Factory Digital Twin: What It Is and How to Start

Digital twins are often pitched with a 3D model of a factory moving on a big screen. As a result, many teams assume a digital twin must be a large, expensive project, while others call an ordinary dashboard a digital twin. Both assumptions blur the investment decision.

A factory digital twin is a digital representation of a machine, line, or process that is kept up to date with data from its real-world counterpart, and used to understand, predict, or decide something about that physical asset. This article covers what is and is not a digital twin, its levels, when the investment makes sense, and how to start from the IoT data you already have.

What a digital twin is, and what it is not

A digital twin has three parts: a real physical asset, a digital model of its structure and behaviour, and an automatic data flow that keeps the two in sync. Academic work on manufacturing distinguishes three levels of integration by the direction of that data flow:

Term	Data flow	Example
Digital model	Manual in both directions	A line simulation in a spreadsheet filled in by hand
Digital shadow	Automatic from asset to model, back through people	A real-time machine dashboard; the operator decides
Digital twin	Automatic in both directions	The model calculates a setpoint and sends it back to the machine

For manufacturing there is also a standard framework, ISO 23247, whose first part was published in 2021 and sets out the general principles and requirements.

Things often called digital twins that are not: a 3D animation with no live data, a one-off simulation from plant design that is never updated, and a boxed product sold off the shelf. A real-time dashboard is not a digital twin either, but it is the foundation.

Levels of a factory digital twin

Level	Question it answers	Example
Status	What is happening now?	A line view showing each machine's status, load, and alarms
Diagnostic	Why did it happen?	Tracing an energy spike to a specific machine, shift, and batch
Predictive	What will happen?	Estimating the days left before pump vibration crosses its limit
Prescriptive	What should we do?	Testing a compressor sequence in the model before applying it, then sending the setpoint

Each level builds on the one below. A predictive model without reliable status data is guessing. In practice the status level is an operational dashboard organised around the asset structure, and a predictive example for rotating machines is covered in our article on predictive maintenance.

Scope grows in steps too: one component, one machine, one line, then the whole plant. Start narrow and low, and move up once the value is proven.

When a factory digital twin is worth it

A digital twin makes sense when at least one of these holds:

• Wrong decisions are expensive, such as downtime on a key line, high-value rejects, or energy that is a large share of production cost.
• The decision repeats, like tuning parameters every batch or scheduling utilities every day, so small gains add up.
• Trying things on the floor is costly or risky, such as changing a recipe, adding a machine to a line, or changing the operating sequence.
• There are many similar assets, so a model built for one pump can serve dozens of others.

Hold off if the basic data is not yet reliable, a simple dashboard and alarms already answer the question, or nobody will look after the model once the project ends.

Starting from the IoT data you already have

1. Pick one question and one asset. For example: why compressor power rises on the night shift, or when a pump bearing needs replacing.
2. Inventory what is already recorded: PLCs, power meters, sensors, logsheets, maintenance records, and production data in the ERP. Much of it often exists already, just scattered.
3. Clean up the asset structure and naming: plant, line, machine, then measuring point, with consistent units and clocks. This structure is the skeleton of the twin.
4. Add context such as product, batch, shift, and maintenance schedule. Without context, data is just a stream of numbers.
5. Build the status and diagnostic levels first, then measure the benefit.
6. Add a model for one predictive question, test it against historical data, and run it alongside manual decisions for a while.
7. Close the two-way loop last, with safe limits and operator approval.

If the data does not exist yet, extra sensors or a gateway for older Modbus equipment are usually enough for the status level, without replacing machines.

Things to keep in mind

• Models age. Machines are serviced, parts replaced, recipes changed. Without recalibration, the model describes a machine that no longer exists.
• Accuracy is limited by data. Uncalibrated sensors or unsynchronised clocks make model output misleading.
• Two-way control needs safeguards. Interlocks and safety logic stay in the PLC, setpoint limits are fixed, and network access is secured.
• Budget for maintaining the model, not only for building it.

Frequently asked questions

How is a digital twin different from a simulation?

A simulation tests scenarios on a model fed with data by hand. A digital twin stays in sync with live data, and simulation can be one of the things it does.

Does a digital twin have to be 3D?

No. Many useful digital twins are just numbers, charts, and calculation models. A 3D view helps with layout or training, but it is not a requirement.

How does a digital twin differ from SCADA?

SCADA supervises and controls the process as it runs. A digital twin adds a behavioural model to explain, predict, and test scenarios, and SCADA is often one of its data sources. See also SCADA vs IoT platforms.

Building a factory digital twin with INCLUDE

INCLUDE does not sell a digital twin as a boxed product. What we provide is the data foundation and a team to build the model on top of it.

On the INCLUDE platform, data from sensors, meters, and PLCs arrives over MQTT or HTTP, including Modbus RTU devices through IncludeGateways, and is stored as time-series history. Devices can be grouped by asset, and dashboards are assembled by drag and drop from 88 widgets. Custom JavaScript widgets and rule chains that run on every incoming reading are available on the Pro plan and above. Actuator control, such as switching a pump or changing a setpoint from the dashboard, provides the path back to the floor. The InAsset solution already runs a small predictive example: the 30-day vibration trend is projected toward the ISO 10816 limit to estimate the days left before a machine fails.

Models, logic, and screens specific to your question are built through our IoT & AI software development service, including integration with your ERP or MES. Historical data can be imported (Pro plan and above) without triggering alarms, so past records can be used to test the model.