MR-J2S Servo Axes: Hitting 5 um CNC Punch Accuracy

Overview: The Real Constraint Is Mechanical, Not the Drive

A four-axis machine built around Mitsubishi MR-J2S amplifiers and an A-series controller is being compared against a Fanuc or Siemens CNC for a high-speed FPC punching machine with a punch-center tolerance of 5 um (0.005 mm). The reported failure mode is not servo following error - it is machine squareness. That is the normal outcome. On a point-to-point punching machine the servo loop resolution is one to two orders of magnitude finer than the achievable mechanical accuracy, so the drive brand is rarely the limiting term in the error budget.

A general-purpose CNC (Fanuc, Siemens) buys you high-order contouring: look-ahead block processing, tool-radius compensation, 3D interpolation smoothing, and a mature G-code interpreter. A punching machine that indexes to X/Y coordinates and fires a ram does not consume any of that. What it consumes is repeatability, thermal stability and geometric alignment.

Error Budget for a 5 um Punch-Center Tolerance

Build the budget before choosing components. Treat 5 um as the total allowed deviation of the punch center from nominal, then allocate. A conservative split for a two-axis positioning table (values below are an allocation example, not a manufacturer spec - verify each term against component certificates):

Note that squareness and thermal drift alone can eat the whole budget. On a high-speed punch running continuous strokes, screw heating is a real term - the screw grows axially and the error is proportional to travel.

Component Selection: Screw, Guide, Preload

For simple positioning at this tolerance the practical selection is:

• Ballscrew: C5 accuracy grade minimum. Use a preloaded nut (double nut or oversized-ball preload) to remove axial backlash. Fix both ends and pretension the screw if travel and duty are high; a fixed-supported arrangement leaves thermal growth uncontrolled at the free end.
• Linear guides: P (precision) grade rails and blocks, with light or medium preload. Zero or light preload gives lower friction but allows block deflection under punch reaction load; medium preload increases stiffness and reduces pitching under the ram impulse. For a punch, favor medium preload on the axis that absorbs the reaction.
• Bearings: Preloaded angular-contact pairs at the fixed end. Any axial float here appears directly as position error and will not be repeatable, so it cannot be compensated.
• Base: Rail mounting faces machined or scraped in one setup. Squareness that is not built in must be scraped in or shimmed in - it cannot be tuned out of the servo.

Servo Resolution and Command Chain (MR-J2S)

Compute the mechanical resolution per encoder count before worrying about tuning:

Resolution [mm/count] = Ballscrew lead [mm] / N_counts_per_rev

Worked example (ASSUMPTION: N = 131072 counts/rev - confirm from
the encoder spec of your exact MR-J2S model and motor suffix):
  Lead 10 mm : 10 / 131072      = 0.0000763 mm  = 0.076 um/count
  Lead 5 mm  : 5  / 131072      = 0.0000381 mm  = 0.038 um/count

Even with a 10 mm lead, feedback resolution is roughly two orders of magnitude finer than a 5 um tolerance. The bottleneck is the command path, not the feedback path:

1. Confirm the amplifier interface variant. MR-J2S is supplied in different command-interface versions (pulse-train/analog vs. bus-type). Check the model suffix on the nameplate and match it to the A-series module you are using - a positioning module outputs pulse train, a motion controller uses the servo bus. Mixing these is a hardware-level incompatibility, not a parameter issue.
2. Set the electronic gear. Configure the drive's electronic gear numerator/denominator so one command pulse equals a clean machine unit (0.1 um or 1 um). Choose lead and gear ratio so the ratio reduces to exact integers - a non-integer ratio produces cumulative rounding drift over long moves. Verify the parameter numbers in the amplifier instruction manual for your model.
3. Check pulse frequency headroom. With 0.1 um command units, 10 m/min feed = 166.7 mm/s = 1.667 Mpulse/s. Confirm both the module's maximum output frequency and the drive's maximum input frequency (and use a differential line driver, never open collector, at these rates).
4. Set the in-position window tighter than your allocated servo error term, and require in-position before the punch output fires. Do not fire on "move complete" from the command generator alone.

Tuning and Compensation

• Gains: Run auto-tuning with the real payload mounted, then switch to manual mode and raise position/speed loop gain until you see the onset of hunting, back off, and add a notch filter if machine resonance shows up. Record the load-to-motor inertia ratio - a very high ratio on a light punching table limits achievable gain and lengthens settling time.
• Settling time drives throughput: On a high-speed punch, cycle time is dominated by settle-to-in-position, not by traverse. Higher stiffness (medium preload guides, larger screw diameter, short overhangs) reduces settling more than higher gain does.
• Backlash compensation: Apply only after mechanical backlash is minimized. Compensating more than a few microns hides a mechanical defect that will drift.
• Pitch error compensation: Build a table from laser interferometer data along each axis, both directions. This removes screw lead error and mount-induced error but not guide straightness.
• Squareness: Measure with a granite square and indicator, or with a diagonal (Pythagorean) laser check across the XY envelope. Correct mechanically. Some controllers support a squareness/cross-compensation term - if yours does not, it must be scraped or shimmed.

When to Use an Index Cam Instead

If the punch positions are a fixed, repeating pattern, a mechanical index cam drive gives repeatability set by cam geometry and hardened stops rather than by a servo loop plus screw plus guide stack. It removes settling time, backlash and thermal screw growth from the error budget entirely and typically runs faster on a fixed pitch. The trade is zero flexibility - changing the pattern means changing hardware.

Commissioning Verification Sequence

1. Backlash: Indicator on the table, approach a point from both directions repeatedly. Record the reversal step.
2. Unidirectional repeatability: 10 approaches to the same target from the same direction, indicator or laser. Spread must be well under 5 um. This is the accuracy floor.
3. Linear accuracy: Laser interferometer over full travel, both directions, at production feed and after thermal soak. Build the pitch error table from this run.
4. Squareness: Granite square plus indicator, or diagonal laser measurement. Correct mechanically, then re-run step 3.
5. Thermal: Run a full production duty cycle for 1-2 hours, re-measure a reference position hourly. Log the drift; if it exceeds the thermal allocation, add screw pretension, screw/nut cooling, or a warm-up cycle before production.
6. Process verification: Punch a test array, measure hole-center positions on a CMM or vision system, and compare against the commanded coordinates. This is the only number the customer cares about.