Troubleshooting guide

Error Compensation on the CNC YK5120 Speed Cork Machine

This page is for maintenance engineers and process planners who run the CNC YK5120 speed cork machine and see indexing drift, taper, or pitch error after a few thousand cycles. It maps five common symptoms to their root causes and to the checks that actually move the number back.

Get Instant Quote Talk to an Engineer

±0.005 mm tolerance100% inspectionISO 9001:2015NDA on request

CNC YK5120 speed cork machine error compensation checks on a gear cutting setup

On this page

Symptoms, Causes, and What to Do First

Read the left column first. The middle column is the most likely cause on a machine that has already been aligned once.

Symptom	Likely cause	First action
Indexing drift grows over a shift	Thermal growth in the work head	Log spindle and table temperature every 30 minutes
Constant offset on every part	Tool wear or a wrong tool offset	Re-measure the tool and reset the offset
Pitch error changes along the axis	Lead screw pitch error, not yet mapped	Run a laser interferometer pass on that axis
Taper across the gear face	Squareness error between axes	Check column-to-table squareness with a granite square
Chatter marks at one index angle	Backlash in the worm or index servo	Measure backlash at that angle, not at zero
Roundness drift on the bore	Spindle radial runout or thermal bend	Indicate the spindle nose hot, not cold

Fix the hardware, then compensate the rest

Error compensation on the CNC YK5120 speed cork machine works on repeatable error only. Measure with a laser and a ballbar, fit a small model, and verify on new gears. If backlash or stiffness is the real problem, no table will cover it.

Get Instant Quote Talk to an Engineer

Why this machine drifts

Why the CNC YK5120 Speed Cork Machine Loses Accuracy

Gear cutting machines like the YK5120 lose accuracy in two ways. The first is geometric: the axes are no longer square, parallel, or at the right pitch. The second is thermal: the machine was aligned cold and then heats up during the first two hours of a shift. Both show up as the same complaint from the shop floor, which is that parts were good yesterday and are not good today.

The kinematics of this machine are a chain. The bed carries the column, the column carries the work head, and the work head carries the gear blank. Every link in that chain has a small angular error. Under load, those errors add up rather than cancel out. A 20 µm squareness error at the column can become a 60 µm pitch error at the tooth flank once the blank is 300 mm from the center of rotation.

That is the argument for software error compensation instead of a rebuild. You measure the actual error field with a laser interferometer, a ballbar, and a set of test gears. You fit a model. You push the model into the controller as a compensation table. The machine then moves to a corrected position for every commanded point.

The limit is important. Compensation corrects repeatable error. It cannot correct random error. If backlash changes every cycle because a bearing is failing, no table will fix it. Fix the hardware first, then compensate what is left.

On the YK5120, the compensation model usually covers three groups. Geometric error covers squareness and parallelism between axes. Thermal error covers the drift that follows spindle and table temperature. Load error covers deflection under cutting force, which matters most on the finishing pass.

1
Compensate repeatable errorIf the same point is off by the same amount twice, a table can fix it.
2
Fix random error firstBacklash that changes cycle to cycle is a hardware problem.
3
Measure hot and coldA cold alignment and a hot alignment are not the same alignment.

Measurement

How to Measure Error Before You Compensate It

You cannot compensate what you have not measured. The order matters, because each instrument answers a different question and a bad order wastes a day.

Start with a laser interferometer on the linear axes. It gives you positioning error, reversal error, and straightness along each axis. Run the axis at the feed rates you actually cut at, not at a slow setup jog. Error at 50 mm/min and error at 3,000 mm/min are different numbers on the same machine.

Then run a ballbar circle test. A ballbar tells you squareness, backlash, and servo mismatch in one setup. If the circular trace looks like a lemon, you have squareness error. If it has flat spots at the quadrant changes, you have backlash or friction. If it is an ellipse, you have a scale or servo gain mismatch.

For the rotary and work head side, use test gears. Cut a set of gears, measure lead and profile on a gear measuring center, and compare against the commanded values. This is the only measurement that includes the full kinematic chain the way the part sees it.

Record temperatures at the same time. A spindle that is 8 °C warmer at the end of the run will have moved the tool relative to the work. If your error plot and your temperature plot rise together, you are looking at a thermal problem.

1
Laser firstPositioning, reversal, and straightness on each linear axis.
2
Ballbar secondSquareness, backlash, and servo mismatch in one circle.
3
Test gears thirdLead and profile error through the full kinematic chain.
4
Log temperatureSpindle and table, every 30 minutes, same sheet as the error data.

