PROJECT DESIGN REVIEW — MINUTES OF MEETING
Project: rotary indexing unit for a coating cell, C-axis
Review subject: reducer family choice and backlash specification
Date: 2 June, 14:00 to 15:40, meeting room 3
Document ref: DMR-2026-0602-CAX04
Attendees
| Name | Role |
|---|---|
| P. Moretti | Lead mechanism designer (chair) |
| H. Andersson | Machine servicing engineer |
| S. Devi | Servo and motion engineer |
| J. Novak | Quality / metrology |
| L. Costa | Procurement |
1. Purpose of the review
The rotary axis positions a 38 kg tooling plate carrying three coating heads, indexing through 120 degrees in 1.9 s. The draft drive is a 1.1 kW servomotor through a planetary reducer of ratio 10 with a specified backlash of 8 arc-minutes. After a first build the head held position to about 0.06 degrees, against a customer target of 0.03 degrees, and the coating pass showed a repeating seam line at the joint. The review was called to decide whether the backlash spec should be tightened, the reducer family changed, or the servo control changed to hide the mechanics.
2. Review comment CA-01: what the 8 arc-minute backlash actually costs
S. Devi turned the backlash number into a position error at the head. Eight arc-minutes of backlash at the reducer output is one eighth of a degree, 0.125 degrees in round numbers, and at the 238 mm pitch radius of the tooling plate that is 0.52 mm of tangential play at the coating head. The customer target of 0.03 degrees is four times smaller than the backlash alone, before the servo adds its own error. The conclusion of the comment was blunt: the drive was specified on the motor side, where backlash is irrelevant, and the eight arc-minute number dominated everything at the head. None of the tuning in the world closes a mechanical gap that opens in both directions of rotation.
J. Novak measured the actual backlash on the first build with a dial gauge at the plate edge and recorded 9.4 arc-minutes with the locking pin out, close to nameplate, and the seam line repeated at exactly the axis joint, confirming the backlash was the visible offender. H. Andersson added that a backlash this size also shows up as a lash click when the head reverses direction during the coating pass, which the customer’s acoustic inspection had already flagged.
3. Review comment CA-02: the reducer families on the table
P. Moretti laid out three reducers with the same ratio 10 and output torque class, and the review compared them across the columns that matter. First, a standard planetary, backlash 8 arc-minutes, price 410 euros, and a stiffness that the servo could not see through. Second, a low-backlash planetary, backlash 3 arc-minutes, price 780 euros. Third, a harmonic drive, backlash 1 arc-minute, price 1,240 euros, with a different torque rating and a different limitation, its flexible spline. The review noted the numbers before the discussion: at the head radius, 8 arc-minutes is 0.52 mm, 3 arc-minutes is 0.20 mm, and 1 arc-minute is 0.07 mm. Only the harmonic drive puts the mechanical error below the 0.03 degree target converted to 0.21 mm at the head, and even then the servo budget must share the remaining margin.
4. Review comment CA-03: backlash is not the only number a servo cares about
S. Devi warned the meeting not to buy the lowest backlash number alone. A drive with tiny backlash but low torsional stiffness resonates with the servo loop, and the reflection of that compliance limits how high the gains can go. The maker’s data sheet lists a torsional stiffness per reducer: the standard planetary at about 9 N-m per arc-minute, the low-backlash at 14 N-m per arc-minute, and the harmonic at 22 N-m per arc-minute at the output. The servo engineer’s rule of thumb pulled from the previous machine: the first resonance of the mechanical train should sit above 25 Hz, or the gains have to drop. With the head inertia of 0.47 kg-m-squared reflected through ratio 10, the standard planetary gives a first torsional resonance near 18 Hz, the low-backlash near 23 Hz, and the harmonic above 30 Hz. The review therefore wrote two criteria on the board: backlash at the head plus resonance against the servo gains, not one column of the datasheet.
5. Review comment CA-04: can the servo tuning hide the backlash layer
S. Devi had been asked three times whether a gain setting could mask the lash, and the answer the review heard was delivered with a shrug. In a position loop with a one-way index, the backlash is present only at reversal, and if the axis always approaches from the same side, the servo can hold a constant offset and the lash never appears. The coating pass, however, scans back and forth across the head seam, so the axis reverses at the seam twice per part, and the lash opens there regardless of tuning. The reviewer closed the option: no gain table removes a 0.52 mm physical gap that appears on every reversal, and any attempt to pre-load the lash with a bias torque heats the motor and wears the reducer. The meeting crossed the tuning-column off the list and moved to mechanical and procurement columns.
6. Review comment CA-05: the harmonic drive and its flexible spline limit
H. Andersson raised the one real reservation about the harmonic drive. The flexible spline is a thin shell, and it has a stated life in cycles under load. On a coating cell running two shifts, the axis reverses about 14 times per minute, roughly 840 times per hour, and the catalogue life of the selected harmonic size under the 12 N-m mean output torque is 20,000 hours, which is 24 months at this duty. The standard planetary has no such cycle limit, only bearing and tooth wear, and it would outlive the machine. The review wrote this against the price and noted that the coating cell is rebuilt every three years, so a 24-month reducer life is a replacement within the machine life, an accepted cost if the selection is right.
