What 'Continuous' Actually Means in Electrical Monitoring
Most monitoring systems poll once per second and call it continuous. Real continuous monitoring captures 8,000 samples per second. Here is what you miss at 1 Hz.
"Continuous monitoring" appears in every BMS product brief, every DCIM spec sheet, and every RFP response in the data center industry. The term has been used to describe systems that poll at 1 Hz, systems that poll at 0.1 Hz, and systems that poll at 10 Hz. It has been applied to SNMP traps that fire on threshold crossings and to Modbus registers read once per minute.
The word "continuous" in these contexts means "the system is always running." It does not describe the temporal resolution of the measurement. This distinction matters because the temporal resolution determines which physical phenomena are observable, and the phenomena that cause equipment failure occupy a specific frequency range that most monitoring architectures cannot resolve.
Defining the measurement gap
The core issue is quantitative. A 60 Hz power system produces 60 complete electrical cycles per second. Each cycle lasts 16.67 ms. Within each cycle, the current waveform contains harmonic components whose amplitudes encode information about the health of the equipment drawing that current.
A monitoring system's sampling rate determines the cycle-resolution it can observe:
| Polling rate | Cycles observed per second | Waveform resolution per cycle | Harmonic content resolvable | Observable phenomena |
|---|---|---|---|---|
| 0.017 Hz (1/min) | 0.017 of 60 | None (single aggregate value) | None | Energy consumption only |
| 1 Hz | 1 of 60 | None (single aggregate value) | None | Load level, phase balance, power factor at fundamental |
| 60 Hz | 60 of 60 | 1 sample per cycle | None (requires >1 sample per cycle for waveform shape) | Cycle-to-cycle RMS variation |
| 1 kHz | 60 of 60 | ~17 samples per cycle | Up to 8th harmonic | Coarse harmonic content |
| 8 kHz | 60 of 60 | 133 samples per cycle | Up to 66th harmonic | Full harmonic decomposition, transient characterization |
The distinction between 1 Hz and 8 kHz is not a matter of precision (more decimal places on the same measurement). It is a matter of dimensionality. At 1 Hz, the measurement is a scalar: the RMS value of whatever happened during the polling instant. At 8 kHz, the measurement is a vector: 133 time-domain samples that decompose via DFT into amplitude and phase at every harmonic order through the 66th.
The scalar tells you how much. The vector tells you what.
Why the waveform shape matters
Electrical equipment health manifests in the frequency domain, not the time domain. Two rectifiers can draw identical RMS current while having completely different harmonic spectra: one healthy, one degrading. The difference is in the waveform shape, in how the current is distributed across harmonic frequencies.
The mechanism is impedance change. Every degradation mode changes the impedance of one component in the power path, which changes the current waveform at the frequencies where that impedance matters.
Example: capacitor ESR increase in a 6-pulse rectifier. The DC bus capacitor presents an impedance of Z_C = ESR + 1/(j * 2 * pi * f * C) at each frequency f. As ESR increases from 15 to 45 milliohms (typical end-of-life trajectory for aluminum electrolytic capacitors at 85C over 5-8 years), the capacitor's impedance at 300 Hz (5th harmonic) increases by approximately 0.3%, while its impedance at 420 Hz (7th harmonic) increases by approximately 0.2%. These are small changes. But they shift the commutation overlap angle by 2-4 degrees, which redistributes 1-3% of the harmonic current from the 5th order to the 7th order. Over three weeks, this produces a measurable, monotonic shift in the H5/H7 ratio.
The RMS current does not change. The power factor does not change. The voltage does not change. The 5th-to-7th harmonic ratio changes, and that change becomes visible only when the waveform is decomposed into its harmonic components, which requires sampling at a rate that resolves those components — at least 3 kHz, per the Nyquist–Shannon sampling theorem for the 25th harmonic at 1,500 Hz. Harmonic measurement methodology for power-quality monitoring is standardized in IEC 61000-4-7, which the industry uses for compliance reporting.
At 1 Hz, this degradation is invisible regardless of how precisely the RMS measurement is made. The information does not exist in the RMS value. It exists in the waveform shape.
