Public MCIFT software
MCIFT 0.1.0a1 is an experimental deterministic Python toolkit for multichannel anomaly detection and condition monitoring.
Pin the exact release for reproducibility.
Requires Python 3.11–3.13 and NumPy 1.26 or newer. This is an experimental pre-release, not a production-ready or safety-certified system.
python -m pip install "mcift==0.1.0a1"Inspectable behaviour, not a claim of general validity.
- explicit reference fitting
- separate calibration and evaluation
- deterministic exchange scoring
- telemetry and vibration profiles
- six inspectable evidence gates
- caller-owned chronological history
- schema-4 safe serialization
- explanation and evidence output
- no runtime telemetry or network calls
From installation to limitations.
Initial frozen IMS Set 2 result.
The evaluation contained 536 recordings: 438 screening positives, 5 persistent warnings and 3 high-confidence warnings. The first persistent and high-confidence warning occurred 49 recordings, or 8 hours 10 minutes, before the terminal recording. This is not necessarily lead time before physical fault onset.
- Full six-gate pipeline
- 0 / 96 No screening, persistent, or high-confidence positives on the healthy holdout.
- G6-only diagnostic
- 8 / 96 Isolated conventional-vibration gate activations. These were not full MCIFT warnings.
- Evaluation result
- 438 → 5 → 3 Screening positives → persistent warnings → high-confidence warnings across 536 evaluation recordings.
- Localization
- Failed Top-1 accuracy: 0.0 · Top-2 bearing-1 coverage: 0.0
The complete six-gate system produced no positives on the 96-recording healthy holdout. G6 alone activated on 8 recordings, but the remaining MCIFT gates prevented those isolated activations from becoming screening, persistent, or high-confidence decisions.