Selection guide
Selecting a Modbus TCP HMI Panel for an Existing Controller
Short answer
A replacement HMI qualifies when its Modbus TCP client driver reproduces the working panel's tag map exactly: data area, raw address, word order and write behavior. A connection that polls proves reachability, not correctness — prove one known tag and one supervised write before rebuilding the screens.
On this page
You are replacing a local HMI that already exchanges data with a controller over Modbus TCP. The replacement qualifies at the tag level, not the connector level: its Modbus TCP client driver must reproduce the working panel's tag map exactly — same data area, same raw address, same word order for 32-bit values, same write behavior. A datasheet line saying the panel supports Modbus TCP promises only a TCP session on port 502; it says nothing about whether 40010 in the old project becomes wire address 9 or wire address 40010 in the new one. The checks below give the numbers, the standard behind each, and the order to run them in.
What a Modbus TCP driver must actually provide #
Modbus TCP carries the same requests as serial Modbus inside an MBAP header over TCP, and the standard server port is 502. One side is the client and initiates every transaction; the other is the server and only answers. In a local-screen arrangement the panel is the client — open the working project and confirm which side initiates before shortlisting anything.
Record the unit identifier the working connection carries. The MBAP header includes it, and a gateway uses it to choose which serial slave a request is for. A direct controller usually accepts one value or ignores the field, but a gateway in the path will misroute on a mismatch.
A panel whose driver list offers only a Modbus TCP server — a device that expects to be polled — cannot replace a client. Decision: shortlist only panels with a Modbus TCP client driver in which the unit identifier, timeout and retry count are all settable.
Record the tag map in raw terms #
The Modbus application protocol specification defines four data areas, each addressed 0 to 65535 on the wire:
| Data area | Access | Read function | Write function |
|---|---|---|---|
| Coils | read/write | 01 | 05 or 15 |
| Discrete inputs | read-only | 02 | none |
| Input registers | read-only | 04 | none |
| Holding registers | read/write | 03 | 06 or 16 |
The familiar 40001-style numbers are one-based notation that configuration tools add on top: zero-based wire addressing runs 0 to 65535 while one-based numbering starts at 1, which is why the same register wears two names across packages. Build a tag sheet that records, for every tag: data area, raw wire address, width in registers, data type, word order for anything wider than 16 bits, scaling with its engineering units, and read/write permission including whether the write is momentary or maintained.
One float, three failure modes #
Take a pump discharge pressure the controller stores as an IEEE 754 single-precision float across two holding registers. The working project calls them 40010 and 40011 — one-based notation. Subtract the base: 40010 − 40001 = 9 and 40011 − 40001 = 10, so the wire request is function code 03, start address 9, quantity 2.
With the pump at 1000.0 psi, the registers return 0x447A and 0x0000, because IEEE 754 encodes 1000.0 as 0x447A0000 and Modbus sends the high byte of each register first. Three ways a replacement gets it wrong:
- Word order. A driver that assembles the pair low word first builds 0x0000447A, a subnormal value: 17530 × 2^-149 ≈ 2.46 × 10^-41 psi. The screen shows 0.0000 while the pump is at 1000 psi.
- Address base. Type 40010 into a zero-based driver and the request goes to wire 40010, 4001 registers past the value. The controller answers with an exception — 0x83, code 02, illegal data address — or returns a different variable that happens to live there.
- Data area. Read input register 30010 instead — same offset 9, wrong area — and you display whatever that word holds, which is nothing the screen should show.
Qualify on one value you can hold still, a setpoint or a frozen counter, and do not proceed until the panel, the controller and the instrument agree. A live connection with a dead number is a real failure mode, not a theoretical one — the same class of fault as why HMI indicators stay off over Modbus RTU polling.
Writes are a separate qualification #
Function code 06 writes exactly one holding register; function code 16 writes 1 to 123 registers in one request, and a 32-bit value has to go out that way. Servers implement the two independently, so a working read at an address proves nothing about writing it — the controller may accept 03 and reject 06, or act on a 16-written value and ignore a 06-written one.
Prove one permitted write per data type, under supervision, and watch the readback change. Do the write tests with the machine in a state where an unexpected output cannot move it, and keep interlocks in the controller program, never in the panel. If the driver will not let you choose the write function and the controller rejects one of them, that panel is out.
