IMDTouchKnowledge BaseShop main store

Selection guide

Choosing an External HMI Architecture for a Robot Controller

HMI and touchscreens Published 7 min read

Short answer

Connect the HMI directly over Modbus TCP only when the controller's server option is enabled, the documented register map covers every needed value, and the panel has a matching client driver. Otherwise use a PLC gateway that exchanges fieldbus I/O or TCP messages with the robot and publishes normalized tags. Direct PROFINET fits only with the fieldbus board installed and a display side that consumes cyclic data.

On this page
  1. The three architectures
  2. Direct Modbus TCP: three separate gates
  3. Direct PROFINET: cyclic data without a server
  4. The PLC gateway and what it must add
  5. Worked example: sizing the poll load
  6. Commissioning the layers in order
  7. Choosing the panel once the architecture is fixed
  8. Questions asked

Choose the connection architecture before you choose the panel. If the robot controller has its Modbus TCP server option enabled, the documented register map covers every value you need, and your panel has a matching client driver, connect the HMI directly: that is the shortest data path and the fewest devices to maintain. If any one of those three gates fails, or the project spans several robots and a control-room screen, put a PLC in between. It exchanges fieldbus I/O or TCP messages with the robot and publishes normalized tags that any panel can read. Direct PROFINET is the middle path, viable only when the controller has the fieldbus board installed and the display side can consume cyclic process data.

The three architectures #

Architecture Controller requirement Display-side requirement Where the mapping lives Choose it when
Direct Modbus TCP Modbus TCP server option enabled Panel with a Modbus TCP client driver Panel tag addresses against the robot register map One robot, about 30 tags or fewer, monitoring plus simple commands
Direct PROFINET Fieldbus board installed A panel or PC runtime acting as the fieldbus controller Cyclic I/O mapped from the controller's device description A fixed set of status words and commands with a required update time
PLC gateway Any installed robot option: fieldbus I/O or TCP messaging Any panel with a driver for that PLC Robot-to-PLC exchange plus normalized HMI tags Several robots, large tag counts, scaling, command sequences, control-room screens

The panel brand does not appear in the decision columns on purpose. A given panel family succeeds or fails on whether the controller option, the driver and the map line up, and the same panel family can sit behind all three architectures.

Direct Modbus TCP: three separate gates #

Gate one is the controller. The Modbus TCP server on this class of robot controller is an ordered option, typically delivered as an option card, and the Ethernet connector on the base unit does not by itself make the robot a server. Read the option list at the controller before anything else is purchased. The server listens on TCP port 502, the port registered for the protocol.

Gate two is the map. Demand the documented register map before designing screens. Modbus data units are 16-bit registers, so a 32-bit value occupies two consecutive registers and the word order must match on both ends. Treat 32-bit exchange as unproven until your controller manual documents it end to end: on some robot controllers the fieldbus path carries 16-bit words cleanly while 32-bit support stops at a controller data type the robot logic cannot consume. The safe default is to split values into two 16-bit words, or scale the value into one word.

Gate three is the panel and its engineering tool. The panel needs a Modbus TCP client driver, and the tool version is part of the compatibility: a vendor application note for connecting CP600 operator panels to motion products over Modbus TCP requires Panel Builder 600 version 1.80.00.34 or later. Check the driver list for your exact tool version, and when the driver is missing entirely, work the alternatives in when the HMI software has no driver for your PLC.

Watch the addressing base as well: protocol addresses start at 0 while many tools display 40001-style references, and a one-register offset shifts every value on the screen. Fixing Modbus TCP addressing errors between an HMI and a PLC covers that class of fault in detail.

Direct PROFINET: cyclic data without a server #

The fieldbus board in the robot controller exchanges cyclic process data: the controller publishes a device description, the far side maps the I/O words, and the exchange runs at the configured cycle time with deterministic behavior. This suits a fixed window of status words and command bits — robot state, cycle running, fault present, start and hold — refreshed continuously without polling logic.

Two limits decide this path. Values outside the mapped window still need messaging, and the display side must act as the fieldbus controller, which most operator panels cannot. In practice the path then ends at a PLC anyway, which is the gateway architecture with a fieldbus on the robot side.

The PLC gateway and what it must add #

The gateway PLC talks to the robot over whatever option is installed — cyclic I/O patterns for compact states and handshakes, TCP messages for larger or on-demand transfers — and presents ordinary PLC tags to the panel. The PLC is where you convert widths and signedness, apply scaling and engineering units, and keep raw communication data separate from the tags the screens read. The intermediate controller is often a Modbus TCP client toward other equipment on the same line, and setting up a CODESYS Modbus TCP client to talk through a gateway walks through that side of the job.

Four things the gateway must add, none of which the direct path gives you:

  • A heartbeat word the robot increments every cycle, so the panel can tell a steady process value from dead communications.
  • Stale marking: when the heartbeat stops, affected values display as stale, never as current numbers.
  • Command handshakes: every operator command passes through request, permissive, acknowledgement and timeout logic, never a direct write into an action bit.
  • Multi-robot aggregation: each robot connects over its own option while the panel reads a single server. For a screen far from the line, the aggregation also raises the considerations in connecting an HMI to remote PLCs over a WAN.

One safety boundary holds in every architecture: HMI commands are operational, not safety, so safe-stop and guarding functions stay in the robot's safety circuit and never in panel logic.

