A digital twin is a live model of a physical thing, updated from that thing's own instrumentation. The value comes from asking questions of the model that you cannot safely ask of the plant: what happens if we run this line 12% faster, what fails first, where does the queue form.
The drift problem
Plants change. An operator reroutes a conveyor, a fixture is replaced with a slightly different one, a changeover procedure gets an undocumented workaround. The twin does not notice. Six months later the model and the plant disagree, and nobody knows by how much.
Making drift visible
- Reconcile continuously. Compare predicted throughput against measured throughput and alert on divergence, not just on failure.
- Treat model changes as changes. If the plant is modified, the twin update belongs in the same work order.
- Publish confidence. A twin that reports how stale its assumptions are gets trusted appropriately.
Start narrow
Whole-factory twins are usually where these projects die. A twin of one bottleneck cell, kept accurate, pays for itself and teaches the team what maintaining a model actually costs. Expand from something that works.