When a Thai factory connects a legacy instrument to a newer PLC or monitoring system, choosing an industrial protocol gateway from a list of supported protocols can leave important questions unresolved. A Modbus RTU flow meter may communicate with an EtherNet/IP PLC while the 32-bit word order, stale-value behavior or write permissions are still wrong. This guide helps maintenance leaders, controls engineers and integrators specify the conversion, compare proposals and accept it through FAT and SAT.
The selection decision in one sentence
Define which device reads or writes which data, what that data means and what should happen when communication fails before choosing a model. An RFP needs the protocol roles, physical interfaces, data map, type and byte order, update time, timeout, diagnostics, backup and local responsibilities. A product statement that it can convert protocol A to B is a useful shortlisting criterion. It is not proof that the proposed configuration will work with your equipment.
This article concerns protocol conversion between unlike industrial communications. The industrial network construction guide covers plant topology and OT/IT segmentation. The industrial IoT gateway selection guide covers transport to higher systems, buffering and replay. A protocol converter should not be assumed to replace a historian, switch, firewall or data-retention service.
Draw the two endpoints and every handoff
Even the label “Modbus compatible” leaves open whether the link is RTU or TCP, which node initiates a transaction and whether reads and writes are both supported. Draw the exact equipment models and firmware, serial settings or IP details, physical ports, protocol role on each side and final consuming application. When the instrument manual is missing, do not fill the mapping table by guesswork. Compare values read from the device with the local display and a known process condition.
A path such as “serial meter → converter → PLC → SCADA” contains several separate claims. Reading the meter does not demonstrate that the PLC receives the intended input image. Displaying a number in SCADA does not demonstrate that it is the instantaneous rather than accumulated flow or that its quality state is handled. Give each handoff its own row in a data-flow table. This makes scaling, rounding, holding and latency visible.
Verify the role for the exact model
An EtherNet/IP adapter and scanner play different roles. If the PLC and gateway are both configured as scanners, both initiate connections and neither provides the required adapter endpoint. PROFINET likewise distinguishes controller and device. A Modbus client initiates the request. Check the proposed model and firmware for the required role, connection count, data size and update method in the manufacturer documentation and then on the bench. Moxa’s MGate 5105-MB-EIP documentation is an example of product-specific role and diagnostic information; a family-level logo is less useful for procurement.
State what must not cross the gateway
Read-only monitoring and bidirectional setpoint control have different failure consequences. If writes are unnecessary, disable them in the specification and test that unauthorized writes fail. If a conversion device is proposed inside a high-speed control loop or safety-related path, ask the controls and safety owners for a separate timing and failure-mode assessment. A successful monitoring application does not establish suitability for closed-loop control.
Make the data map the center of the RFP
Conversion moves meaning, not merely packets. Source registers or bits must be mapped to receiver memory without losing sign, unit, scale or status. The Modbus Organization defines application functions and the protocol data model, but each instrument assigns its own process meaning to addresses. “Supports Modbus” therefore does not mean that every meter can use the same configuration.
For every point, record the source device and address, function, source type, byte and word order, scaling, engineering unit, valid range, invalid value, scan interval, receiver address and type, write permission, owner and test method. A 32-bit integer or floating-point value occupying two 16-bit registers requires checking word as well as byte order. HMS publishes a separate support explanation for endianness on an Anybus Modbus Client Gateway; that is precisely the level of detail the RFP should request.
| Item | RFP response | FAT observation |
|---|---|---|
| Meaning | Tag, unit, scale, sign and limits | Compare known source and PLC values |
| Encoding | Type, word order, invalid value | Inject boundaries and negatives |
| Time | Source interval, poll rate, timeout | Log a step change and disconnection |
| Quality | Stale, invalid and device fault representation | Check value and status after cable removal |
| Writes | Permitted tags, access and range | Test both allowed and refused commands |
Plausible numbers can hide a bad conversion
An integer scaled in tenths may produce a changing display even when interpreted incorrectly. A wrong register area may also yield a credible-looking value over part of the operating range. A test must cover zero, near-limits, negative values where applicable, invalid codes and fast transitions, rather than a single typical reading. If these conditions cannot be generated safely on the real machine, use a simulator and identify which evidence came from simulation.
