Industrial Twin Lab

Synthetic Fault Laboratory

P-101 is a fictional, synthetic teaching fixture. The scenarios below are not field incidents or production observations.

Rare failures are difficult to study from operating history alone. The Fault Lab introduces controlled modifications into a twin or its inputs so detection behavior can be examined without damaging P-101 or disrupting production.

Synthetic fault evidence path

  1. Normal Twin
  2. Fault Injection
  3. Synthetic Operating Scenario
  4. AI Model
  5. Detection Capability
The isolated sequence examines model sensitivity. It neither injects a physical fault nor establishes real-world detection performance.

P-101 synthetic fault scenario catalogue

Canonical machine failure modes

  • bearing-degradationbearing degradationProgressive degradation of bearing condition.Affected signals: bearing-de-temperature, bearing-nde-temperature, axial-vibration, radial-vibration
  • impeller-degradationimpeller degradationLoss of hydraulic performance from impeller condition.Affected signals: flow, motor-power, discharge-pressure
  • cavitationcavitationVapour-cavity formation associated with inadequate suction conditions.Affected signals: suction-pressure, axial-vibration, radial-vibration
  • suction-restrictionsuction restrictionRestriction upstream of the pump inlet.Affected signals: suction-pressure, flow, motor-power
  • seal-leakageseal leakageLoss of process fluid at the pump seal.Affected signals: flow, discharge-pressure
  • motor-degradationmotor degradationReduced motor efficiency or electrical condition.Affected signals: motor-current, motor-power, speed

Sensor and communication faults

  • bias
  • drift
  • dropout
  • communication loss
The machine list is sourced only from the canonical fictional P-101 fixture. The four supplied input faults are separate from machine failure modes.

Candidate scenarios are limited to the canonical P-101 failure modes—bearing degradation, impeller degradation, cavitation, suction restriction, seal leakage, motor degradation—plus bias, drift, dropout, communication loss. Bearing-degradation injection may test explicitly declared friction increases of 5%, 10%, or 20%; these are assumptions, not observed events. Each scenario requires a mechanism, magnitude, affected signals, operating regime, duration, seed or deterministic configuration, twin version, and expected limitations.

Simulation validity comes first

Synthetic data is evidence from a model, not evidence from reality.

A model can learn an artifact of the simulator, a convenient injection shape, or a boundary condition that never occurs in service. Detection on a synthetic P-101 bearing-friction ramp shows sensitivity to that scenario—not proven sensitivity to real bearing degradation. Evidence must be labeled by origin and triangulated with physical reasoning, historical observations, maintenance findings, or separately validated tests.

Useful fault studies include negative controls and sensitivity analysis. The lab should vary load, flow, ambient temperature, sensor noise, missingness, and model parameters to ask whether a candidate detects the mechanism or merely the scenario template. Simulation uncertainty propagates into every resulting confidence statement.

Fault injection never occurs in the physical control system in Phase 1. The lab is isolated, read-oriented, and conceptual. Human engineering review decides whether a synthetic finding merits inspection, data collection, a safer test, or no action.

Conceptual demonstration — synthetic fixture results.

Qualification boundary

Simulation sensitivity is not field validity

Engineering review must document simulation validity, the domain gap, injection assumptions, provenance, limitations, and validation against physical evidence. These controlled scenarios are not field incidents; their only immediate claim is sensitivity to a declared model and injection configuration.