Root mean square RMS

What it is (precise)

The name is the recipe read backwards: square the waveform, take the mean over a cycle, take the root. The squaring is not a mathematical convenience — it is the physics of heat. Instantaneous power in a resistance is i² × R, so the average heating over a cycle tracks the mean of the squared current, and the square root simply returns the result to current units.

The consequence is that P = I²(rms) × R holds exactly, and the working power expressions — P = V × I × cos φ for a single-phase circuit, P = √3 × V(L-L) × I(line) × cos φ for three phases — are true only when the voltages and currents in them are RMS. Feed peak values into either one and the answer comes out a factor of two high (√2 × √2), which is the kind of error that survives a design review because every individual number in the calculation looked plausible.

For an undistorted sinusoid the ratios are fixed: RMS = peak/√2 = 0.707 × peak, and peak = √2 × RMS. The plant's numbers follow directly. A 690 V line-to-line PCS output swings to about ±976 V; a 400 V auxiliary bus peaks at 566 V; a 230 V line-to-neutral control supply at 325 V.

The definition also handles DC without a special case — the RMS of a constant current is the current itself — which is why a fuse's pre-arcing time-current curve, plotted against RMS current, can serve the AC and DC sides of the plant from a single characteristic: melting a fuse element is I²t heating, and RMS is the current axis heating understands.

That split — heating versus instantaneous value — is the clean way to sort every electrical quantity in the project. Anything that responds to average power over seconds and minutes is an RMS-denominated quantity: conductor ampacity, transformer temperature rise and MVA rating, switchgear continuous current, fuse melting.

Anything that responds to the waveform's instantaneous height reads peaks instead: insulation stress and withstand levels, semiconductor blocking voltage, the modulation limit of the converter, the first asymmetric loop of a short-circuit current. Datasheets follow the same split, mostly without saying so.

The datasheet convention — AC figures are RMS unless marked

Take a 3,600 kVA PCS block at 690 V: the implied line current, S/(√3 × V), is about 3,000 A, and every one of those three numbers is RMS. The convention is universal because the ratings are thermal: the 690 V is the RMS the windings are wound for, the 3,000 A is the RMS the busbars and cable lugs must carry without exceeding their temperature class, and the kVA is their product times √3.

When a figure on the same sheet is not RMS, it carries its own label — rated impulse withstand voltage is a peak, short-circuit making current is the peak of the first asymmetric loop, and the DC input range is DC. A reader who assumes RMS-unless-stated and then actually checks the stated exceptions gets every comparison on the sheet right.

Measurement inherits the convention. The quantities the project is contractually held to — voltage, current, power and power factor at the point of interconnection, the energy the capacity test settles on — are RMS values computed by revenue meters and Class A power-quality instruments from sampled waveforms.

The handheld multimeter that disagrees with them during commissioning is often not broken: an average-responding meter reads the rectified average scaled by 1.11, a calibration that is exact for a clean sinusoid and wrong for anything else. On inverter-fed circuits, UPS outputs and harmonic-rich auxiliary feeders, specify true-RMS instruments and treat any average-responding reading as unlabelled.

The boundary the convention cannot cross is the DC bus. The same 3,600 kW moving through a block is roughly 2,700 A of DC at a 1,330 V nominal string voltage and roughly 3,000 A of AC RMS at 690 V — and no fixed factor converts one current into the other, because the DC voltage moves with state of charge and temperature while the AC voltage does not.

Each side's copper is sized to its own current and both heat by I²R; the error to avoid is reading a DC current spec against an AC one, or an AC voltage against a DC one, as though the numbers shared a definition. They share units, which is what makes the mistake easy.

Peak, the DC bus and the √2 the plant is built around

An inverter synthesizes its AC output from the DC bus, and it cannot produce an instantaneous voltage the bus does not reach. For a 690 V RMS line-to-line output the waveform peaks near 976 V, so the bus must sit above that height — before modulation and control margin — at every instant of every cycle.

Run the comparison against the battery's own numbers and the geometry of the DC window appears: a 1500 VDC-class string at 1,150-1,330 V nominal clears the peak comfortably, but the low-SOC floor of roughly 900-1,040 V, reached cold and under load, straddles it. That is the physical reason a PCS publishes a full-power DC range narrower than its operating DC range — the VDC window entry shows the structure — and why the bottom of the battery's voltage window buys continued operation but not rated output at rated AC voltage.

Peaks also own the insulation. Creepage, clearance and withstand-test levels for the 690 V system are coordinated against the waveform's peaks and the switching transients on top of them, not against the RMS label on the nameplate; the semiconductors' blocking voltage is margined over the DC bus, an instantaneous quantity throughout. So the √2 is built into the hardware twice — once between the RMS rating and the waveform the machine must synthesize, once between that waveform and the insulation that contains it — and the nameplate shows none of it. RMS on the front page, peaks in the design.

Crest factor — when √2 stops being the number

The ratio of peak to RMS has a name, crest factor, and √2 is merely its value for one waveform: the undistorted sinusoid. DC has a crest factor of 1; a distorted current can carry a substantially higher one, gaining peak height without gaining heating value, because peak and RMS respond to waveform shape independently.

Real PCS current always carries some harmonic content, and the percentage distortion is worst at light load — so the √2 shortcut from measured RMS to assumed peak is least reliable exactly where a lightly loaded plant spends its idle hours. The spectra, the limits and the mitigation belong to the harmonics and total harmonic distortion entries; what this entry owns is the rule that peak-from-RMS is a waveform-shape assumption, to be checked before it is used.

The practical bite is in measurement and specification. Average-responding meters err off-sine, as above; true-RMS instruments have their own crest-factor limits, beyond which they clip the peaks and under-read the RMS — worth checking before trusting a reading on a waveform a converter built. And when a study or a test procedure needs a genuine peak — for insulation coordination, for instrument sizing, for a protection setting — the defensible route is to measure or compute it from the actual waveform, not to multiply an RMS reading by 1.414 and hope the shape cooperates.

Common pitfalls

The expensive one is comparing a DC value with an AC RMS value digit-for-digit. A 1,330 V bus against a 690 V output looks like a two-to-one margin; the honest comparison is bus against peak — 1,330 V against 976 V — and at the cold, loaded, low-SOC corner the margin is a fraction of what the RMS labels suggested.

The mirror errors are just as common: applying √2 to a figure that is already a peak, or to a DC figure, and double-converting a number somebody upstream already converted. The fix is procedural, not mathematical — every voltage and current in a calculation gets a label (RMS, peak or DC) before any arithmetic happens, and any unlabelled AC figure from a datasheet is read as RMS, which is what its author meant.

Two smaller traps. "RMS power" is not a physical quantity: the kW and MW on every datasheet and settlement statement are average real power, computed from RMS voltage, RMS current and the power factor between them — a document that says "RMS power" means average power, and a calculation that actually takes the RMS of an instantaneous power waveform produces a number with no engineering use.

And the √2 relation itself is the sine-only special case of crest factor, not a property of AC: on any waveform a converter has shaped, the trustworthy RMS is the one a true-RMS instrument measured, and the trustworthy peak is the one you looked at.