Treat writes as a separate acceptance stream
Read accuracy does not prove write safety. A setpoint path needs permitted tags, ranges, operator authority, local/remote selection, equipment interlocks and an explicit decision on retries after interruption. If replaying an old command would be hazardous, the converter must not replay it or the controller must reject expired command generations. Distinguish “command received” from “action executed.” For read-only projects, verify the disabled state and access controls at acceptance.
Specify freshness, not a marketing speed
The useful update interval is the sum of the instrument’s own refresh, shared serial-bus traffic, polling, conversion, PLC scan and upstream acquisition. Polling a meter every 100 ms cannot create new information if the meter updates once a second. Polling many serial nodes too aggressively can instead increase timeouts or disturb existing traffic. The correct target comes from the process use case and the installed equipment, not an assumed universal rate.
Put two distinct requirements in the RFP: the maximum acceptable age of the value at the receiver and the maximum time to declare a communication failure. Measure both during normal operation, simultaneous point changes and a nonresponsive node. Record input change, gateway acquisition and receiver update against a common time base. Compare the largest observed delay and sequences of late values as well as averages.
Decide what a timeout looks like to an operator
Holding the last value keeps a display calm but can make stale data look current. Forcing zero can confuse a broken link with a real zero-flow event. For each tag, choose “last value plus stale quality,” “invalid value plus fault state” or a defined fail-safe behavior. The PLC logic, HMI and alarm procedure must interpret it in the same way. Agree with operations whether restoration clears the alarm automatically or requires acknowledgement.
Check installation and maintainability at the cabinet
Confirm space, power, grounding, temperature, condensation, vibration, cable routing, port orientation and access to terminals in the actual panel. Environmental limits depend on the exact model and mounting arrangement; never copy one model’s range to all candidates. Before modifying a working line, agree on the allowed shutdown window, switchover sequence, backup and rollback. A spare unit is only useful if its firmware, configuration, license, IP parameters and certificates can be restored consistently.
Include the physical layer in scope
RS-485 polarity, termination, shield and grounding, line length, baud rate, parity and slave IDs can block the project before software mapping starts. The Ethernet side may require a switch port, VLAN, IP allocation, time source and firewall change. Attach cabinet photos, drawings, terminal lists and network diagrams to the RFP. Ask bidders to state missing inputs and assumptions. A cheap converter cannot meet a commissioning date if no one owns the cabling or network change.
Compare lifecycle, not only availability today
Request product lifecycle status, replacement model, firmware and vulnerability update policy, configuration-tool OS support, export format, local spare route and support hours. One small converter may carry data for several machines. Repair time is shaped by the local team’s ability to replace and validate it, not merely the existence of a redundancy option on a datasheet. Specify support languages and night-shift response if Thai operators must handle initial incidents.
Security: conversion is not a network boundary
A gateway passes traffic. Even with two network interfaces, it is not by itself a complete OT/IT security boundary. NIST SP 800-82 Rev.3 discusses segmentation and isolation based on mapped, necessary OT data flows while preserving performance, reliability and safety. Place the converter in a reviewed topology with firewall rules, a DMZ where appropriate, restricted management access and logging. Do not mark a plant “segmented” merely because the converter plugs into two LANs.
Specify default-password replacement, unused-service shutdown, allowed management hosts, configuration change logs, backup, firmware update and rollback, and account removal when contractors leave. If the upper interface uses OPC UA, the OPC Foundation’s specifications describe application and user authentication and message protection. “OPC UA capable” still does not identify the deployed certificates, endpoints, profiles or security mode. Test these with the receiving product and rehearse certificate renewal.
