Machine intelligence

Know the mechanism. Then notice when it changes.

An ordinary controller asks one question: did the following error exceed the limit? KineticON also asks a better one — why is this axis behaving differently from itself yesterday? To answer it, the controller first has to know the mechanism it is driving.

In short: AxisDNA turns commissioning into a scientific characterisation and keeps the raw evidence; AutoDynamics builds the compensation that mechanism needs; ShadowTwin runs an expected-state twin in real time and flags drift long before a fault threshold; QuietHold keeps a resting machine cool without ever trading away physical integrity.

AxisDNA™ · commissioning as science

Every axis has a fingerprint.

KineticON Studio characterises the mechanism instead of asking the engineer to guess: travel, direction, counts per millimetre, encoder scale, backlash, hysteresis, static and viscous friction, cogging, force ripple, the resonance spectrum and structural modes, settling behaviour, acceleration and jerk capability, payload response, brake release and engage delay, thermal drift, current requirements and the following-error envelope.

The result is stored as AxisDNA — derived parameters and the raw measurement evidence that produced them. Change the mechanics and the relevant characterisation is invalidated, not silently kept.

AXISDNA · VERIFIED Ten properties measured on commissioning, stored with the raw data that produced them.

AXIS Z · AXISDNA FINGERPRINT● MEASURED

AutoDynamics™ · compensation architecture

Not “here are some PID gains”. The compensation this mechanism needs.

AxisDNA feeds AutoDynamics, which uses automatic system identification to construct the whole compensation architecture appropriate to the mechanism — and shows the frequency-response, Bode and FFT evidence behind every choice.

Machine learning may assist analysis and parameter discovery. The real-time controller remains deterministic. No generative model sits inside a safety-critical servo loop, and no trajectory is ever hallucinated.

gain schedule
velocity feed-forward
acceleration feed-forward
friction compensation
cogging compensation
notch filters
command shaping
harmonic cancellation
cross-axis feed-forward
payload compensation
thermal compensation

ShadowTwin™ · expected versus actual

The machine knows when it no longer feels like itself.

ShadowTwin predicts what the machine should look like right now — from the commanded trajectory, known mechanics, AxisDNA, payload, motor current and torque, motor and load encoders, temperature, vibration, brake state and acceleration — and compares that expectation with what is measured.

Drag the slider through six months of a Z axis whose bearing friction slowly rises. The measured current climbs away from the axis’s own expected envelope while the ordinary fault threshold is never crossed. A conventional controller stays silent until something breaks. ShadowTwin notices.

Service days

SHADOWTWIN: NO LONGER FEELS LIKE ITSELF

AXIS Z · MOTOR CURRENT · ONE MOVE● SHADOWTWIN

expected (twin) with bandmeasuredfault threshold

day 120 · cruise current above this axis’s own baseline · peak still below fault threshold

Current creeping up

bearing friction, guide contamination, a dragging brake

Resonance shifting

belt tension, loosened mounts, coupling degradation

Settling changing

payload anomaly, mechanical misalignment

Offsets drifting

thermal expansion of the structure

MOTOR TEMPERATURE · 2 h AT REST● QUIETHOLD

Horizontal axis at rest: holding current lowered to what the measured load actually needs, with stationary state re-verified continuously.

QuietHold™ · thermal intelligence

A stationary machine should not cook itself.

QuietHold combines stationary-state verification, brake status, load characteristics, the holding-current requirement, motor and drive temperature, power-supply loading and mechanical orientation, and chooses between full hold, reduced hold, brake-supported hold, a zero-torque stationary state and adaptive thermal hold.

The payoff is lower heat and noise, longer component life, a lighter cabinet load, less fan cycling and better dimensional stability. The rule that governs it is simpler: holding torque is never reduced to save energy if physical integrity depends on it.

This is a vertical axis with a brake

Condition, gantry and sensor intelligence

Maintenance from evidence, not from a calendar.

Not “service every six months”, but: this Z brake has completed 1.7 million cycles and its measured release latency has increased from X to Y.

Condition intelligence

Lifetime metrics per axis

Distance travelled, motion hours, acceleration cycles, high-torque time, thermal cycles, brake cycles, limit events, following-error events and abnormal load events — the raw material of real predictions.

Gantry intelligence

Built by gantry builders

Automatic dual-axis squaring, skew detection, cross-axis coupling and feed-forward, dual-encoder control, racking prevention, compliance estimation, independent-home correction and a collision-aware split-axis service mode.

Sensor fusion

Position is more than one number

Motor and load encoders, linear scales, resolvers, absolute encoders, current and torque sensing, accelerometers, temperature and load cells, with motor-loop, load-loop and gantry-differential feedback where the machine needs it.

AI-assisted engineering without AI-controlled chaos

KineticON Studio may use intelligent analysis extensively — detecting resonance, classifying abnormal current, suggesting filters and tuning, interpreting logs, comparing AxisDNA changes, explaining faults, drafting commissioning procedures. It helps engineers understand and configure. It does not replace the deterministic motion kernel, and it never writes a trajectory into a servo loop.

Frequently asked

Frequently asked questions

What is a servo digital twin?

It is a model that runs alongside the real axis and predicts what its current, torque, position and dynamics should be for the motion being executed. KineticON’s ShadowTwin compares that expectation with the measured signals continuously, so gradual deterioration becomes visible before an ordinary fault limit is reached.

What does AxisDNA measure?

Travel, direction, counts per millimetre, encoder scale, backlash, hysteresis, static and viscous friction, cogging, force ripple, resonance spectrum and structural modes, settling, acceleration and jerk capability, payload response, brake release and engage delay, thermal drift, current requirements and the following-error envelope — stored with the raw measurement evidence.

Does KineticON use machine learning inside the servo loop?

No. Machine learning may help analyse data and discover parameters offline or in the engineering tools, but the real-time controller remains explicit, deterministic and bounded. No generative model is placed inside a safety-critical servo loop.

How can holding current be reduced safely on a vertical axis?

Only when the axis is verified stationary and the brake’s engagement is proven. Without that proof QuietHold keeps full holding torque, because integrity outranks energy saving.

What is gantry squaring?

On a gantry driven by two parallel motors, the two sides must stay aligned so the beam does not rack. KineticON squares the two axes automatically, detects skew, controls with both encoders and corrects independent homes.

References

Sources

Architecture →Core, KineticFabric, TimeFabric, ProcessSync, ToolSafe and safety. Motion integrity →MotionProof, StateWitness and FrameProof. Developers →API, simulation, ControlGraph and the machine manifest.

This week in the field

11 – 17 September 2026

Newest PubMed records in robotic manipulation, automated laboratories and precision control — refreshed weekly. Listing only; inclusion is not endorsement.