
Most plant networks did not grow by design. They grew by addition. A new packaging line needed a switch, a vendor needed remote access for a drive, a historian needed data from three cells, and each request was solved on its own. Years later the result is a flat network where a laptop in the front office sits in the same broadcast domain as a safety controller. OT network segmentation is the discipline that undoes that drift by grouping assets according to risk and function, then controlling every path between those groups.
This article explains what segmentation means on a plant floor, how zones and conduits work, where the Purdue Model still helps, and how engineering teams can plan a design that operations will accept.
What OT Network Segmentation Means in Practice
Segmentation divides a control network into smaller sections so that traffic between them is deliberate rather than automatic. Inside a section, devices communicate freely because they need to. Between sections, traffic passes through a defined checkpoint that can inspect, allow, or deny it.
The goal is containment, not isolation. Control systems still need data to move upward for reporting and downward for recipes and setpoints. A sound design keeps those flows intact while removing the thousands of paths nobody planned and nobody monitors.
Segmentation also supports availability, which is usually the first concern in industrial network security. A misconfigured device that floods a network with broadcast traffic affects one area rather than the whole facility, and maintenance in one cell no longer creates risk in another.
Why Flat Industrial Networks Create Risk
Lateral Movement
When every device shares one address space, an intruder who compromises a single workstation can scan and reach controllers, drives, and HMIs without crossing another control point. Most publicly documented industrial incidents involved movement from a business system into control networks rather than a direct attack on a controller.
Unclear Ownership
Flat networks blur responsibility. IT teams manage switches carrying traffic they cannot interpret, and controls engineers make changes that IT never sees. When an outage happens, troubleshooting takes longer because no one can describe the expected traffic pattern.
No Place to Observe
Monitoring requires chokepoints. If traffic never crosses a boundary, there is nowhere to install inspection or logging. Segmentation creates the observation points that detection tools depend on.
The Purdue Model as a Reference Frame
The Purdue Model describes manufacturing systems as layers, from field instruments up to enterprise applications. It predates modern networking and few plants match it exactly, but it remains a useful shared vocabulary between IT and controls staff.
Level | Typical Systems | Traffic Characteristics | Segmentation Priority |
Level 0 | Sensors, actuators, field instruments | Deterministic and continuous | Physical and cell level separation |
Level 1 | PLCs, safety controllers, RTUs | Cyclic control traffic, low latency | High, protect from general access |
Level 2 | HMIs, local SCADA servers, alarm systems | Operator sessions and polling | High, separate by process area |
Level 3 | Historians, batch servers, engineering workstations | Reporting and file transfer | Medium, controlled upward flow |
Level 3.5 | Industrial DMZ services, replicated data, jump hosts | Brokered traffic only | Critical boundary layer |
Level 4 and 5 | ERP and corporate business systems | Internet facing and user driven | No direct path to Levels 0 to 2 |
The practical takeaway is straightforward. Business systems should not open a session directly to a controller. Data should be handed off at a buffer layer where both sides can be inspected.
Zones and Conduits Explained
The IEC 62443 series introduced zones and conduits to describe segmentation without depending on specific products or network technologies. Engineers find the terms useful because they map to how plants are actually operated.
What Defines a Zone
A zone is a group of assets that share the same security requirements. Grouping is based on consequence, not convenience. Assets that would cause similar harm if compromised belong together. A water treatment plant might define one zone for chemical dosing, another for filtration, and another for the pumping station, because the impact of a failure differs in each case.
Useful zone boundaries usually follow process areas, safety system separation, vendor supported equipment that cannot be patched, and areas with different uptime requirements. A zone should be small enough to describe in a single diagram and stable enough that it does not change every time a project is commissioned.
What Defines a Conduit
A conduit is the controlled communication path between zones. It carries only traffic that has been documented and approved, and it is defined by its endpoints, the protocols it carries, the direction of the connection, and the controls applied to it.
Documenting conduits is where most teams learn how their plant really works. Engineers often discover engineering laptops polling controllers across three process areas, or a reporting server reaching devices nobody remembered installing. The inventory alone is valuable before any hardware changes.
Where the OT Firewall Belongs
An OT firewall enforces conduit rules. The important difference from a standard corporate firewall is protocol awareness. Industrial traffic uses protocols such as EtherNet/IP, Modbus TCP, PROFINET, and OPC UA, and a device that cannot interpret them can only filter by port and address. Deeper inspection allows rules that permit read access while blocking program downloads, which matters when a vendor requires connectivity but not control authority.
Placement follows the zone map. Boundary devices sit between the business network and the plant. Internal firewalls or layer 3 switches with access control lists separate process areas. Smaller managed devices can protect individual cells or vendor supplied skids.
Failure behaviour deserves attention during design. If a firewall reboots during a production run, the design should define whether traffic stops or continues. That decision belongs to operations as much as to the network team, and it should be tested before commissioning rather than discovered during an event.
Building a Segmentation Plan
Successful projects follow a predictable sequence.
1. Inventory every connected asset, including vendor equipment and wireless devices.
2. Map actual traffic flows using span ports or passive collection rather than assumptions.
3. Group assets into zones based on consequence and operational grouping.
4. Document each required conduit with protocol, direction, and owner.
5. Apply rules in monitor mode first so unexpected flows surface without stopping production.
6. Enforce the rules, then review them during every change window.
The monitoring step is the one most often skipped, and skipping it is the most common cause of a failed cutover. Passive observation across a full production cycle, including shift changes, weekend runs, and monthly reporting jobs, reveals traffic that a single afternoon of capture will miss.
Segmentation also depends on the systems being segmented. Controller standards developed during PLC and DCS programming, the server and client architecture chosen during SCADA development, and the layout of electrical and control panels all influence how cleanly zones can be drawn and where network hardware can physically be installed.
Common Mistakes to Avoid
Designing zones around switch locations rather than process function produces a diagram that looks organised but protects very little. Copying a corporate segmentation standard directly into the plant causes outages, because patch cycles and session timeouts do not suit control traffic. Leaving remote access outside the zone model creates a path that bypasses every boundary that was carefully built. Treating segmentation as a one time project rather than a maintained standard allows the network to drift back toward flat within a few years.
Moving From Diagram to Deployment
OT network segmentation succeeds when the design reflects how the plant runs rather than how the network happened to be wired. Start with an accurate inventory, define zones by consequence, document conduits honestly, and enforce in stages.
Atlas OT works with facility and engineering teams on control system architecture, network design, and electrical, instrumentation, and controls documentation across industrial environments. To review your current architecture and plan a segmentation approach for your site, contact us.
Frequently Asked Questions
Does segmentation require replacing existing network hardware?
Often not. Many managed switches already support VLANs and access control lists. Replacement is usually driven by unmanaged switches, end of life hardware, or the need for protocol aware inspection at key boundaries.
How many zones should a facility have?
There is no fixed number. Many mid sized plants land between five and fifteen zones. If a zone cannot be explained in a single sentence, it is probably too broad.
Will segmentation slow down control traffic?
A properly designed layout adds minimal latency, because most control traffic stays inside its own zone. Latency problems usually come from routing traffic that should never have crossed a boundary.
How does segmentation relate to IEC 62443?
The standard uses zones and conduits as the basis for assigning security levels. Segmentation is the structure on which the remaining requirements are built.
Can segmentation be phased in?
Yes. Most facilities start with the boundary between business and plant networks, then separate process areas during scheduled outages.

