Work log — Machine tool project, rotary table calibration week
Engineer: H. Gallo (machine tool metrology)
Project: HMC-610 machining centre, 4th axis rotary table
Log period: Day 01 to Day 06
Document ref: WL-2026-0911-RT05
Day 01 — morning stand-up, the index that drifts by thirty microns
Stand-up 08:25. The HMC-610’s rotary table stopped holding its angular position: the customer’s bore pattern jobs show a 30 micron positional error across a 400 mm bolt circle when the table indexes through 90 degree steps, and the trend is worse after warm-up. The table is a helical worm-drive rotary unit, 1:90 ratio, hydraulic clamp, and the quality engineer wants to know whether the error is the worm’s accumulated pitch, the encoder’s mounting, or the clamp that lets two microns in and the parts feel thirty.
Morning I set up the double-ballbar on the equator of the table and the electronic level on the face, because a rotary axis error has three masks: the positioning error, the axis tilt, and the radial runout, and I wanted them separated before I started adjusting anything. First pass at 90 degree steps, cold: the index error came back at plus 8 arc-seconds at 90, minus 12 at 180, plus 5 at 270, minus 9 back to zero, a pattern that repeats per revolution and points at the worm wheel’s accumulated pitch error rather than a one-shot clamp slip. The clamp repeatability check on top of it, ten clamps at zero degrees: plus or minus 2 microns on the ballbar, clean, so the clamp is not the thief.
Evening note: the repeating pattern is the clue. A server-side error, or a mis-mounted encoder, repeats on a fixed offset, a worn worm wheel shows a once-per-revolution slowly varying swing, and this pattern, asymmetric in sign around each 90, is the signature of accumulated pitch error in the worm wheel. Tomorrow I measure the encoder and the worm separately so I am not correcting a symptom.
Day 02 — the encoder vs the worm, and where the error actually lives
Morning, I uncoupled the encoder story from the worm story before touching the machine. The direct-reading encoder on this table reads from a ring on the table spindle, so it sees the true table angle, good, the encoder is not the suspect for the repeating pattern. The suspect is the worm wheel itself. I ran a 360 point continuity check with the ballbar at 5 degree steps, a slow sweep with the clamp off and the table driven by the servo. The plot wrapped on itself so the wheel’s accumulated pitch error emerged as a smooth once-per-revolution sinusoid with a 22 arc-second peak-to-peak swing, and the 90 degree samples sit on that sinusoid with the plus and minus signs from Day 01. Root cause identified: the worm wheel has 20 arc-seconds of accumulated pitch error, from the supplier’s grinding or the wheel’s thermal history, and the machine control does not yet know about it.
Afternoon I checked the reverse direction to separate backlash from pitch. Bi-directional index at each 90 degree step, approach from plus and minus, and the backlash band came back at 3 arc-seconds measured at the table face, negligible, so the pitch error is the whole story and the correct lever is the control’s pitch error compensation table, not a mechanical reset of the worm mesh. The machine control accepts a rotary error compensation array, one value per 5 degrees, and I mapped the 72 readings into the compensation table before end of shift, leaving the roll-over point at zero so the compensation wraps cleanly across 360.
Evening thought: the satisfying part is the error was measured, not assumed, the 22 arc-second sinusoid turns a workshop rumour about worn worms into a compensation table a control actually uses. The risky part is the roll-over, a bad wrap at 360 degrees turns a useful table into a machine that miangles one index, so the wrap check is Day 03.
Day 03 — the wrap check and the verification run
Morning, the roll-over check I flagged at the Day 02 close. I indexed the table through the full 360 in 5 degree steps with the compensation active, both directions, and watched the residual plot. The key moment was the 355 to 360 to 5 degree transition: the compensation wrapped with no step, no doubled correction at the seam, and the residual across the full revolution came back at plus or minus 1.5 arc-seconds instead of the plus or minus 11 of the uncompensated sinusoid. The wrap is clean, and the compensation table is now a closed loop that the control never has to re-derive.
Afternoon, the verification that matters to the customer, not the metrology department: the bore pattern job. A ring with twelve 12 mm bores on a 400 mm bolt circle, indexed at 30 degree steps, machined in alloy steel with the compensation active. The CMM read-back on the finished ring: bore-to-bore angular spacing within 0.9 arc-second of the 30 degree datum, against a 5 arc-second tolerance, and the positional error across the circle dropped from 30 microns to 4 microns, inside the customer’s 8 micron print callout. The 30 micron complaint that opened the week is gone, and the machine now indexes better than it did the day it was installed.
Evening note, partly dry, partly honest: the worm wheel did not improve, the control simply learned its rhythm. The compensation is a live table, not a one-time fix, and if the wheel wears another 10 arc-seconds over the next two years the same 72-point measurement will re-map it, which is exactly why I wrote the calibration procedure into the machine’s PM instead of leaving it in a drawer.
