Oct 29,2026
The memo I sent my own production managers after a $38 part took a $4,180 week out of our August. Reproduced here because three customers asked for a copy.
This is written from the owner's chair, not the technician's. It answers the three questions a plant manager actually asks: what breaks, what it costs, and what is the smallest routine that stops it breaking. Every number below comes from our own service tickets, our own time sheets and our own trial logs — not from a brochure.
Yes, these machines need a daily check. But it has to be five minutes, seven checks, and nothing that requires a screwdriver. A routine that is thorough enough to be respectable is usually too long to survive a busy week.
We proved that on ourselves, at our own cost. Version 1 of our SOP took 22 minutes: window out, mirror inspection, coolant top-up, full power test, written report. Within a month, 41% of shifts had signed it — and the signatures clustered on quiet mornings. Version 2 dropped to five minutes and seven look-only checks. Over the next 214 days, 96% of shifts were logged with photo evidence, and unplanned stops across seven machines fell from 14 a year to 3. Same machines, same operators, same shift patterns. Only the routine changed.
The table below is the failure distribution from our field service tickets on 500W–2000W pulsed units, 2023–2025. Percentages are share of tickets, rounded. Costs are parts plus labour, and deliberately exclude downtime — downtime gets its own table in section 3.
| Failure mode | Share of tickets | Typical repair cost | Caught by a 5-minute check? |
|---|---|---|---|
| Contaminated or burned protective window | 41% | $38–$120 (part only) | Yes — 20 seconds with a torch |
| Focus lens or collimator damaged behind a dirty window | 14% | $850–$2,400 | Yes — same check, one shift earlier |
| Clogged radiator or loaded filter mat (air-cooled units) | 12% | $0–$180 labour | Yes — 40 seconds, visual |
| Loose or dirty QBH connector, overheat trip | 8% | $900–$3,200 | Partly — seat and lock only |
| Galvo bearing wear, sooted mirror | 7% | $1,100–$2,600 | Partly — listen for pitch change |
| Cable strain, torch damage, coolant seep (water-cooled) | 6% | $60–$400 | Yes — 40 seconds, visual |
| "No fault found" — wrong preset, mode or focus | 12% | $0 part, 2–6 h lost | Yes — 10-second test burn |
Table 1 — Failure distribution, TIANYU field service tickets 2023–2025. Directional data from our own installed base, not a warranty of outcome.
Read the first two rows together, because they are the same story told twice. 55% of everything we repair starts as a $38 window nobody looked at. The protective window is a sacrificial part — it is supposed to get dirty, that is its entire job. It sits between the workpiece and the optics precisely so that spatter, metal dust and fume land on a $38 consumable instead of an $1,800 lens.
Five minutes per shift, two shifts, 300 working days: 50 hours of operator time a year. At a loaded $12 an hour, that is $600. Against it, one median lens event in our shop — part, labour, and 4.5 hours of stopped production at $310 an hour — lands at $2,175. One event costs more than three and a half years of doing the check.
| Line item (7 machines, annual) | Before the SOP | With the 5-minute SOP | Delta |
|---|---|---|---|
| Unplanned stops per year | 14 | 3 | −11 |
| Hours of production lost | 63 | 9 | −54 |
| Repair spend (parts + labour) | $2,140 | $410 | −$1,730 |
| Routine labour (50 h × $12) | $0 | $600 | +$600 |
| Downtime cost (hours × $310) | $19,530 | $2,790 | −$16,740 |
| Total annual cost | $21,670 | $3,800 | −$17,870 |
Table 2 — Our own before/after figures, seven pulsed units, one 300-day year. Downtime valued at a $310/hour shop rate.
$17,870 a year across seven machines is $2,553 per machine — about 5.7 times the cost of performing the check. The routine pays for itself in roughly twelve days. Framed honestly, this is not a maintenance programme with a return attached. It is a return with a maintenance form attached.
