Touch-Sensing Offset Bias After a Fixture Change
A robotic cell welds a 300 mm fillet on a fabricated bracket at 22 mm/s. Touch sensing uses three search points per seam, a 25 mm/s search speed and a 60% torque limit. After a fixture change, every weld is consistently 2.5 mm off toward the web, and the error repeats on both left-hand and right-hand parts. The offset is calculated in the user frame but added to the position in the tool frame. The operator re-taught the reference point after the fixture change but did not update the search schedule offsets, and a single offset register is shared by both mirrored seams. The robot is a 6-axis machine with 1.6 m reach, and the part locates on two pins that were replaced with undersized dowels.
1. Explain why a constant 2.5 mm shift appears on both mirrored parts, and why a frame-transform error is not the best explanation.
Read the worked answer
A frame-transform error converts a translation that was computed in one frame into a different direction when it is applied in another. The resulting error direction depends on tool orientation. On mirrored seams the tool orientation differs, so a user-frame-versus-tool-frame mistake would normally make the error point in different directions on the left-hand and right-hand parts, or at least change magnitude with orientation.
Here the error is 2.5 mm in the same direction on both mirrors, which points to a common bias rather than a rotating transform error. The likely source is the reference/search setup: the operator re-taught the reference point but left the search schedule offsets unchanged, so every search result inherits the same systematic shift. Undersized dowels on the locating pins add part-to-part variation, but they cannot produce a repeatable 2.5 mm bias alone unless the part is consistently pushed the same way.
A quick discriminator: if the error flips direction on mirrored parts, revisit the frame logic first; if it stays the same, revisit the reference and search schedule.
2. Describe a dry-run verification using a marker or scribe, and state what measurement proves the offset is correctly applied.
Read the worked answer
Run the program with the arc disabled and a marker or pointed scribe in place of the wire. Jog the robot to the reference point and mark it, then step through the search and the offset shift and mark the post-offset TCP position. Repeat at the seam start, mid-length and end.
Measure the perpendicular distance from each mark to the joint centerline with a rule or feeler gauge. If the marks sit 2.5 mm off the centerline in the same direction, the search/offset logic is wrong; if the marks sit on the joint centerline but the welded bead does not, the problem is downstream in the weld schedule or the tool frame.
Also compare a hand measurement of the fixture datum with the touch-sensed position: touch the datum plate three times and record the position values. A repeatable but biased result confirms a search bias, while scattered values suggest the pins or part location are moving. Only after these agree should the arc be re-enabled.
3. What programming changes would prevent one offset register from serving two mirrored seams?
Read the worked answer
- Allocate a separate offset register for each seam, for example PR[11] for the left-hand seam and PR[12] for the right-hand seam; never share a register across mirrored geometry unless the intent is identical.
- Compute the search offset in the user frame and apply it in that same frame. If the robot allows a frame argument on the offset instruction, state the user frame explicitly rather than relying on the active frame at run time.
- Store the raw search positions in local registers and calculate the offset from them, so an operator re-teaching one point does not silently corrupt the other seam.
- Add a plausibility check: if the computed offset exceeds about 3 mm, stop the program and raise an alarm instead of welding.
- Re-teach the reference point and verify both seams with the marker dry run before the first production part.