7. Review comment CA-06: procurement compared lead times
L. Costa laid out the supply numbers. The standard planetary was in stock at 410 euros, next day. The low-backlash planetary had a 9-day lead at 780 euros. The harmonic drive had a 21-day lead at 1,240 euros, and the supplier offered only a 12-month warranty against a 36-month line. Against the daily cost of the rejected coating seam, which the plant controller had already valued at 220 euros per reappearance plus the rework time, the price difference between the low-backlash and the harmonic was recouped in about two seam-free days. The procurement view changed the tone of the meeting, because a 460 euro upgrade that removes a recurring 220 euro-per-occurrence defect pays for itself inside a week.
8. Disagreement that had to be resolved: harmonic drive or low-backlash planetary
The room split cleanly down the middle for a full twenty minutes. J. Novak argued for the harmonic drive because the spec says 0.03 degrees and only one arc-minute of backlash satisfies the number without leaving the servo to invent the rest of the margin. H. Andersson argued for the low-backlash planetary because the cycle-life limit of the flexible spline would force a reduction replacement inside the machine life, and because the coating seam disappears at 3 arc-minutes once the axis stops reversing through the backlash, which the head already does on its normal pass. S. Devi broke the tie with the resonance number: the low-backlash planetary sits at 23 Hz first resonance, uncomfortably close to the servo’s 25 Hz rule, while the harmonic at over 30 Hz gives the gains real headroom. The vote landed with the harmonic drive, eight to three, on the reasoning that a mechanically true axis is safer than a mechanically borderline one tuned to the limit, with the 24-month reduction life written into the spares plan.
9. Risks and mitigation
Three risks were logged. Risk 1: the 21-day harmonic lead delays the rebuild; mitigation is to place the order immediately and run the old axis at a reduced cycle in the interim. Risk 2: the flexible spline could fail inside two years under an unseen overload; mitigation is a torque-limiting coupling set at 18 N-m at the reducer input and an annual spline inspection. Risk 3: the servo gains tuned today may not fit the stiffer drive once installed; mitigation is a re-tune session with the resonance measured, booked two days after installation.
10. Actions and owners
| ID | Action | Owner | Deadline |
|---|---|---|---|
| ACT-1 | Place harmonic drive order, ratio 10, 1 arc-minute | L. Costa | 4 June |
| ACT-2 | Fit torque-limiting coupling at 18 N-m | H. Andersson | 28 June |
| ACT-3 | Measure first resonance and re-tune servo gains | S. Devi | 3 July |
| ACT-4 | Verify backlash at plate edge, target under 1.5 arc-minutes | J. Novak | 4 July |
| ACT-5 | Add flexible spline inspection to annual service | H. Andersson | from 8 July |
11. Decision summary
Adopt the ratio 10 harmonic drive with one arc-minute of backlash. Fit a torque-limiting coupling to protect the flexible spline. Re-tune the servo against the measured resonance. Verify the assembled backlash at the plate edge below 1.5 arc-minutes, which is 0.10 mm at the head radius, inside the customer’s 0.03 degree target. Plan the flexspline replacement at 24 months.
12. Post-meeting addendum, 20 July
The harmonic drive indexed two weeks without a seam. The verified backlash at the plate edge measured 1.2 arc-minutes, the first resonance landed at 29 Hz and the servo holds the plate within 0.025 degrees, inside the target. The coating supervisor asked what had changed on the machine and the only honest answer was that the machine had stopped lying about where it was pointing.
Glossary
Backlash: the angular play between reducer input and output when rotation reverses. Arc-minute: one sixtieth of a degree. Torsional stiffness: how much output torque per arc-minute of twist the reducer carries. First resonance: the lowest natural frequency of the motor, coupling, reducer and load train. Flexspline: the thin flexible shell of a harmonic drive that carries the teeth and limits cycle life.
Appendix. The selection worksheet the review walked through
The minutes close with the worksheet that the review used, so the reasoning is repeatable rather than remembered. Column one, the duty: 1.1 kW servomotor at 3,000 rpm nominal, ratio 10, output speed 300 rpm at the indexing rate, mean output torque 12 N-m, peak 24 N-m at the head start and stop. Column two, the error budget: customer position target 0.03 degrees, converted to 0.21 mm at the 238 mm head radius; the servo allocated 0.015 degrees of the budget, leaving the reducer to carry the other half. Column three, the candidates with their arc-minute figures against that half-budget: standard planetary 8 arc-minutes fails the reducer half alone, low-backlash planetary 3 arc-minutes passes on paper but drags the resonance to 23 Hz, harmonic drive 1 arc-minute passes the reducer half with margin and keeps the resonance above 30 Hz. Column four, the life and price: standard planetary unlimited cycles at 410 euros, low-backlash planetary unlimited at 780 euros, harmonic drive 20,000 hours at 1,240 euros with the flexspline planned out at two years.
The chair’s closing remark was recorded as agreed on the board: a reducer is specified twice, once for the torque it carries and once for the truth it tells the position loop, and this machine was failing on the second sentence, not the first. The minutes are signed by all five attendees, and the harmonic drive order went out the same afternoon.
