ECSS Fault Tree Analysis — Procedure Execution (space-systems/ecss/q40-12-fta-procedure)
Use when the task is the procedure clause of ECSS-Q-ST-40-12C — running the adopted IEC 61025 method on a tree that already exists: construction checks, cut set expansion, qualitative and quantitative evaluation, importance, sensitivity, and the record that leaves the desk.
Domain quick reference
- The procedure runs in one order and each step consumes the one before it. The cut sets come out of the construction, the probability comes out of the cut sets, and the importance and sensitivity numbers come out of the quantification. A number produced out of order has no traceable basis.
- Only coherent gate logic is expanded here: AND, OR and k-out-of-n. A k-out-of-n gate expands to every k-subset of its inputs, so a two-of-three vote yields three order-two cut sets. A non-coherent gate needs a different expansion and is refused rather than approximated.
- Construction is graded before expansion: an input that names nothing, an identifier used for both a gate and a basic event, a gate with a single input, a voting gate whose k sits outside one to n, and any cycle are all defects that make every downstream number meaningless.
- A cut set is minimal when no proper subset of it is also a cut set. A superset left in the list double counts in the quantification and inflates every importance share, so minimisation is part of the expansion rather than a tidy-up afterwards.
- The qualitative product stands on its own: the order profile, and the order-one cut sets, which are single point failures. That result exists before any probability is attached and often decides the design change.
- Three quantifications answer different questions. Inclusion-exclusion over the minimal cut sets is exact but costs two to the power of the cut set count, so it is bounded. The rare-event sum is an upper estimate that over-counts the overlap. The min-cut-upper-bound sits between them and stays affordable on a large tree.
- Importance is not one measure. Birnbaum is the swing in the top event between the contributor being certain and being impossible, and ignores how likely it is. Fussell-Vesely is the share of the top event carried by the cut sets that contain it. Criticality weights Birnbaum by the contributor's own probability. A design change is argued on Birnbaum; a data improvement is argued on Fussell-Vesely.
- Sensitivity is the honest treatment of a weak input: scale one basic event probability across a range and report how far the top event moves, so the reader sees which number the conclusion actually rests on.
Workflow
- Validate the tree. Reject dangling references, repeated inputs, duplicate identifiers, malformed voting gates and cycles before anything is expanded, and list any gate the top event never develops.
- Expand to minimal cut sets, minimising as you go so the intermediate lists never carry supersets, and stop with an explicit error if the expansion runs away rather than truncating silently.
- Evaluate qualitatively: build the order profile and name every order-one cut set as a single point failure.
- Quantify. Take the exact value where the cut set count allows it, and carry the rare-event and min-cut-upper-bound values alongside so the reader can see the size of the approximation.
- Rank the contributors under all three importance measures, and state which measure the recommendation is argued on.
- Sweep the sensitivity of the basic events whose data basis is weakest, and report the factor at which the conclusion changes.
- Assemble the record: tree, cut sets, order profile, every probability with its method, the importance table, the sweep, and the verdict against the declared target.
Pitfalls
- Quoting the rare-event sum as the answer. It omits the overlap terms, so it is an over-estimate that grows with the number of cut sets, and on a tree with one cut set it is identical to the exact value, which is exactly why it looks safe until it is not.
- Comparing a quantified probability with a target by bare arithmetic. The exact value is a sum of signed products, so a case that sits exactly on its target can land a few units in the last place above it; the comparison absorbs that representation error while the target stays untouched.
- Taking a product over an unordered set of events. Float multiplication is not associative, so an unordered product can differ between runs and between machines; every product here is taken over a sorted event list so a rerun reproduces the same value.
- Reading a low Fussell-Vesely share as a robust contributor. The share is weighted by how likely the event already is, so an event with a very low probability and a huge Birnbaum swing looks unimportant right up to the point the probability estimate turns out to be wrong.
- Leaving an undeveloped gate on the sheet. It appears in the drawing, carries no cut sets and silently contributes nothing, so the tree reviewers approve is not the tree that was quantified.
Behavior contract (gate 3)
The construction validation, cut set expansion and minimisation, order profile, three quantification methods, three importance measures, sensitivity sweep and reporting verdict are exercised by the gate 3 contract test: scripts/test_q40_12_fta_procedure.py against scripts/q40_12_fta_procedure_logic.py (stdlib unittest, offline). Run: python3 scripts/test_q40_12_fta_procedure.py
Compliance
- ECSS standards are freely downloadable (ESA); cite the source and paraphrase per standards-map.yaml.
- compliance: STANDARDS-REF, gated: false.
