UPS behaviour under the load you actually have
An uninterruptible power supply is a power converter plus stored energy plus a set of transfer rules. The brochure lists VA and minutes. The process cares whether voltage stays inside the ITI (CBEMA) envelope, whether the bypass is synchronized, whether the battery string can deliver current at the temperature you have, and whether nonlinear load current makes the inverter current-limit before the watt rating is reached. UPS and critical power investigation is about that behaviour on the load you run, not a resistor in a factory.
How the investigation is actually run
Walk the output distribution before you touch the UPS menus. Many “UPS failures” are loads that were never on the inverter, or were moved during a weekend outage and never moved back. Label what you find. Then capture input and output during a planned transfer if the site will allow one, and during whatever nuisance event they already have. The mode bits — online, battery, bypass, fault — must be in the same time base as the voltage waveform.
Battery work is physical: temperature, connection infrared, and a discharge that the process can tolerate. A displayed minute count is not a test. Nonlinear load is a current-limit test: if the inverter hits current limit while watts look modest, the load mix is the story, not a “weak UPS” slogan. Source distortion that forces bypass belongs in the power-quality file, not in a firmware complaint.
Write what the process saw: stayed inside a named voltage envelope, rebooted, or transferred so slowly that a ride-through spec was fiction. Name the next owner of the work: settings, batteries, load moves, or upstream electrical treatment. Do not publish an uptime percentage.
Symptoms
Servers reboot during a “successful” transfer. Lighting or controls drop while IT stays up — different distribution, different ride-through. The UPS goes to bypass on motor starts or on elevator banks sharing a feeder. Runtime on the display does not match a controlled discharge. Fans and heat rise while kW looks modest: crest factor and current limit. Network gear that browns out because it was never on the UPS the drawing claimed — a critical digital infrastructure adjacency problem.
Causes
Load that is not the design mix: more switched-mode supplies than the harmonic budget, more inrush than the static-switch rating. Bypass out of sync or a source too distorted for the inverter to stay on (see harmonic distortion). Battery aging, imbalance, or temperature. Firmware that transfers on a threshold that is tighter than the process needs — or looser, so the load sees a sag the UPS “ignored.” Neutral issues on the output. Maintenance bypass left in a state nobody labelled.
Investigation
Log input voltage, output voltage, current, and UPS mode bits through a sag, a transfer test, and a slice of real process. Capture waveform at transfer, not only an event flag. Measure actual load crest factor. Walk the output panel: what is really on the inverter, what is on the same utility bus, what is on a different UPS. Infrared on batteries and power connections belongs in the same visit as thermography as evidence.
A conclusion states observed mode changes, whether the output stayed inside a named envelope, and which limit was hit (current, sync, battery voltage, thermal). It does not invent minutes of runtime or a percentage of “uptime.” Next actions are settings, load moves, battery work, or source treatment — scoped in writing. Related: interpreting power-quality data.