Panther MIMO Sine control.
Swept sine on multiple exciters, where magnitude is only half the specification. The phase relationship between control points is the other half, and controlling it is what separates true MIMO sine from several single-axis loops running side by side.
| Step | What happened |
|---|---|
| Field | Rig characterized under random excitation |
| Model | Characterization turned into a plant model |
| Sine ID | Panther's broadband sine identification run against it |
| Match | multiple coherence 0.99996 |
Dense, closely spaced modes that nobody designed to be identifiable. The plant model is a property of the rig, not of the test that measured it.
Is the table translating, rotating, or tearing itself apart?
A swept sine test on one shaker asks a simple question at every frequency: is the control accelerometer at the level the profile calls for? Put the same article on three shakers and the question changes shape. Each shaker now moves every accelerometer, and the test is specified by three amplitudes and the phase relationships between them.
Those relationships determine whether the table is translating, rotating, or tearing itself in opposite directions at 250 Hz. Panther MIMO Sine controls both. Every control channel carries a magnitude reference and a phase reference across the sweep, each with its own alarm and abort bands, and the loop drives the measured complex response vector onto that complex reference.
No matrix inversion anywhere in the sine loop
MIMO Random and MIMO Sine share a platform and a philosophy, but not a control law.
Cross-spectral matrices on a bin grid
Random works on cross-spectral density matrices estimated over a bin grid, with averaging and an ensemble from which to compute coherence.
One complex measurement per channel, per block
At the frequency the sweep is sitting on. The controller maintains a complex impedance matrix that is the inverse: it maps control error directly to the drive correction that will fix it, and improves that matrix from what it just observed by a rank-one multiplicative update. Nothing has to be well-conditioned for the next block to be computed, so the loop degrades gracefully on rigs that would stop a matrix-inversion approach dead.
Bounded correction
The drive change at any step is capped as a fraction of the current drive vector. The impedance update is protected by deadbands that leave it untouched when the observation carries no information, plus a guard that discards a row whose correction is implausible. This is what makes a loop that starts knowing nothing about the plant safe to point at a shaker.
Phase as a first-class abort
Phase can be checked absolutely against stored bands or relative to a nominated reference channel, with aborts requiring consecutive out-of-band blocks rather than a single excursion. Tolerance bands are per-segment and asymmetric, and widen across profile steps so a stepped reference does not abort at its own discontinuities.
Limiting that preserves shape
When a limit channel takes control, a single scalar moves the commanded level. Phase is untouched and the ratios between channels are untouched, so the shape of the commanded motion survives the limit event intact and only its magnitude moves. That is the correct behavior for a multi-exciter test, and it is not what falls out of limiting each channel independently.
Transforms, not special cases
Both the response-side and drive-side transforms are complex, so a transform entry can apply a phase shift, which is exactly what a rotational degree of freedom or a quadrature combination requires. The loop never knows whether it is controlling accelerometers or derived degrees of freedom; reference, error, impedance and tolerances all live in the transformed space.
The established control law, with headroom
As with MIMO Random, the control law is the one Spectral Dynamics established for multi-exciter sine testing, and it is not changed lightly: every deliberate difference is recorded with the measurement that justified it. What has changed is the platform beneath it. The loop computes in 64-bit floating point throughout, where earlier multi-exciter sine control was constrained to single precision by the signal processors of its day, and it runs on modern vectorized linear algebra in front of a 24-bit acquisition subsystem with phase matching to within one degree.
Broadband identification before the sweep starts
A sine controller has to learn the plant before it can control it. The traditional answer is to excite hard at the first sweep point and learn from that, but at a single frequency the shakers are driven in a correlated way, so the measurement is close to rank-deficient and the off-diagonal terms cannot be cleanly separated. Panther measured that effect directly: a substantial off-diagonal overshoot that the loop then had to unlearn during the sweep.
A short broadband pretest instead
- Covers every frequency in the band rather than one.
- Runs well below test level.
- Merges into the pre-shaker checkout a real test requires anyway, so its marginal cost in article exposure approaches zero.
Something sine control has never had
Multiple coherence, as a quality metric, before the test runs. A loose fixture, a bad cable, a saturating amplifier or a nonlinearity all show up in coherence. Until now, a sine operator had no equivalent instrument.
Multiple coherence between the plant Panther's broadband sine identification recovered and the field characterization of a customer rig.
Physical channels resolved to three degrees of freedom, with dense, closely spaced modes.
A plant characterized by a random field run drove a sine identification. The model belongs to the rig.
A characterization of your installation is a reusable asset
The most consequential result from the MIMO Sine program so far is not a control number. It is a validation of the simulation approach that underpins the whole development: the control algorithms are developed and proven against real, densely moded field dynamics rather than against rigs chosen for being well behaved.
Take field dynamics and geometry, simulate the rig to high accuracy, and prove the controller against it before it ever meets your article.
More on the simulated test system behind Panther development.
The same setup and display framework as MIMO Random
MIMO Sine shares the Panther MIMO platform's setup and display framework, including the spatial view and self-contained reporting, so a lab running both applications configures them the same way.
- One channel table, typed. Control, monitor, limit, abort and reference channels are declared in a single table, and the controlled set is derived from the typing rather than declared separately.
- Geometry-driven transforms. The usual answer to more accelerometers than shakers is not a reduced control law but a transform, which makes three sensors on two shakers a square problem in transformed space. A reduced path is available for the cases that genuinely need it.
- Drive protection. Drive limits are checked on physical outputs after the output transform, one test per output, with the attenuator and phase ramped by the hardware rather than stepped.
- Alarm and abort on both quantities. Magnitude bands are per-segment, asymmetric and multiplicative on the reference; phase bands are additive in degrees. Both widen across profile steps.
Panther MIMO Sine
| Capability | Implementation |
|---|---|
| Controlled quantity | Complex response vector: magnitude and inter-channel phase |
| Reference | Magnitude and phase per control channel across the sweep, with independent asymmetric alarm and abort bands per segment |
| Control law | Adaptive complex impedance loop, slew-limited; no matrix inversion in the loop |
| Plant identification | Broadband pretest below test level, with multiple coherence as a quality metric |
| Transforms | Complex input and output transforms; rigid-body DOF control |
| Channel types | Control, monitor, limit, abort, reference |
| Limiting | Scalar on commanded level; inter-channel phase and ratios preserved |
Frequently asked
Does Panther MIMO Sine control phase as well as magnitude?
Yes. Every control channel carries a magnitude reference and a phase reference across the sweep, each with its own alarm and abort bands, and the loop drives the measured complex response vector onto that reference.
Why is there no matrix inversion in the loop?
The controller maintains a complex impedance matrix that maps control error directly to the drive correction, and updates it from each observation. Nothing has to be well-conditioned for the next block to be computed, so the loop degrades gracefully on difficult rigs.
Why identify the plant with a broadband pretest?
At a single sweep frequency the shakers are driven in a correlated way, so the measurement is close to rank-deficient. A short broadband pretest below test level covers every frequency and gives multiple coherence as a quality metric before the test runs.
What happens when a limit channel takes control?
A single scalar moves the commanded level. Phase and the ratios between channels are untouched, so the shape of the commanded motion survives the limit event.
Control the phase, not just the level.
Panther MIMO Sine runs on Spectral Dynamics Panther controllers. Also in the series: MIMO comes to Panther and Panther MIMO Random.
