AI4Manufacturing/IMS-annotated

Roles

Roles: reasoning view of IMS — annot is the source label (ball / inner_race / normal / outer_race), kept machine-parseable as the gold for verification and reward parsing; the model reads query + image, where the image is an envelope spectrum with the theoretical fault frequencies marked. The reasoning column is filled on all 542 records and is the SFT imitation target for this repo; the query enumerates the closed set of labels the answer must come from, and annot remains the string that is scored. metadata is provenance (split keys, source identities) and must never be fed to a model.

IMS / NASA-Bearing — fault from the envelope spectrum, with reasoning

The IMS reasoning track with the reasoning column filled. Same images, same queries, same gold labels, same splits: field-for-field identical to IMS except reasoning (asserted at build time).

Records: 542 (splits {'train': 434, 'test': 108}); labels {'normal': 195, 'inner_race': 36, 'ball': 30, 'outer_race': 281}; evidence_tier {'confirmed': 542}.

Schema (7-field unified record)

field	meaning
query	the classification instruction (one of 30 deterministic paraphrases per representation)
image	the rendered signal image (bytes embedded)
annot	gold fault class: normal / inner_race / outer_race / ball
reasoning	the half-template chain-of-thought for this record; it closes on FINAL ANSWER: <annot>
cate / task	C / T-C1 (signal fault classification)
metadata	JSON string: representation, set, timestamp, time_frac, channel, bearing, bearing_group, rpm, fs, fr_hz, features, fault_freqs, computed_verdict, computed_snr, evidence_tier, line_evidence, image_sha256, split

How reasoning was produced

Half-template, not a language model. Each paragraph is assembled by a deterministic renderer from three pools drawn independently per record — 14 openers × 7 physics-knowledge blocks × 26 argument skeletons — so the surface form varies widely while the argument stays checkable. No API, no sampling, no model output.

Every number is measured, never copied from the answer. All amplitudes, frequencies and peak counts come out of the record's line_evidence — the per-line measurement taken on the plotted envelope spectrum. The generator ran behind three gates, all of which had to be clean or nothing was written: (1) the paragraph ends on the gold, byte for byte; (2) every label-bearing field is deleted and the paragraph re-rendered — the bytes must be identical, which proves no number came out of the answer; (3) writing checks — nothing finer than 0.01 (the y axis cannot resolve it), no detector internals (family scores and thresholds are not on the figure and a reader cannot recompute them), no internal line keys, and numerals that agree with the families actually scored.

Re-checkable. The measurement the paragraph quotes ships with the record: metadata.line_evidence holds the per-line frequencies, amplitudes and SNRs and the noise floor, so every number in the text can be verified against the figure it was read off.

Closing. Every paragraph ends with a blank line and FINAL ANSWER: <annot>, matching the record's gold label exactly — re-verified for every record at build time.

Figures (style pool)

The spectra in this repo come from the figure-style pool: each record is drawn in one of 11 styles. What the style varies: colours, canvas, title placement and presence, tick spacing and minor ticks, grid, spines, fill, the fault-frequency marker drawn as a line or as a band, label placement and orientation, one colour per family plus a legend, font family, scientific notation, and tick direction. The style is picked by sha256("figstyle|" + image_path) % 11 -- deterministic, reproducible, and drawn independently of the opener / knowledge / skeleton pools.

What never varies is the content: the curve; the plotted band; which lines are marked (six on these figures -- the four families plus 2xBPFO and 2xBPFI); and the y-limits, autoscaled once per record and reused by every style, so a peak is the same height whichever style drew it. No threshold bar and no noise-floor line is ever drawn, and no marker is styled differently for being the answer. The point is that a model cannot pass by memorising one house plotting style. The -perception repo's images are unchanged.

Three things made these figures and no single pin covers all three, so provenance.json carries one for each:

• Signals and manifest -- forge_agent 0574e0e25d.
• The curve and the producer's plotting parameters -- forge_agent 733a683. This pin fixes what is plotted; it does not produce the image bytes below.
• The published image bytes -- figure_manifest_sha256 = 203e42bb5cd0491fa8a61faf569d0891c3f44db62d5b3b07b2961571e791c2db, the roll-up of one sha256 per image across the whole style pool. The style layer that draws them (_figstyle_corpus/_src/style_pool.py) lives outside this repository and is not under version control, and the pool has been re-rendered since the commit above -- so this checksum, not a commit, is what identifies these exact images.

Provenance & reproducibility

Generated deterministically by forge_agent/examples/ims/convert.py (0574e0e25d) → forge_model/IMS/convert_ims.py (e023f6cb11); see provenance.json.

Gold = the documented end-state defect (readme / manufacturer teardown): Set 1 → bearing 3 inner-race + bearing 4 roller(ball); Set 2 → bearing 1 outer-race; Set 3 → bearing 3 outer-race. normal = the early files of each run; fault = the late files of the failed bearing (per-set window from the degradation onset). A computed evidence_tier (confirmed/weak/absent) flags detectability.

Caveats

• Evidence-gated, conflict-free release — every image supports (and never fights) its label. IMS faults are WEAK run-to-failure signatures (the dataset's own reference paper, Qiu/Lee/Lin JSV 2006, studies weak-signature detection), so we curate by a computed evidence_tier: the spectrum/reasoning track keeps only confirmed records (the label-independent envelope-spectrum detector independently finds the documented fault → faithful compute-then-check CoT); the perception tracks keep confirmed + non-conflicting weak — weak records where the detector confidently found a different pattern than the gold are dropped (notably ball windows scoring as cage: ball faults are cage-modulated, so the single-label gold and the detector legitimately disagree there).
• inner_race is the scarce class — excluded from the first release, REINSTATED here. The original band-limited detector confirmed only ~8 inner-race spectrum windows ("too few to form a class"); the current full-band demodulation search recovers IMS's weak Set-1 signature and confirms 36 windows — more than the published ball class (30) — so the same evidence standard that excluded it now reinstates it. It remains the weakest class: ~78% of its raw windows are absent-tier (dropped per-record by the gate), and Set 1 ran two degrading bearings (inner-race B3 + ball B4) on one shaft. outer_race (Sets 2-3) is the clean, strong class.
• Few distinct bearings — each fault class comes from one run-to-failure bearing, so a strict bearing-wise split is impossible within a class; the split is file-stratified. Treat cross-bearing generalization claims with care.

Source & license

Source: IMS / NASA-Bearing — NSF I/UCR Center for Intelligent Maintenance Systems (imscenter.net) with Rexnord Corp.; three test-to-failure runs on Rexnord ZA-2115 bearings at 2000 rpm. Reference: H. Qiu, J. Lee, J. Lin, J. Sound and Vibration 289 (2006) 1066–1090. Distributed via the NASA Prognostics Data Repository.