
What Is FL-net? Shared Process Data, Network Roles and Commissioning
FL-net, also identified as OPCN-2, is an open industrial network for exchanging control data over Ethernet. Its controller-level model lets participating PLCs, CNC systems and other compatible controllers maintain a common process-data map. A device publishes its assigned region; the other participants receive that region rather than needing a separate request for every value. [1]
A useful starting question is not “Does this controller have an Ethernet socket?” but “Can this interface join this version of FL-net, and do both applications agree on the data map?” A link light answers neither question. The distinction matters when replacing a module, connecting a machine cell to an existing line, or tracing a value that looks plausible but has stopped updating.
Scope: this guide concentrates on controller-to-controller cyclic exchange. Version 3 device-level functions are treated separately. The opening illustration is conceptual; the depicted equipment is not a verified list of compatible products.
Start with the interface, not the connector
FL-net adds an FA link protocol above UDP/IP. A standard Ethernet interface needs an appropriate FL-net implementation to become a participant. On a PC that may be compatible software or a dedicated interface; on a PLC it must be supported by the actual CPU or communication module. A general communication port is not a promise of protocol support. [5]
Two real examples make the version issue concrete. Mitsubishi’s QJ71FL71-T-F01 is a MELSEC-Q module specified for FL-net Version 2.00 with 10BASE-T/100BASE-TX. Mitsubishi Electric Engineering’s ER-1FL2-T serves the MELSEC iQ-R platform and is specified for Version 3. The shared network name does not make these modules mechanically or programmatically interchangeable. [7] [3]
The ER-1FL2-T documentation expressly excludes connection and communication with Version 1. Do not infer a migration path from the revision number alone. Record the complete part number, supported host platform, firmware, operating mode and the vendor’s interoperability conditions before changing a working network. [3]
The same rule applies to drives: consult the actual option manual rather than treating a manufacturer name as a protocol specification. Drive option listings are useful for locating a part designation, but they are not evidence that every inverter in a series supports that card, firmware or operating mode.
Version 3 changes the scope, not just the label
Version 3 extends FL-net toward device-level I/O and permits coexistence with ordinary TCP/IP traffic. That capability must not be projected back onto a Version 2 installation. Conversely, a restriction in an older dedicated-LAN manual should not be presented as a universal rule for every Version 3 product. Use the operating conditions of the actual mixed installation. [1] [5]
Device-level I/O can introduce ownership and master-role concepts. That does not turn controller-level cyclic exchange into a permanently centralized polling system. A description of “masterless FL-net” needs to say which service and network level it describes. [12]
What replaces a polling master?
In controller-level cyclic operation, transmission rights move between participants using a token. A node sends its allocated data when entitled to do so, then the right passes on. Unlike a simple polling master–slave arrangement, a single permanent controller does not have to request every other node’s process data. FL-net also provides a separate message service for non-cyclic exchanges. [2]
The token order is logical; it is not an instruction to wire the Ethernet ports into a physical ring. Consider three controllers connected to one switch. Their transmission sequence may circulate even though every cable terminates at that switch. Network participation recovery also says nothing, by itself, about whether a machine can continue safely after losing one controller’s data.
Cyclic exchange broadcasts the assigned process-data regions to the participating network. Message transfer handles supported transactions as needed; it should not be mistaken for an unlimited file channel. For example, the QJ71FL71-T-F01 specification limits message data to 1,024 bytes. Check the destination’s supported operation and address interpretation rather than assuming every device implements every transaction. [13] [12]
Read the capacity figures as a shared allocation
The ER-1FL2-T documentation specifies 254 nodes: 249 for control and five reserved for diagnosis. Its common cyclic map contains 8,192 bits in Area 1 and 8,192 16-bit words in Area 2. These are logical data spaces, not two physical memory chips shared electrically by all controllers. [3]
Do not multiply the common map by the node count. Its logical payload capacity is 8,192 / 8 + 8,192 × 2 = 17,408 bytes, or 17 KiB. The regions published by individual nodes are allocated within that map. The arithmetic does not include Ethernet, IP or UDP overhead, and it does not describe a single packet or a guaranteed update interval.
Worked example: give each producer an unambiguous region
Imagine a small cell containing a PLC, a CNC and an industrial robot controller, all with compatible FL-net interfaces. For this hypothetical training example, assign 32 status bits and 16 data words to each participant. The following ranges are zero-based decimal offsets, not addresses copied from a particular machine.
| Producer / IP address | Area 1 bit range | Area 2 word range |
|---|---|---|
| 1 — PLC 192.168.250.1 | 0–31 | 0–15 |
| 2 — CNC 192.168.250.2 | 32–63 | 16–31 |
| 3 — Robot 192.168.250.3 | 64–95 | 32–47 |
When the PLC publishes its region, the CNC and robot consume the received copy. They do not become additional writers of the same region. A command and its acknowledgement should therefore be designed as distinct application fields: the requesting controller owns the command, and the responding controller owns the acknowledgement. Receiving the request is not evidence that the physical action finished.
