Work log — Indexer project, gear backlash control week
Engineer: T. Nowak (precision mechanisms)
Project: FBL-4 four-bar rotary indexer, cam and gear-blended drive
Log period: Day 01 to Day 06
Document ref: WL-2026-0911-BK06
Day 01 — morning stand-up, the backlash that shows up in the dwell
Stand-up 08:30. The FBL-4 rotary indexer parks its fixture at four stations, and the customer’s elevator assembly is rejecting parts because the fixture rocks 0.12 mm in the parking position, visibly, at the end of each index. The rocking is the classic signature of backlash somewhere in the gear train that decelerates into the dwell, and the customer’s inspector has been chasing the clamp, but the clamp is a cam lock that measures clean. My suspicion from the report is the gear pair just upstream of the output, and the week’s job is to isolate the backlash to a member, correct it, and prove the fix on the rocking fixture.
Morning I measured the backlash the honest way, with the output locked and a torque wrench on the input. The total train backlash came back at 0.35 degrees of input rotation before the output moved under a 4 N·m preload, which at the indexer’s effective radius means the 0.12 mm rock the customer reported. Then I split the train at the couplings and measured each stage: the input bevel pair carries 0.07 degrees, the helical pair 0.11, and the final spur pair into the output carries 0.16 degrees, nearly half the total, so the spur pair is the first target, and a 0.16 degree backlash on a quality-cut spur pair is high, it should be closer to 0.05 after proper center distance setting.
Evening note: the split told me where to start, but not why the spur pair is loose, and the why matters, a worn tooth flank asks for a different fix than a center distance that drifted when the machine was bolted down. Tomorrow I inspect the mesh and the mount before I shim anything.
Day 02 — the mount that carried the center distance with it
Stand-up: a field report beat me to it, the indexer had been through a service four months prior, and the service sheet said the output cover was removed and re-seated. That is the smell. Morning I pulled the output cover and the spur pair guard, and the spur gear mount told the story: the two-bearing pillow block that carries the input gear of the spur pair was sitting 0.14 mm further from the output shaft centre than the drawing’s center distance asks for, and the mounting bolts were not torqued to the block’s 95 N·m spec, they came out at 40 and 55 N·m, and the block had walked on its slots. The backlash in the spur pair was not worn teeth, it was a walking gearbox mount, and the teeth were fine, a clean flank check with the dye showed zero pitting.
Morning repair: I unbolted the pillow block, cleaned the slots, and set the center distance with a pin gauge, adjustable housing with two dowel pins, and I re-bored the dowel holes 0.2 mm to take new oversized dowels so the setting cannot walk again. The block went back with the center distance set to drawing, and the bolts torqued to 95 N·m in a two-pass criss-cross sequence. Re-measured the spur pair backlash cold: 0.035 degrees, down from 0.16, a clean number that sits just inside the 0.04 target line, and the split measurement now shows the total train at 0.18 degrees, with the remaining backlash concentrated where the pairs are correctly set.
Afternoon I re-checked the whole train with the torque wrench method and the total dropped from 0.35 to 0.19 degrees, then I ran the indexer at reduced speed into the dwell and watched the fixture: the visible rock shrank from 0.12 mm to 0.05 mm, worth the work but not yet quiet, which tells me the residual 0.19 degrees is now the bevel and helical pairs, not the spur, and whether they matter depends on the load cycle. The customer’s rejection criterion is 0.08 mm, so the fixture is close but I cannot hand it over on a close.
Evening thought: the walking block is an assembly-floor failure that a torque wrench would have prevented, and the fix cost a re-bore and two dowels instead of a gear set. The lesson I will put in the log is that backlash investigation has to start at the mount, because the gearbox does not wear its backlash, the workshop installs it.
Day 03 — the preload question, and a decision about the bevel pair
The residual 0.05 mm rock at the dwell sent me to the bevel and helical pairs, and I spent the morning deciding whether they matter in this load cycle. The indexer parks against a hard stop in the cam phase, so the gear train is not the positioning element, the cam is, but the residual backlash lets the fixture settle and the 0.05 mm shows up as the visible nudge during the dwell. The choice: leave 0.19 degrees total and let the cam hold the station, or preload the bevel pair on its preloaded bearings and take the residual to zero at the cost of slight bearing warmth and a hint of gear whine on the index.
The decision, recorded at 11:30, was to preload the bevel pair only as far as the elastic deflection of the mounting allows, not to rely on a spring-loaded anti-backlash gear, which on a four-bar indexer carrying a live fixture is more fragility than precision. I set the bevel pair’s preload on the adjustable bearing stack by adding one 0.5 mm shim to the preload washer pack, which took 0.06 of the remaining 0.07 bevel degrees out, leaving the bevel effective zero under the 4 N·m test and the helical pair carrying the last 0.08 degrees, a number the cam phase holder absorbs without complaint.
