Practical implementation.
WHAT TO DECIDE FIRST
What question the twin must answer.
WHY
It determines what to measure, what to model, and how accurate it must be.
WHAT TO AVOID
Building a comprehensive model and then seeking a use.
WHAT TO START WITH
One asset, one question, and the minimum that answers it.
WHAT DATA TO COLLECT
Only what the model needs.
WHY
Sensors, transmission and storage all cost, and unused data is pure cost.
WHAT MODEL TYPES EXIST
Physics-based, derived from engineering principles Data-driven, learned from observation Hybrid, combining both
WHAT PHYSICS-BASED MODELS PROVIDE
Behaviour in conditions never observed.
WHAT DATA-DRIVEN MODELS PROVIDE
Accuracy in observed conditions, with no need to model mechanisms.
WHAT HYBRID PROVIDES
The strengths of both, at greater complexity.
WHAT TO VALIDATE
The model against reality, continuously.
WHAT DRIFT MEANS
The model diverging from the asset as it ages or changes.
WHAT TO DO ABOUT IT
Monitor error, and recalibrate.
WHAT TO NEVER DO
Act on a model whose accuracy is unverified.
