SINUMERIK 840Di Reversal Jerk on Heidenhain Linear Scale Axes: Backlash, Exact Stop, and Kv Tuning

1. Problem Statement

On a Pegard 5-axis + spindle machine controlled by a Siemens SINUMERIK 840Di PL, the X axis is equipped with a Heidenhain external linear scale as the position feedback element. The configured mechanical backlash value in the controller is 0.02 mm. Under the following conditions the axis produces a measurable transient spike:

• Operating mode: 0.01 mm handwheel (JOG incremental)
• Command: traverse in the negative direction by 0.01 mm
• Observed behavior: the dial indicator first shows a displacement of approximately 0.1 mm, then settles back to 0.01 mm
• Same-direction moves: displacement is correct, no overshoot
• Reverse-direction moves: 0.1 mm fluctuation on the dial (overshoot of ~10× the commanded distance)

The same 0.1 mm reversal spike is not observed for unidirectional moves. Mechanical slide clearance has been measured at less than 0.02 mm and is therefore ruled out as the root cause. Reducing the position controller gain MD32200 from 0.5 to 0.1 and reducing the drive parameter MD1407 from 12.66 to 10 produced no measurable improvement.

2. Why the Reversal Spikes: Mechanics of Backlash Compensation

When a CNC axis reverses direction the controller must first take up the lost motion (mechanical backlash) before the actual load-side position begins to move. SINUMERIK handles this in two stages:

1. Detected lost motion between the motor-side feedback and the load-side feedback (when both are active).
2. Compensating travel in the form of a programmed position offset, injected as soon as the direction reversal is detected.

If the commanded reversal step is small — for example 10 µm in handwheel mode — and the configured backlash is also small (20 µm), the compensator must still deliver the full lost-motion distance before the 10 µm command becomes visible. The axis will therefore look like it traveled 0.1 mm and then came back to 0.01 mm, because the displayed actual value is the difference between the load-side position and the internally compensated motor position.

The key insight from the field is:

If the axis is on a closed-loop direct-measuring system (linear scale) and the mechanical lost motion is already small, additional software backlash compensation introduces a transient that is larger than the physical defect it is trying to fix.

Setting the backlash value to zero eliminates the corrective travel, and the reversal spike disappears immediately. The trade-off is loss of compensation for thermal growth of the ballscrew and for the actual physical lost motion still present in the drive train.

3. Linear Scale vs. Motor Encoder: Feedback Architecture

The 840Di (and the 840D sl family) supports dual-encoder axes. For each axis there can be up to two feedback channels:

Relevant 840Di machine data for encoder configuration includes:

• MD30200 $MA_NUM_ENCS — number of encoders assigned to the axis (1 or 2)
• MD30240 $MA_ENC_FEEDBACK_POL[] — feedback polarity of each encoder
• MD31020 $MA_ENC_RESOL[] — encoder resolution
• MD32110 $MA_ENC_FEEDBACK_POL[] — feedback polarity for position control
• MD32900 $MA_DYNAMIC_MATCH_ENABLE — dynamic matching of two encoders

When the linear scale is the position-control feedback (the typical configuration on a precision 5-axis), the actual position reported to the operator and the position used for the position-control loop are derived from the scale. The motor encoder is used only by the speed and current loops inside the drive. In that case the controller already "sees" the true load-side position, and the closed loop is no longer blind to the backlash of the ballscrew.

4. The Two Backlash Values per Axis

For every axis the SINUMERIK 840D/840Di-sl data block contains two backlash values, indexed [0] and [1]:

This split is a frequent source of confusion. On an axis with a linear scale, the value that drives the reversal transient is $MA_BACKLASH[1]. Verify the active feedback path on the operator panel under Diagnosis > Axis Status (this is also where the active encoder selection is shown for cross-checking). The PLC can switch the active feedback system through the interface DB; on a Pegard 5-axis the active encoder is normally fixed by configuration, but a configuration error in MD30200 or in the NC-PLC interface can leave the axis running on the wrong feedback path with the wrong backlash value active.

5. SINUMERIK 840Di Backlash Machine Data

6. Drive Parameter MD1407 (Sinamics / 611U-Compatible)

MD1407 is a drive-level parameter visible in the 840Di commissioning tree. On Sinamics S120 drive objects that interoperate with the 840Di (NCU software 1.5 / 2.5 and later) it normally maps to a torque / current loop proportional gain of the inner speed controller, with typical values in the 8–20 ms integral-time / gain range. The original value on the reported machine was 12.66, reduced to 10 during diagnosis. Because MD1407 acts inside the speed-control loop, it cannot remove a reversal transient whose source is in the position-control loop (i.e. backlash compensation). Reducing it generally softens the response of the inner loop and may slightly delay the moment at which the overshoot peaks, but does not eliminate the overshoot itself. This is exactly the symptom observed on the field: 0.5 → 0.1 on MD32200 and 12.66 → 10 on MD1407 produced no improvement.

