Segmentation, monitoring and patching decisions all get stuck on the same question: which devices are on site, which firmware are they running, who are they talking to? A zone diagram drawn without answering this question stays on paper. On an OT network, starting to collect this information from existing records and from passive observation that doesn’t send packets to the network reduces the risk of affecting devices. Below we explain how we set up passive asset discovery, where we put the sensor and where the method gets stuck; the target readers are security and automation teams starting an OT inventory.
Why active scanning is risky in OT
On the IT side, asset discovery is habitually done with active scanning: ICMP sweeps, port scans, service and operating system fingerprinting. When you unleash the same scanner on the production network, the picture changes. The TCP/IP stacks of older-generation PLCs, communication modules and embedded HMIs were written with limited resources, assuming a certain traffic pattern. Packets with non-standard flags, half-open TCP sessions and dozens of connections opened in a short time can have these consequences on such devices:
- The connection table filling up: Many industrial devices accept only a few TCP sessions at a time. If the scanner consumes these slots, the HMI or SCADA server can’t connect to the device.
- The communication module faulting: Odd packets sent for operating system detection restart some modules or put them in an error state.
- Time-critical traffic being disrupted: On a segment carrying real-time I/O, extra load can produce timeouts in the control loop.
- Impact on the safety system: Loss of communication on a safety PLC (SIS) takes the process to a safe state by design. That means an unplanned shutdown.
Not every device reacts this way; the effect of active queries varies by device, protocol and operating conditions. Because how each device will react isn’t known in advance, the first step is reviewing existing records and passive observation that doesn’t send packets. The passive method covers only the traffic the sensor sees; setting up port mirroring or a physical TAP also requires separate preparation. From an enterprise cybersecurity perspective, the purpose of the inventory work is also to reduce this unknown.
Passive discovery: SPAN, TAP and sensor placement
In passive discovery, a copy of existing traffic is sent to the analysis sensor. There are two basic ways to get the copy:
| Method | Advantage | Limitation |
|---|---|---|
| SPAN (port mirroring) | No extra hardware needed, configured remotely | Drops packets under overload; errored frames aren’t copied; can load the switch CPU |
| Passive TAP | Lossless copy, independent of the network device | Installation requires briefly cutting the link; separate hardware for each link |
| Aggregation TAP / packet broker | Collects many points into a single sensor | Cost and additional management overhead |
On a Cisco IOS-based distribution switch, a local SPAN session is defined like this:
monitor session 1 source interface GigabitEthernet1/0/1 - 12 both
monitor session 1 destination interface GigabitEthernet1/0/48
!
show monitor session 1
The destination port is connected only to the sensor and carries no other traffic. If the total traffic of the source ports exceeds the destination port’s capacity, loss is unavoidable; then you split the sources or move to a TAP. The first place to look in the console is the show monitor session output and the output drop counter on the destination port. If the counter is increasing, the picture the sensor sees is incomplete.
Where should the sensor go?
A common mistake is attaching the sensor only to the firewall at the IT/OT boundary or the core switch. From that point, SCADA, historian and remote access traffic is visible; PLC–I/O communication inside the cell isn’t. PROFINET real-time traffic (Ethertype 0x8892) and EtherNet/IP I/O traffic mostly don’t leave the cell switch and aren’t routed. A practical placement order:
- The transition between level 3 and level 2: supervisory traffic between SCADA, HMIs, engineering workstations and PLCs.
- The uplink ports of cell or line switches: the routed part of intra-cell traffic and broadcast traffic.
- In critical cells, the cell switch itself: Layer 2 traffic between the PLC and field devices.
Before doing the placement, the topology drawing should be up to date. If Purdue levels and cell boundaries are unclear, the existing IT/OT segmentation structure needs to be mapped first. If the drawing is wrong, the sensor is in the wrong place too.
Asset information that can be extracted from industrial protocols
The strength of passive discovery comes from industrial protocols carrying device identity within themselves. But this information is visible only when the relevant message travels on the network. Which information appears under which condition is below:
| Protocol (port) | Information that can be extracted | Condition for visibility |
|---|---|---|
| Modbus TCP (502/tcp) | Master–slave relationships, unit ID, function codes; vendor, product code and revision via FC 43/14 | Read Device Identification is visible only when someone queries it |
| S7comm (102/tcp) | CPU type, order number, firmware, module information | When engineering software or an HMI performs an SZL read |
| EtherNet/IP (44818/tcp, 2222/udp) | Vendor ID, device type, product code, revision, serial number, product name | When a List Identity response travels on the network |
| PROFINET DCP (Layer 2) | Station name, vendor ID, device ID, IP settings | When an engineering tool or controller sends DCP Identify |
| General (ARP, DHCP, LLDP, SNMP) | MAC/OUI, IP, hostname, switch port neighbors | Continuous |
The real column to look at in the table is “condition for visibility.” An S7 CPU’s firmware version enters the traffic only when the engineering workstation connects to that CPU and reads diagnostic information. For a PLC nobody has connected to for months, what the passive sensor knows is the IP, the MAC, the vendor via the OUI, and its communication relationships. Nothing more. S7comm-plus and encrypted sessions used by the newer S7-1200/1500 family narrow the decodable area even further.
