Computational representation of a physical installation, fed continuously by measurements taken from it and holding a feedback loop by which its outputs influence operations.
Three cumulative conditions distinguish a digital twin from a mere model or simulation: a representation, continuous feeding by real measurements, and a feedback loop by which the model's output influences operations. It is the third condition that carries the legal consequences, since it moves the model from study tool to organ of command. The crossing usually happens in discrete steps, decision support, then a recommendation followed by habit, then automatic control of certain parameters, with no formal decision sanctioning it. The failure mode proper to this object is divergence, a growing and silent gap between the state of the model and the real state of the installation, produced by wear, repairs and process changes, with no triggering event and no alarm. A twin also produces, through simple technical necessity, a timestamped log of measurements, computations and setpoints issued, which permanently documents what the operator knew and what it decided.
The damage is observed at the physical end of the chain, a machinery breakdown or an injury, while the cause sits in the model, a link that is neither an asset, nor a product in the old sense, nor an attacked information system.
digital twin, réplique numérique, modèle miroir