This is a ready-to-use worksheet for deriving and documenting a defensible compression, or exception-reporting, deadband for one process historian tag. Compression settings are configuration, not raw fact: a deadband chosen without arithmetic against the parameter’s control limits can smooth a real excursion out of the record at the moment of capture, and nobody signed off on that. This worksheet turns “the deadband looks reasonable” into a calculation an inspector, or the next engineer, can follow. Replace every <<FILL: ...>> placeholder with your own values. A fully worked specimen follows. This is general guidance to adapt and verify against your own validated process limits, not legal or regulatory advice.
Header
| Field | Entry |
|---|---|
| Worksheet title | Historian Compression / Exception-Reporting Deadband Qualification |
| Worksheet number | <<FILL: FORM-ID>> |
| Version | <<FILL: version>> |
| Historian / system | <<FILL>> |
| Tag name | <<FILL>> |
| Prepared by (Process / area SME) | <<FILL>> |
| Date | <<FILL>> |
1. Inputs
| Field | Entry | Source |
|---|---|---|
| Engineering units | <<FILL>> | Tag register |
| GxP classification | <<FILL: GxP-critical / GxP-supporting / non-GxP>> | Tag register |
| Validated control range | <<FILL: low to high>> | Process validation report / control strategy |
| Alarm or action limit (if tighter than control range) | <<FILL>> | Alarm philosophy document |
| Target detectable excursion (magnitude) | <<FILL>> | Risk assessment or process SME judgment, tied to potential product or process impact |
| Target detectable excursion (minimum duration) | <<FILL>> | Risk assessment; how briefly a deviation could matter and still need to be provable |
| Sensor / signal noise floor (typical peak-to-peak) | <<FILL>> | Instrument specification or historical data review |
| Current scan / sample rate | <<FILL>> | Tag register |
2. Formula and method
The deadband defines how far the signal must move from the last stored value before the historian records a new point. Set it well inside the target detectable excursion, and comfortably above the noise floor, so genuine noise does not force excessive storage while a real, significant excursion cannot be absorbed into a straight line between two stored points.
A useful working range, subject to engineering judgment and the specifics of the parameter: the deadband should be small enough that the target excursion is at least several times larger than the deadband (so the excursion clearly produces new stored points rather than sitting at the edge of detection), and large enough to sit comfortably above the noise floor (so storage is not driven purely by sensor noise). There is no single regulatory number for this ratio; document the reasoning for the specific parameter rather than applying a generic rule uncritically.
Record the calculation:
| Step | Value |
|---|---|
| Target detectable excursion | <<FILL>> |
| Proposed deadband | <<FILL>> |
| Ratio: target excursion / proposed deadband | <<FILL: calculate>> |
| Proposed deadband compared to noise floor | <<FILL: above / at / below>> |
| Conclusion | <<FILL: deadband accepted / rejected / revise>> |
3. Sensitivity check
Confirm the proposed deadband against at least one excursion smaller than the target, to see how close to the edge of detection the setting sits.
| Excursion magnitude tested | Duration | Would it generate a new stored point at the proposed deadband? | Notes |
|---|---|---|---|
<<FILL: at target>> | <<FILL>> | <<FILL: Y/N>> | <<FILL>> |
<<FILL: below target>> | <<FILL>> | <<FILL: Y/N>> | <<FILL>> |
4. Acceptance criteria
- The proposed deadband is smaller than the target detectable excursion by a documented, reasoned margin, not a copied rule of thumb.
- The proposed deadband is above the signal’s noise floor, so storage volume is not driven purely by noise.
- The sensitivity check shows the target excursion reliably generates new stored points at the proposed setting.
- For GxP-critical tags, disabling compression entirely (storing every sample) is considered and the decision to use a deadband instead, if made, is separately justified.
- The final deadband is recorded in the tag register and implemented only through the tag configuration change request, under change control.
5. Disposition
| Field | Entry |
|---|---|
| Final deadband selected | <<FILL>> |
| Compression disabled instead (Y/N) | <<FILL>> |
| Change control reference for implementation | <<FILL>> |
| Prepared by (signature, date) | <<FILL>> |
| Reviewed by, Automation (signature, date) | <<FILL>> |
| Approved by, QA (signature, date) | <<FILL>> |
References
21 CFR 211.68 (accuracy of automatic equipment); 21 CFR Part 11. EU GMP Annex 11, clause 5 (accuracy of data captured by a computerised system). GAMP 5 (2nd ed., 2022), for the risk-based approach to configuration decisions.
Confirm the current version of each reference before issue.
Filled specimen
A worked calculation for a dissolved oxygen tag on a bioreactor.
| Field | Entry |
|---|---|
| Historian / system | Site historian, bioreactor suite |
| Tag name | BR101.DO.PV |
| Engineering units | % of saturation |
| GxP classification | GxP-critical |
| Validated control range | 25% to 55% (around a 40% setpoint) |
| Alarm limit | 20% low, 60% high |
| Target detectable excursion (magnitude) | 3% |
| Target detectable excursion (duration) | 15 seconds or longer |
| Noise floor | Approximately 0.3% peak-to-peak, from probe specification |
| Current scan rate | 1 second |
| Step | Value |
|---|---|
| Target detectable excursion | 3% |
| Proposed deadband | 0.1% |
| Ratio: target excursion / proposed deadband | 30:1 |
| Proposed deadband compared to noise floor | Below (0.1% versus roughly 0.3% peak-to-peak noise) |
| Conclusion | Revise: 0.1% sits below the probe’s own noise floor, so ordinary sensor jitter, not real process movement, would drive most stored points, inflating storage without adding real detection margin against the 3% target; a 0.5% deadband keeps the ratio comfortably above 5:1 against the target excursion and sits clearly above the 0.3% noise floor |
Sensitivity check at the revised 0.5% deadband: a 3% excursion (the target) generates multiple new stored points well before reaching the deadband threshold at each step, confirming detection. A 1% excursion, below the target, still generates stored points, giving margin beyond the minimum requirement.
| Field | Entry |
|---|---|
| Final deadband selected | 0.5% |
| Compression disabled instead | No, justified deadband used |
| Change control reference | CC-2026-0512 |
| Prepared by | M. Osei, Process SME, 08 September 2026 |
| Reviewed by, Automation | J. Marsh, 09 September 2026 |
| Approved by, QA | R. Gomez, 10 September 2026 |
The first candidate, 0.1%, looked conservative on paper, tighter is not automatically better, because it sat too close to the sensor’s own noise floor to add real protection while inflating storage. The worksheet caught that before it went into production, which is the point of doing the arithmetic instead of picking a small-sounding number.
Common inspection findings this worksheet prevents
- A compression deadband set from a vendor’s out-of-box default with no calculation against the parameter’s own control limits.
- A deadband so close to the sensor noise floor that it provides no real protection while still discarding data.
- No documented target excursion or duration, so nobody can say what the deadband was actually meant to catch.
- A deadband changed later with no re-derivation of the arithmetic behind the new value.
How to adapt this worksheet
- Pull the validated control range and alarm limits from your process validation report or alarm philosophy document, not from memory.
- Set the target detectable excursion from your own risk assessment; a tighter target than the example here is appropriate for a more sensitive parameter.
- Use this worksheet alongside the Historian Tag / Point Configuration Change Request form to implement the resulting deadband under change control.
- For a tag where the arithmetic keeps landing close to the noise floor no matter what deadband is tried, treat that as a signal to disable compression for that tag rather than force a number that does not hold up.