Dq Transform (Three-Phase Current Id/Iq)

Resolves three-phase current U/V/W into Id / Iq components through the Clarke → Park transform, and additionally derives electrical frequency f_e, electrical angle θ_e, mechanical speed, torque, the EPVA fault indicator, order colormaps, and a harmonic-injection analysis that normalizes vibration/noise by the same current f_e so acoustic orders can be compared against electrical orders on one axis.

U/V/W (A) ──Clarke──▶ i_α/i_β ──Park (rotate by θ)──▶ Id/Iq

Clarke uses the power-invariant (Concordia) form: i_α=√(2/3)(i_U−½i_V−½i_W), i_β=√(2/3)·(√3/2)(i_V−i_W). The vector magnitude |Idq|=√(i_α²+i_β²) is frame-independent — changing the frame, δ or the pole-pair count never moves it. That property is the best self-check you have for a mis-typed parameter.

This page documents every parameter on the property panel: what it means, how to choose a value, what a wrong value looks like on screen, and how the parameters interact. For the underlying concepts (Clarke/Park/FOC/PLL/EPVA, how to measure rotor angle, how harmonic injection works) see the companion document “Three-Phase Current Id/Iq Analysis · Fundamentals”.

Before you start: what to select in the Source pool

At least 3 current channels (U/V/W). With fewer than 3 the analysis aborts and the event log reads “Three-phase current required”.
For true rotor dq (Frame C): also select a measured rotor-angle channel (encoder, resolver-decoded angle, or the electrical-angle ramp exported by the MCU).
For harmonic-injection targeting / correlation: select the vibration and microphone channels too — every non-current channel is treated as an NVH sensor for order targeting.

How the three phases are identified

There is no U/V/W selector on the panel; the engine resolves them in this order:

Channel name contains current_u / current u, or ends with _u or space u, or equals u (same for V/W, case-insensitive);
No match → falls back to source channels 0/1/2.

The fallback is silent: if the first three channels are not U/V/W (or are swapped or wired with inverted polarity), you still get plots — with wrong numbers. Check the balance figure on the last log line: balance = rms(i_U+i_V+i_W) / rms(i_U), which should be near 0 for a balanced set. Above 10% the log prints “⚠ too large (check channel mapping / polarity / zero-sequence)”. The other symptom is a non-constant |Idq| and an i_α-i_β trajectory that is not a circle.

Parameter overview

Parameter	Default	Visible when	Affects
Frame (A/B/C)	`C`	always	the definition of Id/Iq (the whole analysis)
PLL Bandwidth (Hz)	`5`	Frame = B or C	smoothness vs. tracking speed of θ̂ / f_e
Pole Pairs	`4`	Frame = C	speed, absolute torque, mechanical-angle conversion
Align Offset δ (°)	`0`	Frame = C	split between Id and Iq
Flux Linkage ψ (Wb)	`0`	Frame = C	torque in N·m vs. relative torque
Max Order	`10`	“Order colormap” ticked	order-map ceiling + search ceiling for the order table / overlay
Angle Channel (-1=auto)	`-1`	Frame = C	measured vs. synthetic angle
Angle Full-scale	`6.283185` (2π)	Frame = C	channel value → radians
Angle is mechanical	off	Frame = C	whether to multiply by pole pairs
Plot types (9 tick boxes)	all off	frame-dependent	which curves / colormaps this run produces
Target orders	blank (auto)	“order overlay” ticked	which orders get overlay + null map

Frame (A / B / C)

The most important parameter — it defines what “Id/Iq” actually mean. Same currents, different coordinate system you observe them from.

Value	Definition of Id/Iq	What the curves look like	Angle needed	Typical use
A Stationary Park vector	Id = i_α, Iq = i_β	still sinusoidal at f_e	no	EPVA fault diagnosis (healthy = circular i_α-i_β)
B Synchronous (SRF-PLL)	Park with the PLL-estimated θ̂	fundamental collapses to near-DC; harmonics/ripple emerge	no (self-estimated)	harmonics, ripple, modulation
C Rotor frame (default)	uses the measured rotor angle if present, otherwise falls back to the synthetic PLL angle; then subtracts δ	near-DC, and (with a true angle) genuine field/torque components	yes (otherwise it degrades to B)	FOC quantities, speed, torque

The curve labels state the active frame: Id (iα) / Id (synchronous) / Id (rotor, measured) / Id (rotor, synthetic).