What the candidate driver must show #
| Requirement | What to verify | How to prove it |
|---|---|---|
| Client role | Modbus TCP client driver; unit identifier, timeout and retries settable | Configure one read; the panel must initiate and the controller answer |
| Address entry | One-based and zero-based entry both supported, or the convention documented | Read the same known register entered both ways |
| 32-bit interpretation | Word-order setting for floats and long integers | Read the worked-example float and compare with the controller |
| Write functions | Codes 06 and 16 selectable | One supervised write per data type, readback verified |
| Capacity | Tag and screen limits above the working project with margin | Import the full tag list into the configuration software |
| Diagnostics | Per-tag status, timeout and retry counters visible | Pull the link mid-run; the screen must flag communication loss |
Scan rate, grouping and timeouts #
Function code 03 returns up to 125 registers per request, so contiguous tags should travel in one block. Thirty holding-register tags scattered across a 40-register span cost 30 requests per cycle at a 500 ms rate — 60 transactions per second — where one 40-register block costs 2.
Set two poll tiers: 500 ms for the values an operator watches on the main screens, 5 s for configuration and setpoint lists. Set the driver timeout to 1000 ms; a healthy local link answers in single-digit milliseconds, so recurring timeouts point at request load or a server limit, not distance. A Modbus TCP server also accepts a finite number of simultaneous TCP sessions — find the controller's documented session count before planning a second panel on the same server.
Dedicated panel or panel PC #
| Option | Choose it when | You accept |
|---|---|---|
| Dedicated HMI panel | The job is a like-for-like local operator screen | Rebuilding the project in one vendor's configuration software |
| Panel PC running HMI software | You expect to change display hardware, swap HMI packages, or run other applications | Operating-system and software maintenance as permanent support work |
For a straight replacement of a working local screen, take the dedicated panel. Take the panel PC only when software portability or the extra applications earn the operating-system upkeep. The platform changes none of the driver checks above.
Qualification sequence #
- Back up the working panel project and the controller configuration with their own engineering tools — the same discipline as how to back up a PLC program the right way — and record IP addresses, unit identifier, poll intervals and timeouts.
- Build the tag sheet from the working project: data area, raw wire address, width, data type, word order, scaling with units, and access with write action.
- Configure the candidate and read one stable, known value. Fix addressing and interpretation before anything else moves.
- Prove one write per data type under supervision, watching the readback and the controller's response.
- Import the full tag list, group contiguous reads, and apply the poll tiers and timeout from the scan-rate section.
- Run the finished application under normal load and watch the timeout counters for a full shift.
- Close out the physical side — cutout, supply, environmental rating, screen size — using planning an HMI terminal replacement from an older operator panel when the mounting changes with the panel.
Questions asked
The replacement panel connects and polls, but every 32-bit value is absurd. What do I check first?
Word order. If a value stored as 1000.0 psi displays as roughly 2.46 × 10^-41, the driver is assembling the two registers low word first. Find the 32-bit or float word-order setting in the driver's tag configuration, flip it, and re-read the same registers before touching addresses or scaling.
Is 40010 the same register in every HMI package?
No. In one-based notation 40010 is the tenth holding register, wire address 9. A driver set to zero-based entry treats 40010 as the literal wire address 40010 — 4001 registers away. Record the raw wire address for every tag during migration, not just the notation the old project displayed.
How many panels can poll one Modbus TCP server?
As many as the server accepts. Each panel opens its own TCP session, and controllers publish a simultaneous-session limit — some accept one, some several. Find that number in the controller documentation before planning a second panel; if the limit is one session, a single client must read and republish for the others.
Why does a write fail when reading the same address works?
Reads and writes are separate function codes, and a server can implement one without the other. Reading uses function code 03; a single-register write uses 06 and a multi-register write uses 16, and some controllers accept only one of the pair. Select the write function the controller implements, then confirm the controller program actually acts on the register.
Does the unit identifier matter on a direct Ethernet connection?
Often not, until a gateway appears. The MBAP header carries a unit identifier that a gateway uses to route the request to a serial slave behind it. A direct controller usually accepts one value or ignores the field. Copy the setting from the working project; if a gateway sits anywhere in the path, the value must match the slave it routes to.