IMDTouchKnowledge BaseShop main store

Troubleshooting

WinCC Professional intermittent PLC connection loss diagnosis

Industrial networks Published 9 min read

Short answer

Treat the aborted-connections value as a running count of dropped sessions, not fault code 5. Capture the channel log, a packet capture and the CPU's online diagnostics while the link is down, because a live HMI session proves only the HMI's path. The TIA Portal scan is a trigger, not a repair: the durable fix is normally the runtime computer's path, a stale ARP entry or PLC connection resources.

On this page
  1. Why does the HMI stay online while WinCC Professional loses the PLC?
  2. What does 'aborted connections 5' mean in WinCC diagnostics?
  3. Why does a TIA Portal network scan bring the connection back?
  4. Where are the WinCC Professional channel log files?
  5. What do I capture before the connection recovers?
  6. Can fixed 10.16.73.xx addresses still break delivery?
  7. Is the PLC running out of connection resources?
  8. Which fix ends the dropouts for good?
  9. Questions asked

That aborted-connections reading is a running count of the sessions your WinCC Professional channel has dropped, not a fault code, and the network scan in TIA Portal is a trigger that restarts the channel, not a repair. Because the HMI keeps communicating, the PLC is reachable; what fails is one of the things only the runtime computer touches: its network path, its address resolution, its project channel or the PLC's connection resources. Capture the channel log, a packet capture and the CPU's online diagnostics in the failed state first, then run the scan once and record what changes on the wire. The sections below give the order, the numbers and the decision for each layer.

Why does the HMI stay online while WinCC Professional loses the PLC? #

The HMI and the WinCC runtime hold two separate sessions with their own paths, so a live HMI screen proves only that the PLC answers the HMI, and it says nothing about whether the WinCC channel can reconnect on the runtime computer's route.

Two clients maintain two independent session states: the panel holds its own connection with its own supervision, and the WinCC S7 channel holds another. When a dropout hits one and not the other, the shared plant network is largely fine and the suspects narrow to what the failing client alone uses: the runtime computer's adapter and driver, its switch port, its neighbor table, its project channel configuration, and its share of the CPU's connection pool. That the affected PLC changes between incidents reinforces this: a per-PLC cause would repeat on the same CPU every time.

Before trusting the HMI as the healthy reference, confirm its values are refreshing and not frozen; the panel-side version of that check is described in PanelView Plus logic module shows question marks on live values. The Ethernet-layer checks themselves overlap with Diagnosing intermittent HMI to PLC Ethernet comms hangs - same physical path, different client at the end of it. Note also that a correctly set PG/PC interface proves the engineering path only: TIA Portal reaching the CPU for a scan or a download rides the engineering stack, while WinCC runtime rides the route the operating system selects for the SCADA channel.

Decision: treat a live HMI session as proof of the HMI's path alone and start every incident from the runtime computer's own chain.

What does 'aborted connections 5' mean in WinCC diagnostics? #

It is a running count of connections the channel has aborted since the counter started, so 5 means five drop events, not error code 5, and the only reliable use of the number is to trend it against the times your PLCs actually dropped.

Worked example: at 06:00 on Monday the diagnostics read aborted connections = 5, the figure from the failure report. By 14:00 the counter read 9, and the operator log recorded dropouts at 08:12, 09:47, 11:03 and 13:30. The arithmetic: 9 - 5 = 4 aborted sessions in the shift, and 4 aborted sessions over 4 logged dropouts = 1 abort per dropout. A one-to-one step means the counter tracks channel aborts cleanly; a jump of two or more during one incident would mean reconnection attempts are aborting too, which moves the suspicion toward addressing or resources.

Decision: record the counter and a timestamp at every incident. A counter that steps once per outage is your severity gauge; a counter that stays flat during a reported loss tells you the tags were stale on screen rather than disconnected.

Why does a TIA Portal network scan bring the connection back? #

The scan is a diagnostic trigger, not a repair: it sends discovery requests to every address in the range, that burst of traffic refreshes address resolution and shakes loose a stuck session, and whatever it changes on the wire is the evidence you need.