This allocation uses 96 bits and 48 words: 96 / 8 + 48 × 2 = 108 bytes of logical process data. It leaves the rest of the map unallocated. Adding the fourth machine is an allocation and timing exercise, not simply a matter of finding a spare switch port.
A configuration trap: “Area 1 is a bit area” does not necessarily mean its setup fields use bit units. In Mitsubishi manual SH-080350E-L, printed p.3-24, lengths are expressed in 16-bit words. Our 32-bit regions therefore have size 2 words, with starting word offsets 0, 2 and 4. Area 2 starts at 0, 16 and 32 words, or 0000H, 0010H and 0020H. Keep these shared-map offsets separate from module buffer addresses and local CPU registers. [4]
Before connecting an HMI, finish the data dictionary. Each field needs an owner, offset, length, type, engineering unit, validity rule and restart behavior. That is how “word 1” becomes an agreed measurement rather than an attractive but ambiguous number on a screen.
A correct packet can still carry the wrong interpretation
Extend the example with a pressure measurement. Suppose the sensor represents 0–200 bar as 4–20 mA, and the PLC has been correctly scaled. A 12 mA input represents (12 − 4) / 16 × 200 = 100 bar. Our application chooses to publish that pressure at Area 2 word 1 as an unsigned integer in 0.1 bar units. The transmitted application value is therefore 1,000; the receiver calculates 1,000 × 0.1 = 100.0 bar.
The location and scale are project choices, not a built-in FL-net pressure object. A receiver configured for 1 bar per count would display 1,000 bar despite receiving the intended integer. Check the data contract before replacing hardware. For signed values, floating-point values or numbers spanning two words, agree the representation and byte/word interpretation explicitly.
For this example, reserve word 0 for an application sequence counter. The producer advances it when it publishes a new application sample; the receiver watches it alongside node participation and measurement validity. A data logger should retain the value, time and quality together. The counter is our proposed field, not a universal FL-net watchdog register.
This arrangement distinguishes a steady pressure from an application that has stopped producing samples. It still needs defined handling for counter wrap and controller restart, and it is not proof that the sensor is calibrated. Successful delivery and measurement accuracy are different acceptance criteria. In particular, do not keep labeling the last readable number as a live measurement after its freshness limit expires.
Build a timing budget around the application
An Ethernet line rate and an end-to-end control response are different quantities. When assessing real-time Ethernet, ask when the source task updates its data, when that data reaches the interface, when the network distributes it, and when the receiving task actually uses it. Treat the following numbers as assumptions for a worked budget, not FL-net specification limits.
Illustrative bounds: 8 ms waiting for the source task; 2 ms for source CPU-to-module transfer; 12 ms for the next network update and its delivery; 2 ms for destination module-to-CPU transfer; 10 ms waiting for the receiving task. The sum is 34 ms. It is a bound only for this assumed path when each component meets its stated bound. Sensor filtering, actuator response, network faults and rejoin behavior are excluded.
A PID loop or a servo application may impose different timing and synchronization requirements. Measure the proposed system at its intended load before using a network-cycle claim as a control-performance guarantee. Reducing the minimum frame interval is not a safe shortcut when another participant cannot handle the resulting rate. [5]
Keep node identity separate from the data location
The conventional FL-net IP pattern is 192.168.250.n. Control nodes use 1–249; 250–254 are reserved for maintenance/diagnostic tools. The Version 2 common manual specifies 255.255.255.0 as the subnet mask. Support for a different network-address range depends on the implementation, so check the vendor documentation rather than improvising an office-LAN address plan. [5] [8]
A node number, an IP address, a MAC address and a process-data offset answer different questions. Record them separately. The node and IP identify a participant, the MAC identifies its Ethernet interface, and the shared-map offset identifies a field. Replacing a module may preserve the logical node assignment while changing its hardware address; do not confuse an inventory change with a changed measurement address.
