A line stops on a Sunday shift and the maintenance screen shows one red word: station failure. The PLC is in RUN, the machine is dead, and nothing was touched since Friday. Such faults are rarely caused by a dead controller — usually one IO device has stopped answering within its watchdog window. The same symptom — BF LED on, inputs frozen, outputs at substitute values — can come from a bad RJ45, a duplicated device name, an aggressive update time, or a GSDML file that no longer matches the firmware.
This guide gives a fixed diagnostic order that works on S7-1200, S7-1500 and third-party devices — from the LEDs through naming and cycle settings to the diagnostic buffer and OB86. If the fault ends in a failed module or CPU, our PLC collection lists tested SIMATIC spares, including the 6ES7517-3UQ10-0AB0 CPU 1517-3 PN/DP.
PROFINET cyclic data runs inside an Application Relation (AR) between controller and device. Output frames go out every update cycle, input frames come back on the same cycle, and a watchdog timer runs in parallel. If no valid frame arrives within the watchdog time, the device drives its outputs to substitute values and the controller logs a station failure. Every diagnosis therefore comes down to one question: why did the frame not arrive in time?
Ninety seconds with a torch saves an hour. LINK must be green on every port in use; a dark LINK means no electrical connection. RX/TX flickering is normal traffic. BF (bus fault) means the AR is not established: steady red is a hard failure, flashing red usually means the device is waiting for configuration or a name. On S7-1500 the older SF indication is replaced by MAINT; yellow means a maintenance demand, not necessarily a lost link. Compare the pattern with the device manual — blink codes differ between families.
Copper segments are limited to 100 m. Check that the cable is industrial Cat5e or better and that the shield is bonded at the connector, not left floating. Then look at the counters instead of the cable: in TIA Portal open Online & diagnostics > PROFINET interface > Port statistics and read CRC errors, discarded frames and link-down events. Any non-zero CRC count on a stable line points to a damaged pair, a crushed cable or a connector contaminated with coolant. Link-down events that increment in step with machine movement point to a flexing cable or a loose gland. Replace the patch lead before you replace anything expensive.
PROFINET addresses devices by name, not by IP: the controller resolves the name at startup and then assigns the IP. Names must be unique in the IO system and follow IEC 61158-6-10: maximum 240 characters, each label up to 63, only lower-case letters a–z, digits, hyphen and period, no underscore, space or umlaut, and no leading or trailing hyphen or dot. A name such as Line2_ET200SP is silently rejected. Duplicate names are worse: two devices answer, the AR collapses, and the fault appears on both. After replacing a device, remember the name lives in the device, not on the memory card — reassign it, or configure topology-based replacement so the controller passes it on by neighbour detection.
This is the most misdiagnosed setting in the field. The watchdog is not entered directly: watchdog time = update time × accepted update cycles, default 3 cycles. The update time is a reduction ratio (1, 2, 4 … 512) of the send clock — with a 1 ms send clock you can set 1 ms to 512 ms. A device set to 1 ms with three accepted cycles fails after 3 ms of silence, which a burst of IT traffic can exceed. If the line drops only under load, raise the update time to 4 or 8 ms, or increase the accepted cycles, and watch whether the interval between failures lengthens.
If the AR never establishes at all, stop looking at the cable. Compare the GSDML version in the project with the device firmware; a device updated to newer firmware but still represented by an old GSDML produces a mismatch that looks exactly like a hardware failure. Check that the slot and submodule layout matches the physical rack — a missing or extra module prevents the AR from starting. Then open the topology view and compare configured partner ports with the real cabling. A swapped port rarely stops data exchange, but it breaks topology-based replacement and hides the real fault behind neighbour mismatch entries.
The diagnostic buffer gives the event class; OB86 shows how the operating system classified it. Load OB86: without it a station failure sends the CPU to STOP, which is why some lines die completely instead of degrading. Its start information carries Event_Class and Fault_ID, and the combinations are unambiguous:
| Event_Class | Fault_ID | Meaning |
|---|---|---|
| B#16#39 | B#16#CA | PROFINET IO system failure |
| B#16#39 / 38 | B#16#CB | IO device failure / return |
| B#16#38 | B#16#CC | Device returned, still faulty or maintenance required |
| B#16#38 | B#16#CD | Device returned, configured and actual configuration differ |
| B#16#38 | B#16#CE | Device returned, module configuration error |
A 38CB event that repeats every few minutes is the signature of an intermittent link or an overloaded cycle, not a dead device. Log the timestamp of each 39CB and 38CB pair; the interval between them is the real outage time, and it is usually far shorter than the operator's report suggests.
If the failure appears only when a nearby motor starts, when a VFD ramps, or when a welding cell fires, you have an EMC problem. Route PROFINET cable at least 200 mm from motor and drive output cables, cross power cables at 90 degrees, and bond shields to the cabinet backplate at both ends. Use managed industrial switches: PROFINET RT frames need priority handling, and an unmanaged switch drops them under load. Disable Energy Efficient Ethernet on ports carrying cyclic traffic, keep cameras and office traffic off the segment, and check that IGMP snooping is not filtering device discovery frames.
On a ring, the MRP reconfiguration time is 200 ms, so the watchdog must be longer than 200 ms or the controller declares a station failure every time the ring recovers — even though redundancy worked. Reach that margin with the accepted cycles: 2 ms × 110 cycles gives 220 ms. Configure exactly one MRP manager per domain, leave the rest as clients, and make sure both ring ports are the ones wired into the ring. A ring closed through a non-MRP switch behaves like a line with a loop, and the resulting broadcast storm looks like total network collapse.
Replace connectors, patch leads and surge-damaged ports first; they are consumables. Replace an interface module when port statistics show CRC errors only on that device while the same cable works elsewhere, or when the device fails to keep its name after a power cycle. Replace the CPU only after you have proved the backplane and the power supply. For ET 200SP stations, the 6ES7155-6AU01-0BN0 interface module is the usual swap, and a spare 6ES7131-6BF01-0BA0 digital input module on the shelf turns a two-shift wait into a twenty-minute repair.
Record the port statistics of every station once a quarter; a slowly rising CRC count predicts a failure months before it stops the line. Keep GSDML files and firmware versions in the same archive as the project, so a replacement device can be commissioned without hunting for downloads. Label every cable at both ends with the device name, not the cabinet number. Check cabinet filters annually — a device running at 60 °C in a dirty cabinet loses its link under load long before it fails outright. Finally, document each device name and IP in the drawings; the next technician will need them at 2 a.m.
Why does the fault appear only when the machine starts moving? Vibration and EMI are the usual answers. A marginal connector opens under acceleration, and an unshielded run near a drive cable corrupts frames during ramps. Watch the link-down counter while the axis moves.
Can I use a normal office switch in the PROFINET line? It may work on a bench, but it has no priority handling for cyclic frames and no diagnostics. In production use a managed industrial switch and keep IT traffic off the segment.
The device name is correct, but the station still fails. Check the IP subnet, the GSDML version and the configured slot layout. Name resolution can succeed while the AR is rejected because the configuration does not match the hardware.
What does a 38CB event mean? The device has returned after a failure. If it repeats, the cause was never fixed — treat the 39CB and 38CB timestamp pair as your real fault log.
