Engineering resource

Industrial sensor PCBA: calibration and drift

An industrial sensor PCBA needs more than a successful communication test before its measurements can be trusted. A board can report plausible numbers while using the wrong calibration coefficients, measuring through an uncharacterized fixture, or applying a correction intended for another input range. The manufacturing release should define exactly what is calibrated, what is verified afterward, and which evidence travels with each assembled unit.

This guide focuses on transferring an approved measurement design into repeatable production. It does not assign an accuracy rating to an unspecified sensor. Begin with the product’s industrial assembly requirements, then make calibration a controlled operation with named inputs, outputs and acceptance decisions.

Draw the boundary around the calibrated system

Mark the signal path from the physical sensing element through excitation, conditioning, conversion and firmware to the reported engineering value. Identify where the reference stimulus enters that path. Applying an electrical signal at a board connector exercises a different boundary from applying pressure to a completed pressure sensor. Neither result should silently stand in for the other.

Analog Devices distinguishes internal ADC calibration from system calibration in AN-1464. The former addresses errors within the converter; the latter can include external circuitry in the calibrated path. The listed converter families also differ in supported modes and required sequences. Use the exact device documentation when implementing the procedure rather than copying a generic zero-and-span command sequence.

Put this boundary in the purchase specification. If the assembler receives only the board and a sensor simulator, the deliverable is an electrically calibrated board under the agreed conditions. If the actual sensing element arrives later, assign responsibility for integrating and verifying it. A serial number linking the two parts is useful only when the associated calibration remains applicable to that pairing.

Separate adjustment from independent verification

Use different steps for recording the initial response, calculating a correction, storing it, and judging the corrected result. This separation prevents a test program from declaring success simply because it calculated numbers. Retain the initial readings even when the final unit passes; they help engineering recognize a population shift before routine adjustment conceals it.

For a straight-line response, two distinct reference inputs can determine slope and intercept. That mathematical fit alone says nothing about the response between the inputs. Include independent verification points selected by the design owner, and document why they are sufficient for the specified measurement range. A curved response, hysteresis or unstable reading requires investigation or a different model rather than extra decimal places in the same two coefficients.

Keep a distinction between a repeat of the reference input and an independent check of the measurement system. Repeating an identical setting can reveal short-term repeatability, but a shared reference error can affect both adjustment and verification. Ask the test engineer how the reference equipment, connection method and uncertainty support the acceptance limit.

A worked example for reviewing the software

Consider a fictional transmitter whose input span is represented by values from zero to one hundred engineering units. The fixture applies a reference input of ten units and records a raw result of twelve. At a reference input of ninety units, the raw result is ninety-four. These values are illustrative software-review data, not specifications or measured factory results.

A correction written as corrected value equals a times raw value plus b requires a equal to eighty divided by eighty-two, approximately 0.97560976. The intercept b equals ten minus a times twelve, approximately negative 1.70731707. Applying that correction to a raw reading of fifty-three produces fifty engineering units. The two endpoint observations therefore define a consistent linear mapping.

Now suppose the independent reference input of fifty units produces a raw reading of fifty-four. The corrected result is approximately 50.97561, leaving a positive residual of about 0.97561 engineering units. Whether that residual passes depends on the approved acceptance limit and measurement uncertainty. Do not move the limit or reuse that point to refit the model without recording an engineering change.

Turn this example into a software test with expected intermediate results. Check the sign of the intercept, the order of operations, numerical rounding and the behavior when both reference readings are identical. A zero denominator should generate an explicit failure. Also test negative inputs and out-of-range values when those conditions belong to the product specification.

Characterize temperature behavior separately

A room-temperature correction does not establish performance throughout the operating environment. Analog Devices’ AN-1377 illustrates temperature compensation for the AD7403 and AD7405 using local temperature information and characterized drift behavior. Its coefficients and results belong to that investigation. They should not become default correction values for a different sensor board.

For production planning, define which temperatures are part of design characterization and which are part of routine screening. Record the temperature being measured: chamber air, a sensor near the converter, and the sensing element can represent different conditions. Specify the stabilization criterion and where its evidence is recorded. A timer alone is not proof that every relevant part has reached the intended condition.

