How Can You Detect a Stuck Sensor?
From the earlier reference collection. Read the current software blog.
Repeated identical values can flag a sensor for review, but repetition alone does not prove a fault.
Look for suspicious repetition
A sensor that reports the same value over and over may be stuck, disconnected from its sensing element, or publishing a cached value. It may also be measuring a genuinely stable condition, or rounding small changes to the same displayed number. Repetition is a useful warning signal, not a diagnosis.
A simple check counts consecutive equal values or measures how long a series remains unchanged. For example, an hourly reading of 12.3 for six hours can be flagged if the site usually varies, while the same run at a stable, controlled site may be less unusual. Any rule should fit the sampling interval and the sensor’s expected behavior.
Check the precision and context
Before raising an alert, inspect the precision of the published data. A sensor may vary slightly while a dashboard rounds every value to one decimal place. In that case, repeated displayed values conceal variation in the underlying measurements. If raw readings are available, compare them too, and document whether the check uses raw or rounded values.
Context can make a flat series plausible. Water conditions may remain steady for a period, and a low-resolution instrument may not show subtle changes. Compare the suspect sensor with nearby observations only after considering differences in depth, timing, location, and measurement method. A neighbor that changes does not automatically prove the first sensor is defective.
Use multiple checks
A practical monitor can combine repetition with other signals: missing updates, impossible values under the system’s stated limits, abrupt jumps, or a mismatch between sample and receipt times. Treat each as evidence to review. For example, unchanged values plus normal update timestamps may suggest a sensor or rounding issue; an unchanged timestamp may instead indicate that the feed stopped updating.
Keep the checks explainable. Record the rule, such as “flag after eight identical raw samples,” alongside the flagged period. Avoid presenting a score as a probability of failure unless it has been validated for that purpose. A status such as “review recommended” communicates uncertainty more honestly than “sensor broken.”
Review before changing the data
When a series is flagged, inspect station metadata, calibration records, maintenance notes, and the source feed. If possible, compare with an independent measurement under similar conditions. A technician may then confirm a fault, identify a display-rounding issue, or conclude that the stable values were plausible. Until there is evidence, preserve the values and mark the concern rather than deleting or replacing the data.
A useful system separates the observed data from quality notes. That lets later readers see what the instrument reported and why a reviewer questioned it. Repeated values deserve attention when they conflict with expected behavior, but only contextual review can establish what happened.