What a wrong choice looks like

Chose A but want harmonics: Id/Iq are large sinusoids and the fundamental buries everything → switch to B.
Chose C without an angle channel: the log says the measured angle was not found and the synthetic PLL angle is used; labels change to rotor, synthetic. The Id/Iq split then has no absolute meaning (the synthetic angle’s zero is arbitrary), and absolute torque is not trustworthy — the result differs from Frame B only by the constant δ.
Chose B/C and expected sinusoids: near-DC curves are correct, not a failure.

Interaction

A hides the PLL-bandwidth row, all rotor rows (pole pairs / δ / ψ / the three angle rows) and the “speed & torque” output.
B adds only the PLL bandwidth.
C shows PLL bandwidth + pole pairs + δ + ψ + angle channel/full-scale/mechanical + speed & torque.

PLL Bandwidth (Hz)

Loop bandwidth of the SRF-PLL. The engine uses a standard second-order loop with damping ζ=0.707 and ω_n=2π·bandwidth, giving k_p=2ζω_n and k_i=ω_n². It is the “how fast may θ̂ change” knob.

Values: default 5 Hz. Steady state or slow ramps: 1–5 Hz. Fast run-ups/run-downs: 10–30 Hz. There is no universal value — run once at 5 and look at the f_e curve before adjusting.
Too low: f_e is smooth but lags, speed steps are rounded off; θ̂ falls behind the real angle, so Id/Iq in Frames B/C stop being DC and show a slow beat.
Too high: f_e gets spiky, θ_e jitters, the Id/Iq noise floor rises, and order maps smear because harmonics leak into the tracked angle.

Two hidden behaviours (they look like bugs)

1. In Frame A with “Order colormap” ticked, the PLL still runs. Order normalization needs f_e, so the PLL runs in Frame A too and uses exactly this bandwidth value — while the row itself is hidden. To change it, switch to B or C, edit, and switch back.

2. The frequency is clamped. The seed fundamental f₀ is the median per-sample angle increment over the whole record (log: “PLL seed fundamental … Hz”), after which the loop frequency is clamped to 0.5–1.8 × |f₀| (active when |f₀| > 0.5 Hz) to block the 1/5-subharmonic and 2nd-harmonic lock attractors. The cost: if the electrical frequency varies beyond that range within one analysis, the f_e curve is flattened against the clamp limits and the fundamental drifts off order 1 in the order map. Fix: analyse wide run-ups in segments.

The “PLL seed fundamental” log line is a free sanity check: it should be close to your known electrical frequency (f_e = speed × pole pairs / 60). If it is not, suspect channel mapping or polarity before PLL tuning.

Pole Pairs

Pole-pair count p (= poles ÷ 2, from the nameplate). Frame C only. Three uses:

Speed: n_mech = f_e × 60 / p.
Absolute torque: T = 1.5 · p · ψ · Iq.
Mechanical-angle conversion: with “Angle is mechanical” ticked, θ_e = p · θ_m.

Wrong value (uses 1&2): Id/Iq are unchanged; only the speed and torque curves are off by an integer ratio. Measured 4000 rpm but the curve shows 3000 rpm → p is 4/3 too large.
Wrong value (use 3, the harsh one): with “Angle is mechanical” ticked, a wrong p corrupts θ_e completely and Id/Iq turn into a visible beat instead of DC — the most obvious alarm there is.
Back-check: p = f_e × 60 / measured speed should come out an integer.

Align Offset δ (°)

Frame C only: the angle actually fed to Park is θ = θ_ref − δ. It only rotates the split between Id and Iq and never changes |Idq|. Physically it aligns the d axis with the magnet flux — “Id = field, Iq = torque” only holds once aligned.

