PLC Data Logging: Capture the Evidence Between Scans

A diagnostic log should distinguish layers. For a motor that stops intermittently, record the operator or sequence request, the final PLC command, every critical permissive, drive or contactor feedback, fault code and a process variable such as speed or current. If only the motor command is trended, you can see that the PLC turned it off but not why. If only feedback is trended, you cannot tell whether the controller requested the stop.

A performance log may emphasize rate, cycle time, queue depth, energy or control error. A quality log may need recipe, batch, measured conditions and product identifier. A compliance record can require validated timestamps, audit trails and controlled configuration. These purposes have different integrity and retention needs; do not let a convenient HMI trend become an undocumented system of record.

Tag naming and metadata make later analysis possible. Store a stable identifier, human-readable description, engineering unit, source, scale and quality meaning. When equipment is duplicated, include asset context without embedding a temporary screen label. A trend called Value1 is almost useless six months later; Reactor2.TemperaturePV with °C and a known source can be interpreted.

Keep the set small enough to review during the event window. Begin with the causal chain, then add supporting variables when a hypothesis needs them. This staged method also reduces controller, network and storage load. In a persistent historian, collect high-value standard tags continuously and enable higher-resolution temporary traces for difficult commissioning or intermittent faults where the platform supports it.

If a sensor chatters for 100 milliseconds, one sample every minute cannot diagnose it. If tank level changes over hours, millisecond samples waste bandwidth and storage while adding little information. Identify the shortest event, response time or oscillation period relevant to the question, then choose a resolution with enough points to distinguish its shape and order. Use a controller trace or event capture for very fast logic that a supervisory historian cannot economically sample.

Periodic sampling records on a time interval. Task-linked sampling records every task cycle or every nth cycle. Change-based recording stores when the value changes beyond a defined condition or deadband. Triggered capture keeps a pre-event and post-event window around a fault. These methods can be combined. Each creates different gaps and interpolation assumptions, so the trend viewer should not imply continuous measurement where only sparse samples exist.

The sample timestamp can come from the controller, gateway, server or client. Network delay means arrival time is not always measurement time. For event-order analysis across devices, synchronize clocks and record source timestamps plus quality where available. Define timezone and daylight-saving handling for human reports. A plot with unlabeled local time can be ambiguous during clock changes or cross-site review.

Test the capture path under load. Controller tasks, communications, database writes and visualization can compete for resources. Official CODESYS trend documentation, for example, warns that excessive recording or storage work can contribute to task timeouts on constrained targets. Reduce variables, adjust intervals or move the workload rather than risking the control task for a prettier chart.

If command and feedback fall together, the controller may have intentionally stopped the actuator; inspect permissives and sequence state before replacing hardware. If command remains true while feedback falls, investigate the output path, drive, contactor, wiring or load. If both remain true while the process value collapses, the measurement or mechanical process may be the problem. The aligned pens turn one symptom into a smaller diagnostic branch.

For intermittent faults, add an environmental or operating correlate: vibration, current, temperature, line rate or product size. Repeated alignment can establish a useful hypothesis. It is not proof of causation. Inspect and reproduce the suspected connection, component or process condition where safe. Document the physical proving test that confirms or rejects the trend-based hypothesis.

Use event markers for operator actions, alarms and sequence transitions. A graph alone may not reveal that a mode changed or an acknowledgement occurred. Overlaying or correlating event history helps reconstruct the narrative. Time synchronization is essential if the drive, PLC and SCADA generate events on different clocks.

After a repair, capture the same signals over the same operating condition and compare before and after. Confirm the original mismatch is gone without introducing oscillation, longer cycle time or new alarms. A post-fix trend is stronger evidence than a single successful restart because it tests the repaired boundary over time.

Generate known raw and engineered values at zero, midpoint and span. Confirm stored types, scale and units. Use a changing pattern rather than a flat line to expose delayed or swapped tags. If two registers form one value, choose a pattern that differs across both words. Reconcile the source tag, stored value and displayed trend separately.

Create a timed pulse or state transition and record it at the source and historian boundary. Measure effective sample spacing and event order under ordinary and high load. Verify timezone and daylight-saving presentation if humans use local time. When multiple devices contribute, compare their clocks and define the acceptable synchronization error for the analysis.

Interrupt the communication or storage path. The display should show bad quality or a gap according to policy, not an ordinary zero or smooth invented line. Restore the path and verify buffering, duplicate handling and chronological order. Fill a test buffer or exceed a small retention limit in a safe environment to prove oldest/newest policy and alerting.

Finally retrieve the data using the same query, trend or export workflow operators and analysts will use. Access controls, tag aliases, aggregation and report timezone can introduce errors after storage succeeds. Keep a small acceptance dataset with known values so the end-to-end path can be rerun after upgrades or mapping changes.