Lab 3 · Motion error and backlash

Goal#

Measure the backlash of demo-2axis-vision (the configuration says 2.4 µm) and answer: why is unidirectional positioning accurate while reciprocating motion is not?

Steps#

1. Devices → Virtual devices → axisX → Error model, confirm backlashUm = 2.4.
2. Unidirectional repeatability: move X from 50 mm to 100 mm, always from the same side, 30 times, measuring the actual position with vision each time. Compute the standard deviation.
3. Bidirectional repeatability: alternate 50 → 100 → 50 → 100, 30 times. Compute again.
4. Compare. Half the difference estimates the backlash.
5. Set backlashUm to 50 and repeat steps 2 and 3. See how far the two curves separate.
6. Enable Compensation → Backlash, enter your measured value, repeat step 3.

Pass criteria#

Item	Criterion
Unidirectional repeatability (1σ)	< 1.5 µm
Bidirectional, uncompensated (b = 2.4)	clearly worse than unidirectional
Backlash estimate	2.4 ± 0.5 µm
Bidirectional after compensation	close to the unidirectional figure

Common mistakes#

• Concluding from a unidirectional test only. Almost every "accurate in the lab, inaccurate on the line" story starts here.
• Confusing repeatability with accuracy. Repeatability says "the same every time"; accuracy says "the same as the truth". A machine can repeat to 1 µm and be 20 µm off.
• Compensating with the wrong sign. The error doubles, which is at least obvious.
• Forgetting to home. Homing repeatability also lands in your measurement.

Review questions#

1. Why does backlash compensation never fully remove bidirectional error? (Hint: backlash is not a constant — it depends on load and position.)
2. Which can be removed by a single constant, pitch error or backlash? Which needs a lookup table?
3. If your application only ever moves one way, is compensation worth it? Work the cost-benefit.
4. Using the error budget: on this machine, how much does the final 3σ figure improve if backlash drops from 2.4 µm to 1 µm? Is that worth a new ballscrew?