Robot repeatability loss: how to measure it and causes

The robot has nailed the same point for three years and this week it starts "landing wrong": one part comes out fine, the next with 0.3 mm of deviation, the one after fine again. No alarm. That's repeatability loss, and it's one of the symptoms where the most time gets wasted, because almost nobody measures it before they start swapping parts.

Repeatability is not the same as accuracy

Confusing the two is the most expensive starting mistake. They are two different properties:

• Repeatability: if you send the robot to the SAME programmed point 20 times, does it always reach the same physical spot? The scatter among those 20 arrivals is the repeatability. A mid-size industrial robot is typically ±0.03 to ±0.1 mm from factory.
• Accuracy: if you ask it to go to a coordinate X,Y,Z, does it reach that actual coordinate or an offset one? A robot can be very repeatable (always the same spot) but not accurate (that spot is 2 mm off). Accuracy is corrected with mastering and calibration; repeatability is not.

The distinction drives the diagnosis: if the robot ALWAYS lands wrong in the same place, it's miscalibration or a badly defined TCP and you fix it with software. If it lands DIFFERENTLY each time at the same programmed point, you have a mechanical or thermal problem and recalibrating won't fix anything. If you're unsure whether it's calibration, first review what mastering and axis calibration are.

How to measure repeatability without lab equipment

You don't need a laser tracker to get a useful answer on the floor. With a hundredths dial indicator mounted on a fixed stand off the robot:

1. Mount a rigid reference tip or ball on the flange (no tooling with play).
2. Program an approach point and a measurement point, always entering from the SAME direction and speed. The approach matters: coming in from two different sides gives different readings because of backlash.
3. Touch off the measurement point against the indicator. Move the robot away to a distant position, come back and repeat. Do 10-20 cycles.
4. Log each arrival reading. The difference between the highest and lowest is your practical scatter.
5. Repeat with the robot cold (start of shift) and warm (after 2-3 hours of production). If the scatter changes a lot, you have a thermal component.

This method won't give you a certifiable number, but it tells you what you need: how much it moves, whether it worsens with temperature, and which axis shows it most (move one axis at a time to isolate it).

Causes ordered by frequency

1. Temperature and lack of warm-up

The number one cause, and the most ignored. A cold robot doesn't repeat like a warm one: gearboxes and structure expand. If you measure the first parts of the shift and those from mid-morning, you'll see the difference. The "fix" isn't a spare part: it's a warm-up cycle before producing critical dimensions. Many cells that fail in the early hours have this, not a fault.

2. Loose tool or fixture

Before touching the robot, tighten and check. A flange with loose bolts, a fixture that took a hit, a tool changer worn at the coupling: all of that adds scatter that looks like the robot and isn't. It's the first thing to rule out because it's the cheapest to fix.

3. Backlash in the transmission

If repeatability slowly worsens over months and shows more when you change the approach direction, suspect mechanical play in a gearbox or belt. Here it pays to measure the mechanical play in the joints before going further. Growing backlash isn't fixed by recalibrating; you only mask it temporarily.

4. Unstable encoder or feedback

Less frequent, but when it happens the scatter is erratic with no clear pattern against temperature or direction. A noisy encoder or resolver, or a damaged feedback cable, can cause position jumps. It usually comes with micro-stops or intermittent warnings.

5. External vibration or floor

A robot bolted to a vibrating base, or a floor mount that has loosened, adds scatter that isn't in the robot. Check the base anchor bolts and whether heavy machinery nearby is pounding the floor.

What to check and in what order

1. Rule out the cheap stuff first: tighten flange, tool and fixture. Check the base anchor bolts.
2. Separate thermal from mechanical: measure cold and warm. If scatter appears only cold, it's warm-up, not a fault.
3. Isolate the axis: move one axis at a time to the same point and measure. Locate where the scatter originates.
4. Test the approach direction: if readings differ when entering from opposite sides, there's backlash in that axis.
5. Check feedback: inspect encoder and battery status if the scatter is erratic with no pattern.
6. Check calibration: only if the robot ALWAYS lands in the same wrong spot (that's accuracy, not repeatability).

Common mistakes on the floor

• Recalibrating to mask backlash: the new TCP works a few days and comes back worse. You're correcting a growing symptom.
• Not measuring cold and warm: good gearboxes get swapped because the "fault" was lack of warm-up.
• Measuring from different directions: the measurement itself introduces error and confuses the diagnosis.
• Blaming the robot without checking the tooling: many "repeatability loss" cases are cell mechanics, not the robot.

When to stop and call a specialist

If you've ruled out tool, anchoring and warm-up, and the scatter persists and grows, the source is internal mechanics (gearbox, bearing, backlash) or feedback. That calls for a corrective diagnosis with controlled measurement to identify the axis and decide on intervention before the scatter ruins production. If repeatability degrades repeatedly, a preventive maintenance plan with periodic dimensional checks keeps you from getting to this point.