You can take a first look at raw traffic with open-source tools. Zeek comes with built-in Modbus and DNP3 parsers; CISA’s ICSNPP project offers additional Zeek parsers for protocols such as S7comm, EtherNet/IP and PROFINET. A quick relationship map comes out of tshark, Wireshark’s command-line tool:
# IP conversation pairs and volumes
tshark -r cell3.pcapng -q -z conv,ip
# Layer 2 endpoints (MAC addresses)
tshark -r cell3.pcapng -q -z endpoints,eth
# Modbus requests: source, destination, function code
tshark -r cell3.pcapng -Y "modbus" -T fields -e ip.src -e ip.dst -e modbus.func_code | sort | uniq -c
Commercial OT visibility platforms do the same job continuously and automatically, but they hit the same protocol limit. Behind the sentence “automatically discovers all assets” there’s always a condition: the relevant message traveling on the network.
Comparing passive, selective active and manual methods
Passive discovery alone doesn’t produce a complete inventory. Two methods fill the gaps. Selective active querying is a slow, targeted query using the device’s own protocol: an FC 43/14 request to a Modbus device, List Identity to an EtherNet/IP device, an accessible node scan with the vendor’s own tool on Siemens devices, and so on. These are messages the device already answers in normal operation; they aren’t random like a port scan. Still, the first attempt is made in a maintenance window, with operations approval and first in a test environment or a non-critical cell.
Manual inventory collects information never visible on the network: devices talking over serial lines, operator panels not connected to the network, I/O modules inside cabinets, physical location and the responsible person. This information doesn’t come without opening the cabinet door.
| Criterion | Passive | Selective active | Manual |
|---|---|---|---|
| Impact on production | None | Low, controlled | None (requires site access) |
| Firmware/serial no. | Conditional | Usually yes | Yes (label, panel) |
| Communication relationships | Yes | No | Partly (from documentation) |
| Devices not connected to the network | No | No | Yes |
| Continuity | Continuous | Periodic | Periodic |
The working order is: a few weeks of passive observation first, then targeted active queries for the gaps, and finally verification with a site walk. If passive observation is kept long enough to cover shift changes, recipe changes and at least one maintenance activity, engineering traffic is captured too. Keep it short and the firmware column stays empty.
Fields to keep in the OT asset inventory
Whether the inventory is useful depends on choosing fields based on downstream processes. Vulnerability matching needs product and firmware information, incident response needs the owner and location, and segmentation needs communication relationships. The minimum field set:
- Identity: Tag name, vendor, model, order number (e.g. Siemens MLFB), serial number.
- Software: Firmware version, HMI or SCADA software version, operating system and patch level.
- Network: IP, MAC, VLAN, the switch and port it’s connected to, PROFINET station name.
- Communication: Protocols it uses, devices it talks to, its role (master/slave, client/server).
- Operational context: Line, cell, process function, criticality, whether it carries a safety function.
- Responsibility: Maintenance owner, machine builder (OEM), support contract, remote access path.
- Source and date: By which method (passive/active/manual) and when the information was verified.
The last item is skipped in most inventories. Did “Firmware V2.8” come from a site walk two years ago, or from an SZL response captured yesterday? In a vulnerability decision, this difference directly changes the outcome. These fields are the input for OT vulnerability and patching decisions.
Keeping the inventory current and verifying it
An inventory left as a one-off project output drifts away from the site within a few months. A new cabinet is added, a module is replaced, an OEM connects a laptop to the cell. If there’s a permanent passive sensor, updates can be tied to automatic events:
- When a previously unseen MAC/IP appears on the network, a record is opened and its owner is sought.
- When a device’s firmware or product information changes (e.g. a different revision in the List Identity response), it’s compared with the change log.
- When a new communication relationship appears (e.g. a PLC talking to an office IP for the first time), the event is passed to security monitoring; this is also the foundation of OT attack detection work.
If an up-to-date device and connection inventory is missing, record the current state of the network with a factory OT infrastructure discovery.
Three criteria are enough for verification at first: comparing the number of active ports in switch MAC address tables with the number of network-connected devices in the inventory, matching records on a few randomly chosen cabinets with a site walk, and tracking the share of records with the firmware field filled in over time. If you repeat these comparisons at every maintenance shutdown, drift shows up early. If you don’t, the inventory quietly goes stale.
The inventory is also a prerequisite for zone and conduit design and risk assessment under IEC 62443. Teams that want to structure the asset list within this framework can proceed with IEC 62443 consulting, and those planning on-site discovery and assessment work with the OT/ICS security service. For the general framework of the method, NIST’s SP 800-82 Rev. 3 OT security guide is the reference source.
Frequently asked questions
Does passive discovery find all OT assets?
No. A passive sensor finds only devices that generate traffic and are in the segment it sees. Selective active queries and a site walk are essential for serial-line devices, panels not connected to the network and devices that haven’t communicated for a long time.
Should SPAN or a TAP be preferred?
For temporary discovery and low-traffic segments, SPAN is usually enough. On critical links that need lossless, continuous visibility, especially if the switch is under load, a passive TAP gives more reliable data.
How long should passive observation last?
There’s no fixed period. Observation should cover shift changes, recipe or product changes and at least one maintenance or engineering activity. If it doesn’t, only routine traffic is seen and information such as firmware stays empty.
Is selective active querying completely risk-free?
The risk is low but not zero. Even though the query uses the device’s own protocol and is done slowly, the first run should be in a maintenance window, with operations approval and tried first on a non-critical cell.