IMDTouchKnowledge BaseShop main store

Troubleshooting

Diagnosing intermittent HMI to PLC Ethernet comms hangs

HMI and touchscreens Published 8 min read

Short answer

Stop changing the topology. Put one packet capture where it can see the PLC-to-HMI unicast exchange — a mirror port or a tap, not a spare switch port — run it through at least three hangs with device logs on the same clock, set every panel update cycle to 1000 ms, and confirm the three devices hold unique IP addresses. Whether panel requests stop or PLC responses disappear decides which side you repair.

On this page
  1. What the isolation test proves
  2. Record four things at the same hang
  3. Lock down addressing and the single path
  4. Put the capture where it can see unicast
  5. Read the capture at the stop time
  6. Set polling to 1000 ms for the diagnostic run
  7. Change the driver, block layout and firmware last
  8. What each finding means
  9. Questions asked

An intermittent hang that appears only with the HMI connected is a proven trigger, not a proven cause. Four candidates are still standing: the panel's communication stack stops requesting, the PLC stops answering, an address conflict on a three-device network breaks delivery, or the robot program shuts the cycle down on stale panel data. You have already spent the topology moves — shared switch, daisy-chain through the robot controller, and the panel bridging the two — and all three produced the same hang. Stop rearranging and start measuring. One capture decides it: put a packet capture where it can actually see the PLC-to-HMI unicast exchange, run it through the hang with device logs timestamped to the same clock, and set every panel update cycle to 1000 ms while you do it. Then read who stopped talking.

What the isolation test proves #

Running the machine for an extended period with the panel disconnected and seeing no fault proves participation, not cause. Pulling the panel changes three things at once: the traffic on the wire, the number of open connections the PLC holds, and the data the robot program sees. To make the isolation mean something, run it as a matched pair — the same machine loading, one full production shift per configuration, one change at a time. If the hang recurs with the panel present and never without it across matched runs, the panel exchange is the trigger and the capture is the next step, not another cable.

One boundary while you work: the panel data channel is not a safety path. The robot must stop safely on communication loss by design, and the safety functions stay out of these diagnostic changes.

Record four things at the same hang #

The fault is intermittent, so the evidence is whatever was being logged when it happened. Before the next event, set up recording that lands on one clock:

  • The robot stop indication and the exact program condition that requested it
  • The panel's connection state and its communication fault log
  • The PLC's diagnostic log entries
  • The machine state at that second — cycle position and what the robot was executing

Read the timestamps, not the codes. A stop that follows a stale-data or timeout condition in the robot program is the application reacting to missing updates; treat it as an exchange or polling problem. Several connections or devices failing within the same second is a shared network event or resource exhaustion on one device. Sync the panel, PLC and capture clocks to one reference and note the offset — a log you cannot align to the second is commentary, not evidence.

Lock down addressing and the single path #

Three devices, three unique addresses, one subnet. Verify what is actually configured in each device, not what the drawing says. Then read the Address Resolution Protocol table from any device on the network: each IP address must map to one stable hardware address. Repeat the check during a hang. A mapping that changes between readings means two devices claim the same address, and duplicate addressing produces exactly this failure shape — intermittent delivery that pings fine whenever you happen to test the survivor.

The panel's second port in bridge mode forwards whatever appears on it. With nothing attached there is no loop today, but the capture should see one path and one path only, so leave the second port empty or disable bridging for the diagnostic runs. New patch leads routed away from power wiring remove the two most common physical causes, a damaged cable and a loose connector. Where the devices expose link error counters, those counters decide whether a physical fault remains: a port logging rising receive errors during the hang is a connector or cable, and no amount of retransmission analysis replaces that number.

Put the capture where it can see unicast #

A switch learns where each device sits and forwards steady-state unicast only to the destination port. A laptop on a spare port sees broadcasts and its own traffic, never the exchange you came for — which is why plugging into the switch has not settled this yet. Two placements work: a managed switch with port mirroring configured on the panel's port, or an inline tap on one link. Mirror the panel's port and you get both directions of its conversation plus everything it emits on its own.

Before waiting on an intermittent fault, prove the capture works: record a few minutes of healthy running and confirm you can see request and response pairs. Then let it run through at least three hangs, noting each stop time to the second. One event can be a coincidence; three with the same signature is a diagnosis. A capture of a three-device network is small, so record continuously and cut out the interesting minute afterward.

Read the capture at the stop time #

This list decides the repair:

  • Requests from the panel stop, other traffic continues. The panel's driver stopped asking. Pull its communication fault log for the same timestamp — a panel timeout error that clears only by power-cycling the panel, documented as PLC-001 on one vendor's operator interfaces, sits on this branch. The next moves are panel firmware, then the spare panel.
  • Requests continue, PLC responses disappear. The PLC stopped answering. Its diagnostic log and its communication loading are the subject; find what else it was executing in that same second.
  • Both continue, then retransmissions and duplicate requests burst. Frames are being lost between the endpoints. Retransmissions tell you an acknowledgment never arrived, not why — link error counters, connectors and the cable path own this branch.
  • A connection reset lands at the hang. One endpoint terminated the session deliberately. Match the direction of the reset to the side that closed first and read that side's log.
  • Broadcast volume jumps in the same second as the hang. On a three-device network, a flood means a loop or a faulting interface. The source hardware address in the flood frames names the emitter; compare frames per second just before and during the event, and let the coincidence — not any fixed threshold — convict it.

