ECSS Mechanical Testing of Metals -- Fatigue Testing (space-systems/ecss/q7045-fatigue-testing)
Use when the methods clause of ECSS-Q-ST-70-45 calls for fatigue data: a set
of specimens has been cycled at a controlled frequency, some failed and some
were stopped at a declared run-out, and the question is what stress-life
relation those points support and which of them were allowed to help
establish it.
Domain quick reference
- A run-out is a lower bound, not an observation. The specimen did not fail,
so its recorded life is the point at which somebody stopped the machine;
letting it into a least-squares fit pulls the curve towards a life that was
never measured, always in the unconservative direction.
- A run-out still has to be consistent with the curve it is excluded from. If
it was stopped before the fitted relation places failure, it survived
nothing the curve did not already expect and it constrains nothing.
- The declared threshold is a contract in both directions. A point marked
run-out below it was stopped early, and a failure recorded beyond it says
the threshold in the report is not the threshold that was used.
- The curve belongs to a specimen form. A notched set without its stress
concentration factor cannot be carried to a different notch, and a factor
declared against an unnotched form means the two records disagree.
- Frequency travels with the data. Outside the band where heating and rate
effects are negligible, the set is internally consistent and not comparable
with anything else in the allowables.
Workflow
- Validate the set: positive stress amplitudes, at least one whole cycle per
point, and an explicit boolean run-out marking on every point.
- Separate failures from run-outs and fit log10 life against log10 stress
over the failures alone, requiring at least three of them.
- Refuse a fit that rises with stress amplitude or that has no stress spread
to fit against; neither describes a fatigue curve.
- Reconcile each run-out with the declared threshold, then with the fitted
curve, and report the ones that constrain nothing.
- Report failures recorded beyond the declared threshold as a contradiction
in the record rather than silently accepting them.
- Compare the cycling frequency with the band and the specimen form with its
notch factor, and attach both to the curve.
- Invert the curve onto the run-out life to report the stress the data
places there, and onto any design stress that was asked about.
Pitfalls
- Fitting the run-outs with the failures. It is one line of code and it moves
the whole curve to the right.
- Reporting the run-outs as an endurance limit. A set stopped at a threshold
shows where the machine stopped, not where the material stops failing.
- Quoting a life from the curve far outside the stress range that was tested.
A log-log line extrapolates cheerfully and without warning.
- Mixing notched and unnotched points into one set. The fit succeeds, the
scatter looks like material variability, and the notch effect disappears
into the curve.
- Comparing a set cycled well above the band with one inside it. Self-heating
changes what is being measured, and neither set is wrong on its own.
Behavior contract (gate 3)
The S-N validation, the censored log-log fit with its quality and refusals,
the life and stress inversions, the run-out threshold and consistency
findings, the frequency band, the specimen form and notch-factor
reconciliation and the usable-curve verdict are exercised by the gate 3
contract test: scripts/test_q7045_fatigue_testing.py against
scripts/q7045_fatigue_testing_logic.py (stdlib unittest, offline). Run:
python3 scripts/test_q7045_fatigue_testing.py
Compliance
- ECSS standards are freely downloadable (ESA); cite the source and
paraphrase per standards-map.yaml.
- compliance: STANDARDS-REF, gated: false.
