Digital twin architecture: Layers, components, and types

Digital twin architecture is the layered technical structure that connects a physical asset to its live virtual model: the sensors and connectivity that capture data from the real world, the systems that process and model it, and the interface that makes it usable. This guide covers that technical architecture specifically.

If you're looking for how digital twins are used in buildings, construction, or the built environment, see digital twins for architecture, engineering, and construction instead. "Digital twin architecture" gets used both ways, and this page focuses on the systems-architecture reading.

For a general definition of what a digital twin is before diving into structure, see what a digital twin is.

Understanding this architecture is not just theoretical: in a McKinsey survey of industrial respondents, 86% said a digital twin is applicable to their operations, and 44% had already implemented one (McKinsey, 2024).

The layers of a digital twin architecture

A digital twin architecture is typically described as a stack, running from the physical world to the interface a person interacts with. While implementations vary, most digital twin systems include six recurring layers:

1. Physical layer. The real-world asset, machine, product, or environment being twinned, along with the sensors and instrumentation attached to it.

2. Data acquisition and connectivity layer. IoT devices, sensors, and edge hardware that capture data at the source and transmit it, often filtering or pre-processing at the edge before it moves further up the stack.

3. Integration layer. The middleware and APIs that pull data in from multiple sources - sensors, but also existing systems like MES or ERP - and normalize it into a consistent format the rest of the architecture can use.

4. Digital model layer. The virtual representation itself: the geometry, structure, and logic that mirror the physical asset. This is often built from CAD or BIM source data and is the core of the "twin."

5. Analytics and AI layer. Simulation, machine learning, and analytics run against the digital model and incoming data to generate insight; detecting anomalies, predicting failures, or testing what-if scenarios.

6. Visualization and application layer. The interface where people actually see and interact with the twin: dashboards, 3D visualizations, or applications built on top of the model and its analytics. This is where real-time 3D and converted CAD/BIM data live in the stack, turning the digital model and its data into something a person can see, navigate, and act on.

Data doesn't just flow in one direction through this stack. A defining feature of a digital twin, as opposed to a static model, is that insight generated in the analytics layer can inform action back on the physical asset. It is a bidirectional link rather than a one-way feed.

Key components of a digital twin

Regardless of how the layers above are implemented, most digital twin architectures share the same functional components:

Data sources. Sensors, IoT devices, and existing enterprise systems that generate the raw data feeding the twin.

Connectivity. The networking and protocols (often edge-to-cloud) that move data from where it's generated to where it's processed.

The virtual model. The digital representation of the physical asset's structure, geometry, and behavior, frequently originating from CAD or BIM data.

The analytics and simulation engine. The software that processes incoming data against the model to produce insight, run simulations, or apply machine learning.

The user interface. The dashboard, visualization, or application through which a person interacts with the twin's output.

The bidirectional data link. The mechanism connecting insight generated by the twin back to the physical asset or to the people who work with it, closing the loop between digital insight and physical action.

The four types of digital twins

One common taxonomy classifies twins by technical scope:

A second, related taxonomy classifies twins by what they represent in a business context: product twins, asset twins, factory twins, and end-to-end twins that span the full chain from product to factory to supply chain.

Both frameworks describe the same underlying concept, but from different angles. One is scoped by technical complexity (component to process), the other by business scope (product to enterprise). In practice, a component or asset twin in the first taxonomy usually maps to a product or asset twin in the second, while a system or process twin maps to a factory or end-to-end twin.

Reference architectures and standards

Several named reference models give structure to how digital twin architectures are designed:

ISO 23247 is an international standard specifically for digital twin frameworks in manufacturing, defining a reference architecture for how manufacturing digital twins should be structured and how their components should interact.

The Digital Twin Consortium publishes reference architecture guidance intended to be vendor-neutral, aimed at helping organizations design interoperable digital twin systems rather than adopting a single vendor's proprietary structure.

Platform-specific reference architectures, such as Microsoft's Azure Digital Twins, offer a concrete, implementable version of the layered architecture described above, built around that platform's specific services. These are useful as a working example of the architecture in practice, though they represent one vendor's implementation rather than a neutral standard.

Together, these give architects and engineers a starting point for design decisions, rather than requiring every organization to define its own reference model independently.

Digital twin vs. BIM vs. simulation

Because these terms often appear in the same conversations, it's worth being precise about what each one is.

BIM (Building Information Modeling) is a structured design dataset, most associated with architecture, engineering, and construction. It's typically static or semi-static, representing how something was designed and built at a point in time.

Simulation models expected behavior, usually as a standalone exercise using assumed or historical inputs, to test a hypothesis rather than to reflect live, real-world conditions continuously.

A digital twin differs from both because of the architecture described above: it's a live model, continuously connected to real-world data through the layers from physical asset to visualization. It often uses CAD or BIM data as its starting geometry and simulation as one of its analytical tools, but what makes it a digital twin specifically is the ongoing, bidirectional connection to the physical world.

Where real-time 3D fits in the architecture

The visualization and application layer is where a digital twin's underlying data and analytics become usable by a person.

Getting to that layer usually starts with the same problem: the digital model layer is often built from CAD or BIM source data, which tends to be detailed, heavy, and structured for engineering workflows rather than for real-time interaction. Making that data usable in the visualization layer requires converting and optimizing it into a real-time-ready 3D model that can run responsively, whether that's a desktop application, a web-based dashboard, or an immersive environment.

Unity's role in a digital twin architecture sits specifically at this layer. Converting CAD and BIM data for real-time 3D is one part of building the visualization layer described above. This isn't the only way to build that layer, and other visualization approaches exist depending on the platform and use case.

For a look at how these architectural layers come together in a specific industry, see digital twins in manufacturing, and for a broader survey of where digital twins are applied, see digital twin use cases across industries.