Trace the measurement from the instrument through the input module, controller, HMI, historian, and report. At each boundary, record the value, unit, range, transformation, and timestamp. The same physical reading can look different because one system stores raw current, another stores an engineering value, or a conversion was applied twice. A label alone does not prove which value is present.
Before changing a live configuration, follow the site’s change-control and safety procedures. Compare readings at a stable, known condition and avoid making control decisions from an unverified test value.
Find where the engineering value is created
Create a short signal map. For each layer identify its input and output: a transmitter might output a 4–20 mA signal; an analog module might return counts or engineering units; controller logic might scale the value; and the historian may store the controller’s output or apply another transformation. Actual behavior is equipment and configuration specific.
For a hypothetical linear transmitter configured for 0–100 units over 4–20 mA, the ideal interpolation is:
engineering value = 0 + (current mA − 4) / 16 × 100
At 12 mA the ideal value is 50 units. This formula applies only to the stated linear range and valid configuration. Check under-range, over-range, bad quality, and device failure behavior separately; do not turn a fault current into an ordinary process value.
If the module already supplies 50 engineering units and controller logic treats that number as raw current or raw counts, a second scaling operation will be wrong. Conversely, a historian may hold raw counts while the display labels them as engineering units. Inspect configuration and logic rather than assuming that a familiar tag name identifies the representation.
Separate a conversion from a relabel
A conversion changes the numeric value according to a defined relationship; a relabel changes only the text displayed next to it. Relabeling 100 °C as 100 °F is not a temperature conversion. Temperature conversions also involve an offset: 0 °C corresponds to 32 °F, so multiplying by a factor alone is insufficient.
Unit families carry further context. Gauge and absolute pressure use different references; converting between them requires a reference pressure. Volumetric flow at operating conditions is not interchangeable with a standard-condition flow merely by changing the label. The right correction requires knowing what the measurement represents and what conditions the calculation assumes.
Check square-root extraction once
For an ideal differential-pressure flow relationship at fixed relevant conditions, normalized flow is proportional to the square root of normalized differential pressure. If the DP signal is 50% of its configured span, the corresponding ideal flow is √0.5 ≈ 0.7071, or 70.71% of full-scale flow. Treating the same DP fraction as a linear flow gives 50%: the difference is 20.71 percentage points of full scale, about 29.3% below the ideal flow value.
This example assumes the 4–20 mA output is linear in DP, the ranges align, and the ideal relationship applies. It does not mean every transmitter output needs another square root. Extraction can be performed in a transmitter or control system; applying it twice is an error. Some configurations provide a flow variable that already incorporates the relationship, and low-flow cutoffs or compensation can change behavior. Check the actual instrument manual and configuration before changing logic.
Verify at three points
- Source: Identify the measured quantity, instrument range, output mode, and quality or diagnostic status.
- Controller: Inspect input-module scaling and program logic. Where safe and authorized, compare the displayed value with a known test point or verified reference.
- Historian and report: Trace the stored tag and any query or dashboard transformation, including units and rounding. Confirm all values refer to the same time and operating condition.
An endpoint check alone is insufficient for nonlinear scaling: linear and square-root curves both meet at normalized zero and full scale. Check a documented intermediate point such as 50% DP as well. An out-of-range signal, a changed transmitter range, or mismatched timestamps may explain a discrepancy without a unit-conversion error.
For datasets used by an assistant, attach the physical quantity, engineering unit, source, range, scaling version, timestamp, and quality status. Ask the assistant to flag incompatible units and missing metadata; a model cannot infer whether a value is raw or already scaled from a number alone. Preserve the original data and document any correction so historical values are not silently reinterpreted.