Modeling

Building the Compensation Model for the CNC YK5120 Speed Cork Machine

A useful compensation model has to do three things. It has to predict error at a point the machine has not visited. It has to update when the machine warms up. And it has to be cheap enough to run inside the control cycle.

The common approach is a multi-body kinematic model. You describe the machine as a chain of rigid bodies. Each body has a nominal position and a small error term for each degree of freedom. You multiply the transforms along the chain and read the tool position error at the end. The method is standard in machine tool metrology and it fits a machine with one rotary work head well.

The model has six error terms per joint in the general case. In practice, on a machine like this, most of them are small enough to drop. Keep the terms that your measurement data can actually see. A model with twenty coefficients fitted from four data points is not a model, it is a guess.

Thermal terms need a different treatment. Geometric error is roughly constant over a shift. Thermal error moves. Fit it as a function of temperature, or as a function of time since cold start with temperature as the input. A simple linear fit against spindle temperature often removes 60 to 80 percent of the drift on a small gear machine.

Test the model on data you did not fit it with. Cut a gear, measure it, and compare. If the residual is not smaller than the raw error, the model is adding noise, not removing it.

1
Keep the model smallOnly fit coefficients your measurement data can resolve.
2
Fit thermal separatelyGeometric error is constant; thermal error is not.
3
Validate on new dataAlways hold back a gear or an axis run for testing.

Shop floor procedure

Step by Step: Bringing the Machine Back Into Tolerance

Do these in order. Skipping a step usually means you compensate a hardware fault and it comes back in a week.

1
1. Baselines the machine coldPark the machine overnight. Record spindle, table, and ambient temperature. Measure positioning error on the linear axes with a laser interferometer at the cutting feed rate. Save this as the cold baseline.
2
2. Repeat the measurement hotRun the machine for two hours under a typical cutting load. Repeat the laser run and the temperature log. The difference between the cold and hot plots is your thermal error budget.
3
3. Check the mechanicsIndicate the spindle nose for radial and axial runout. Check backlash on each linear axis with a dial indicator at the feed rate you cut at. Anything above 5 µm needs a mechanical fix before you write a compensation table.
4
4. Run a ballbar circle testSet the ballbar at a 150 to 300 mm radius. Run at 500 mm/min and at 2,000 mm/min. Compare the two traces. A shape that changes with feed rate points to a servo or gain problem, not a geometric one.
5
5. Cut and measure test gearsCut three gears at different index positions. Measure lead, profile, and pitch on a gear measuring center. Log the data against the commanded values. This is the number you are trying to fix.
6
6. Fit and load the compensation tableFit the geometric terms from the laser data and the thermal terms from the hot and cold plots. Load the table into the controller. Keep the model to the terms your data supports.
7
7. Verify on a production batchCut a full batch and measure a sample. Compare the residual error against the step 5 numbers. If the residual is not smaller, roll the table back and re-check step 3.

FAQs

Questions Engineers Ask About Error Compensation

How often should the compensation table be refreshed?

On a machine that runs one shift a day, a full re-measurement every six months is usually enough. Machines that run three shifts, or that cut hard materials, should be checked quarterly.

Any time you change a bearing, a lead screw, or the work head, treat the table as invalid. A mechanical change moves the error field, and the old table will now be wrong in a new way.

Can compensation replace a rebuild?

For repeatable geometric and thermal error, yes. A well-fitted table can recover most of the accuracy a worn machine has lost, at a fraction of the cost of a rebuild.

It cannot recover lost stiffness. If the machine deflects under load in a way that changes with the part, compensation will not help. That is a mechanical problem and it has to be fixed mechanically.

What tolerance can we actually hold after compensation?

It depends on the machine and the part. On a well-maintained gear machine, compensation typically brings the residual error down to a third or a quarter of the uncompensated value.

For parts that need tight limits, we machine and inspect to ±0.005 mm on our own 5-axis and mill-turn equipment, with 100% inspection before shipment. On gear cutting specifically, the achievable lead and profile error depends on the machine condition, not on the compensation table alone.

Does the compensation table slow down the control cycle?

A small table does not. The control has to look up a correction for every commanded point, which is a few extra arithmetic operations.

A large table with hundreds of coefficients can add measurable cycle time and can also add numerical noise. Keep the model as small as the data allows.

What temperature should we use as the model input?

Spindle temperature is the single best input on most machines. It responds fastest to cutting load and it sits closest to the error source.

If the machine has a long bed, add one or two table or bed sensors. The bed moves slower than the spindle, and a single sensor will miss the lag.

How do we know the model is wrong rather than the machine?

Cut a gear at a point you did not use to fit the model. If the residual error is larger than the raw error was, the model is wrong.

A model that fits its own data well and fails on new data is usually too complex. Drop terms until it generalizes.