Values: default 0. Fill it in only after a one-time magnet-alignment calibration told you the offset.
Wrong value: Id and Iq bleed into each other — current that should sit on the q axis lands partly on Id, so Iq drops, Id rises and relative torque reads low; yet |Idq| still equals √(Id²+Iq²). That is exactly the dividing line: a healthy |Idq| with a suspicious Id/Iq split points at δ or the angle zero, not at the data.
Interaction: with a synthetic angle (no measured channel) the zero of θ is arbitrary, so δ merely rotates everything and has no absolute meaning; do not read absolute torque in that case.

Flux Linkage ψ (Wb)

Permanent-magnet flux linkage from the motor datasheet. Frame C only; it decides whether torque is absolute or relative:

Value	Curve	Formula	Unit
`0` (default)	T_rel (∝Iq)	1.5 · p · Iq	rel
> 0	T (torque)	1.5 · p · ψ · Iq	N·m

Magnitude: typically 0.02–0.3 Wb. Entering mWb as Wb makes torque 1000× too small.
With 0 the log reminds you that torque is relative and that ψ yields N·m.
Model limits: this is the surface-mount PMSM torque equation, without the reluctance term (interior PM machines add 1.5·p·(L_d−L_q)·Id·Iq), and the absolute value further assumes δ is correctly aligned. If either does not hold, read the torque curve as a trend, not a calibrated value.

Max Order

Ceiling for electrical order (order = f / f_e, fundamental = 1). Default 10. One parameter, three jobs:

ceiling of the order colormap axis, grid step fixed at 0.02 order;
search ceiling of the problem-order table in the log, grid step 0.5 order, counting only orders ≥ 2;
search ceiling when the electrical↔acoustic overlay picks its target order automatically.

Panel trap: this row is only shown while “Order colormap (Id/Iq)” is ticked, yet it still governs jobs 2 and 3. If you cannot find the row, tick “Order colormap” first; you may untick it afterwards, the value is kept.

Too small: the order of interest never appears in the table (e.g. order 24 on an 8-pole machine), and the order map is cut off at the top.
Too large: at 0.02 steps each order costs 50 grid points, so the map gets slower; and everything above (f_s/2)/f_e has no physical data, showing up as a large blank band at the top.

Measured rotor angle (three parameters)

Frame C only. A real angle is what makes Id/Iq genuine field/torque components; without one the engine silently falls back to the synthetic PLL angle. Acquisition requirements for the angle signal (DC-coupled voltage channel, same clock and sample rate as the currents, range covering the ramp) are in the fundamentals document.

Angle Channel (-1 = auto)

-1: auto-detect by channel name, keywords angle, position, theta, θ, resolver, encoder, rotor, pos_, eangle, elecangle, winkel plus the Chinese equivalents (channels already taken as U/V/W are skipped).
≥ 0: zero-based index into the source-pool channel order.
Not found: the log states the fallback to the synthetic θ and the label becomes rotor, synthetic. Not an error — but you are looking at an approximation.
Pointing at the wrong channel: θ_e is no longer a clean ramp but a messy waveform, Id/Iq look like noise, yet |Idq| stays perfectly normal (it is frame-independent). “|Idq| fine, Id/Iq garbage” almost always means the angle input.

Channels whose names look like an angle are automatically excluded from the NVH sensor set (an angle ramp is not vibration), so they never pollute the order table.

Angle Full-scale

The channel value that corresponds to one full turn. Conversion: θ = value × 2π / full-scale.

Angle signal form	Enter
radians (0–2π)	`6.283185` (default)
degrees (0–360°)	`360`
normalized (0–1)	`1`
analog ramp voltage (0–V_max)	the ramp peak, e.g. `5` or `3.3`

Half the true value: the angle runs twice as fast → the fundamental is no longer DC in dq but leaves a residual sinusoid at 1× f_e.
Twice the true value: the angle runs at half speed → the same residual, opposite direction.
Hard criterion: the ramp rate of θ_e must equal the f_e reading (or f_e/p if the channel carries a mechanical angle). If it does not, full-scale or the “mechanical” tick is wrong.

Angle is mechanical (× pole pairs)

Ticked = the channel carries a mechanical angle θ_m; the engine multiplies by the pole pairs.
An MCU angle computed for FOC is usually already electrical → leave unticked. Encoder/resolver mechanical position → tick it.
Should be ticked but is not: θ_e runs at 1/p of the electrical frequency → Id/Iq oscillate strongly at (p−1)·f_e.
Ticked when it should not be: the angle runs p× too fast — equally obvious oscillation.
This tick binds the pole-pair count into the angle path: once ticked, a wrong p no longer merely scales speed/torque, it destroys Id/Iq.

