Feature Factory
P-101 is a fictional, synthetic teaching fixture. Its signals and engineered relationships are not field evidence.
Raw signals acquire meaning through units, locations, operating context, time windows, and physical relationships. The Feature Factory records those transformations so sensors and engineered features can compete as carefully as algorithms.
P-101 feature transformation sequence
- Temperature → DE temperature delta ambient, NDE temperature delta ambient, Rolling Mean 30m
- Vibration → Vibration RMS, Vibration Kurtosis
- Engineering relationships → Pressure Ratio, Flow / Speed, Power / Flow, Twin Residual
For vibration, a waveform can yield RMS, peak, kurtosis, crest factor, FFT, spectral energy, envelope spectrum, order components, and bearing frequencies. For process equipment, measurements combine with engineering relationships to produce pressure ratio, efficiency, heat rate, normalized power, ΔP, ΔT, flow per speed, power per flow, measured minus expected, and measured minus twin prediction.
Phase 1 publishes feature definitions and lineage records. It does not ingest waveforms, compute these transformations, or measure their diagnostic performance.
P-101 feature lineage
Pressure Ratio
Discharge pressure relative to suction pressure.
- Feature ID
- pressure-ratio
- Feature group
- process
- Source signals
- suction-pressure, discharge-pressure
- Unit
- dimensionless ratio
- Formula / transformation
- discharge pressure ÷ suction pressure
- Window
- Aligned sample; no rolling window.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
Flow / Speed
Flow normalized by rotational speed.
- Feature ID
- flow-per-speed
- Feature group
- process
- Source signals
- flow, speed
- Unit
- m³/h per rpm
- Formula / transformation
- flow ÷ speed
- Window
- Aligned sample; no rolling window.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
Power / Flow
Electrical power normalized by flow.
- Feature ID
- power-per-flow
- Feature group
- process
- Source signals
- motor-power, flow
- Unit
- kW per (m³/h)
- Formula / transformation
- motor power ÷ flow
- Window
- Aligned sample; no rolling window.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
DE temperature delta ambient
Drive-end bearing temperature above ambient.
- Feature ID
- bearing-de-delta-ambient
- Feature group
- physics
- Source signals
- bearing-de-temperature, ambient-temperature
- Unit
- °C
- Formula / transformation
- bearing DE temperature − ambient temperature
- Window
- Aligned sample; no rolling window.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
NDE temperature delta ambient
Non-drive-end bearing temperature above ambient.
- Feature ID
- bearing-nde-delta-ambient
- Feature group
- physics
- Source signals
- bearing-nde-temperature, ambient-temperature
- Unit
- °C
- Formula / transformation
- bearing NDE temperature − ambient temperature
- Window
- Aligned sample; no rolling window.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
Vibration RMS
RMS vibration for each bearing-housing channel. The fixture supplies scalar RMS levels, not waveform samples.
- Feature ID
- vibration-rms
- Feature group
- vibration
- Source signals
- axial-vibration, radial-vibration
- Unit
- mm/s RMS
- Formula / transformation
- per-channel square root of the mean squared waveform velocity; requires waveform samples absent from this fixture
- Window
- Declared vibration analysis window; duration not specified in Phase 1.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
Vibration Kurtosis
Per-channel waveform distribution-tail indicator; cannot be inferred from the two scalar RMS levels.
- Feature ID
- vibration-kurtosis
- Feature group
- vibration
- Source signals
- axial-vibration, radial-vibration
- Unit
- dimensionless
- Formula / transformation
- per-channel fourth central moment divided by variance squared; requires a nonconstant waveform absent from this fixture
- Window
- Declared vibration analysis window; duration not specified in Phase 1.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
Twin Residual
Measured value minus digital-twin prediction; unavailable until an independently validated prediction model exists.
- Feature ID
- twin-residual
- Feature group
- physics
- Source signals
- motor-power, flow, speed
- Unit
- kW
- Formula / transformation
- motor power − digital-twin motor-power prediction conditioned on flow and speed
- Window
- Aligned sample; no rolling window.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
Rolling Mean 30m
Thirty-minute rolling temperature mean; requires a timestamped history absent from this single snapshot.
- Feature ID
- rolling-mean-30m
- Feature group
- temporal
- Source signals
- bearing-de-temperature
- Unit
- °C
- Formula / transformation
- mean of bearing DE temperature over the trailing 30 minutes
- Window
- Trailing 30 minutes.
- Applicable regime
- Within the declared fictional P-101 operating envelope only.
- Provenance
- Industrial Twin Lab fictional engineering fixture; no plant-derived feature values.
- Leakage assessment
- Past or contemporaneous source values only; future values, maintenance labels, and held-out outcomes are prohibited.
- Validation status
- Not validated in Phase 1
Residual Intelligence
Residual Intelligence
Residual = Physical Measurement − Digital Twin Prediction
A residual asks whether P-101 behaves as the current twin expects under the same flow, speed, pressure, and ambient conditions. That deviation may carry more diagnostic information than a raw value because ordinary regime changes are represented explicitly. It is not automatically an anomaly: residual usefulness depends on twin validity, sensor quality, alignment, calibration, and the uncertainty of both measurement and prediction.
Feature competition must therefore record source signals, transformations, windows, units, missing-data behavior, software version, and leakage controls. A candidate that gains accuracy by using unavailable future data, a maintenance label, or a regime proxy has failed even if its metric improves. A smaller feature set may be preferred when it preserves detection performance with fewer sensors, lower inference cost, and clearer physical interpretation.
Conceptual demonstration — synthetic fixture results. P-101 feature definitions are illustrative and require validation before any physical-machine use.
Interpretation boundary
Traceable features are candidates, not conclusions
Feature provenance must bind sources, units, windows, code, and missing-data behavior. Performance remains conditional on operating regime, leakage controls, twin fidelity, and engineering validation. Feature generation may support a hypothesis; it does not establish causality.