Worked example: sizing the poll load #

All numbers here are planning values; measure your controller's actual response before committing a line to them. The basis is a quiet LAN where one Modbus TCP request-and-response round trip, including the controller's turnaround, takes 3 ms.

A screen needs 24 monitored values refreshed every 100 ms:

  • One request per tag: 24 tags × 10 updates/s = 240 requests/s. Occupied time is 240 × 3 ms = 720 ms out of every 1000 ms, a 72% load on the connection. A retry storm, a second panel or fast screen objects push it into communication errors.
  • With the 24 words mapped contiguously, a driver that groups contiguous addresses reads them as 3 blocks of 8 registers (16 bytes of data each): 3 × 10 = 30 requests/s, and 30 × 3 ms = 90 ms per second, a 9% load. Most panels group contiguous addresses; confirm it in the driver settings.

The decision from the arithmetic: when you go direct, design the robot-side map contiguous and verify block reads in the panel. If the controller's register map is fixed and scattered, the gateway wins — the PLC collects the scattered words once per cycle and hands the panel one clean contiguous block. Before attaching a second panel to the same server, read the documented limit on simultaneous connections; that number is controller-specific and cannot be derived from the protocol.

Commissioning the layers in order #

Test the chain one layer at a time:

  1. Change a known robot value and confirm the raw word in the PLC, or the raw tag in the panel monitor on the direct path, changes with it.
  2. Compare raw and normalized tags to prove width, signedness and scaling.
  3. Confirm the normalized tag moves before touching any screen object.
  4. Check the displayed value and units at minimum, normal and maximum operating points.
  5. Run each command with permissives satisfied and blocked; confirm acknowledgement, rejection and timeout each display.
  6. Pull each communication link: values go stale, affected commands inhibit, and the failed layer is identified on screen.
  7. Restore the link: recovery is automatic, no command stays latched, and no old value presents as current.

Choosing the panel once the architecture is fixed #

On the direct path, any panel with a Modbus TCP client driver whose address range covers the robot map will do, and the criteria are the ones in selecting a Modbus TCP HMI panel for an existing controller. On the gateway path, any panel with a driver for that PLC works, which widens the choice to nearly every panel on the market.

When the operator station becomes a PC-based runtime at the machine, industrial panel PCs put the visualization software on a touchscreen computer at the cabinet, and the software decision follows choosing a PC-based HMI runtime for a Modbus TCP PLC.

The decision in one line: one robot, an enabled server option, a complete map and a matching driver — direct Modbus TCP. A fieldbus board already installed and a fixed cyclic word set — direct PROFINET. Anything larger, several robots, or a control-room screen — the PLC gateway, with the panel chosen afterward.

Questions asked

Does the robot's built-in Ethernet port support Modbus TCP without the option?

No. On this class of controller the Modbus TCP server is an ordered option, typically delivered as an option card, and the base Ethernet connector does not make the robot a server. Read the option list at the controller first. If the option is absent, the direct path is closed and the fieldbus or gateway path applies.

How many values can one panel read directly from the robot?

The limit is transaction rate, not tag count. In the worked example, 24 words refreshed every 100 ms as three block reads ran at 9 percent connection occupancy, while one request per tag ran at 72 percent and failed under stress. Keep the direct path to roughly 30 tags, keep the map contiguous, and bench-test the robot's actual response rate first.

Should the HMI write start commands straight into robot registers?

No. A direct write to an action bit gives the operator no acknowledgement, no timeout and no indication of a rejected command. Route every command through request, permissive, acknowledgement and timeout logic in the gateway PLC, or in panel script on the direct path, so a refused or lost command is displayed as such.

How do I tell a steady value from dead communications?

Map one word that the robot increments every controller cycle. The panel alarms when that heartbeat is unchanged for 1 second, which is orders of magnitude longer than any healthy cycle and short enough to flag stale data before an operator trusts the screen. Without it, a frozen number and a stable process look identical.

We run several robots and want one screen in the control room. Which architecture?

The gateway. One PLC aggregates every robot over its own installed option, converts the data into a single normalized tag set, and the screen reads one server instead of many different maps. It also gives one place for alarm states, communication health and future robots, which no direct panel connection offers.

Keep reading

  • How-to

    Connecting One NS-Series HMI to a Redundant CS1D and a CP1H

    Yes. One NS-series panel hosts both controllers: a duplex CS1D pair whose Ethernet units keep a single system IP address through CPU changeover, so the panel never selects a CPU, plus a CP1H reached through its CIF41 as a second host. Both links run FINS over Ethernet simultaneously.

    HMI and touchscreens

  • Troubleshooting

    Why Vijeo Designer cannot connect to an M580 PLC

    The HMI fails to reach an M580 because one leap hides which layer broke. Match the project target to the HMIGTO, prove the computer-to-panel transfer path and the panel-to-PLC runtime path as two separate tests, add the Modbus TCP equipment the controller actually serves, then prove communication with one forced read and one approved write before any faceplate. Each layer gives a pass or a named fault.

    HMI and touchscreens

  • Selection guide

    Choosing a PC-based HMI runtime for a Modbus TCP PLC

    No — C-more projects deploy to C-more panel hardware, so a PC-only interface needs a PC runtime instead. Run AdvancedHMI, a PyQt5 front end with pymodbus, or an Excel VBA socket client as the Modbus TCP client, with the PLC acting as the server on TCP port 502.

    HMI and touchscreens