You do not have to check this by hand. Whenever a measured angle is used, the log carries a measured-angle check: angle rate xx Hz / PLL f_e xx Hz = ratio line:

≈1 → the angle convention is right;
≈1/p → the channel actually carries a mechanical angle with this box unticked; the log says so and states that Id/Iq will show a residual oscillation at (p−1)·f_e;
≈p → the pole-pair factor was applied once too often (box ticked wrongly, or full-scale entered as 1/p of the truth);
anything else → check full-scale, then the angle channel, then the pole-pair count.

The reference is the PLL's own f_e (the PLL runs anyway in Frame C); the angle rate is the median per-sample increment, so the ramp wrap-around does not disturb the verdict.

Select plot types (which plots this run)

This group picks which plots this run produces — the analysis can emit a lot of curves and colormaps at once, enough to swamp a single Data Viewer, so nothing is ticked by default: tick what you need, and the selection is stored with the node. Ticking nothing still gives you the diagnosis: balance, EPVA modulation depth and the problem-order table always go to the log.

Output	Default	Produces	Requires / interacts
Current curves (Id / Iq / Idq)	off	three curves in A (footer single value = RMS)	—
Electrical frequency f_e	off	f_e vs. time in Hz (sign = direction of rotation)	only if the PLL ran (see below)
Electrical angle θ_e	off	ramp wrapped to (−π,π], label states the source (measured/synthetic/PLL)	same as above
Speed n_mech & torque	off	speed in rpm + T or T_rel	Frame C only
Colormap vs. time	off	three time-frequency maps for Id / Iq / Idq (linear A)	—
Order colormap (Id/Iq)	off	two order maps (order = f/f_e)	needs f_e → forces the PLL to run, even in Frame A
Correlate vibration & noise	off	one order map per non-current channel, normalized by the current f_e	must be combined with “Order colormap”, otherwise nothing is produced
Electrical↔acoustic order overlay (phase)	off	per target order: normalized current, normalized sensors, phase-difference curves + one null map	needs θ; see target orders below

The two interactions people trip over:

1. “Correlate vibration & noise” alone does nothing. Its output lives inside the order-colormap branch — tick both to get vibration/noise order maps.

2. Frame A without “Order colormap” has no angle and no f_e. So the f_e and θ_e curves are absent, the overlay and null map are skipped entirely, and the problem-order table never appears in the log. Nothing failed — that frame simply never computes an angle. To get them in Frame A, tick “Order colormap” (it starts the PLL as a side effect).

Target orders (blank = auto)

Only shown while the overlay is ticked. Separate with commas, semicolons or spaces, e.g. 6, 12, 24.
Blank = use the top entry of the problem-order table.
At most the first 3 are used; the rest are ignored (each target costs a curve set plus a null map).
Non-numeric or ≤ 0 entries are dropped — if a value you typed produced no plot, check the separators first.
Target orders are not limited by “Max Order” (order 24 works while Max Order is 10); only the automatic pick stays within Max Order.

Fixed values the engine uses that are not on the panel

Item	Value	Effect
Spectrum size	4096	colormap frequency resolution Δf = f_s/4096
Window	Hanning	spectral leakage in the colormaps
Overlap	50%	colormap time resolution
Max time bins	600	excess frames are max-reduced (peak hold, not averaged)
Max colormap frequency	2000 Hz	ceiling of the vs.-time colormap axis
Remove DC	off	each phase has its record mean removed before Clarke
Order-analysis decimation	to about 24 kHz	applied before coherent demodulation; it only lifts the noise floor slightly

Where to look: curves, colormaps, log

Curves and colormaps

All go to the Data Viewer. Current colormap values are linear amperes and are never converted to dB(SPL).

