The operational problem
Rotating equipment changes with load, speed, maintenance state and environment. A useful early-warning method must separate degradation from normal regime changes while keeping false alarms low enough for operators to act.
Early-warning research for predictive maintenance
MCIFT explores transparent candidate features that may help identify meaningful deviation before a machine event. The work is an evaluation hypothesis, not a certified diagnostic or a replacement for established condition monitoring.
Relevant operational signals
- Vibration waveforms and spectra
- Bearing and gearbox condition indicators
- Temperature, pressure and flow
- RPM, load and operating regime
- Maintenance actions and confirmed event records
How MCIFT maps the problem
The proposed mapping represents agreement, periodic structure, changing boundaries and propagation as ordinary numerical features. Exact protected formula details are not published here. Candidate features remain inspectable and can be frozen before evaluation.
What must be compared
- Ordinary alarm thresholds
- Trend and residual analysis
- Spectral, envelope and order monitoring
- Established condition-monitoring features
- A model using the same input data without MCIFT features
Potential applications
- Pump condition and cavitation review
- Turbine and rotating-equipment monitoring
- Bearing and gearbox pattern change
- Efficiency drift and flow disturbance
- Earlier maintenance triage
Limitations
A deviation is not automatically a fault. Load changes, sensor replacement, speed variation and maintenance can move the reference state. Root cause and safety decisions require established engineering procedures.
Current evidence status
The website demonstrates schematic and computational mappings using illustrative data. No claim of superior warning time, diagnostic accuracy or industrial readiness is made before a controlled comparison.
Data needed for validation
- Time-aligned telemetry across normal and abnormal periods
- Operating-regime and load context
- Maintenance logs and confirmed event windows
- Sensor quality and missing-data records
- A holdout period not used to tune the mapping
Suggested evaluation metrics
- Warning lead time at a fixed false-alarm budget
- False alarms per operating hour
- Missed-event rate
- Precision and recall for review alerts
- Performance stability across regimes
- Calibration and uncertainty
Test one bounded machine decision.
Select one asset and compare frozen MCIFT candidate features with the monitoring method already used. The first deliverable is an evidence report, including neutral and negative results.
Discuss a validation study