The sensor seemed to be working

Direct answer

A signal can respond to process variations and maintain a reasonable appearance while the instrument's physical condition changes slowly. Build-up, drift, loss of calibration or an installation condition can shift the measurement without producing an obviously absurd value.

Data validation must therefore examine time scales longer than one shift and relate the signal to maintenance, inspections, redundant variables and process conditions.

Observability mechanism

Within a short window, a biased signal can retain daily variations and visual correlations with other variables. Over a longer window, the gradual shift may appear as a process trend even though it partly originates in the instrument.

The historian stores values and timestamps. It does not always store the sensor's physical condition, maintenance intervention, regime change or the explanation for a manual action. The numerical history must be complemented by the history of the instruments and the operation.

What to review

• the signal trend over windows of hours, days and weeks;
• changes in offset, drift, saturation or noise;
• calibration, cleaning, maintenance and replacement;
• availability and synchronisation with other signals;
• consistency with balances, redundant variables and field observations;
• changes in regime, ore and operating conditions;
• shift notes and manual actions.

The initial question should not be only whether the signal varies. It should be whether it varies in a way that is consistent with the process, the instrument and its maintenance history.

Operational implications

Using months of historical data to model or compare performance without reviewing measurement quality can introduce relationships that appear physical but mix process response with instrument bias. The analysis output must state which intervals are usable, which require review and which uncertainty remains.

Detecting a bias does not automatically mean discarding the entire history. Treatment depends on the cause, duration, possibility of corroboration and intended use of the data.

Common mistakes

• validating a sensor only because it does not produce out-of-range values;
• observing only short windows;
• cleaning a signal without retaining a record of the change;
• imputing values without declaring the affected period;
• using historical data without recovering the instrument history.

Sources and evidence status

Discovery source LinkedIn post dated 10 Aug 26.

Use of the source The post provides a drift pattern associated with adhered material and the need to expand the observation window.

Publication status PUBLISHED. The account in the post is treated as editorial experience, not as a publishable case with an identified client.

Editorial scope Vocabulary and acceptance criteria must be adjusted to each plant's control system.