Data Center UPS Monitoring

Power remains the leading cause of significant data center outages. According to the Uptime Institute's 2025 Annual Outage Analysis, power issues drove 45% of impactful outages in 2025, most often from UPS issues, and more than half of operators reported their most recent significant outage cost more than $100,000. One in five said it cost more than $1 million.

So the obvious question: if every operator I talk to monitors their UPS systems, why do UPS-related outages keep showing up at the top of the list?

The short version is that "UPS monitoring" and "UPS health monitoring" are not the same thing. Research Jon Chu and I published last year in our white paper on UPS power infrastructure challenges kept landing on the same finding: standard UPS monitoring is built to tell you whether a unit is on. It is not built to tell you whether the unit is healthy. Those are different questions, and they need different measurements.

What your UPS monitoring actually checks

Worth being specific about this, because most operators I talk to assume their monitoring covers more than it does. Here's what your standard monitoring actually checks via SNMP or Modbus:

Parameter	What it tells you	What it misses
Online/offline status	Whether the UPS is active	Nothing about internal component health
Battery charge level	Whether stored energy is available	Battery impedance changes indicating degradation
Output voltage	Whether output is within spec	Subtle waveform distortion from rectifier issues
Output current	Whether load is within capacity	Harmonic content changes indicating degradation
Temperature	Whether cooling is adequate	Localized heating from component stress
Alarm status	Whether any threshold has tripped	Degradation below alarm thresholds

Every parameter on that list is valuable, and you should keep monitoring them. Together they answer one question well: is this UPS currently working? But look at the right column. That's the degradation gap. Standard monitoring catches the moment of failure. The weeks of degradation that lead up to it sit outside what these parameters can see.

Why it is a physics problem, not a configuration problem

UPS rectifiers don't fail instantaneously. They degrade. Electrolytic capacitors in the DC link lose capacitance over time, switching devices develop timing drift, and these changes show up in the electrical waveform long before they affect performance.

Here's where the resolution problem comes in. At 1 Hz polling, your BMS captures one data point per second. Power runs at 60 Hz, which means each polling sample is one snapshot out of 60 cycles, and the harmonic signature changes that indicate rectifier degradation live in the sub-cycle detail between samples. You could tune your alarm thresholds to any sensitivity you want and it wouldn't matter, because the information isn't in the data stream you're monitoring. It lives at frequencies above what 1 Hz monitoring is designed to read.

I've had this conversation more times than I can count with operators who spent months tuning their alarm thresholds, trying to catch degradation earlier. It wasn't a threshold problem. The data they needed wasn't there to begin with.

What 8 kHz waveform monitoring sees

The work Jon and I did kept turning up the same thing: the degradation modes that cause UPS outages express themselves in the electrical waveform weeks before they affect performance. Capacitor ESR rises, commutation timing shifts, and harmonic content at the 5th and 7th orders drifts from baseline. These shifts are gradual, sit below every alarm threshold, and fall outside what standard monitoring is designed to see. Published research on capacitor health monitoring backs this up: harmonic-domain measurement is the most reliable way to predict ESR-driven failures in rectifier circuits.

At 8 kHz (8,000 samples per second, continuous), these waveform changes get captured on every electrical cycle. In Verdigris field deployments, that's given us a typical 21-day advance warning window before functional failure on degrading rectifiers, validated across multiple customer sites.

Status checking vs. health monitoring

The difference is in what question you ask.

Standard monitoring asks: "Is this UPS on, charged, and within spec?" If all three answers are yes, the UPS gets called healthy.

Waveform monitoring asks a different question: "How is this UPS behaving compared to its own baseline and compared to its peers?" A unit can be on, charged, and within spec while its harmonic profile diverges from fleet norms. That divergence is the early warning.

Here's what's detectable per failure mode:

Degradation mode	Harmonic indicator	Detection method
Capacitor degradation	Elevated 5th + 7th order THD	Trend analysis against unit baseline
Switching device wear	Harmonic envelope shift at characteristic frequencies	Fleet-wide comparison
Phase imbalance	Asymmetric harmonic content across phases	Per-phase waveform analysis
Cooling degradation	Temperature-correlated harmonic drift	Combined thermal + electrical trending

Each of these is detectable weeks before the unit fails. Standard monitoring catches none of them until the degradation produces symptoms severe enough to trip an alarm. By then your lead time is typically minutes to hours, not weeks.

How the two systems work together

Waveform monitoring isn't a replacement for your existing UPS monitoring. You still need the SNMP/Modbus data for BMS integration, capacity tracking, and basic alarming. Status checks serve their purpose well.

What waveform monitoring adds is the pre-failure detection layer, telling you which specific units in your fleet are degrading while there's still time to schedule maintenance during a planned window rather than at 2 AM when the NOC is scrambling.

This is what we built Verdigris EV2 meters for: non-invasive install on UPS output circuits, no downtime, no modification to existing systems, with derived signals flowing into whatever observability stack your team is already using. The Signals platform analyzes the waveform data and generates early warnings with diagnostic context: which unit, what is changing, and how it compares to the rest of the fleet.

Standard monitoring tells you the UPS is on. Waveform monitoring tells you it's healthy. You need both. Together, they close the gap between "working" and "working safely."

The question I'd ask

If your UPS fleet is critical to your operations (and it is, or it wouldn't be a UPS), do you know the health status of each unit right now? Not the alarm status. The health status.

If you can't answer that with per-unit specificity, there's a detection gap between what you're monitoring and what you actually need to know. The Uptime Institute data suggests that gap is where a lot of expensive outages live.

For the full technical detail on physics-based UPS failure detection, see the UPS Failure Prediction solution page.

Request a demo to add waveform-level health monitoring to your UPS fleet.