The same request-versus-response reading applies to CIP traffic when connecting a CompactLogix PLC to an HMI over EtherNet/IP: the protocol differs, the question of who went quiet first does not.

Set polling to 1000 ms for the diagnostic run #

Set every active screen and every background read — alarms, recipes, clock synchronization — to a 1000 ms update cycle before the capture starts. Community guidance from one PLC vendor's support portal is direct in the useful direction: refresh cycles of 100-200 ms with a large tag count are not recommended and cause communication problems, and 1 s is where a stable reading starts. A stripped runtime is not automatically a light one; screen complexity is a poor proxy for request volume, because background tasks keep polling on an empty screen. Count the real request rate in the capture, not in the project tree.

Worked example with illustrative numbers: a screen carrying 20 variables at a 1000 ms cycle asks for 20 ÷ 1.0 s = 20 reads per second. The same screen at 100 ms asks for 20 ÷ 0.1 s = 200 reads per second — ten times the transactions arriving at the PLC's communication handler, for values that change once per machine cycle. Decision: if the hangs stop at 1000 ms, the mechanism is request volume against PLC communication resources. Keep bulk data at 1000 ms permanently and poll only the handful of values that genuinely need to be fresh.

Change the driver, block layout and firmware last #

A single shared data block with fixed absolute addresses is a legitimate interface, and its characteristic failure is consistent, not intermittent: a wrong offset or type fails every read, every time. This hang started after commissioning and early production had run clean, which argues against the driver as the first suspect. Verify that every panel access still matches the compiled block layout after each rebuild — an address that no longer lands on the intended member is a real defect, just usually a visible one.

When the capture puts the stop on the panel side, two moves remain, in order: update the panel to the vendor's current firmware release, because communication-stack fixes are exactly what release notes list; then swap in the spare panel, which separates hardware from runtime in a single test. For the baseline settings on the Siemens side of the family, see connecting an HMI panel to an S7-1500 PLC before you change driver parameters.

What each finding means #

Finding at the hang Meaning Action
Panel requests stop; other traffic normal Panel communication stack stalled Panel fault log, firmware update, then spare panel
Requests continue; responses gone PLC stopped answering PLC diagnostic log and communication loading; re-check poll rate
Retransmission burst Frames lost between endpoints Link error counters, connectors, cable path
Connection reset One endpoint closed the session Match reset direction to the closing side
Broadcast flood in the same second Loop or faulting interface Read the flood's source address; check the bridged port
Hang gone at 1000 ms polling Request volume against PLC resources Keep 1000 ms; poll only critical values faster

The capture is the decision-maker because every other test you have run changed multiple variables at once. Three hangs with a clean signature ends the guessing, and the branch you land on is the repair.

Questions asked

The hang clears when I power-cycle the panel. Does that prove the panel is faulty?

No. A restart proves the panel's driver had stopped requesting and that cycling it reopens the connection; it does not separate a hardware fault from a runtime stall. Use the power cycle as the workaround that limits downtime, log each event's timestamp, and let the capture decide. If requests stop while other traffic continues, the firmware update and the spare panel are the tests that isolate it.

The same hang appears on a second machine with a different layout. Same cause?

Probably, and that is useful. Two machines with different designs sharing one symptom point at what they share: the panel model, the driver settings, and the commissioning template — update cycles, address plan, block layout. Diff the communication configuration of both projects before touching hardware. The setting that differs between a machine that halts and one that only stumbles is a strong suspect, and poll rate is the first place to look.

Why does my laptop capture show only broadcasts and its own traffic?

Your port receives frames addressed to it plus broadcasts; the switch does not copy PLC-to-HMI unicast to spare ports. Mirror the panel's port on a managed switch or insert an inline tap. One more trap: a laptop with a second active network adapter on the same subnet can send replies out the wrong interface, so disable the adapter you are not using before you capture.

Should I simplify the network before running the capture?

No. Capture the fault in the arrangement that produces it. The hang has already survived a switch, a daisy-chain and a bridged panel, so simplifying first risks recording a network that no longer exhibits it and proves nothing about the field fault. Get the signature of three events first, then change one variable at a time, starting with the 1000 ms polling test.

Keep reading

  • Troubleshooting

    Why HMI Indicators Stay Off Over Modbus RTU Polling

    Indicators on internal tags animate from HMI memory with no serial traffic; an indicator bound to a PLC tag needs a completed Modbus RTU read of the right address in the right data area. Confirm the PLC bit changes, then read the request and response, then fix the address at one known point.

    HMI and touchscreens

  • Troubleshooting

    Why an HMI Freezes or Updates Slowly Over Serial

    A panel-wide freeze or slow updates over serial is a link fault — driver, PLC port, wiring, or startup timing — not a screen-object fault. Test the two field-proven fixes first: a 10-second startup delay before the first poll, and a port set to four-wire RS-422 at 19,200 baud. Fix an isolated stale tag separately, by address and data type.

    HMI and touchscreens

  • How-to

    Testing HMI Simulation Against a Live PLC

    A standalone HMI simulator proves screens and navigation, not communications. Test against a live controller only through a documented online-simulation mode or a physical panel, and verify every read at two data states and every write at the controller before release.

    HMI and touchscreens