7. Position Loop Gain MD32200

MD32200 $MA_POSCTRL_GAIN — the position-controller gain factor K_v — sets the proportional relationship between the position error and the velocity command issued to the drive. Units are (m/min) / mm. Typical values:

The position loop bandwidth in rad/s is approximately K_v / 60 (for K_v in (m/min)/mm). Reducing K_v reduces the closed-loop stiffness and therefore the reversal spike would actually be expected to increase in some configurations, not decrease. This matches the field observation: lowering MD32200 from 0.5 to 0.1 did not solve the issue. The right way to attack a reversal transient that is dominated by backlash compensation is not to reduce the loop gain, but to remove or shape the compensating travel itself.

8. Exact Stop Limits: MD36000 and MD36010

The "in-position" detection of the 840Di is governed by two thresholds per axis:

On the reported Y axis these were set to 1.0 mm and 0.5 mm respectively — i.e. very loose. These values do not directly cause the 0.1 mm reversal spike (the spike is a real position overshoot, not a coarse-stop advance). However, leaving the limits loose will mask the symptom in part-program execution because the controller considers the axis "in position" long before the transient settles. Tightening them to 0.5 mm / 0.1 mm — or better, to 0.02 mm / 0.005 mm on a precision 5-axis — ensures that block advance is correctly delayed until the reversal transient has decayed, and that the G09 / G60 exact-stop instructions behave as expected.

9. Reversal Compensation Ramp (MD32457)

The contributor's note that "MD32457 limits the rate of application of the reversal" is consistent with the SINUMERIK backlash compensation ramp that prevents the compensating travel from being injected as a velocity step into the position loop. Functionally it behaves like a velocity limit on the compensating travel and is expressed in mm/min. The trade-off is the same as for any input shaper:

• Aggressive (high value): the compensation is applied in one cycle of the position controller — the reversal transient on the load is fast but the displayed actual value will follow the compensating travel immediately.
• Soft (low value): the compensation is spread over several position-controller cycles — the transient is gentle but the axis can be off-position for tens of milliseconds after a reversal.

If you cannot zero the backlash value (for example because thermal compensation requires the residual value to be retained), MD32457 is the correct knob to soften the spike. A starting point on a precision linear-scale axis is 50% of the present value, then iterate against the dial-indicator reading in 0.01 mm handwheel mode.

10. Step-by-Step Diagnostic Procedure

The following procedure has been used in production to isolate and resolve the same class of reversal transient on 840Di-controlled machines.

1. Verify the active feedback path. On the operator panel, open Diagnosis > Axis Status. Confirm that the load-side encoder (linear scale) is shown as the active position feedback. If it is not, the axis is closing its position loop on the motor encoder and the linear scale is only used for display — in that case, the entire 0.1 mm spike is real mechanical lost motion, not a compensation artifact.
2. Read the active backlash values. Display $MA_BACKLASH[0] and $MA_BACKLASH[1] for the affected axis. The active index matches the active encoder. On a direct-measuring axis [1] is the relevant value.
3. Temporarily zero the backlash value. Set MD32450 = 0 (or the relevant [0] / [1] sub-index) and re-test the 0.01 mm reversal. The spike should disappear, confirming that the compensation travel is the source.
4. Measure the physical backlash directly. With the compensation value still zero, push the axis against a fixed stop with a dial indicator, apply ~10 N of force in the reverse direction, and read the indicator. This is the true lost motion of the drive train. On a Pegard 5-axis with a Heidenhain LC/LS scale, values of 0.05 – 0.15 mm are typical of a preloaded ballscrew and nut.
5. Re-enable backlash only if the measured physical lost motion is non-zero and unacceptable. If the physical value is < 0.005 mm, leave MD32450 = 0 and the position loop on the scale will absorb the rest. If the physical value is larger, re-enter the measured value in MD32450 and use MD32457 to soften the reversal ramp.
6. Tune MD32457 if required. Iterate on MD32457 to find the lowest value that still allows the dial indicator to settle at the commanded position within one handwheel step (typically ~50 ms at 0.01 mm).
7. Re-check the position loop gain. Restore MD32200 to the value the machine was originally commissioned with (0.5 is conservative for a 5-axis slide). Lowering K_v trades bandwidth for compliance and is the wrong response to a backlash-induced reversal spike.
8. Tighten the exact-stop limits. Set MD36000 to 0.5 mm and MD36010 to 0.1 mm as a baseline. On a precision 5-axis with K_v = 1.0, MD36010 = 0.005 is achievable but must be verified against the steady-state following error.
9. Document the final values in the machine's commissioning record with the date, the operator panel version (840Di Operate 4.x / 4.5 / 4.7 / 4.8 — different SP releases default to different backlash algorithms), and the build of the Sinamics drive firmware.

