1. Declare the analysis window.
Record asset/tag identity, time zone, start and end, sample interval or event semantics, engineering unit, rated basis, data-quality flags, and the source-system export. Never combine timestamps whose clock or daylight-saving treatment is unresolved.
2. Reconstruct categorical state before calculating a KPI.
Use the Machine State Log Analyzer when the export contains state changes rather than evenly sampled values. Confirm whether each row means “state became active now,” remove duplicate retransmissions only with evidence, and supply a closing timestamp if the final state needs duration. Treat RUNNING, IDLE, STOPPED, FAULT, unavailable, and communications-loss boundaries consistently.
3. Build time at load from a stable basis.
The Equipment Load Profile Analyzer expects equally spaced percent-of-rated-load readings. DOE guidance describes subdividing metered load into bins, counting readings, and converting the count to hours at load. Siemens condition-monitoring material likewise uses load-band operating hours to expose pump operating profiles. A mixed rating, changed sensor scaling, or irregular interval invalidates a simple count-times-interval histogram.
4. Measure excursions with an explicit hold convention.
The Threshold Excursion Duration Analyzer assigns each reading to the interval that follows it. That left-hold assumption is common for state-like exports but is not automatically correct for every analog recorder. If crossing time matters, retrieve higher-frequency evidence or use a justified interpolation method. Keep unavailable or suspect periods unknown rather than silently inside the limits.
5. Separate level limits from rate limits.
A value can remain inside its high/low band while changing too quickly. The Rate-of-Change Screener calculates change per actual elapsed time. Rockwell Automation documents rate alarming as process-variable change over a configured evaluation period against separate positive and negative limits; it also warns that resolution and a short period can create nuisance rate events.
6. Join evidence only after each boundary is sound.
State time may explain whether an excursion occurred during startup, steady operation, idle, or shutdown. Load distribution may show that one operating band dominates exposure. Rate events may identify transitions that a level-duration summary hides. These are investigation pointers—not proof of mechanism, damage, efficiency loss, remaining life, or maintenance need.
7. Preserve negative and uncertain results.
No detected event can mean the tag stayed within declared rules, the sample interval missed a short event, the sensor was stale, the limit was wrong for the active mode, or the window was not representative. Record which interpretation remains possible and what evidence would distinguish it.
Minimum review record
- Asset, tag, unit, rated basis, location, and operating mode.
- Source export, time zone, clock check, sample/event semantics, and window.
- Missing, suspect, flatline, maintenance, bypass, and communications periods.
- State vocabulary and final-row closure treatment.
- Load bins and interval basis.
- Lower/upper or rise/fall limits, source, revision, and applicability.
- Reconstructed results and unresolved conflicts.
- Qualified reviewer, next evidence request, and separate operating decision.
Public sources
- U.S. DOE Energy Information Handbook publishes the count × interval method for hours-at-load histograms.
- Siemens PumpMon monitoring guide documents load-distribution and operating-hours histograms for asset-management review.
- Rockwell Automation Logix 5000 process-control reference defines rate-of-change evaluation and the effect of the evaluation period.
- Rockwell FactoryTalk Optix rate-of-change alarm documentation distinguishes rate limits and polling time.
ReliabilityBench processes pasted values locally and applies declared arithmetic conventions. It does not supply safe operating limits, validate sensor calibration, configure protection, or authorize operation.