The Eaton UPS Sizing Mistake That Cost Us $12,000 (And the Pre-Check List That Prevents It Now)
In March 2019, I was managing a data center expansion for a manufacturing plant outside Cleveland. Nothing huge—20 racks, some network gear, environmental monitoring for the industrial floor. The critical load totaled about 13kW.
I ran the numbers. Specified a 20kVA Eaton UPS. That gave me roughly 35% headroom, which felt responsible. Order went in. Install was scheduled.
Then the load test happened.
The chiller kicked on during the first full-load simulation, and the UPS transferred straight to bypass. About 90 seconds of dark in the server room. Ninety seconds that cost us roughly $4,000 in lost production time and recovery efforts.
But the real damage came after. Because the error wasn't what I thought it was.
What I Actually Got Wrong
I assumed I'd undersized the unit. That's the obvious conclusion, right? Buy a bigger box next time.
Not quite. When the electricians pulled the logs, the UPS was running at 117% of its rated capacity at the moment of transfer. My 20kVA unit was trying to carry 23.4kVA of load.
Here's where it gets embarrassing: my load calculation wasn't wrong. The 13kW figure was accurate. But I'd done the math in kilowatts and bought a unit rated in kilovolt-amps. For the IT load alone, the power factor was around 0.92, which is fine—13kW ÷ 0.92 = 14.1kVA. Comfortable.
What I forgot was the mechanical load. The plant insisted on backing up the environmental controls too. Those compressors run at a power factor closer to 0.72. And I never accounted for the starting current.
Two separate mistakes, same root cause: I treated "13kW" as a single number. It wasn't. It was a collection of loads with different characteristics, and I'd applied a one-size-fits-all multiplier. (I should mention: the chiller's inrush current was the actual trigger. That motor draws 6x its running current for about 3 seconds on startup. The UPS saw that spike and noped out.)
Never expected the fix to be a load study, not a bigger UPS. Turns out the right 20kVA unit—correctly configured for the actual load profile—would have handled it fine. We'd just specified the wrong input parameters.
The Other Three Things Nobody Warned Me About
Once we started untangling the mess, we found three more issues sitting quietly in the spec, waiting to cause problems.
1. The Zenith manual transfer switch wasn't rated for inverter output.
We'd specified a Zenith manual transfer switch for maintenance bypass. Standard practice. What nobody checked was whether the switchgear was rated for the synthesized sine wave coming out of the UPS. Some manual switches are designed for utility-grade pure sine wave. When fed a UPS output, the contacts can chatter or arc. Ours did both during the second test. Replacement switchgear: $2,800 and three days of schedule slip.
2. The onboard battery charger was set to default.
The UPS came with a charger configured for a 10% charge rate. That's fine for shallow discharges. But our batteries took a 40% depth-of-discharge hit during the transfer event. At 10% charge rate, they needed 14 hours to recover to 90% capacity. Had anyone configured the charger for a higher rate (within the battery's spec, obviously), we could have cut that to 6 hours. This one didn't cost us money directly—but it did mean the UPS was running at reduced autonomy for most of a day. (Note to self: add "charger configuration" to the commissioning checklist.)
3. We tested every circuit breaker under power, but not without it.
This one still bugs me. During commissioning, we ran every live test on the schedule. Load tests, transfer tests, fault simulation—all with the system energized.
What we didn't do was verify the circuit breaker coordination in a de-energized state. That's the test where you confirm that the breakers you think are coordinated actually trip in the sequence you expect. Per NFPA 70 (National Electrical Code) Article 700.32, selective coordination is required for emergency systems—but verifying that coordination is a separate step from designing it. We found out during a parallel maintenance window that two upstream breakers were misrated for the UPS feeder. $800 in replacement breakers and another day of downtime.
If you've ever looked up how to test a circuit breaker without power, you know the procedure is straightforward—trip each breaker, verify continuity, check the trip curve against the coordination study. It takes maybe 45 minutes for a system like ours. We skipped it because it wasn't in the commissioning template.
What It Actually Cost
Running total:
- UPS upgrade (20kVA to 30kVA): $5,300
- Rewiring and re-termination: $2,700
- Replacement transfer switch: $2,800
- Breaker replacements: $800
- Emergency service call for charger configuration: $600
- Total: $12,200
Plus five days of schedule slip and a very uncomfortable conversation with the plant manager.
The bitter part? Every one of these errors was preventable with a 30-minute pre-check. Not a deep engineering analysis—just someone walking through the spec with a list and asking "did we verify this?"
The Checklist We Use Now
I built this after the third project where something similar (though smaller) happened. It's not glamorous. It's not even particularly technical. But it works.
Pre-purchase phase:
- Load profile broken down by type (IT, mechanical, resistive, etc.)—not just total kW
- Power factor per load type, with kVA calculated separately for each
- Starting current analysis for any motor-driven load (chillers, pumps, compressors)
- Run the Eaton UPS calculator (or equivalent) with the actual load profile, not the simplified version
Pre-install phase:
- Power quality requirements of every device downstream—does anything need pure sine wave?
- Transfer switch rating matched to UPS output waveform, not just current
- Battery charger configured for expected depth of discharge and recovery time
Commissioning phase:
- Breaker coordination verified in de-energized state before live testing
- Full-load transfer test with actual inrush conditions simulated
That's nine items. Five minutes of verification beats five days of correction—I can promise you that from experience.
For what it's worth, the Eaton 9PX series we ended up with has been solid since (the 30kVA unit, not the 20kVA one that started this whole mess). Their sizing calculator—the Eaton UPS selector tool—would have flagged the power factor mismatch if I'd used it with accurate inputs. I just didn't know to input the mixed-load profile correctly.
Looking back, I should have paid for a proper load study before ordering anything. At the time, the $1,800 quoted price seemed excessive for "just running some numbers." It wasn't. $1,800 in pre-work would have saved $12,200 in rework.