
Choose the setting against the required evidence. A smaller setting still needs checks for collection timing, storage behaviour and display interpretation.
Can a maximum interval replace event recording?
Do not assume it can. Test the shortest required event explicitly and confirm what the interval guarantees.
How often should settings be reviewed?
Review them when the instrument, range, process, collection path or intended use changes. Repeat the comparison before accepting the revised record.
Need help defining what your plant history must preserve? Our automation services can help scope the recording requirements and verification work.