Parameters of the Perceiving Machine at High Speed

• 1
  What the perceiving machine at high speed actually senses
• 2
  Sampling rate, resolution and bandwidth of the sensing loop
• 3
  Latency and filter delay: where the loop loses time
• 4
  Stability limits when speed rises
• 5
  What to check and how to tune for a real job
• 6
  Sensing and servo parameters compared

A perceiving machine at high speed is not one sensor. It is a chain: glass scale or encoder, a read head, an interpolator, the drive current loop, and the spindle or table that finally moves. Each link adds delay. At 200 mm/s of feed, a 1 ms delay already means 0.2 mm of position error before the tool touches the part. That number is why the chain matters more than any single spec on a brochure.

The three quantities that carry most of the information are position, force and vibration. Position tells the control where the axis is. Force, measured on the spindle or the tool holder, tells it how hard the cut is pushing. Vibration, read from accelerometers on the headstock, tells it whether the cut is stable. Position is fast and cheap. Force and vibration are slower but explain why a cut fails.

Speed changes the weighting. Below 100 mm/s you can ignore most transport delays and treat the loop as static. Above roughly 300 mm/s, every millisecond of latency becomes a positioning error, and the control has less time to correct. The machine is still perceiving, but it is perceiving the past.

• 1
  Position loopScale to control, typically 1–4 kHz update on modern drives.
• 2
  Force loopSpindle or table dynamometer, bandwidth often below 1 kHz.
• 3
  Vibration loopAccelerometers on the head, useful for chatter detection.

Sampling rate sets the ceiling. A loop that samples at 1 kHz sees the world every 1 ms and cannot correct anything faster than about 100 Hz in a stable way. The usable control bandwidth is usually one tenth to one fifth of the sampling frequency, so a 4 kHz position loop gives you roughly 400–800 Hz of real correction. Push the gain past that and the axis starts to ring instead of following.

Resolution is separate from accuracy. A 0.1 μm scale resolution does not mean the machine holds 0.1 μm; thermal drift, ball screw pitch error and servo following error sit on top of it. For most production parts, a scale resolved to 0.5 μm with a following error under 5 μm is more useful than a fine scale on a loose axis.

Bandwidth also depends on mass. A small 5-axis trunnion can be tuned to 80–120 Hz of position bandwidth. A 4,000 mm gantry cannot. Its first structural mode may sit near 20 Hz, so the loop must stay well below that or the machine will excite its own frame.

• 1
  Rule of thumbControl bandwidth ≈ sampling rate ÷ 5 to ÷ 10.
• 2
  Small rotary axes80–120 Hz position bandwidth is realistic.
• 3
  Large gantriesStructural modes near 20 Hz cap the loop.
• 4
  Following errorKeep under 5 μm on axes used for finishing.

Every filter in the chain costs phase. A low-pass filter set at 500 Hz to kill noise will also delay the signal by roughly 1 ms. Add encoder interpolation, fieldbus transfer and drive processing, and total latency on a typical machine lands between 1 ms and 5 ms. At 400 mm/s, 5 ms is 2 mm of unrecovered position.

This is why feed-forward matters more at speed. Instead of waiting for the error to appear and then correcting it, the control uses the programmed path to predict the next command. Velocity feed-forward removes most following error on a constant feed. Acceleration feed-forward removes the lag at corners. Both rely on knowing the machine mass and stiffness, so they must be re-tuned if you change fixtures or add a heavy vise.

Look-ahead is the other lever. A control that reads 200 blocks ahead can slow down before a tight corner rather than overshoot it. With 20 blocks of look-ahead, high-speed finishing on a 2 mm corner radius will show visible gouging. With 200 blocks, the corner is clean. Block count is a real parameter, not a marketing number.

• 1
  Typical latency1–5 ms from sensor to motion, depending on drive and bus.
• 2
  Filter trade-offLower cutoff means less noise and more delay.
• 3
  Feed-forwardCuts following error without raising gain.
• 4
  Look-aheadNeeds 200+ blocks for tight 3D corners.

Raise the gain and the axis follows better, until it does not. The limit is set by the lowest mechanical resonance in the loop. On a mill with a long tool, that resonance can drop below 100 Hz, and the control must back off to stay stable. The same machine with a stubby tool may allow twice the gain and half the following error.

Thermal behavior is the slow limit. A ball screw grows about 12 μm per meter per degree Celsius. Run a machine hard for two hours and the screw can be 3 °C warmer than the scale, which is 36 μm of drift per meter. No bandwidth fixes that. It is handled with cooling, pre-tension, or by keeping the scale on the same thermal path as the screw.

The practical boundary is this: speed and accuracy trade against each other through the loop, not through the spindle. If the part needs ±0.005 mm, the limiting factor is usually the sensing and servo loop at the required feed, not the cutter. Test at the real feed before committing a process.

• 1
  Chatter edgeLowest structural mode caps usable gain.
• 2
  Tool overhangLong tools push resonance down and force lower gain.
• 3
  Thermal driftAbout 12 μm per meter per °C on steel screws.

Start with a circular test at the feed you actually run. A ballbar or a simple circular cut shows following error, reversal spikes and servo mismatch in one picture. If the circle looks round at 50 mm/s and egg-shaped at 300 mm/s, the problem is the loop, not the geometry.

Then check the scale mounting. A scale that is not parallel to the axis within 0.1 mm over its length will produce a periodic error synchronized with the screw pitch. It looks like a mechanical fault but it is a sensing fault. Clean the read head, verify the gap, and re-run the test.

Finally, tune at temperature. Do the gain and feed-forward work after 30 minutes of warm-up, not cold? A machine tuned cold will drift as it heats, and the operator will blame the program. On parts held to ±0.005 mm, we warm up spindles and axes before the first cut and re-check the first article.

• 1
  Circular testRun at production feed, not at a safe slow speed.
• 2
  Scale alignmentKeep parallelism within 0.1 mm over the full travel.
• 3
  Warm-upTune and inspect after the machine reaches steady temperature.