The decisive split is between two outcomes. First, basic delivery was broken and the scan's traffic refreshed the runtime's address resolution, so delivery works again for everyone. Second, delivery never broke and only the channel was stuck; the scan's traffic gave the channel's retry a fresh state to negotiate. You cannot tell them apart from the screen - the packet capture from the checklist below tells you, because the first outcome shows requests leaving for a hardware address that never answered until the scan fired, and the second shows replies arriving all along with only the session setup failing.

Decision: run the scan once per incident, after the failed-state capture, and never on a timer - a scheduled scan hides the failure interval the diagnosis depends on.

Where are the WinCC Professional channel log files? #

Open C:\Program Files (x86)\Siemens\WinCC\diagnose on the runtime PC in a text editor, open the S7 channel log named like SIMATIC_S7-1200_S7-1500_Channel_0X.LOG, and read the entries flagged ERROR inside the failure window first.

The location moves with version and installation type: runtime-only Professional installations document it under Siemens\Automation\SCADA-RT_V1x\WinCC\diagnose, where the channel log looks like CCS7PLUSCHANNEL_01.LOG, and older installations use C:\Program Files\Siemens\WinCC\diagnose. Identify the file your installed channel actually writes rather than assuming any of these names. Read the entries whose timestamps fall inside the dropout window and line them up with your incident log.

Decision: ERROR-flagged lines at the dropout time hand you the channel's own failure reason. No entries at all in that window means the channel never registered an abort, so raise its diagnostic logging for the next incident and treat that loss as a display-side stale-value problem until proven otherwise.

What do I capture before the connection recovers? #

Everything diagnostic depends on the failed state, because once the channel recovers the stale neighbor entries, half-open sessions and occupied resources that caused the dropout are gone, so work down this list before you touch the scan.

  1. Stamp the incident: PLC address, WinCC connection name, last tag update time, tag quality, and the aborted connections reading.
  2. From the runtime computer, ping the PLC across the dropout and note whether requests leave and replies return; a ping with no reply is inconclusive on its own when a filter blocks it.
  3. Start a packet capture on the runtime computer that spans the dropout and the recovery.
  4. During the dropout, read the operating system's neighbor table with arp -a and compare the hardware address shown for the PLC with the one the PLC actually has.
  5. Open the CPU's online diagnostics during the dropout and note occupied versus free connection resources.
  6. Run the TIA Portal network scan once, still capturing, and mark the timestamp.

One incident, one time-correlated record: when the scan fires at step 6, the capture shows which of the three layers moved.

Can fixed 10.16.73.xx addresses still break delivery? #

Yes: a fixed IP plan decides only addressing, and delivery on a local subnet is decided by ARP, so a stale or duplicated hardware-address entry on the runtime computer sends your WinCC traffic to a station that never answers it.

Pin the subnet mask first, because the 10.16.73.xx report does not include one. If every station uses 255.255.255.0, the range is one on-link subnet: 256 addresses minus the network address 10.16.73.0 and the broadcast address 10.16.73.255 leaves 254 hosts, and every delivery is an ARP decision. If even one device carries 255.255.255.128 instead, the range splits into 10.16.73.1-127 and 10.16.73.129-254, 128 addresses each, and traffic across that boundary needs a gateway. A device with the wrong mask then reaches some PLCs directly and others not at all - a different PLC each time, exactly as reported.

Fixed addressing also does not clean the runtime computer's neighbor table. During a dropout, if the entry for the PLC's address shows a hardware address that never answers, and the capture shows your S7 frames going to that silent address until the scan refreshes it, you have found the layer. Find what poisoned the entry - a second station answering to the PLC's address, or a stale entry left over from a hardware swap - and correct that.

Decision: correct the wrong mask or the duplicate on the device that owns it; never paper over it with a permanent static neighbor entry or an extra route.

Is the PLC running out of connection resources? #

Read the CPU's online diagnostics during the dropout: an existing HMI session keeps its resource while the WinCC channel's reconnection fails, so the same CPU can show the HMI healthy and the SCADA client blocked at the same moment.

Every HMI session, SCADA channel and engineering station draws from the same model-specific pool, and a CPU does not normally drop a healthy session on its own - so a pinned pool with the HMI still holding its resource is a consistent picture of this fault. There is no universal resource count to design against; the online diagnostics show occupied versus free on your CPU while the fault exists, which beats any secondhand number.

Decision: if occupied sits at the limit during dropouts, clear the dead sessions - cycle the affected WinCC channel, close unused engineering sessions - and reconcile the configured client count with the CPU's documented capacity before considering bigger hardware. If resources are demonstrably free during a dropout, the resource layer is exonerated in one reading and the channel layer owns the case.