Three categories of phenomena 1 Hz misses
Sub-cycle transients (millisecond timescale)
A voltage transient lasting 5 ms has a 0.5% probability of being captured by a 1 Hz polling system (5 ms / 1,000 ms polling interval). In practice, most short transients are never recorded. At 8 kHz, the same 5 ms transient produces 40 samples, sufficient to determine peak magnitude, rise time, decay constant, and oscillatory behavior.
GPU workload transitions produce load steps that complete in 8-10 ms. These create transient voltage disturbances at the PDU output that stress capacitors and insulation. At 1 Hz, these transients are sampled approximately once per 100-120 occurrences. The other 99% are missed entirely.
Harmonic trending (day-to-week timescale)
Degradation trends unfold over days to weeks, but the measurement that reveals them must be made at sub-cycle resolution. The trend is in the harmonic amplitude, which can only be computed from a waveform captured at sufficient resolution.
A useful analogy from medicine: blood pressure trending (days-to-weeks timescale) requires an individual blood pressure measurement (seconds timescale). You cannot trend blood pressure by measuring body weight. Similarly, you cannot trend harmonic amplitude by measuring RMS current. The quantity being trended and the aggregate measurement operate at different levels of abstraction.
Physics-validated anomaly detection (statistical)
Fleet-level physics-validated anomaly detection requires computing spectral features for each unit and comparing them against the fleet distribution. This is a statistical technique: the unit whose 7th harmonic sits at the 98th percentile of the fleet distribution is flagged, regardless of whether any absolute threshold has been crossed.
At 1 Hz, there are no spectral features to compare. The fleet distribution is over RMS values, which are far less discriminating. Two rectifiers can have identical RMS current (within measurement uncertainty) while having 7th harmonic amplitudes that differ by a factor of 2. The spectral anomaly is detectable. The RMS anomaly is not.
The Nyquist argument
The case for continuous waveform capture is not a product argument. It is a sampling theory argument.
The diagnostically useful information for equipment health lives in harmonic components up to the 25th order (1,500 Hz on a 60 Hz system). Per the Nyquist-Shannon theorem, resolving those components requires sampling at a minimum of 3 kHz. At 1 Hz, you are undersampling by a factor of 3,000. No signal processing algorithm can recover information from samples that were never taken.
This is the same principle that makes a 44.1 kHz audio recording faithful to the original sound (Nyquist limit of 22.05 kHz, above the human hearing range) while a 1 kHz recording captures nothing recognizable. The physics does not negotiate.
At 8 kHz, every electrical cycle is fully resolved. A UPS rectifier whose 5th harmonic amplitude shifts by 2% over three weeks is visible in the per-cycle spectral data, even though the BMS reading (RMS current, voltage, power factor) shows no change. The degradation information exists in the waveform shape, not in the aggregate values. A 1 Hz system captures the aggregate. An 8 kHz system captures the shape.
This is not a comparison of system quality. A 1 Hz BMS captures exactly the data it was designed to capture and reports it accurately. The degradation information is absent from that data, because it exists at frequencies that 1 Hz cannot sample.
What "continuous" should mean
A system that is always running is operationally continuous. A system that captures every electrical cycle at waveform resolution is metrologically continuous. The distinction matters because equipment health detection requires the second definition.
Proposed operational definition: continuous electrical monitoring captures sufficient data on every electrical cycle to compute per-cycle harmonic amplitudes through at least the 25th order (1,500 Hz on a 60 Hz system). This requires a minimum sampling rate of 3 kHz per Nyquist, with practical implementations at 8 kHz or above to provide margin for anti-aliasing and amplitude accuracy.
By this definition, systems polling at 1 Hz, 10 Hz, or even 60 Hz (one sample per cycle) are not continuous in the diagnostic sense. They are continuous in the operational sense (always running) but not in the metrological sense (capturing the full information content of each cycle).
This is not a criticism of 1 Hz systems. It is a description of what they measure. For load tracking, energy accounting, capacity planning, and basic fault detection (open circuit, ground fault, overcurrent), 1 Hz is entirely adequate. For degradation detection, it is structurally insufficient, and no amount of algorithmic sophistication can recover information that was never captured.
Request a demo to compare what 8 kHz measurement reveals on your circuits.