Buy usable diagnostics
An incident needs more information than a red LED. Check whether engineers can see read timeouts, CRC or protocol exceptions, peer loss, out-of-range maps, PLC connection state, last good update and configuration version. Moxa documents protocol diagnostics and traffic monitoring for specific products; verify which events your chosen model records, retention and export. Logs trapped inside a supplier screen are weak evidence for a night-shift fault or future supplier change.
Build a comparable RFP response sheet
For each requirement, ask bidders to mark compliant, conditional or noncompliant; identify model and firmware; cite a manual page; identify license or configuration work; propose a FAT demonstration; and list limitations. A model name alone does not expose differences in tools, installation, spares and support.
| ID | Procurement condition | Evidence |
|---|---|---|
| IF-01 | Ports, protocol versions and both roles | Model-specific datasheet and diagram |
| MAP-01 | All tag types, scales, units, destinations and writes | Editable map and test capture |
| TIME-01 | Freshness, timeout and stale-value handling | Load-aware test plan |
| FAIL-01 | Cable loss, power loss and restart behavior | State transition and reproduction steps |
| SEC-01 | Authentication, management and allowed paths | Configuration and operating procedure |
| OPS-01 | Backup, replacement, diagnostics and support | Restore test and support matrix |
Acceptance means the measured result meets a stated limit and evidence can be exported. If the RFP says “under one second,” define the source refresh, concurrent point count and timestamp boundaries. Otherwise bidders may measure different things. Choose limits from the plant’s control and monitoring needs.
FAT: break meaning, timing and connections on purpose
Prepare a real source or simulator, converter, receiving PLC or test client, and synchronized logging. Freeze firmware, map revision, configuration file, test values and pass criteria. Capture both input and output for each scenario. One successful normal reading cannot pass all points or failure modes.
- Connection: establish the specified roles and change startup order.
- Types: check zero, negative values, near-maximum values, two-register word order and bit boundaries.
- Conversion: follow known scale, unit, rounding and invalid codes from source to receiver.
- Timing: measure freshness when all points change and one node stops responding.
- Loss: disconnect serial and Ethernet sides separately; log values, quality and alarms.
- Recovery: reconnect, restart and cycle power; inspect stale values and pending writes.
- Bad input: use missing addresses, wrong function codes and unauthorized writes; inspect rejection and diagnostics.
- Operation: restore a spare from backup and check map and firmware versions.
Link each result to action time, source value, destination value, status, log, configuration version and tester. When a test fails, investigate the meter map, wiring, poll design and PLC setting before attributing it to a product defect. Any manual workaround or external script needed for acceptance becomes part of the supported configuration and handover scope.
SAT closes the difference between bench and production
SAT uses the actual cable route, electrical environment, running serial bus, PLC scan, network restrictions and maintenance procedure. Agree on the shutdown window and rollback trigger before cutover. Use change control so that commissioning does not introduce an unplanned control modification. Potentially disruptive fault injection may have to occur only during an approved stop or on a safe test arrangement.
Check three outcomes: real plant values match instrument displays; alarms and stale-value behavior support the operators’ response; and the local team can replace and restore the unit. If a night shift owns first response, hand over a Thai- or English-language diagnosis sheet stating who to call and which logs to collect. A solution recoverable only while the supplier is onsite has not been handed over operationally.
Preserve acceptance evidence as a maintenance asset
Deliver final topology, port and IP lists, permitted flows, point map, configuration export, firmware revision, spare ID, log procedure, FAT/SAT evidence, changes and support contacts. Store credentials separately in an approved location. State which tags must be retested after a PLC update and which configuration changes trigger reacceptance. This prevents a future line from running on an undocumented setting nobody can safely reproduce.
Compare total scope rather than a device price
Compare hardware, tools, license, panel work, wiring, network changes, FAT/SAT, production stops, spare, updates and training. There is no defensible universal price in this article: the installed equipment, missing manuals, point count and use for control or monitoring change the work. Request separate lines for equipment, mapping, commissioning and support, with exclusions explicit.