Day 04 to Day 05 — stability through the shifts and the temperature band
Days 04 and 05 ran production with the compensated table. Day 04: forty-two parts through the bore pattern, every CMM read within the 8 micron callout, first-pass yield at 100 percent for the day. Day 05, running through the afternoon warm-up to 26 C table skin: the point index offsets held within 1.2 arc-seconds cold-to-warm, the compensation table’s values are stable, the machine’s thermal growth is being managed by the existing tool-length compensation rather than leaking into the rotary axis.
One warm note for the log: the machine’s health report card now shows the rotary axis out of the red, and the clamping force check at each 90 degree index returned the same 2 micron repeatability as Day 01, confirming the clamp was never at fault and stays untouched, the discipline that kept me from dismantling a healthy part of the machine.
Day 06 — handover and the calibration record
Handover day with the customer’s metrologist in the room. I walked him through the measurement deck, the 360 point plot, the compensation table, the wrap check, and the verification ring, and he watched a fresh 5 degree index land at 0.8 arc-second from command. He signed the calibration certificate with the residual band plus or minus 1.5 arc-seconds written in, and the machine’s compensation file now lives in two places, the control and the engineering archive, so a future controller swap does not resurrect the 30 micron complaint from the memory of the old files.
Closing the log with the numbers the job owed: 22 arc-second peak-to-peak wheel pitch error measured, 72-point compensation mapped at 5 degree intervals, residual from plus or minus 11 to plus or minus 1.5 arc-seconds, backlash 3 arc-seconds, bore pattern positional error from 30 to 4 microns on a 400 mm circle, and first-pass yield on the pattern at 100 percent for two shifts. The worm wheel is what it is, and now the machine tells it where to stand.
Attachment A — calibration record table
| Item | Before | After |
|---|---|---|
| Wheel pitch error | 22 arc-seconds p-p | mapped, compensated |
| Index residual | +/- 11 arc-seconds | +/- 1.5 arc-seconds |
| Backlash | 3 arc-seconds | 3 arc-seconds (untouched) |
| Bore pattern error | 30 microns | 4 microns |
| Clamp repeatability | +/- 2 microns | +/- 2 microns |
| Pattern first-pass | falling yield | 100 percent |
Attachment B — rotary calibration check list
1. Measure indexing error at 90 degree steps with a rotary ballbar, cold and warmed. 2. Check clamp repeatability separately before adjusting the drive. 3. Run a 360 point sweep at 5 degree steps with the clamp off. 4. Fit the once-per-revolution sinusoid and record peak-to-peak pitch error. 5. Map the 72 values into the control compensation table. 6. Verify the 360 degree roll-over leaves no seam. 7. Re-measure the customer’s critical pattern and record the residual.
Glossary
Accumulated pitch error: the progressive angular error of a worm wheel around one revolution. Ballbar: an instrument that measures the radius change as an axis moves, here reading table index error. Compensation table: a control array that adds a stored correction at each index position. Roll-over: the transition of the compensation across 360 degrees back to 0. Backlash: the lost motion when the drive reverses direction.
Week-end reflection — the log as a metrology sermon
Friday evening, and the machine is running clean. Three things from this week I am writing down so the next calibration does not start from zero. First, measure before you blame, the repeating index pattern told me in ten minutes that the error was the worm wheel’s accumulated pitch and not the clamp, and the temptation to tighten a healthy clamp would have cost the customer a reinspection and me a week of shame. Second, a compensation table is only as good as its wrap, a clean 360 degree seam is worth a full re-measurement, because one doubled correction at the roll-over silently reintroduces the failure on a single index a day later. Third, record the before numbers, the 22 arc-second sinusoid and the plus or minus 2 micron clamp repeatability are the only reason anyone will believe the machine improved, and they are the baseline a future engineer needs when the wheel wears another ten arc-seconds and the pattern drifts again.
The machine now indexes where it is told, the bore pattern ships at 100 percent first pass, and the measurement deck sits in the archive for the day the worm wheel turns its next ten arc-seconds. That is the day I told the customer to call, because the compensation is awake now, and it will tell us when the wheel has moved.
One last arithmetic note, because the seminar asked for it: at the machine’s 600 mm table radius with a job riveted near the edge, a 22 arc-second index error deflects the tool tip by about 64 microns, which is why the 400 mm bore circle, radius 200 mm, showed the 30 micron number it did, 22 arc-seconds times 200 mm radius works out almost exactly to that 30 microns. The compensation cuts the angular error to about 1.5 arc-seconds, so the same radius now moves the tip by about 4 microns. Every time a machining centre drifts, it is worth remembering an arc-second is not a metrology curio, at the end of a radius it is a micron walking toward the tolerance band.
