IMS Bearing Early-Warning Benchmark

Result summary

What was tested

IMS Set 2 is a physical bearing experiment run to its endpoint. This publication evaluates derived observations from one run. Recordings originate from sampled vibration waveforms, but this page publishes only derived scores. The experiment endpoint is known; the exact physical damage-onset time is not labelled. The detector uses all four IMS Set 2 bearing channels; Bearing 1 is associated with the endpoint failure.

Result timeline

Accessible timeline summary

Warning persistence rule

An isolated threshold crossing is recorded as a warning candidate. MCIFT first crossed the threshold for the sustained warning event at recording index 532. Index 533 was the second and index 534 the third consecutive positive observation, at which point the warning was confirmed. Recordings are spaced approximately 10 minutes apart: index 532 to 533 is one interval and 533 to 534 is another, approximately 20 minutes in total. Earlier shorter candidate runs did not become official alerts.

Methodology

Comparison with baselines

MCIFT began its stable run 50 minutes earlier relative to the endpoint and had fewer crossings before calibration ended. The conventional method had fewer fragmented positive runs. Export verdict: not established. Without a labelled damage onset, valid precision, recall and F1 cannot be calculated.

Detailed score table

Derived benchmark artifacts

Reproducibility and provenance

Benchmark repository

https://github.com/corpobear/MCIFT-Benchmarks

Git commit

731c249288d67dac4e3a10792980ec99be1de34d

Configuration

ims-set2-v1.yaml

Execution started (UTC)

2026-07-18T18:31:18.864811+00:00

Execution completed (UTC)

2026-07-18T18:33:52.434327+00:00

Seed

20260101

Dataset manifest hash

ec44367d57a672dec140eeec2870d1916f80672d6a5cdc2d8b5fd2ca43aaec21

Status in export

completed-unapproved

SHA-256 · summary.json

1c6a08585b88013ebce956988f2189fd06fba2eb86a1d1ad0c965508bf4b5b08

SHA-256 · scores.csv

d6c717e0bfeb3d7a97e57910feae7a0fc02a890c917db4fc8b10676082727ed8

IMS RESULT DATA HANDOFF

This folder contains verified derived IMS benchmark data for another agent to
process. It contains no website files, raw vibration recordings, Azure resource
identifiers, credentials, tokens, signed URLs, or environment-variable values.

Files:
- summary.json: exact verified benchmark summary downloaded from private run storage.
- scores.csv: exact verified per-recording derived detector scores.
- comparison.json: additional MCIFT-versus-conventional statistics derived from scores.csv.
- provenance.redacted.json: reproducibility facts with locations and environment details removed.
- data-dictionary.json: column meanings and interpretation constraints.

Scientific constraint:
The IMS experiment labels its final recording but does not provide a verified
physical degradation-onset timestamp. Lead time is therefore measured only to
the final recording and must not be described as proven damage-prediction lead.

Run status is completed but not scientifically approved for publication.

Limitations

• One bearing run was evaluated.
• The exact physical damage onset is not labelled.
• The experiment endpoint is not equivalent to damage onset.
• The result does not prove cross-machine generalization.
• Thresholds may depend on the defined calibration protocol.
• Additional IMS runs and stronger baseline comparisons remain necessary.
• Performance on heterogeneous digital telemetry must be evaluated separately.
• One successful benchmark does not validate the entire MCIFT framework.

Verdict