Log (Event Viewer)

These are always written and cost no plot slot — check them first when a parameter looks wrong:

PLL seed fundamental … Hz — compare against your known electrical frequency.
the angle-source line: measured rotor-angle channel used, or fallback to the synthetic θ.
EPVA modulation depth std(|Idq|)/mean = x.x% — ≈ 0 when healthy; a rising value suggests a modulating fault (inter-turn short at 2f_e, broken rotor bar at 2·s·f_e, eccentricity …). Locate the sidebands in the |Idq| colormap / order map.
the problem-order table: order | NVH amplitude | current amplitude | correlation | advice. Correlation is the Pearson correlation between the current and NVH amplitude tracks of that order; |corr| ≥ 0.5 counts as “electrically driven”, and the top entry is flagged as the preferred injection order. With no vibration/noise channel selected, the table says so.
Predicted optimum injection at order k: amplitude ≈ …, phase ≈ …° — the deepest point of the null map (only when the overlay ran).
the final line Frame=… U=… V=… W=… N=… balance=…% — channel mapping and balance self-check.

Symptom → cause quick reference

What you see	Most likely	What to do
Log “balance … ⚠ too large”, \|Idq\| not constant	phase mapping/polarity wrong, or real zero-sequence	verify the first 3 source channels are U/V/W; check clamp orientation
Id/Iq in Frame B/C beat slowly instead of being DC	PLL bandwidth too low, or angle full-scale / mechanical tick wrong	check the θ_e ramp rate against f_e; if it matches, raise the bandwidth
f_e flattened against two horizontal limits	PLL frequency clamp (0.5–1.8 × seed) is active	analyse wide run-ups in segments
f_e reads about 1/5 or 2× the true value	PLL locked to a sub/super harmonic (bad seed)	check the seed line in the log; rule out channel mapping first
Id/Iq are noise but \|Idq\| is fine	angle input (wrong channel / full-scale / mechanical tick)	look at θ_e — is it a clean ramp?
Speed curve off by an integer ratio	wrong pole-pair count	p = f_e×60 / measured speed, rounded to an integer
Torque unit is rel, not N·m	flux linkage ψ left at 0	enter the nameplate flux linkage in Wb
“Correlate vibration & noise” ticked but no vibration order maps	“Order colormap” not ticked	tick both
No f_e / θ_e / order table in Frame A	Frame A computes no angle by default	tick “Order colormap”, or use Frame B
Large blank band at the top of the order map	Max Order exceeds (f_s/2)/f_e	lower Max Order
Order map almost entirely dark	\|f_e\| < 1 Hz in those frames (not locked / machine stopped)	cut the record to the running section
Cannot find the “Max Order” row	it only appears with “Order colormap” ticked	tick it, edit, untick again if you like
Order table says there are no vibration/noise channels	only currents selected in the source pool	add the vibration/microphone channels

Shortest path to a result

Currents only: select U/V/W, choose Frame B → Id/Iq/|Idq| plus EPVA modulation depth and the balance check. No angle, pole pairs or flux linkage needed.
Harmonics / orders: tick “Order colormap” (and “Correlate vibration & noise” if wanted) and set Max Order.
“Which order should we inject?”: add the vibration/noise channels to the source pool and read the problem-order table in the log; tick the overlay for the null map and the predicted optimum injection phase.
True rotor dq / absolute torque: wire the measured angle (DC-coupled, synchronously sampled), choose Frame C, set the angle channel/full-scale/mechanical tick, then pole pairs and flux linkage.

θ_e is always the foundation to verify: a uniform ramp whose rate equals f_e (or f_e/p for a mechanical angle) means the angle is right and everything downstream — Id/Iq, torque, order phase — is meaningful. Jumps, breaks or a mismatched rate mean fix the angle first and ignore the results.

Scope and known limits

Order maps come from an STFT normalized frame-by-frame by that frame’s f_e, not from angular-domain resampling; under hard acceleration the order lines smear slightly more than a true order-tracking method.
The torque model is surface-mount PMSM without a reluctance term, and absolute values also assume a completed magnet-alignment calibration for δ.
The PLL frequency clamp limits how much the electrical frequency may vary inside one analysis (0.5–1.8×).
The harmonic-injection output is a prediction of the optimum injection amplitude/phase, derived from the measured complex transfer ratio H at baseline. The injection itself happens in the drive/inverter and needs a before/after test-bench verification.