How to Fix CNC Machines: Symptom, Cause, Action

• 1
  Symptom, likely cause, first action
• 2
  Diagnose before you adjust anything
• 3
  Size and position errors that repeat
• 4
  Errors that drift during the shift
• 5
  Alarms, drives, and the checks that clear them
• 6
  Coolant, chips, and surface finish
• 7
  A seven-step fault routine

Most calls about how to fix cnc machines start with an operator already turning offsets. That is the wrong end. An offset changes where the tool goes, but it does not tell you why the part moved. Write down three numbers before you touch a setting: the measured size, the size the program expects, and how many parts the error has been growing across. Those three numbers narrow the fault to a group in a few minutes.

Then separate repeatable errors from drifting ones. A repeatable error of 0.05 mm on every part is usually a geometry problem: tool length, work offset, or a fixture sitting on a chip. A drifting error that grows from 0.01 mm to 0.04 mm over forty parts is thermal or mechanical. The second kind will come back after lunch even if you dial it out now.

Machine type changes what you check first. On a 3-axis mill the usual suspects are tool length offset and fixture seating. On a 5-axis center, rotary table alignment and post-processor values eat more hours than the spindle ever does. On a mill-turn center, check the sub-spindle sync before you blame the main spindle.

Keep a log. Date, part number, measured error, action taken, result. After three months the log will tell you which machine needs a ball screw and which one just needs a warm-up routine. That log is also what a service engineer asks for first when you hand the machine over.

• 1
  Measure, do not guessDigital caliper and micrometer readings on the same feature, twice
• 2
  One change at a timeChange, cut a test part, record the result, then change the next thing
• 3
  Warm-up firstA cold machine moves 0.01-0.03 mm in the first 30 minutes

A part that is consistently 0.03 mm oversize on a pocket wall points at cutter compensation or tool wear, not at the machine. Check the offset value against the drawing callout, then check whether the operator used radius or diameter compensation. Mixing the two is common when a program comes from another shop. Re-cut one test part after each change.

Tool pullout is the other classic. A roughing end mill in a sidelock holder under heavy axial load can creep 0.02-0.05 mm in a deep pocket. Mark the tool at the holder face with a paint pen before the cut. If the mark moves, the holder or the collet is the problem, not the offset.

On a 5-axis center, position errors on angled faces usually trace to the rotary table, not the linear axes. Indicate the table face and the bore at four points. If runout exceeds the machine spec, no amount of offset editing will hold the part. That is a realignment job, and it needs a service engineer with a ballbar.

When a shop asks us how to fix cnc machines that hold size on one feature and miss on another, we look at the fixture first. Clamping pressure on a thin wall can move the part 0.05 mm or more between rough and finish. Use a torque wrench on the clamps and record the number that holds the part without distorting it.

• 1
  Compensation modeConfirm radius vs diameter before editing any value
• 2
  Tool mark testPaint pen at the holder face reveals pullout in one cut
• 3
  Clamp torqueThin walls move more from clamping than from cutting

Thermal growth is the most misdiagnosed fault on any mill. A spindle running at 12,000 rpm for an hour can grow enough to move Z by 0.02-0.04 mm. If the first part of the morning is good and the twentieth is not, you are watching the machine warm up, not a programming error. Run a 20-30 minute warm-up cycle before the first production cut and the spread usually closes.

Backlash shows up as a position error that changes direction. Command the axis to move 50 mm one way, then back, and measure with an indicator on the table. If the return reading differs from the start by more than the machine spec, the ball screw, nut, or thrust bearing needs attention. Record both directions; a one-sided number hides the answer.

Bearing noise under load is a spindle warning. Run the spindle at 2,000 rpm and 8,000 rpm with no cut and listen. A rumble that appears only at high speed usually means bearing preload loss, and it will get worse with cutting load. Stop running production parts on that spindle until it is checked.

Coolant temperature matters more than most people think. A chiller drifting 3-5 °C over a shift moves the machine frame and the part together, so both go wrong at once and the error looks random. Log the chiller setpoint next to your measurement log for a week. It is a cheap check that explains a lot of strange nights.

• 1
  Warm-up cycle20-30 minutes at production rpm before the first cut
• 2
  Backlash testMove 50 mm and back, compare both indicator readings
• 3
  Spindle listen testNo-load run at two speeds separates bearing noise from chatter
• 4
  Chiller logSetpoint drift of 3-5 °C can move the frame

Read the alarm history before you clear anything. Most controls keep the last 50-100 alarms with timestamps, and the order tells you which one started the chain. Clearing the visible alarm and restarting hides the first event, which is usually the one that matters. Photograph the history screen and keep it with the maintenance log.

Overload trips at idle point at the drive or the cable, not the cutting load. Check the axis load meter with the machine sitting still. A reading that sits high at idle means friction in the ways or a failing bearing. If the load is normal at idle and trips only in cut, look at your depth of cut and feed per tooth first.

Encoder and cable faults are common after a crash or a chip jam near the cable track. Wiggle test the cable while watching the position display; a jump in the readout confirms a break. Do not reroute or lengthen encoder cables to save time. The shielding and length are matched to the drive, and a shortcut here returns as a random alarm weeks later.

Servo tuning is the last thing to touch. If the machine has held tolerance for months and suddenly overshoots, something mechanical changed. Retuning the drive to mask a worn ball screw leaves you with a machine that passes a test part and fails a production run.

• 1
  Alarm history firstThe first alarm in the chain is the useful one
• 2
  Idle load readingHigh load with no cut points at friction, not feed rate

Coolant that smells like a drain and etches the vise is a bacteria problem, not a chemistry problem. Dump the sump, scrape the sludge, and clean the tank before you recharge. Refill with a 6-8% concentration measured with a refractometer, not by eye. Below 5% you get rust and tool wear; above 10% you get skin irritation and foaming.

Chip evacuation drives finish more than most speed and feed changes. Recutting chips in a pocket raises Ra fast and breaks small end mills. Increase coolant pressure at the tool, or switch to a through-tool holder if the machine supports it. On deep pockets, a peck cycle that clears chips beats a faster feed every time.

Finishing parameters have a narrow useful window. For aluminium, Ra 0.8-1.6 μm is a normal as-machined result with a sharp cutter and good chip clearing. If you need Ra 0.2-0.8 μm, plan a separate finishing pass with a smaller stepover instead of pushing the same tool harder. Chasing finish with rpm alone burns cutters.

Tram and level are worth checking once a year or after a move. A machine that is out of level twists the bed and shows taper on long parts. It is a two-hour job with a precision level and it prevents a lot of scrap in the following months.

• 1
  Refractometer6-8% is the working range for general steel and aluminium
• 2
  Chip clearingRecut chips are the top cause of sudden finish loss
• 3
  Separate finish passRa 0.2-0.8 μm needs its own light cut, not more rpm