AI4Manufacturing/XJTU-annotated

Roles

Roles: reasoning view of XJTU — annot is the source label (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 907 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.

XJTU-SY — fault from the envelope spectrum, with reasoning

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

Records: 907 (splits {'train': 690, 'test': 217}); labels {'normal': 416, 'outer_race': 352, 'inner_race': 139}; evidence_tier {'confirmed': 907}.

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
reasoning	the chain of thought for this record, four paragraphs written by a deterministic renderer; it closes on FINAL ANSWER: <annot>
cate / task	C / T-C1 (signal fault classification)
metadata	JSON string: representation, condition, bearing_id, file_number, time_frac, life_files, channel, bearing, rpm, fs, fr_nominal, fr_used, fr_source, features, fault_freqs, computed_verdict, computed_snr, evidence_tier, line_evidence, image_sha256, split

How reasoning was produced

This revision (2026-09-25) replaces the reasoning column; earlier versions remain in the repository history.

A deterministic renderer, not a language model. Each text is assembled by fixed code from the record's metadata and its gold label. No API, no sampling, no model output: rendering the same record again gives the same bytes.

Four paragraphs, separated by blank lines, with no headings inside the text:

1. Figure reading — what the figure is (an envelope spectrum, the shaft rate, where the BPFO / BPFI / BSF / FTF marks sit) and the complete readings at the marked positions, their harmonics and sidebands, and the shaft lines 1x and 2x, all taken from the record's metadata; where one peak serves several positions, that is said. Description only: no explanation, no verdict.
2. Physics — every candidate class explained in one continuous derivation: a damaged spot struck over and over gives a line at the striking rate and its harmonics; damage on the outer race is struck at BPFO with equal force because the spot stays put, so it has no sidebands; damage on the inner race is struck at BPFI and carried in and out of the load zone once per shaft turn, which modulates the strikes and puts sidebands at BPFI ± 1x. Rolling-element damage is not a candidate in this set, so the BSF and FTF marks are described as references. It also says what is not evidence for any class — the shaft lines 1x and 2x, a line at FTF on its own, and a peak where the positions of two parts nearly coincide — and that a sound bearing stays near the noise floor. Every term is defined where it is first used. The content of this paragraph does not depend on the record's label.
3. Linking — the readings from paragraph 1 tied to the mechanisms in paragraph 2: which family's lines rise, why that pattern is the one this class produces, and why the other families do not account for the figure. Only readings that paragraph 1 reports are cited.
4. Conclusion — one sentence, then a blank line and FINAL ANSWER: <annot>, equal to the record's gold label byte for byte.

Variation comes from the image, never from the label. Paragraphs 1 and 2 swap order for roughly half the records (the exact share is given below); the linking paragraph and the conclusion always come last. Every sentence has at least three phrasings, chosen sentence by sentence; the order in which the families are listed in paragraph 1 varies as well. The swap and every choice of phrasing are taken from the record's image hash, image_sha256 in metadata, never from the label.

No criterion in the text. No decision threshold, class boundary or per-class range appears anywhere in the text, and no reading is compared against one. Heights are given as multiples of the noise floor, rounded down to a whole number (in steps of five from 50 up; below 2 the text says "barely above the floor"), so one printed value never stands for readings on both sides of a boundary the label depends on. Strength is otherwise put in words ("rise clearly", "stays near the floor").

Every number is measured, never copied from the answer. Frequencies come from the record's fault_freqs and shaft rate, heights from line_evidence — the per-line measurement taken on the plotted envelope spectrum.

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.

Numbers for this set. 907 records; median length 809 words (whole text, FINAL ANSWER line included); the physics paragraph comes first in 434 of 907 (47.9%). All 907 texts are distinct, and every text ends on its gold label (checked on all rows).

Example. The linking paragraph (paragraph 3) of test record 146 (0-based row of the test split, gold outer_race), verbatim:

Here, 3 lines of the BPFO family stand up together: BPFO about 26 times, 3xBPFO about 13 times, 2xBPFO about 13 times. This fits a defect on the outer race: the spot stays where it is, each element hits it with the same force, and equal strikes repeating at BPFO give BPFO and its multiples. Of the other families, nothing rises at the BPFI positions; nothing rises at the BSF positions.

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 phrasing choices in reasoning.

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/xjtu/convert.py (0574e0e25d) → forge_model/XJTU/convert_xjtu.py (e023f6cb11); see provenance.json.

Gold = the documented teardown failure element (Table 3 of the dataset paper): outer_race = bearings 1_1/1_2/1_3/2_2/2_4/2_5/3_1/3_5, inner_race = 2_1/3_3/3_4. normal = early files (before min(30% of life, the data-driven degradation onset)); bearings whose onset is floor-bound (degrading from day one) contribute no normals. Each record was scored under both the nominal and a spectrum-refined shaft rate (rigs deviate 0.5–2% from nominal); the better envelope-pattern match won.

Caveats

• Evidence-gated, conflict-free release. The reasoning track keeps only confirmed records (the label-independent envelope-spectrum detector independently finds the documented fault). The perception tracks keep confirmed + non-conflicting weak; records where the detector confidently found a different pattern than the gold (e.g. bearing 2_5's healthy shaft harmonic aliasing into BPFI within 1.7%) are dropped — an image should never fight its own label.
• cage is EXCLUDED from this release. XJTU has two cage-failure bearings, but only 6/142 files confirm a cage (FTF-ladder) signature — and 57 fault files score as outer_race instead (a failing cage hammers the outer raceway; 8×FTF ≡ BPFO for this geometry). Retained in the raw form; the cage bearings' certified-healthy early files still serve as normal. Published classes: normal / inner_race / outer_race.
• TRUE bearing-wise split — the first run-to-failure set with enough bearings for it: test = whole held-out bearings (1_3, 2_5 outer; 3_3 inner; plus cage bearing 2_3, which after the cage exclusion contributes early-life normal records only), so evaluation is on unseen bearings; normal appears in both splits from disjoint bearings. Compound-failure bearings (1_5, 3_2) appear only as early-life normal.
• End-of-life masking — in the final ~3% of life, broadband breakdown can mask the discrete fault comb, so late windows are not uniformly confirmed. A property of the physics, not the converter.

Source & license

Source: XJTU-SY bearing datasets — Xi'an Jiaotong University & Changxing Sumyoung Technology; 15 LDK UER204 bearings run to failure under 3 conditions (2100/2250/2400 rpm, 12/11/10 kN); horizontal-channel accelerometer snapshots (25.6 kHz, 1.28 s per minute). Cite: B. Wang, Y. Lei, N. Li, N. Li, IEEE Trans. Reliability 69(1):401–412, 2020 (DOI 10.1109/TR.2018.2882682). Gold labels: Table 3 of Lei et al., J. Mech. Eng. 55(16), 2019 (DOI 10.3901/JME.2019.16.001). Released by the authors for research use (biaowang.tech/xjtu-sy-bearing-datasets).