ECSS Additive Manufacturing — Part Acceptance (space-systems/ecss/q7080-part-acceptance)
Use when the task is the acceptance step of ECSS-Q-ST-70-80C on a lot of built parts -- reading the limits the part class sets, confirming the screens that class owes, and deciding how much of the lot has to be examined before the lot as a whole can be accepted.
Domain quick reference
- Acceptance has two halves. The attribute half asks whether a part meets the density, porosity and surface limits of its class. The lot half asks how many parts are examined at all, how many findings the plan tolerates, and what a passing lot proves about the parts nobody looked at. Both have to close before the lot is accepted.
- A sampling plan is not a discount on inspection. It is a stated probability of accepting a lot that carries a given share of non-conforming parts, so the honest way to size one is backwards: name the limiting quality the project refuses to accept, name the risk it will tolerate of accepting it anyway, and take the smallest sample that holds that risk.
- Drawing parts from a finite lot happens without replacement, so the acceptance probability is hypergeometric and can be summed exactly. The binomial approximation that treats the lot as infinite is optimistic on exactly the small lots additive manufacturing produces.
- Acceptance probability falls as the sample grows, which is why the smallest protective sample is found by bisection rather than by scanning, and why a sample that protects a lot of one hundred also protects a lot of five hundred at barely a larger size. Protection tracks the share of bad parts, not the headcount.
- An acceptance number is a promise to ship a known finding. If it is at least the number of bad parts a limiting lot carries, no sample size makes the plan protective -- not even examining every part -- and the only fix is a smaller acceptance number.
- The most demanding class is not sampled at all. Where an escape is not survivable, every part is examined and the acceptance number is zero, and the sampling machinery is simply not the instrument.
- An attribute that was never measured is open, not passed. An empty field carries no information about conformance in either direction, and a lot closed on missing data is closed on an assumption.
Workflow
- Declare the part class and the lot size. Reject an uncategorized part rather than defaulting it, because the limits, the screens and the plan all hang off the class.
- Where the class is examined in full, take the whole lot and stop sizing; the plan is the inspection.
- Otherwise take the limiting quality and the tolerated consumer risk for the class, count the bad parts a limiting lot carries, and find the smallest sample whose acceptance probability stays under the risk. Floor it at the practical class minimum.
- Where no sample can reach the protection, say so and name the cause -- the acceptance number -- rather than quietly returning the lot size.
- Evaluate the attributes of the examined parts against the class limits, and list every attribute that was not measured as open.
- Confirm the mandatory screens for the class were performed, and close with an acceptance verdict that names each reason: findings beyond the acceptance number, a non-conforming attribute, a missing screen, or a plan that never was protective.
Pitfalls
- Choosing a sample size first and computing its protection afterwards, if at all. A round fraction of the lot has no stated consumer risk, and the plans that feel generous on small lots are the least protective ones there.
- Using a binomial model on a small lot. It assumes sampling with replacement from an infinite population, so it overstates the chance of catching a bad part in exactly the lot sizes a build plate produces.
- Reading a passing lot as evidence about the parts that were never examined. It is evidence about the lot at a stated confidence, and the strength of that evidence is the consumer risk, not the pass.
- Setting an acceptance number above zero on a class whose escapes matter. It tolerates a known finding, and on a small lot it can make protection unreachable at any sample size, which shows up as a plan that never goes protective however much is inspected.
- Treating a mandatory screen as satisfied because a sample was taken. Screens apply to every part of the class, so a sampling plan does not reduce them, and a screen that was skipped is a gap regardless of what the sample found.
- Comparing an acceptance probability with the tolerated risk by bare arithmetic. The probability is a ratio of large integers, so a plan meant to sit exactly on the limit can land a few units in the last place above it; the comparison absorbs that representation error while the risk limit stays untouched.
Behavior contract (gate 3)
The hypergeometric acceptance probability, limiting-quality count, protective sample search, class plans, attribute limits, screening gap and the lot verdict are exercised by the gate 3 contract test: scripts/test_q7080_part_acceptance.py against scripts/q7080_part_acceptance_logic.py (stdlib unittest, offline). Run: python3 scripts/test_q7080_part_acceptance.py
Compliance
- ECSS standards are freely downloadable (ESA); cite the source and paraphrase per standards-map.yaml.
- compliance: STANDARDS-REF, gated: false.
