The Sensor That Read 0.05 mm Differently Every Time: Repeatability vs Accuracy
A laser displacement sensor measuring part height on a check station. The same part, measured 10 times, gave 10 readings: 12.00, 12.03, 11.98, 12.05, 12.01, 11.99, 12.04, 12.00, 12.02, 11.97 mm. The spread was 0.08 mm. The spec called for ±0.01 mm. The sensor’s catalog said ±0.005 mm accuracy. Why the spread?
Accuracy vs repeatability
The catalog’s ±0.005 mm is the sensor’s intrinsic accuracy — how close it reads to a calibrated standard. But that number is measured under ideal conditions: flat target, perpendicular incidence, stable temperature, no vibration. On the shop floor, the repeatability (same part, same sensor, same setup) is limited by everything except the sensor itself.
The 0.08 mm spread wasn’t the sensor. It was the measurement environment.
What causes the spread
| Source | Typical spread | Fix |
|---|---|---|
| Part placement variation | 0.02–0.10 mm | Fixture with hard locators |
| Surface angle (not perpendicular) | 0.01–0.05 mm | Sensor aimed at normal to surface |
| Vibration (machine running) | 0.005–0.02 mm | Stiff mount, sensor trigger on stable |
| Surface finish (matte vs glossy) | 0.01–0.03 mm | Diffuse target, correct sensor type |
| Temperature drift | 0.005–0.02 mm/°C | Warm-up, temp compensation |
| Sensor intrinsic noise | 0.001–0.005 mm | Not the issue here |
What we found
The part was placed in a nest with a spring-loaded plunger. The plunger pushed the part against two locator pins. But the plunger force was 5 N — barely enough to hold the part against gravity. If the operator placed the part slightly differently, the plunger pushed it to a slightly different position. The sensor measured the part’s top surface, which moved ±0.05 mm based on how the part seated.
The fix: a pneumatic clamp (30 N) that pushes the part firmly against hard locators. The clamp activates after the part is placed. The part seats the same way every time. The spread dropped to 0.01 mm.
The sensor aiming
The sensor was aimed at the part at a 15° angle. A laser displacement sensor measures distance along its optical axis. If the surface is tilted relative to the sensor, the reading is wrong by a factor of 1/cos(angle). At 15°, that’s 1.035× — a 3.5% error on a 12 mm measurement, which is 0.4 mm. We re-aimed the sensor perpendicular to the surface (0°). The systematic error disappeared.
For curved or angled surfaces, use a confocal chromatic sensor (not a laser triangulation). It measures along the surface normal regardless of angle. But it costs 3× more. For a flat surface, the laser triangulation at 0° is fine.
Stable measurement trigger
The sensor was taking readings while the machine was still settling. The part arrived, the clamp activated, and the PLC read the sensor 100 ms later. The machine’s vibration (from the conveyor motor) was still present. We delayed the reading by 500 ms after the clamp, and took 5 readings averaged over 100 ms. The average filtered out the vibration. The spread dropped further.
The measurement system analysis
We ran a Gage R&R (Gage Repeatability and Reproducibility) study: 10 parts, 3 operators, 3 repetitions each. The variation had three components:
- Part variation (actual differences between parts): 0.50 mm (the parts vary)
- Measurement variation (same part, same operator): 0.08 mm (before fix), 0.01 mm (after)
- Operator variation (different operators placing the part): 0.05 mm (before fix), 0.005 mm (after, with pneumatic clamp)
After the fix, the measurement system was acceptable (less than 10% of the part variation). The sensor was fine all along. The fixturing and trigger were the problem.
The numbers
| Parameter | Before | After |
|---|---|---|
| Repeatability (10 readings) | 0.08 mm spread | 0.01 mm spread |
| Placement | Spring plunger (5 N) | Pneumatic clamp (30 N) |
| Sensor angle | 15° off perpendicular | 0° (normal) |
| Read delay | 100 ms after clamp | 500 ms, average of 5 |
| Operator variation | 0.05 mm | 0.005 mm |
The measurement rule: If the sensor’s catalog accuracy is better than your measured repeatability, the problem isn’t the sensor. It’s the fixturing, the aiming, or the trigger. Don’t blame the sensor for a part that seats differently every time. Fix the fixture, aim perpendicular, and wait for the vibration to settle.