A few read-only tags and bidirectional control data require different designs and tests. When several lines share a map, versioned templates can reduce repeated work. When every old instrument differs, test a representative unit before buying a fleet of gateways. The best candidate is the one that performs the required conversion and can be maintained at your site, not automatically the one with the longest protocol list.
Common selection failures
Score proposals before opening the prices
Define disqualifying requirements first: required protocol roles, protection against unauthorized writes and a restore method that the local team can execute. Then compare configuration effort, accessible logs, local spares, support and lifetime cost with the same scoring weights for every bidder. Demonstrations should use the same source device or test data and the same injected faults. Mark undocumented claims “unverified” and close them with a test or contract condition instead of assuming compliance.
Before procurement in a Thai plant, confirm who approves a shutdown, owns cabinet drawings and can change instrument settings. If Japan-based engineers design the map while Thai maintenance staff restore the equipment, both teams need the same controlled revision and troubleshooting sheet. A sample unit does not represent the fleet until model, firmware, register map and communication settings are checked against the asset list.
For a flow-meter example, record the meter and PLC display together during stable flow, rising flow and true zero. Disconnect the link while flow is zero and verify that zero and lost communication produce different quality states. Restore power and record when the value became fresh and when the alarm cleared. For a temperature-controller write path, test an authorized change, an out-of-range value, local mode and a command interrupted by link loss. A “sent” message is not evidence that the controller accepted and executed the command. Tie both ends’ screens, logs and configuration revision to the same test ID; synchronize the clocks before interpreting delay.
Buying from protocol logos: roles or data size do not match. Require model-specific proof.
Accepting any moving number: a scale or word-order error remains hidden. Test known values and boundaries.
Treating stale as normal: held values can mislead operations. Send age or quality with the value.
Calling the converter a firewall: allowed traffic and management routes are not defined. Design the security boundary separately.
Buying a spare without a restore test: the team cannot bring the spare online. Prove restoration in FAT and SAT.
Calling simulator FAT an installed-system test: site wiring, noise and PLC behavior remain untested. Close those gaps in SAT.
Conclusion
Industrial protocol gateway selection is a design decision about preserving data meaning and safe failure behavior across unlike communications. Put the topology, roles, map, freshness, quality, write control, diagnostics, security placement and restoration in the RFP. Reproduce them in FAT, then close site differences in SAT. A common input set and fault test make competing proposals genuinely comparable.
If your Thai plant is considering how to connect older instruments with a new PLC or monitoring system, you can contact TOMAS TECH while the connection diagram and data map are still being drafted. The proposed shutdown window and intended monitoring or control use are enough to begin defining an RFP and acceptance criteria.
FAQ: industrial protocol gateway selection
What is an industrial protocol gateway?
It transfers data between different industrial communication methods. Supported protocols, roles, data size and diagnostics differ by model. Check suitability against the actual endpoints and operating purpose.
Does Modbus and EtherNet/IP support mean it will work with every PLC?
No. RTU versus TCP, client versus server, scanner versus adapter, PLC configuration, data length and the point map must align. Specify the model and firmware and bench-test the pairing.
How should protocol gateway cost be estimated?
Separate hardware, configuration, mapping, panel and wiring changes, network approval, FAT/SAT, downtime, spares and support. Include discovery work when legacy documentation is incomplete. A device unit price is not the total project cost.
What is the difference between FAT and SAT?
FAT uses known inputs and injected faults to prove types, meaning, timing, disconnection, recovery and restoration. SAT proves the same behavior with actual equipment, wiring, network and local operating team.
Does a gateway safely separate OT and IT?
No. Approved flows, management routes, firewalls, DMZ placement, monitoring and change control belong in the network architecture. Accept conversion and boundary protection as separate requirements.
Is setting a disconnected value to zero safe?
That depends on the process. Zero may be a valid measurement. Holding the last value also requires an explicit stale quality. Controls and operations owners should choose and test the fail-safe response for each point.
