Anti-interference of CNC Machine Tools: 5 Measures That Hold Tolerance

• 1
  Why the anti-interference of CNC machine tools shows up as dimensional error
• 2
  Shielding: closing the path before it reaches the drive
• 3
  Isolation: breaking the galvanic loop
• 4
  Filtering: removing what shielding and isolation let through
• 5
  Earthing and bonding: the reference everything else depends on
• 6
  Software filtering and diagnostics: the last line of defense
• 7
  Which measure fits which failure mode

Servo drives switch at several kilohertz. Spindle drives switch faster. Every one of those edges radiates, and the cables running to encoders and glass scales act as antennas. When a noise voltage couples onto a 1 V peak-to-peak encoder signal, the drive reads a position that never existed. On a finishing pass at 0.010 mm depth, that is enough to leave a witness mark.

The failure mode is rarely a dead machine. It is drift. A part measures 24.998 mm in the morning and 25.003 mm after lunch, or a 5-axis toolpath leaves chatter only on the side of the part facing the spindle drive cabinet. Operators blame the tool. The tool is fine.

Three coupling paths carry the noise in: conducted along shared power and ground conductors, capacitively across nearby surfaces, and inductively through the loop area between a signal cable and a motor cable. Each path needs a different fix, which is why one measure alone seldom solves the problem.

Shielding stops radiated and capacitively coupled noise by wrapping the sensitive conductor in a conductive layer that is then bonded to ground. For a machine tool, the practical targets are the encoder cable, the glass scale cable, the spindle command cable, and the drive power cable if it runs more than a meter unshielded.

The shield only works if it is terminated at both ends with a 360° clamp. A pigtail drain wire adds roughly 1 μH of inductance per 10 cm, which turns the shield into a radiator above a few megahertz. We see this most often on retrofits where the original cable was replaced with a generic shielded cable and the electrician landed the drain on a terminal block.

Metal conduit and sealed cabinet panels do the same job at the enclosure level. Bond the panel to the cabinet frame at multiple points, not just at the hinge. A single bonding point gives the panel a high impedance at the frequencies that matter.

Shielding is not free. It adds cable cost, bend-radius limits, and connector weight. On a compact 3-axis machine with short cable runs, the benefit is small. On a 4,000 mm traveling gantry where the scale cable runs 6 m beside a spindle cable, it is the difference between holding ±0.005 mm and chasing drift all shift.

• 1
  Terminate shields 360°Use metal cable clamps, not drain wires, at both ends.
• 2
  Separate power and signalKeep at least 200 mm between motor cables and encoder cables.
• 3
  Bond panels at multiple pointsOne bonding point leaves the panel floating at high frequency.

Where shielding keeps noise out, isolation stops it from traveling. The most common hardware is a digital isolator or an optocoupler on the signal lines between the control and the drives, and a DC/DC converter with reinforced insulation on the power side.

The number to check is common-mode transient immunity, usually quoted in kV/μs. A drive cabinet with long motor cables can see 10 kV/μs or more during switching. A 5 kV/μs isolator will pass noise straight through and look like it is doing nothing. Match the part to the measured environment, not to the cheapest option on the distributor page.

Isolation also breaks ground loops. When the machine frame, the control cabinet, and the plant earth are at slightly different potentials, a current flows through the shield and the signal reference. That current shows up as a 50 Hz or 60 Hz beat on the scale reading. A single isolation barrier on the encoder link usually removes it.

Do not isolate everything. Each barrier adds propagation delay, and on a closed-loop position link that delay eats phase margin. Isolate the signal paths that cross between cabinets or between the machine and the plant network. Keep the local servo loop galvanically connected.

Filters handle the noise that is already on the conductor. Three places are worth the effort on a CNC machine: the AC input to the drive cabinet, the low-voltage supply feeding the control and I/O, and the analog front end of any measurement channel.

At the AC input, an EMI filter with common-mode and differential-mode stages attenuates the conducted emissions that the drives generate and that the plant feeds back. Mount it at the cabinet entry, not in the middle of the panel, and keep the input and output wiring separated. Wiring a filter with input and output bundled together bypasses it entirely.

On the DC side, ferrite beads and pi filters on the 24 V rails cut the high-frequency content that rides on the control supply. A ferrite with 100 Ω to 300 Ω impedance at 100 MHz is a reasonable starting point for an encoder supply line. Check the DC current rating first; a saturated bead is a resistor, not a filter.

Analog channels need a low-pass filter matched to the signal bandwidth. A scale or probe channel sampled at 10 kHz does not need content above 5 kHz. Setting the corner at 2 kHz to 3 kHz removes switching residue without adding visible lag to the servo loop.

Grounding gives every signal a defined reference and gives fault current a low-impedance path back to the source. Two different jobs, sometimes in conflict. Safety earth must carry fault current without a dangerous voltage rise. Signal reference must be quiet.

Use a single-point star for the signal reference and bond the machine frame, cabinet, and spindle housing to the plant earth with short, wide conductors. A flat braid is better than a round wire of the same cross-section because the impedance at high frequency is lower. Keep the bonding run under 0.5 m where you can.

The most common mistake is daisy-chaining earth between cabinets. Each segment adds inductance, and the last cabinet in the chain sits on a reference that moves with every drive switching event. Run a dedicated conductor from each cabinet back to the earth bar.

Measure it. A clamp-on earth resistance tester on the bonding path should read well under 0.1 Ω at the frequencies of interest. If the reading drifts when the spindle starts, the bond is inductive and needs to be shortened or widened.

Digital filtering inside the control can reject noise that reaches the encoder or probe input. The usual tools are a moving average, a notch filter tuned to the line frequency, and a median filter for spike rejection on slow channels such as temperature or tool-set probes.

A notch at 50 Hz or 60 Hz with a Q of 10 to 30 removes the dominant line-frequency beat without touching the servo bandwidth. Use it when the disturbance is narrow and known. A moving average over 4 to 8 samples is simpler and works on broadband noise, but it adds group delay that shows up as following error on fast contouring moves.

Modern controls also log following error, current ripple, and encoder signal quality. That data is more useful than any single measurement. A rising following error on one axis, correlated with spindle speed, points at a coupling problem long before the part goes out of tolerance.

Software is not a substitute for hardware. If the noise is strong enough to saturate the encoder receiver, no filter recovers the lost counts. Use software to clean up marginal cases and to confirm that a hardware fix worked.