Which fix ends the dropouts for good? #

Fix the one layer your failed-state capture named: the runtime path, the neighbor entry, or the PLC's connection resources - one controlled change at a time, with the project backed up, and the TIA Portal scan left out of production operation entirely.

  • Path or neighbor fault: the capture showed lost replies, a wrong mask, or frames sent to a silent hardware address. Repair that chain - replace the suspect patch lead, move the runtime computer off the failing switch port, fix the mask, remove the station answering to the PLC's address. If a firewall sits on the runtime computer, open only the ports the S7 channel actually uses; narrowing the exception is the whole change and the protection stays enabled.
  • Resource exhaustion: the diagnostics showed the pool pinned during dropouts. Clear dead sessions, cut duplicate clients, then verify on the next natural dropout window that the occupied figure stays below the limit.
  • Channel layer: the path was clean and resources were free in the failed state. Take the ERROR-flagged log lines and your exact WinCC build to the vendor's release notes and official support before changing anything in the project.

Acceptance test: run the same PLC population, the same HMI sessions and engineering access together for 24 hours. The example shift logged 4 dropouts in 8 hours, a mean interval of 8 / 4 = 2 hours, so 24 clean hours span 12 of those intervals. Pass means the aborted connections counter never moved and the capture shows no channel re-attempts; any increment fails and sends you back to the failed-state capture, never to another scan.

Questions asked

Why doesn't WinCC Professional reconnect by itself once the PLC answers again?

The channel retries on its own schedule, but a retry cannot succeed while the cause is still in place: a stale neighbor entry still points at a silent hardware address, or the CPU's connection resources are still occupied by dead sessions. The network scan appears to fix the fault because it changes exactly that state. Remove the underlying cause and the channel recovers on its next retry without any scan.

Should I schedule a periodic network scan to keep WinCC connected?

No. A scheduled scan hides the fault instead of fixing it, and it destroys the failure-interval data that the diagnosis and the acceptance test depend on. It also puts periodic discovery broadcast traffic onto a production subnet. Use the scan once per incident, after the failed-state capture, and treat a system that needs a scan to stay connected as a system with an unfixed layer.

Will a longer connection timeout in the WinCC project stop the aborts?

Only when the capture shows brief, single-packet losses inside an otherwise healthy path. The per-connection timeout sits with the S7 connection in the project, so extend it after you know the real dropout duration from the capture. When the loss lasts seconds, or when frames were addressed to a silent hardware address, no timeout value helps: the channel is faithfully reporting a genuine delivery failure.

How do I catch a duplicate IP address that only shows up during the fault?

Watch the neighbor table entry for the PLC's address across two separate dropouts. A hardware address that changes between reads, or that matches no station on your records, is the duplicate or the stale entry. Fixed addressing in the project does not rule this out, because the test is taken on the wire during the failure; the entry's age state in the same table shows when it was last refreshed.

Keep reading

  • How-to

    Connecting multiple S7-300 PLCs to WinCC stations

    Connect the six S7-300 PLCs and the three WinCC stations over Industrial Ethernet with configured S7 connections; PROFIBUS DP allows only one cyclic master, so it cannot host three independent operator stations. A CP 343-1 Lean gives each PLC 8 S7 connections and the full CP 343-1 up to 16. Keep everything in one STEP 7 project so NetPro configures all connections in one place.

    Industrial networks

  • Troubleshooting

    Gateway setup when a network scan finds no Linux runtime

    A runtime you can start from the Development Environment but cannot find with Scan for Devices is reachable for management but not for discovery. Install Gateway for Linux on the Debian industrial PC, add a gateway entry in the project that points at that PC's IP address, select it, and scan again; the IPC then appears.

    Industrial networks

  • Troubleshooting

    Why One PLC to Two HMIs Fails Over RS-422 Multilink

    A 1:1 serial test proves one cable and one framing set; it cannot prove a shared n:1 multilink. Fix two HMIs on one RS-422 trunk by running the multilink driver entry on both panels, giving each a unique link identity, matching every serial parameter, and terminating only the trunk ends. The documented limit is 16 displays, so two panels are a configuration problem, never a capacity one.

    Industrial networks