Seven steps, one operator, no tools, no disassembly. Print it, laminate it, tape it to the machine at eye level. Times assume a 500W–2000W pulsed unit; robotic cells add the cell interlock and light curtain to step 1.
| # | Check | Time | Pass criteria | If it fails |
|---|---|---|---|---|
| 1 | Protective window, held up to a light | 20 s | Clear, no haze ring, no spatter spots, no brown tint | Swap window, log it. Do not wipe and continue |
| 2 | Nozzle and air knife | 20 s | No spatter bridging, air flows freely, no deformation | Clean or replace nozzle |
| 3 | Air pressure at the gauge | 20 s | Within ±10% of commissioning value (typically 0.4–0.6 MPa) | Check filter first, then regulator |
| 4 | Cooling: radiator fins, filter mat, coolant level and colour | 40 s | Fins clear, mat not matted, coolant at the mark and still clear | Clean, top up, report any colour change |
| 5 | Torch cable, strain relief, connector seated and locked | 40 s | No kinks, no crushed insulation, connector locked | Tag out. Do not run the shift |
| 6 | Ten-second test burn on scrap | 40 s | Steady pitch, clean line, no spark shower | Stop and escalate to maintenance |
| 7 | Photo of the window plus log entry | 40 s | Photo timestamped, sheet signed by name | No photo = the check did not happen |
Table 3 — The routine. 3.5 minutes of work, 5 minutes including walk-up and log entry.
Step 7 is the one that decides whether the other six ever happen. A signature proves somebody stood in front of the machine. A timestamped photo proves what the window looked like at that moment, which is the only record that survives a dispute three weeks later. We moved from 41% to 96% compliance by adding a photo, not by adding a training session — and I would spend the training budget on the photo habit first.
Over-maintenance is a real cost, and it is the usual reason a maintenance programme dies in month three. Four things that stay off the daily list, no matter how diligent the operator:
Do not solvent-wipe the window every day. Alcohol plus a shop paper towel drags grit across the coating and leaves a film that absorbs more energy than the dust did. Wipe only when the light test shows haze: lint-free wipe, approved solvent, one direction, one pass, then look again.
Do not open the housing or touch the optics. Three of our ten "no fault found" tickets traced back to somebody who "just had a look" at the mirrors. There is nothing inside that improves with daily attention.
Do not run a full-power burn as a daily test. It spends window life and adds thermal cycles to buy information that a ten-second burn already gives you. Save the power test for the monthly slot.
Do not grease or adjust the galvo. It is a sealed, factory-calibrated assembly. Daily intervention there manufactures the failure it is supposed to prevent.
Daily checks catch contamination. They do not catch wear — bearings, seals, connector end faces and coolant chemistry all degrade on a slower clock. Two more cadences, both short enough that people keep doing them:
| Cadence | Time | Tasks | Owner |
|---|---|---|---|
| Daily, per shift | 5 min | The seven checks in Table 3 | Machine operator |
| Weekly, Saturday end of shift | 15 min | Window removed and inspected against a light; filter mat checked; cable and connector end face inspected; log reviewed for repeat entries | Shift leader |
| Monthly, first Monday | 45 min | Radiator blown out from the inside; coolant sampled; galvo noise baseline recorded; power meter reading against commissioning; spare window and nozzle stock counted | Maintenance |
| Quarterly | 2 h | Full optical path inspection, focus calibration check, parameter and firmware backup | Service partner or trained in-house tech |
Table 4 — Four cadences. Only the first one is non-negotiable, and it is the one that saves the money.
Both came from people I respect, and both were wrong in the same way: they priced the routine and forgot to price the event.
"Fifty hours a year of operator time is expensive."
It is $600. One lens event is $2,175 once you count the four and a half stopped hours. You are not choosing between spending and not spending — you are choosing between a small certain cost and a large uncertain one.
"We already have a service contract."
A contract covers the repair. It does not cover the four and a half hours, the missed Friday shipment, or the customer who now wants weekly progress photos. Response time is not uptime.
Do not buy anything for this. Do not write a new policy document. Run it for thirty days on one machine and let the log make the argument for you:
If you want the printable A4 version of Table 3 — same pass criteria, same fail actions, the layout we tape to our own machines — ask us and we will send the PDF. And if you would rather start with the parts that fail most often, a spare protective window and nozzle kit costs less than one hour of your production time.