Afternoon dwell test at full speed: the fixture now rocks 0.02 mm, measured with a lever indicator over twenty cycles, against the customer’s 0.08 mm criterion, and the rock is now a bearing-microscopic settle, not a backlash clunk. Total train backlash under the 4 N·m preload sits at 0.09 degrees, all of it on the helical pair whose open design exists precisely to absorb thermal growth, a pair I am deliberately not touching.
Evening note: the tempting move would have been to anti-backlash the helical pair too, and the resulting zero-backlash train would bind when the box warms and the shafts grow, turning a 0.02 mm rock into a gear galling complaint that would arrive in summer. The residual backlash that I leave is a design feature, and I want that sentence in the release notes so the next engineer does not delete it.
Day 04 to Day 05 — production in the dwell, and a temperature soak proves the choice
Days 04 and 05 the indexer ran in production, eleven hundred cycles at 24 degrees of indexing per 2.1 second cycle, and the fixture rock never left the 0.02 to 0.03 mm band across both shifts. I ran a deliberate thermal proof on Day 05 to check my refusal to kill the helical backlash: after four hours at maximum rate the gearbox skin sat at 48 C, the casing expansion pushed the helix pair center distance by 0.035 mm, and the train backlash softened from 0.09 to 0.11 degrees, exactly the thermal growth the helical clearance allows. The fixture held 0.03 mm at the warm end, so the 0.02 shim decision and the deliberate open backlash on the helical pair were both vindicated by the soak.
One nuisance noted on Day 05: a harmonic at the third station, a lighter rock, 0.02 mm versus 0.03, and I traced it to the station’s own locating pin wear, 0.08 mm slop in the pin bushing, an unrelated fault that the machinery has earned after ten thousand cycles, and it gets its own D-number and its own fix next week, not a backlash concession.
Day 06 — the acceptance run and the log closure
Acceptance with the customer’s process engineer present. The fixture rock on the elevator assembly rejection fixture: 0.02 mm against the 0.08 mm criterion, total train backlash 0.09 degrees at 4 N·m preload, spur pair 0.035 degrees, bevel effective zero, helical 0.08 degrees left open for thermal growth. The process engineer signed the acceptance with the 0.02 millimetre number ringed, and I handed over the re-bored dowel set and the torque sequence sheet as part of the fix package.
The rejection stream that started the week, 0.12 millimetre rocks on eleven percent of assemblies, is closed at zero rejects over eleven hundred cycles, and the reason is honest: the spur pair was vaulted back to its drawing center distance, the bevel pair preloaded into effective zero, and the helical pair deliberately left loose so the box can breathe. A fully zero-lash train would have shipped the 0.02 millimetre fix today and failed a 48 C profit run in July.
Attachment A — backlash before and after the corrections
| Stage | Before | After |
|---|---|---|
| Spur pair | 0.16 degrees | 0.035 degrees |
| Bevel pair | 0.07 degrees | effective zero (preloaded) |
| Helical pair | 0.11 degrees | 0.08 degrees (kept open) |
| Total train | 0.35 degrees | 0.09 degrees |
| Fixture rock | 0.12 mm | 0.02 mm |
Attachment B — backlash investigation checklist
1. Measure total backlash with the output locked and a torque wrench on the input. 2. Split the train stage by stage to size the suspect. 3. Inspect the mounts and bolt torque before touching the gears. 4. Set the center distance to drawing and pin it with dowels. 5. Re-measure the suspect pair cold. 6. Preload only what the design will tolerate indefinitely. 7. Soak the box warm and confirm the residual backlash grows the way a deliberate clearance should.
Glossary
Backlash: the lost motion between a gear pair when the drive reverses. Center distance: the nominal distance between two gear shaft centres. Anti-backlash gear: a spring-loaded gear that removes free play at zero load. Dwell: the stationary phase of an indexer between movements. Cam phase holder: the hard stop that positions the indexer in its parked state.
Week-end reflection
Friday evening. Three bullets for the log. First, backlash is installed, not worn, the walking pillow block produced 0.16 degrees while the teeth were perfect, and a torque wrench plus a pin would have skipped the entire complaint. Second, zero backlash is not the goal, the goal is the right backlash held reliably, the helical pair that I left at 0.08 degrees is the pair that will save this box in summer. Third, measure at the split line, the 0.05 rock after the spur fix told me the bevel and helical still carried residual motion, and chasing it with one more preload would have over-tightened a healthy pair. The indexer now parks on 0.02 millimetres of truth, and the log records why the helix stays loose.
