What is a Digital Twin?

A Digital Twin (DT) of a dynamic system is a simulation model

  1. with real-time information synchronization from the twinned real-world system to the DT,
  2. allowing real-time optimization and (human or automated) operational decision making based on simulation experiments.

The relevant information in the DT about the twinned real-world system is always up-to-date such that the DT can be used for running various what-if scenarios (simulation experiments) for improving the outcomes of the real-world system.

Information synchronization can be based, e.g., on sensor data when the real-world system is physical, or on forms of master-slave data duplication mechanisms when the real-world system is digital or cyberphysical.

The simulation-based operational decision making can be performed either by human users that manually run simulation experiments on the DT and then change behavior parameters of the real-world system or by a digital optimization system running simulation experiments on the DT and then changing behavior parameters of the real-world system. The optimization system may be a separate system or a component of either the real-world system or the DT. The concept of a DT does not imply such a component.

The ability to run fast simulation experiments on a DT on the basis of up-to-date information allows

  1. proactive simulation approaches, e.g., in a manufacturing DT, for testing if the expected resource availability allows to keep promised delivery dates;
  2. reactive simulation approaches, e.g., in a supply chain DT, for figuring out alternative routes in the case of supply chain disruptions.

Originally, the term "digital twin" has been coined in the area of Product Lifecycle Management, see (Grieves 2023), where it referred to virtual models of products. While physical products, and other physical objects, allow for static structure models focussing on their geometry and structural/material composition, it is more challenging to conceive of DTs as virtual models of dynamic systems in the sense of simulation models.

There is no need to restrict the DT concept to physical systems. In fact, it applies to any kind of dynamic real-world system, including digital/cyberphysical systems (such as ERP systems or robots), social systems (such as organizations or markets) and biological systems (such as human beings or eco-systems).

A DT may be a discrete, continuous or hybrid simulation model, depending on the relevance of accounting for discrete or continuous state changes.

Terminological Issues

The meaning of the term "digital twin" is still evolving. Since it has been adopted in different (scientific and professional) communities, it has been defined in different ways creating a number of terminological issues:

  1. In many works (especially in engineering areas) a "digital twin" is defined to be a "virtual replica" or "virtual model" of a real-world entity or system serving "certain purposes" (it's often not explained which ones). As opposed to our more narrow definition above, this popular definition seems to be way too broad, allowing to call any kind of computer model a "digital twin". For instance, according to this blurred definition, a plain 3D model of a vehicle or a building information model (BIM) or an ERP system or a Manufacturing Execution System (MES) would qualify as a DT, despite the fact that there is no need for (and no gain of) such a terminology given the fact that the widely used names "3D model", "BIM", "ERP system" and "MES" are already well-established.

    The term "virtual model" blurs the real meaning of DTs. Using such blurred definitions is quite popular in many engineering works, probably because leaving the real meaning of a term open and vague is easier than committing to a specific meaning (and may be also good for marketing by confusing and impressing readers).

  2. While most DT definitions list simulation as one of the purposes of a DT, they fail to identify simulation as its main purpose. In fact, other notable purposes, such as predictive maintenance or process mining (both requiring to access the event logs of the real world systems), or other forms of data processing for machine learning, need not be associated with the DT, but can be taken care of by other systems connected to the real world system and possibly also to its DT.
  3. Several authors (e.g., Eramo et al 2022) have proposed to make a distinction between "digital shadow" and "digital twin" by associating the feature of real-time information synchronization with the "shadow" and real-time optimization with the "twin", such that a DT would include a "digital shadow" as a subsystem. However, such a distinction does not provide any gain and just leads to terminology bloat.