11. Verification Procedure

After the tuning is complete, run the following test sequence and record the results for the machine log:

1. Dial-indicator test (JOG handwheel): command 0.01 mm in +X, then 0.01 mm in –X, ten times. The indicator must stay within ±0.005 mm of the commanded position at all times. No 0.1 mm spike.
2. Laser linear test (500 mm stroke): traverse the X axis back and forth over the full 500 mm stroke at feed 1000 mm/min. Compare the axis position reported by the scale to the laser reference. With K_v = 1.0 and a linear-scale feedback, the position error should be < 0.01 mm / 500 mm. A 0.05 mm / 500 mm error indicates a geometric or thermal issue, not a control issue, and is not improved by K_v tuning.
3. Block-change test (parts program): write a 10-block program that does only 0.01 mm reversals at F500. With MD36000 = 0.5 the block advance must happen in less than 200 ms per block. If "exact stop not reached" alarms appear, increase MD36000 by 0.05 mm until the alarms disappear.
4. Contour test: run a circular interpolation G02 / G03 program at 1 m/min with a 50 mm radius on the X-Y plane. The circularity error measured with a 3-D probe or a dial indicator must remain < 0.01 mm. The reversal transient will show up as a localized "flat" or "spike" in the contour; the absence of such features confirms the tuning.

12. Positioning Error Compensation: Why 0.05 mm / 500 mm on a Linear-Scale Axis

A residual 0.05 mm error over a 500 mm stroke on an axis that is closed-loop on a Heidenhain linear scale is almost never a control-loop issue. The most common sources, in order of likelihood on a Pegard-class 5-axis:

Of these, thermal expansion is the most common single cause of a 0.05 mm / 500 mm error. For a steel scale on a cast-iron bed the linear coefficient of expansion difference is about 1–2 µm/m·K. Over 500 mm and a 5 K temperature rise, this is 2.5–5 µm — i.e. not 50 µm, so the 0.05 mm error is more likely to come from a coarse alignment error or from a fixture not yet thermally stabilized at the time of measurement. Always let the machine warm up for at least one full thermal cycle (typically 30–60 min) before performing a laser calibration.

13. Compensation Tools Available in the 840Di

Use the SSFK / MSFK compensation tables (System Software Floating-point Kompensation / Measuring-System Floating-point Kompensation) to map a measured error curve and let the controller apply the inverse. For a 0.05 mm / 500 mm error, build the table from a Renishaw laser or Heidenhain KK laser measurement, with 9 to 21 reference points, and verify that the residual error after compensation is < 0.01 mm / 500 mm.

14. Troubleshooting Matrix

15. Safety and Commissioning Notes

• Verify before clearing alarms: alarm 20062 ("Axis %1 exact stop not reached") and 20063 ("Axis %1 exact stop not reached during jerk-limited deceleration") are the typical alarms raised if MD36000 or MD36010 are tightened below the steady-state following error. Do not blanket-acknowledge these alarms — tighten the limits only after measuring the actual following error with $AA_FOLLOW_ERR in the axis service display.
• Direction reversal test under safe conditions: perform the 0.01 mm handwheel test with the spindle in a safe position, the tool clear of the workpiece, and the axis within the mid-stroke of its travel range to avoid the end-of-stroke deceleration region.
• Backlash vs. friction compensation: SINUMERIK also has a friction-compensation block (MD32500-MD32540) which is independent of backlash compensation. Do not confuse the two. The reversal transient on a linear-scale axis is normally a backlash issue, not a friction issue.
• Firmware consistency: 840Di Operate 4.5 SP2 and earlier use a backlash-compensation implementation that injects the full compensating travel in one position-controller cycle. 4.7 and later have an additional smoothing option activated through a separate MD. If the machine runs Operate 4.5, the only available knob is MD32457; on 4.7+ the smoothing option is preferred.
• Mechanical inspection last, not first: always rule out the control-side causes (backlash, gain, exact-stop limits) before disassembling a slide. The field case described here was resolved in software in a few minutes — a slide strip-down would have taken a full day and produced no improvement.

16. Summary of Recommended Values for a Pegard 5-Axis with Heidenhain LC Scale on 840Di

17. FAQ