1. Why separate the factory network
Many plants were built as isolated networks and were later connected to the office to share production data, orders and machine status. The connection is useful, but if it is not designed, the office network and the factory network become one single space. A computer infected through email can then reach a controller it should never see, and a scan run out of curiosity can stop a line.
Segmentation is about deciding in advance what can talk to what. It does not remove the connections that are needed, but makes them visible, limited and controllable. For this reason the work starts from a shared reference model and a map of the real connections, not from an intuition about where to put a firewall. The model helps precisely with this, rather than choosing at random.
2. The Purdue model in brief
The Purdue model is a reference model that describes the plant as a series of levels. The lowest level contains sensors, actuators and devices that measure or act on the process. Above them sit the PLC controllers, then supervisory systems and operator panels, then production management systems. At the top is the corporate network, with management systems and email. It is a common language for designers, suppliers and staff.
The practical value is that every boundary between levels becomes a control point. Whoever designs the network decides which flows may cross it, in which direction and with which protocol. In most cases traffic should flow upward from the factory to the business, while the reverse should be limited to the strictly necessary commands. Anyone reading the model should always ask which flow crosses the boundary and why.
3. Where to draw the boundaries
The most important boundaries are between the corporate network and the production network, and between the production network and the lower levels. For data flows that must cross a boundary, the most common solution is an intermediate zone, often called a demilitarised zone, where the servers that collect data from the factory and make it available to the business are located. In this way the office network does not open direct connections into the factory.
The table shows the most common boundaries and the control that usually accompanies them. It is not a design: every plant has exceptions, older lines and historical connections that must be discussed with those who manage the network and with those who know the machines. Document every exception, with the reason and the person accountable for it. Verify the table with the network manager before applying it.
| Boundary | What it separates | Typical control |
|---|---|---|
| Business and production | Management and email systems from factory systems | Intermediate zone with transit rules, no direct inbound connection |
| Production and supervision | Collection servers from controllers and panels | Rules allowing only known flows, mainly in one direction |
| Supervision and control | Operator panels from PLCs | Access limited to authorised workstations, no internet browsing |
| Isolated devices | Machines with unnecessary network services | Block everything not needed, with documented exceptions |
4. Prerequisites before touching the network
Do not start with the firewall. First you need to know what exists. A list of devices, their addresses, their external connections and the current data flows is the most important prerequisite. Without this list every rule is a guess, and a wrong rule can interrupt a line that nobody realised depended on the network. A half-finished list produces half-finished rules.
It is also necessary to clarify responsibilities. The network manager, the production manager and the supplier that manages the devices must all be known. A moment must be agreed when production can accept a controlled shutdown, and there must be a verified copy of the programs and configurations to restart from. The network manager approves every new rule, in writing.
- Up-to-date inventory of devices, addresses and external connections
- Map of real data flows, with who uses them and why
- Named owners for network, production and maintenance
- Verified copies of PLC programs and HMI configurations
- Intervention window agreed with production
5. The steps for segmentation
Proceed in small, reversible steps. Segmentation done all at once is hard to check and to reverse. Each step must close with a written verification and proof that the line works as before. If a rule has no documented reason, it should not be added, and if a flow has not been identified it must be treated as suspicious until someone explains its purpose.
- Choose a pilot area with few machines and an available manager.
- Create the intermediate zone and move the data collection servers there.
- Define the transit rules, allowing only the flows already catalogued.
- Block direct connections from the office network into the factory.
- Remove unnecessary services on the devices in the pilot area.
- Verify that the line operates correctly and record the exceptions.
- Extend the method to the following areas, one at a time.
6. Common mistakes to avoid
The first mistake is treating segmentation as a network-only project. Those configuring the switches and firewalls must talk to those who know the machines, otherwise the rules block a flow needed by a function nobody had documented. The second mistake is creating overly broad rules, such as allow everything inside the factory, which makes the boundary almost useless. The third is forgetting emergency connections, such as a maintenance modem or a laptop that moves from one network to another.
7. How to confirm the separation works
Segmentation is verified with evidence, not trust. Check that a device in the production zone does not respond to requests from the office network, that the permitted flows really work, and that the blocked ones are logged. Run the test after every change, not only at the end of the project. Also try to connect from outside, as an unauthorised technician would.
Keep the results in a simple document: date, rule tested, result and the person who ran the test. This document also serves to demonstrate how you manage risk if a customer or an authority asks. Keep it in the same place where you keep the plant documentation. Repeat the verification when you change a device, a supplier or a network rule. If something does not add up, investigate before closing.
8. How to get started
This week, do just one thing: make a list of the devices in the smallest and least critical area, with network address, function, owner and external connections. A well-made list of a single area is worth more than an incomplete diagram of the whole plant. Ask the network manager and the production manager to review it together, without trying to map everything in one go.
Once the list is ready, select that same area as the pilot candidate and arrange a meeting with the supplier that manages it. The questions to bring are which flows are needed, which access exists and which devices cannot be touched without a shutdown. The answers will define the first concrete step of the segmentation. Also note the answers that do not arrive, because they are useful information.
Frequently Asked Questions
Is the Purdue model a mandatory standard?
No. It is a reference model, used by many designers to describe the plant and reason about boundaries. It does not prescribe a specific configuration and does not replace risk assessments. Its advantage is that it gives everyone a common language: production staff, the supplier and the IT team can refer to the same level when discussing the same device.
Is a firewall enough to separate the networks?
A firewall is a tool, not a strategy. It applies the rules, but the rules must derive from a census of real flows and a decision about what should pass. The result depends more on the quality of the rules than on the product installed.
Can I segment without stopping production?
In many cases yes, if the work is planned in steps. Network changes can usually be made without interrupting the machines, but some checks on devices require a brief shutdown agreed with production. Start with the least critical areas, prepare a rollback plan and keep the previous configuration to hand, so you can quickly reverse a change.
Who needs to be involved in the project?
At least three roles: the network or IT manager, the production manager who knows the machines, and the supplier that manages them. A data protection consultant is often needed too, if segmentation touches personal data of workers or suppliers. A single point of contact coordinates the decisions and keeps the register of exceptions.
What happens if a machine cannot be isolated?
It must be treated as a documented exception. You can restrict who accesses it, block unnecessary services, control the flows that reach it and keep verified copies. Record in the register why it cannot be isolated and who is accountable for it. It should be reviewed periodically.