Include the operating state that creates heat. A radio transmitting continuously, a powered output stage or a changed excitation setting can alter conditions relative to an idle bench test. Have engineering choose representative states and explain the production check that follows from characterization. Do not promise full environmental accuracy based on a single ambient measurement.

Bind coefficients to configuration and identity

The calibration record should identify the board serial number, hardware revision, firmware version, sensor identity when relevant, input channel, range and configuration. It should also identify the procedure version and fixture. Without those fields, a valid-looking coefficient file can be applied to a different measurement path and still pass a basic communication check.

Specify the stored format, units, scaling and integrity check. If firmware stores fixed-point integers, document how those integers map to the engineering coefficients used in the analysis. Read the values back after programming and verify the decoded result. Comparing only the bytes sent by the programmer confirms the intended command, not the contents retained by the board.

Test the startup path as well as the programming path. Power-cycle the unit and verify that the operating firmware loads the correct record before reporting a calibrated measurement. Define the product’s response to a missing record, invalid integrity check or incompatible version. The factory should see a distinct failure rather than an apparently reasonable default output.

Qualify the fixture before judging the board

A calibration station is part of the measurement chain. Document the reference source, cables, switching, contacts, power supply and environmental measurements that affect the result. Give each station a controlled identity. When a fixture is repaired or a cable is replaced, use the agreed verification procedure before returning it to production.

Use a check artifact or reference unit for station checks only within its established purpose. A reference board can reveal a change from its recorded baseline, but it does not automatically provide an absolute reference for every channel and range. Record its handling history and the conditions under which its baseline remains valid.

When two stations disagree, preserve the raw readings and swap one controlled element at a time under an engineering investigation plan. Check whether the discrepancy follows the unit, connection or station. Avoid repeatedly adjusting the same board on different fixtures until it happens to pass; that sequence can erase useful evidence and produce coefficients tied to a faulty station.

Define rework and recalibration decisions

Replacing a converter, reference component, sensing element or part of the analog path should trigger a review of the existing calibration. The appropriate response depends on what changed and how the original procedure bounded the system. Establish the decision in the rework instruction so operators do not have to infer it from component appearance or location.

Preserve the old record alongside the replacement record with the reason for the change. Link both to the same unit history rather than overwriting the earlier result without explanation. If the board receives a new serial number, document the relationship so field returns can still be interpreted correctly.

Firmware changes also deserve review. A revision that changes scaling, channel selection, filtering or coefficient interpretation may require migration or new verification even when the hardware is untouched. Ask the firmware owner to define compatibility explicitly. A familiar file extension does not establish that two versions use the same calibration format.

Release a test package the assembler can execute

Agree the assembly testing scope before ordering the first production batch. Supply the reference-input sequence, allowed configuration, stabilization rule, sample processing, coefficient algorithm, verification points and failure dispositions. Include example records for a passing unit and several intentional failures. These examples let the assembler validate the station logic before valuable product is used for debugging.

Keep acceptance limits separate from display formatting. A rounded value shown to an operator may conceal a result that lies outside the actual limit. State which numerical value is compared, how uncertainty is considered, and which rule applies at the boundary. The responsible engineer should approve that rule before production begins.

Use the release-file checklist to assemble the firmware image, programming instructions, calibration schema and test procedure alongside the normal board files. Identify who can approve a deviation and how affected units are held. A clear release package gives production an executable decision instead of a general request to make the sensor accurate.

Plan for evidence from later service

A factory result describes the unit under recorded conditions at that time. It does not establish a universal recalibration interval. Decide how field checks, environmental exposure and product requirements will inform later maintenance. Preserve enough original data to distinguish a changed sensor response from a different test setup when a returned unit is evaluated.

For repeat orders, review trends in initial readings, correction values and verification residuals by configuration and station. Investigate changes before assuming that wider adjustments are harmless. Keep the raw evidence, the approved correction and the final verification connected to the unit. That relationship makes calibration useful to manufacturing, engineering and service rather than merely another completed checkbox.

Your next build,
clearly scoped.
Get a Quote ↗