1---2name: q7045-fatigue-testing3description: "Derive a stress-life curve from an S-N set and report what its points are entitled to claim. Use when the methods clause of ECSS-Q-ST-70-45 calls for fatigue data: censor the run-outs out of the log-log least-squares fit instead of letting them pull the curve towards a life nobody observed, reconcile every point against the declared run-out threshold, report a run-out that stopped before the fitted curve predicts failure as constraining nothing, and carry the cycling frequency and the specimen form with the curve. Trigger: ecss, q-st-70-45, sn-curve-basquin-loglog-fit, fatigue-runout-censoring, fatigue-runout-threshold-contradiction, fatigue-cycling-frequency-band, fatigue-notched-specimen-kt."4license: Apache-2.05compliance: STANDARDS-REF6standards:7 - id: ecss8 reference-only: true9gated: false10domain: space-systems11pack: space-systems12compatibility: "agentskills.io SKILL.md; any SKILL.md host (Claude Code, Hermes, OpenClaw)"13metadata:14 source: https://skillmd.com/skills/ashfordeou/q7045-fatigue-testing15 domain: space-systems16 subdomain: ecss17 tags: [ecss, q-st-70-45-mechanical-testing-scope, q7045-fatigue-testing, sn-curve-basquin-loglog-fit, fatigue-runout-censoring, fatigue-runout-threshold-contradiction, fatigue-cycling-frequency-band, fatigue-notched-specimen-kt]18 version: 0.1.019 author: Aero Agent Skills20---2122# ECSS Mechanical Testing of Metals -- Fatigue Testing (space-systems/ecss/q7045-fatigue-testing)2324Use when the methods clause of ECSS-Q-ST-70-45 calls for fatigue data: a set25of specimens has been cycled at a controlled frequency, some failed and some26were stopped at a declared run-out, and the question is what stress-life27relation those points support and which of them were allowed to help28establish it.2930## Domain quick reference3132- A run-out is a lower bound, not an observation. The specimen did not fail,33 so its recorded life is the point at which somebody stopped the machine;34 letting it into a least-squares fit pulls the curve towards a life that was35 never measured, always in the unconservative direction.36- A run-out still has to be consistent with the curve it is excluded from. If37 it was stopped before the fitted relation places failure, it survived38 nothing the curve did not already expect and it constrains nothing.39- The declared threshold is a contract in both directions. A point marked40 run-out below it was stopped early, and a failure recorded beyond it says41 the threshold in the report is not the threshold that was used.42- The curve belongs to a specimen form. A notched set without its stress43 concentration factor cannot be carried to a different notch, and a factor44 declared against an unnotched form means the two records disagree.45- Frequency travels with the data. Outside the band where heating and rate46 effects are negligible, the set is internally consistent and not comparable47 with anything else in the allowables.4849## Workflow50511. Validate the set: positive stress amplitudes, at least one whole cycle per52 point, and an explicit boolean run-out marking on every point.532. Separate failures from run-outs and fit log10 life against log10 stress54 over the failures alone, requiring at least three of them.553. Refuse a fit that rises with stress amplitude or that has no stress spread56 to fit against; neither describes a fatigue curve.574. Reconcile each run-out with the declared threshold, then with the fitted58 curve, and report the ones that constrain nothing.595. Report failures recorded beyond the declared threshold as a contradiction60 in the record rather than silently accepting them.616. Compare the cycling frequency with the band and the specimen form with its62 notch factor, and attach both to the curve.637. Invert the curve onto the run-out life to report the stress the data64 places there, and onto any design stress that was asked about.6566## Pitfalls6768- Fitting the run-outs with the failures. It is one line of code and it moves69 the whole curve to the right.70- Reporting the run-outs as an endurance limit. A set stopped at a threshold71 shows where the machine stopped, not where the material stops failing.72- Quoting a life from the curve far outside the stress range that was tested.73 A log-log line extrapolates cheerfully and without warning.74- Mixing notched and unnotched points into one set. The fit succeeds, the75 scatter looks like material variability, and the notch effect disappears76 into the curve.77- Comparing a set cycled well above the band with one inside it. Self-heating78 changes what is being measured, and neither set is wrong on its own.7980## Behavior contract (gate 3)8182The S-N validation, the censored log-log fit with its quality and refusals,83the life and stress inversions, the run-out threshold and consistency84findings, the frequency band, the specimen form and notch-factor85reconciliation and the usable-curve verdict are exercised by the gate 386contract test: scripts/test_q7045_fatigue_testing.py against87scripts/q7045_fatigue_testing_logic.py (stdlib unittest, offline). Run:88python3 scripts/test_q7045_fatigue_testing.py8990## Compliance9192- ECSS standards are freely downloadable (ESA); cite the source and93 paraphrase per standards-map.yaml.94- compliance: STANDARDS-REF, gated: false.