Panther MIMO Random control.
Multi-exciter random vibration, controlled as a matrix rather than a set of independent loops, with the cross-spectral references, degree-of-freedom transforms and per-line limiting that real multi-axis testing demands.
| Metric | Reads | What it means |
|---|---|---|
| Ordinary coherence | ≈ 0.33 | Each drive explains only part of each response. Throttle gain on this and the loop freezes. |
| Multiple coherence | the truth | How much of the response all the drives explain together. |
SELECTED AUTOMATICALLY · Panther detects the condition and switches to multiple coherence.
Not one scalar problem repeated N times
Single-shaker random control is a scalar problem: one drive, one control spectrum, one loop. Drive several shakers into one structure and every drive appears in every response.
The shakers fight each other
Control what each accelerometer sees in isolation and the rig heats up, the article is over-tested somewhere and under-tested somewhere else, and the plots look fine throughout.
The full cross-spectral density matrix
The reference is not a set of PSDs. It is a matrix specifying, for every pair of control points, both the auto-spectra and the coherence and phase relationship between them. The controller drives the measured response matrix onto that reference.
Matrix power control, in 64-bit floating point
Panther MIMO Random uses the matrix power control algorithm that Spectral Dynamics established for real-time multi-exciter testing, computed in 64-bit floating point on modern multicore hardware.
- Newton MPCA with optimal step. The response error is projected into drive space through the pseudo-inverse of the measured plant, and at every frequency line a cubic is solved for the step size that minimizes the predicted next error. The update is applied to the Cholesky factor of the drive matrix, so the commanded drive spectrum is physically realizable by construction.
- Per-bin regularization. Correction authority is scaled per frequency line by the local condition number, so well-conditioned bins converge quickly while stiff ones are damped rather than driven into oscillation. Bins that persistently worsen are throttled and, if necessary, frozen.
- Coherence weighting. Correction gain at each line is weighted by measured coherence, so the controller does not chase noise where the measurement cannot support a correction.
It reports what it found, before it proceeds
Everything depends on knowing the plant. Panther MIMO measures the full M × N frequency response matrix, every drive to every response.
- Multiple coherence per response, the correct measure when several drives contribute to one sensor.
- Per-bin condition number, with the theoretical accuracy floor it implies, so an ill-conditioned rig is identified before the test rather than diagnosed afterwards.
- Controllability index per channel: how much of that sensor's motion the drive set can actually command.
- Reciprocity and continuity checks, flagging asymmetric drive and response paths, missing signals and reversed polarity before a level is applied.
- Resonance and anti-resonance detection, including structural nodes. An accelerometer on a node cannot be controlled at that frequency by any algorithm, and the operator is told so.
What separates real multi-exciter control from parallel single-axis control
Cross-spectral references
The test specifies the phase and coherence between control points, not just their levels. This is what makes a genuinely multi-axis environment specifiable, and on some geometries it is the difference between a controllable test and an impossible one. Where several sensors measure essentially the same motion, a diagonal reference asks for something no drive set can deliver; a cross-spectral reference asks for what the rig can actually do.
Degree-of-freedom control
Sensor measurements are transformed into rigid-body motions of the table, translations and rotations, and those are controlled. Panther applies the geometric transform on both the response side and the drive side. Response-only control leaves the plant coupled, and the weak axis is quietly under-tested while every displayed metric reads healthy. With both transforms in place the DOF plant is diagonal and each degree of freedom is controlled on its own terms.
Validated against measured field data
Panther MIMO Random's degree-of-freedom control was validated by decoding Jaguar's measured field files from a customer installation and running Panther against a plant built from the same rig.
Settled control spectra vs measured field spectra on a three-DOF test (X, Y and yaw), on all three DOF, from two independently constructed transforms.
Configurations in the nightly regression: 2 × 2 through 12 × 6, square and rectangular, from measured customer plants and synthetic models.
Unit tests that must run green on every build.
A half-decibel movement is treated as a defect, not as noise.
See how the simulated test system and nightly regression work.
Random limiting in MIMO is a different animal
A single scalar applied to the whole reference matrix cannot express what a multi-shaker rig actually needs. A limited run that goes unreported looks perfect on every conventional plot, because the loop tracks the derated reference exactly. The instrumentation to see it is the feature.
- The notch is a curve, not a number. Every frequency line is derated independently, every frame, with recovery proportional to available headroom rather than a fixed percentage.
- Per-line attribution. Panther records which limit channel is holding control at each line, and reports the deepest notch, the frequency it occurred at, and how many lines were derated.
- The ghost trace. The derated target is drawn against the original profile, so the gap between them is the notch depth. Alarm and abort bands move with the derated target.
Aborts that explain themselves
- Line-counting abort channels. A test trips when a defined number of lines exceed the limit, not on the first block, so a single noisy bin cannot end a valid run.
- Every abort explains itself: which channel, what it measured, against what level, at which frequency, and by how much.
- Two-tier safety under a transform: wide physical limits protecting the hardware, tight virtual limits governing control accuracy, with limit channels kept outside the control matrix so protecting the article never costs control channels.
The AI setup agent does the heavy lifting
It runs the system identification, assigns channel roles, verifies from geometry that the actuator layout can resolve the requested degrees of freedom, builds the transforms, and calculates loop gain, initial level, ramp step and iteration limits from what it measured rather than from stored defaults. Every choice is presented with its reasoning, and a predict mode lets an engineer explore a configuration without committing to it. More on the AI setup agent.
Errors on the fixture, results in one document
The spatial view renders per-channel error, coherence, safety state and channel roles on the fixture layout itself. Reports capture the plots, the channel and coherence matrices, the notch record and the setup reasoning in a single self-contained document, alongside machine-readable per-frame exports for teams that do their own post-processing.
The hardware underneath
| Input subsystem | Output subsystem |
|---|---|
| 8 channels, expandable to 64 | Dynamic range > 110 dB |
| Dynamic range > 110 dB · 24-bit ADC | 20-bit DAC · ±12 V peak |
| 262,144 samples/s, simultaneous | 262,144 samples/s |
| Channel crosstalk < −100 dB | Frequency accuracy ±5 ppm |
| Phase match ±1.0° to 100 kHz, within and between units | Output impedance 60 Ω |
| IEPE, TEDS (IEEE 1451.4), 8 gain ranges, auto-ranging | USB, Ethernet, optional Wi-Fi |
Phase match between units to within one degree is what makes multi-exciter control on expanded Panther hardware meaningful. A MIMO controller is only as good as the phase coherence of its front end.
Frequently asked
What does Panther MIMO Random control?
The full cross-spectral density matrix: the auto-spectra at each control point plus the coherence and phase relationship between every pair of control points.
Why does Panther use multiple coherence?
When several drives contribute to one sensor, ordinary coherence reads low even on an excellent measurement, about 0.33 for three drives on a symmetric rig. Multiple coherence measures how much of the response all the drives explain together, and Panther selects it automatically.
How does limiting work in MIMO Random?
Every frequency line is derated independently, every frame. Panther records which limit channel holds control at each line and reports the deepest notch, where it occurred and how many lines were derated.
Why apply the DOF transform on both sides?
Controlling DOF responses against physical drives leaves the plant coupled, and the weak axis is quietly under-tested while every metric reads healthy. With response-side and drive-side transforms the DOF plant is diagonal.
Control the matrix, not just the channels.
Panther MIMO Random runs on Spectral Dynamics Panther controllers. Also in the series: MIMO comes to Panther and Panther MIMO Sine.
