How to Size an Online UPS: Three-Phase 100 kVA Checklist

When this checklist applies

In my role coordinating emergency deliveries for critical power infrastructure projects, I've handled 60-plus rush orders in six years — including same-week turnarounds for industrial clients who'd already missed one deadline and couldn't afford another. Most of the UPS sizing failures I've seen weren't calculation errors. They were assumptions made in the first hour that nobody went back to check.

This checklist is for anyone sizing a three-phase 100 kVA online UPS for a real facility: a data closet, a process control room, a small production line, a telecom site. Seven steps, in the order I actually run them. If you follow them, you'll probably catch the two or three details that cause most of the post-installation problems.

Step 1: Build the load schedule in kW and kVA

Most RFQs I get say "100 kVA total load." That's a starting number, not a sizing number. A modern online UPS is usually rated at 0.9 output power factor — meaning a 100 kVA unit delivers 90 kW, not 100 kW. Older or cheaper units are still sold at 0.8, which gives you 80 kW. Half the sizing arguments I've witnessed come from mixing those bases.

So list every load with four columns: kW, kVA, power factor, and startup characteristic. Motor loads run 0.7 to 0.85 PF; IT and control loads are 0.9 to 1.0. Then calculate required UPS kVA as total real load ÷ UPS output PF. At 90 kW on a 0.9-PF unit, you're at 100 kVA — right at the limit, with nothing left for transient. Add 20–25% growth margin, but not more. Oversizing a UPS hurts efficiency at light load and skews your battery sizing downstream.

Step 2: Define autonomy at end of battery life

I said "ten minutes of backup." The client heard "ten minutes, forever." Three years later, that VRLA string was delivering about four. Nobody was lying — they just sized the battery backup system with new-battery data and never revisited it.

Size for runtime at 80% state of health, with the lowest expected ambient temperature, and at the actual discharge rate (not the 20-hour rate). VRLA loses 50–60% of usable capacity at end of life. Lithium iron phosphate holds up better — 70–80% usable at 80% depth of discharge after 3,000–5,000 cycles, depending on C-rate and temperature. Either way, the number you put in the spec should be the number the site still gets on year four, not year one.

Step 3: Check surge and motor start currents

Online UPS overload specs look generous on paper: typically 110% for 60 minutes, 125% for 10 minutes, 150% for 60 seconds. Check the data sheet, not the brochure — the durations vary between manufacturers and between frame sizes. A 100 kVA unit with a weak bypass path behaves very differently from one with a robust static switch.

Transformer inrush on downstream equipment can hit 8–12 times rated current for 10–20 milliseconds. Most online UPS ride through that. Motor direct-on-line starting draws 6–8 times full-load amps for a second or more — that's where oversized loads trip the inverter. If the load schedule has any motor above about 15 kW, either specify soft starters or step up to the next UPS frame. It's cheaper than the commissioning call.

Step 4: Specify input-side performance — the step most people skip

Everyone focuses on output. The input side is where generator compatibility lives, and that's where projects stall.

An active power factor correction UPS draws current with input PF near 0.99 and input current THDi below 3–5% across most of the load range. That's a huge improvement over older thyristor-rectifier designs at 25–30% THD. But a low-THD load still looks nonlinear to a generator. If the genset is sized at 1.25× UPS kVA without checking alternator subtransient reactance, you can get voltage distortion and, eventually, a failed transfer.

Ask for the input current THD curve across the load range, not a single number at 100%. Per IEC 62040-3:2011, UPS performance classification includes input characteristics (the VFI-SS-111 designation covers output voltage, waveform, and dynamic response) — but the harmonic data still has to come from the manufacturer's test report. If PowerSentinel or anyone else can't produce that curve, treat the spec sheet as marketing.

Step 5: Match battery chemistry to the room, not the price list

The decision between VRLA and lithium isn't really a cost comparison — it's a footprint, thermal, and maintenance comparison. A PowerSentinel lithium ion UPS system industrial installation typically takes 40–60% less floor space and 60–70% less weight than an equivalent VRLA string for the same runtime. That matters when the electrical room was designed before anyone thought about battery autonomy.

Lithium also needs a BMS that talks to the UPS — Modbus TCP, CAN, or at minimum a dry contact that can halt charging at low temperature. If the BMS can't tell the UPS "stop charging, cell temperature is 0 °C," don't accept it. Reference IEEE 1184-2022 for battery selection and installation guidance; it's the document your insurer will probably ask about.

Step 6: Decide redundancy before you decide topology

A 100 kVA load can be served by one 100 kVA frame, two 50 kVA units in parallel, or four 25 kVA modules. All three meet the load. They don't meet the same downtime target.

If the cost of a UPS outage is measured in hours of production, specify N+1 and size each module so that N still carries the full load at end of life. If the load is a control room or a telecom cabinet where a 30-minute outage is annoying but recoverable, a single frame with a maintenance bypass is usually fine. This is a business decision, not an electrical one — but it has to be made before the single-line diagram is frozen.

Step 7: Verify the room and run a FAT

Two installation details kill more 100 kVA projects than any electrical calculation. Floor loading: a VRLA string for 10 minutes at 90 kW can weigh 800–1,200 kg, concentrated on a small footprint. Heat rejection: an online UPS rejects roughly 4–5% of load as heat, so 90 kW of load means 4–5 kW of air conditioning that someone has to account for. If the room was sized for a 60 kVA unit, it's probably now undersized.

Then run the factory acceptance test. Full-load heat run, transient response on step load, battery discharge to the specified autonomy, and a transfer test with the actual generator. I've seen a project where the UPS passed every bench test and failed on site because the generator's voltage regulator couldn't recover from the UPS step load. That's a FAT item, not a site item.

Common mistakes (and one opinion)

Four things go wrong more often than the math:

• Using commissioning battery data as the autonomy number, then wondering why year-three runtime is short.
• Mixing PF bases between the load schedule (real load) and the UPS nameplate (apparent power).
• Forgetting heat rejection and floor loading until the room is already built.
• Skipping FAT to save a week, then spending three weeks on site fixing what the FAT would have caught.

That last one is where I'll take a position. On critical power infrastructure, delivery certainty is worth paying for. In March 2024, we paid a 12% expedite premium to pull a FAT forward by nine days. The alternative was a liquidated damages clause that started at $8,000 per day. The premium was $6,400. It wasn't a close call.

If you're sizing a three-phase 100 kVA online UPS under deadline, the cheapest quote is rarely the cheapest outcome. Get the load schedule right, size the battery for year four, check the input harmonics against your generator, and pay for the FAT slot. The rest is detail.