Define the observable event first.
Write what changed, how it was detected, when it began and ended, and what recovery was observed. Do not begin with a component verdict. Use the event report builder before resets or replacements when approved procedures and safety allow.
Preserve the as-found state and original evidence.
Export native logs, verify timestamp bases, identify software and configuration, and retain trends, waveforms, photos, witness accounts, load, operating phase, and environment. Record every intervention in order. A reset that restores operation is an observed recovery action; it does not finish the investigation.
Make events comparable before counting them.
Apply stable unit IDs and controlled names for phase, code, and recovery. The pattern analyzer counts exact signatures transparently, but raw event counts are not exposure-normalized rates. Compare starts with starts, hours with hours, or another defensible opportunity denominator before claiming one condition increases risk.
Follow the unit, the installation, and the symptom separately.
No-fault-found can mean the shop test did not reveal a defect under its tested conditions; it does not prove the removed unit healthy in every installation. A repeated symptom after a swap can redirect attention to wiring, connectors, power, configuration, or system interaction, but still does not prove a cause. Use the history screener to expose three ratios against a declared local review threshold.
Turn patterns into bounded test questions.
State the target observable symptom, preserve the baseline, select triggers grounded in evidence, define synchronized measurements, change one factor at a time when feasible, and restore the baseline between stages. The reproduction plan builder creates a review draft, not test authority.
Retain negative as well as positive results.
A finite reproduction attempt samples only the declared conditions, duration, repetitions, and measurement capability. Failure to reproduce does not clear an intermittent condition. Record what was not observed and the limits of the attempt so the next event improves the investigation rather than restarting it.
Close the loop under accountable review.
- Preserve the event and as-found state.
- Normalize terms and compare serial, installation, and symptom history.
- Form a bounded, falsifiable test question.
- Approve hazards, procedures, limits, instrumentation, stop criteria, and restoration.
- Run only under qualified authorization.
- Retain source evidence and update the event history.
- Apply the organization’s separate release-to-service and corrective-action process.
Public sources
- Airbus, Intermittent Repetitive Failure — operating-condition triggers, occurrence monitoring, prior actions, resets, and rogue-unit history.
- FAA Maintenance Reliability Program Working Group report — public definitions and discussion of no-fault-found and chronic or rogue units.
- FAA MEDA Users Guide — investigation prompts covering fault isolation, operational tests, inspection, and maintenance-event learning.
- NCMS, intermittent fault detection and isolation study — public discussion of intermittent faults, no-fault-found outcomes, and detection challenges.
- Association for Advancing Automation, isolating intermittent faults — practical industrial troubleshooting context.
These sources establish the investigation problem and useful evidence practices. ReliabilityBench fields, ratios, and screens are declared workflow conventions, not regulatory limits, diagnostic standards, or compliance claims.