Nov 16,2026
A time study from a Tier-1 body-in-white supplier. Cleaning used to consume 22 of the 90 seconds. The fix was not a faster cleaner.
One cell, two robots, a 90-second takt, 1,200 parts per shift, two shifts. The process sheet lists welding, handling and inspection. It does not list cleaning, because cleaning was absorbed into the operator's walk.
Two things were wrong with that line item. It was not in the standard, so it was not measured. And it happened serially — while the operator wiped, the robots waited. Twenty-two seconds of two robots standing still, twice a minute, all shift.
| Cycle element | Before | After | Change |
|---|---|---|---|
| Load / clamp | 18 s | 18 s | — |
| Robot weld 1 | 29 s | 29 s | — |
| Robot weld 2 | 17 s | 17 s | — |
| Fixture wipe | 22 s | 8 s (parallel) | −14 s |
| Unload / inspect | 4 s | 4 s | — |
| Effective takt | 90 s | 76 s | −15.6% |
Table 1 — Time study, 40 consecutive cycles. The 8 seconds sit inside the welding pass, so the cell is not waiting for them.
A faster handheld cleaner would still be serial: pick up, clean, put down, restart. Even at 12 seconds it would cost the cell 12 seconds. Mounting a 200 W head on the existing gantry and firing it during the weld pass costs zero takt seconds, because those seconds were already being spent by the robots.
The catch is geometric: the head has to reach the fixture faces without entering the weld path, and it has to fire on a trigger from the robot controller, not from an operator. That is an integration job, not a purchase.
| Option | Takt impact | Cost | Fits |
|---|---|---|---|
|
1. Standalone booth Operator cleans parts before loading |
Still serial, ~10 s | $16,500 | Low volume, mixed parts |
|
2. Gantry-mounted head Fires during the weld pass, triggered by PLC |
0 s (parallel) | $34,000 | Single-part, high volume |
|
3. Robot-mounted, shared Cleans between welds on the same arm |
+3 s (path sharing) | $27,500 | Multi-part cells |
Table 2 — Integration options. Option 2 costs twice option 1 and pays back four times faster, because it buys takt rather than labour.
The plant tracked three numbers before and after: availability (unplanned stops), performance (takt adherence) and quality (spatter-induced rework). Cleaning affects all three, which is why the business case was easier than expected.
| Metric | Before | After 6 months | Delta |
|---|---|---|---|
| Availability | 82.4% | 89.1% | +6.7 pts |
| Performance (takt adherence) | 91.0% | 97.6% | +6.6 pts |
| Quality — spatter rework | 3.1% | 0.6% | −2.5 pts |
| OEE | 74.7% | 84.7% | +10.0 pts |
Table 3 — OEE before and after, same cell, same product mix. Availability gains came from fewer fixture-related stops; quality gains from consistent cleaning instead of "wipe when it looks bad".
The plant runs 1,200 parts per shift, two shifts, 300 days. A 10-point OEE gain on a 90-second takt is worth roughly 115 extra parts a day at the same headcount — and 115 parts is about $1,840 of contribution at their margin.
| Line item | Value |
|---|---|
| Integration cost (option 2) | $34,000 |
| Extra parts per day | 115 |
| Contribution per part | $16 |
| Daily gain | $1,840 |
| Payback | 18.5 working days |
Table 4 — Payback at their own contribution margin. Note that this is a capacity argument, not a labour-saving one: nobody was made redundant.
Three things, all predictable in hindsight. The trigger signal needed a 40 ms debounce or the head fired on a fixture clamp spike. The extraction hood had to be repositioned twice before it stopped pulling the shielding gas away from the weld. And the protective window needed changing every 90 hours instead of the quoted 160, because anti-spatter oil aerosols faster than rust dust.
None of these were machine faults. They were integration faults, and they cost eleven days of engineering time — which is the number to budget for, not the hardware.
Three conditions have to hold before the parallel trick works, and we check all three before quoting. If one fails, the project becomes a booth.
First, there must be a window. The head has to fire on surfaces the robot is not welding at that moment. On this cell the fixtures sit 400 mm behind the weld path, so a fixed head aimed at the fixture face never crosses the beam. On a tight cell where the fixture is inside the weld envelope, there is no window — and no amount of programming creates one.
Second, the trigger has to come from the controller. An operator pressing a button is a serial action again. The PLC has to fire the laser on a program block, which means the machine supplier and the integrator have to agree on a signal. This is where most projects lose a week: the laser accepts an external trigger, but the robot program has no spare output and no one wants to re-teach the cell.
Third, the extraction has to be local. Cleaning during the weld pass means fume and spatter are being generated in the same enclosure as the weld — which already has extraction sized for welding fume, not for vaporised oil. In this plant we added a 600 m³/h hood at the fixture, and it was the cheapest part of the whole job.
If all three hold, the cleaning disappears from the takt. If any one fails, it does not — and the honest answer is a standalone booth with a 10-second serial cost, which is still better than 22.
Cleaning was 22 seconds of a 90-second takt and appeared on no process sheet. Moving it into the welding pass took it to 8 seconds of parallel time and lifted OEE by 10 points.
The rule: if the cell is takt-limited and runs two shifts, integrate. If it is not, buy the booth, keep the 10 seconds, and spend the difference on something that is actually the constraint.