Backup Power: A Practical Guide to Batteries, UPS & Generators

NFPA 110 classifies emergency and standby power systems along two genuinely separate axes that are easy to conflate into a single number but describe completely different things. Type describes how fast the system must restore power — Type U (uninterrupted, effectively 0 seconds), Type 10 (within 10 seconds), Type 60, Type 120, or Type M (manual start, no automatic time requirement). Class describes how long the system must be able to run at its rated load without refueling or recharging — Class 0.083 (5 minutes) through Class 48 (48 hours) up to Class X (unlimited, i.e. connected to a continuous fuel supply). A complete system specification needs both a Type and a Class stated together (for instance, 'Type 10, Class 2' means power must be restored within 10 seconds and the system must be able to sustain rated load for at least 2 hours) — quoting only one of the two leaves the specification genuinely incomplete.

NEC further separates standby systems into three priority tiers based on what's actually at stake if power is interrupted. Emergency systems (NEC Article 700) cover life-safety loads — egress lighting, fire alarm, and similar — and carry the strictest maximum permitted interruption time (commonly cited around 10 seconds). Legally required standby systems (NEC Article 701) cover loads a jurisdiction's code requires backup for but that aren't immediately life-safety critical (certain HVAC, communications, or industrial process equipment) and get a more relaxed maximum interruption time (commonly around 60 seconds). Optional standby systems (NEC Article 702) cover loads a facility chooses to back up for business continuity reasons — with no code-mandated maximum interruption time at all, since the decision to back them up (and how quickly) is the facility's own choice rather than a life-safety code requirement. Sizing and sequencing a backup system without recognizing which tier a given load actually belongs to risks either overspending on non-critical loads or, worse, under-protecting a genuinely life-safety-critical one.

A facility can have a correctly sized UPS and a correctly sized generator and still have a real gap in coverage if the timing between them doesn't actually add up. The sequence after a utility outage typically runs: the automatic transfer switch (ATS) waits a short confirmation delay before deciding the outage is real, the generator then has to crank and reach rated speed and voltage, and the ATS then has its own transfer time to actually switch the load over — and the UPS's battery autonomy has to cover the entire sum of that sequence, with margin, or the load loses power in the gap between the UPS running out and the generator actually being ready to carry it. This bridge-timing check is exactly why it's treated as a distinct calculation in this category rather than assumed to work out automatically just because both a UPS and a generator are present.

Sizing a UPS's own battery bank for a straightforward, roughly constant-current backup period is a reasonably simple estimate: usable energy is approximately load times time, and nameplate battery capacity is usable energy divided by depth-of-discharge and round-trip efficiency. But a genuinely rigorous battery design — particularly for DC systems with a duty cycle that varies significantly over time, such as an initial high current spike for closing a breaker or annunciating alarms followed by a much lower sustained standby load — needs IEEE 485's full duty-cycle section method instead. That method breaks the discharge into multiple current/duration sections and applies a capacity-rating factor (Kt) specific to the battery product's own published data, accounting for the fact that a battery doesn't deliver a simple fixed amp-hour capacity regardless of how fast it's discharged — faster discharge rates extract proportionally less total capacity than slower ones. The quick UPS estimate is a reasonable first-pass sizing tool; the dedicated battery calculator is the right tool once the duty cycle is well-defined and the design needs to be load-tested and specified against a real product's data sheet.

A battery's nameplate (rated) capacity is never fully usable in practice — depth of discharge (DoD) limits how far the battery can be discharged without significantly shortening its service life, and round-trip efficiency (RTE) accounts for energy lost internally during charge and discharge. Both vary meaningfully by chemistry: lithium-ion batteries commonly tolerate a high DoD (around 90%) with high round-trip efficiency (around 92%), lead-acid batteries are typically limited to a much shallower DoD (around 50%) to preserve cycle life with somewhat lower efficiency (around 85%), and flow batteries can often discharge fully (100% DoD) but at a meaningfully lower round-trip efficiency (around 75%) due to their different underlying electrochemistry. These are chemistry-typical defaults, not universal constants — always confirmed against the specific manufacturer's own data sheet — but the underlying principle holds regardless of chemistry: usable energy is always some fraction of nameplate capacity, never the full nameplate figure, and that fraction is chemistry-dependent.

Depth of discharge also trades off directly against cycle life for most battery chemistries — discharging more deeply on every cycle generally reduces the total number of cycles the battery can deliver before its capacity degrades below a useful threshold, which is why chemistry and DoD selection is as much an economic decision (upfront capacity cost vs. expected replacement interval) as a purely technical one.

A generator's fuel consumption rate genuinely depends on how heavily it's loaded, not just its nameplate rating — running significantly below rated load is often noticeably less fuel-efficient per kWh delivered than running closer to rated capacity, since a generator engine has real fixed losses that don't scale down proportionally with load. Sizing fuel storage and estimating running cost from a generator's actual consumption rate at its expected operating load (rather than assuming a single generic rate regardless of loading) gives a meaningfully more accurate runtime and cost estimate than a flat nameplate-based assumption would.