A commentary from Dr Ebada, a professor of informatics and AI with a background at BMW and founder of digital-twin platform provider HOPn, argues that much of what the industry calls a "digital twin" is in fact a static 3D model with some data overlaid, not a genuine twin. He distinguishes four stages: a simulation, which runs on fixed assumptions; a digital model, which adds structure but no real-world feedback; a digital shadow, which receives live data from the physical system but only one way; and a true digital twin, which closes the loop by learning from physical behaviour, updating its internal state and flagging when a design or process has drifted from reality.
According to the piece, the main obstacle isn't model fidelity but data integrity: noisy signals, drifting components and small dataset inconsistencies propagate across the vehicle lifecycle and undermine trust in the twin. This is described as an organisational challenge as much as a technical one, requiring engineering, logistics, testing, suppliers, aftersales and compliance functions to agree on shared data definitions and interfaces.
Electrification is cited as a case where the stakes are highest: high-voltage battery data spans cell provenance, CO2 accounting, traceability, regional regulation and thermal models, and OEMs that manage these as separate IT tasks rather than a unified digital backbone end up with twins that can't answer basic questions.
The author also warns that AI layered on top of digital twins simply amplifies whatever data quality already exists, meaning fragmented or duplicated datasets will produce unreliable anomaly detection regardless of model sophistication.
Source: Automotive Testing Technology International — read the original article