Wire for the supported physical interface
Choose the Ethernet cable and link mode for the actual interface. For the ER-1FL2-T’s copper Ethernet connection, the stated 100 m segment is the link between the hub/switch and the node, subject to installation conditions. It is not a promise of total plant-wide cable reach. Nor does that one module’s 10/100 Mbit/s interface define the capabilities of every FL-net implementation. [3]
Pay attention to the difference between a switch and a repeater hub. Repeater cascade figures must not be reused as universal limits for switched networks. Equally, closing a cable loop is not a way to help token circulation. A redundant physical arrangement needs compatible network equipment and a deliberately engineered loop-prevention or redundancy mechanism. [3]
Commission in stages, with a defined failure response
Document the baseline. Preserve the controller project, module settings and network configuration. Identify every interface by full part number and revision. Check node numbers and IP addresses for duplicates before connecting a replacement. Do not use a working plant as a trial-and-error compatibility bench.
Prove the data path. Start with read-only status and measurements. Confirm the publisher’s region, transfer length, local CPU mapping, data type and scale. Check both the communication status and a changing application counter. Then verify that the consuming program uses the intended field, not merely that a diagnostic screen updates.
Prove loss and recovery. Under authorized, risk-assessed test conditions, verify what happens when one participant disappears and returns. Does the application reject stale data? Can an old command become effective after reconnection? Record the expected response, the observed response and any restart authorization required. A recovered communication link is not, by itself, permission to resume motion.
Separate a link fault from a data-contract fault
Link is up, but the node never joins. Check the protocol generation and operating mode, node/IP configuration, participation status and module diagnostics. A healthy physical link only establishes the first part of the path; it does not establish an FL-net conversation.
The node joins, but values are missing or frozen. Check allocated transmit length, overlapping regions, CPU-to-module refresh and the application’s execution state. Compare the sample counter with the process value. A number that remains constant is not enough evidence to decide whether the network or the process has stopped changing.
Values update but are numerically wrong. Review decimal versus hexadecimal offsets, bit versus word units, signedness, scaling and multiword representation. Inspect the producer’s integer and the receiver’s integer before comparing formatted engineering values. That separates transfer errors from interpretation errors.
Dropouts appear only during production. Correlate module error counters, link events, power-supply conditions and load changes on a common timeline. Keep evidence from both normal and failing operation. Review the permitted traffic mix and frame-interval settings before adjusting them; changing several variables at once makes the result harder to interpret.
Use packet capture as observation, not as a control command
The Version 2 common guide assigns UDP 55000 to token/cyclic frames, 55001 to messages, 55002 to trigger/participation frames and 55003 to transmission. These are transport-layer ports, not TCP services or serial COM names. Confirm the applicable document for the installed implementation. [8]
For an authorized packet recording, the following are two separate Wireshark display-filter examples. They filter an existing capture; they do not generate traffic or issue controller commands. [9]
udp.port in {55000, 55001, 55002, 55003}
ip.addr == 192.168.250.2 && udp
The first matches UDP packets with a source or destination port in the set. The second narrows the view to UDP involving the example node. A port match is not proof that the payload is a valid FL-net frame. Establish that the capture location can actually observe the relevant traffic; an empty view on an unrelated switch port is not proof of network silence.
Why an Ethernet connection does not create interoperability
The Ethernet layer transports the frame; IP and UDP carry the FL-net protocol above it. Another controller using EtherNet/IP, PROFINET or CC-Link cannot be assumed to join the shared map simply because it has a compatible connector. Interoperability requires the protocol implementation and the agreed application mapping, not just a common cable.
An upper-level integration could use a compatible controller or gateway to expose selected data to SCADA. When presenting it through OPC UA, define units, timestamps and quality as well as values. An edge-computing layer may host that service. This is an architecture proposal, not a claim that a particular gateway supports the required combination, and it is not an instruction to bridge the control LAN transparently into the office network.
Network recovery, cybersecurity and machine safety are separate
Possession of a token is not cryptographic authorization. An industrial cybersecurity design still needs network separation, explicitly permitted data flows and controlled maintenance access. NIST’s OT guidance treats segmentation and isolation as engineering decisions that must account for operational requirements, not merely the presence of an IP address. [10]
Likewise, ordinary cyclic data must not be assumed to implement a machine-safety function. Protocol conformance does not establish a SIL claim for a machine. Decide how the physical process responds to lost, stale or inconsistent information within the safety design, rather than assuming that “communication error” always produces the right stop.
Read a standard reference at the right level
JEMA lists JIS B 3521 and JEM 1479 for the protocol and JEM 1480 for testing. Its July 2026 brochure also identifies IEC 61158 Type 26 and IEC 61784 CPF 21. A communication profile family reference identifies the standardized technology; it is not a blanket approval of every product carrying a manufacturer’s logo. [6] [12]
Use the conformance documentation for the actual device and implementation. Certification can support a protocol-interoperability assessment; it does not validate your application data dictionary, cable installation, operating timing or complete machine. Those remain separate acceptance tasks. [11]