Common Mistakes

Common Mistakes in Industrial Sensor Manufacturing and How to Catch Them

A troubleshooting guide to the errors that quietly wreck sensor yield, calibration schedules, and accuracy scores, each with a symptom, a root cause, and a numeric fix.

The most expensive sensor mistake is confusing accuracy with resolution. Symptom: a pressure transmitter reads to 0.01 percent of span but fails customer acceptance at 0.25 percent. Root cause: resolution is the smallest displayable step, while accuracy folds in nonlinearity, hysteresis, and repeatability. A 0 to 100 bar unit with 0.1 percent nonlinearity plus 0.05 percent hysteresis plus 0.05 percent repeatability has a total error band near 0.12 bar RSS, not 0.01. Fix: quote total error band as the root-sum-square of every term, then verify against the Instrument Accuracy Score before you print a datasheet.

Unit errors dominate calibration math. Symptom: a technician logs drift as 2 percent but the customer spec is 2 percent of reading, not full scale. On a 0 to 250 degC probe reading 40 degC, 2 percent of reading is 0.8 degC while 2 percent of full scale is 5 degC, a 6x difference that can pass or fail a lot. Root cause: mixing percent of reading, percent of span, and absolute units in one log. Fix: force one basis per column and confirm the drift figure against the Sensor Drift Risk tool before dispositioning.

Yield gets overstated when first-pass and final yield are collapsed into one number. Symptom: reported yield of 96 percent, but scrap dollars imply closer to 88 percent. Root cause: rework loops are counted as passes, so a part that failed calibration twice still lands in the good column. If 1,000 units enter and 120 need a rework pass, true first-pass yield is 88 percent even when final yield reaches 96. Fix: track both, and feed the honest first-pass figure into the Pressure Sensor Yield calculator so capacity plans are not built on hidden rework.

Calibration workload is chronically underestimated because setup and soak time get ignored. Symptom: a plan assumes 8 minutes per unit, but the line delivers 22 minutes. Root cause: only the active measurement points were counted, not the 10 minute thermal soak, fixture change, and reference logging. A 5 point calibration with 90 second dwell per point is 7.5 minutes of dwell alone, before soak. Fix: build the estimate in the Sensor Calibration Workload tool with soak and setup as explicit line items, not a hidden allowance buried in the per-unit rate.

Drift assumptions fail when engineers treat drift as linear over the full life. Symptom: a probe rated for 0.1 percent per year is recalled at month 8 for 0.3 percent shift. Root cause: most sensor drift is front-loaded, with 60 to 70 percent of first-year drift occurring in the first 90 days before the element stabilizes. Fix: schedule the first recalibration at 90 days, not 12 months, and use the Sensor Drift Risk score to flag elements whose early shift exceeds one third of the annual budget.

Skipping or shortening burn-in ships infant-mortality failures straight to the customer. Symptom: field returns spike at 2 to 6 weeks with no manufacturing defect found. Root cause: burn-in was trimmed from 96 hours to 24 to hit a ship date, so latent solder and MEMS-die failures never surfaced. Bathtub-curve data shows most early failures appear within the first 48 to 72 hours at elevated temperature. Fix: size the oven against the Sensor Burn-in Capacity tool so the full 96 hour dwell fits the schedule instead of being cut under pressure.

Test time estimates ignore the difference between sequential and parallel test. Symptom: a 12 station bench is quoted at the single-unit test time and misses throughput by 3x. Root cause: instrumentation test time was multiplied per unit without accounting for parallel fixtures or shared reference standards that serialize certain steps. Fix: separate the parallelizable steps from the serial ones, model both in the Instrumentation Test Time calculator, and confirm the reference standard is not a hidden bottleneck across stations.

Firmware effort is the silent schedule killer on smart sensors. Symptom: hardware is ready but the program slips 6 weeks. Root cause: calibration coefficient handling, digital compensation tables, and protocol stacks like IO-Link or HART were scoped as trivial. A temperature compensated smart sensor can need 15 to 25 coefficients per unit, each requiring test vectors. Fix: estimate the work with the Smart Sensor Firmware Workload tool early, and treat compensation-table validation as a first-class test task, not an afterthought bundled into final calibration.

Published 2026-07-01.