CNC System Checking Method: Seven Steps From Alarm to Root Cause
This guide is for maintenance engineers and shop supervisors who need to find a fault on a CNC control without swapping parts at random. It walks through a repeatable CNC system checking method, from reading the alarm log to measuring backlash and spindle runout. By the end you will know which checks to run first, what numbers are acceptable, and when the fault is mechanical rather than electronic.
In this article
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8
What matters before you open the cabinet
Why a fixed CNC system checking method beats experience alone
Most CNC faults are not exotic. A servo alarm, a drifting dimension or a spindle that sounds wrong usually traces back to one of six places: power, signal, parameter, mechanical drive train, tooling, or the part program. The problem is that experienced staff jump straight to the place they have seen fail before, and on a different machine that instinct sends them down the wrong path for two hours.
A documented checking method fixes the order of operations. You start where information is cheapest to collect, and you finish with the checks that require pulling covers or stopping production. That sequence matters on a machine that is mid-job with a hot spindle and a fixture already dialed in.
The method below assumes a vertical or horizontal machining center with a modern control, but the logic carries over to turning centers and mill-turn machines. The parameter ranges and tolerance figures are the ones we use on our own 127 machines across three plants, where we hold ±0.005 mm on production parts.
One rule sits above all seven steps: never swap a board to test a theory. Swap to confirm a diagnosis you already have from measurement.
- 1Cheap information firstAlarms, status bits and operator notes cost nothing to collect.
- 2Measurement before disassemblyDial indicator and ballbar readings tell you where to look.
- 3One variable per passIsolate the fault instead of disturbing five things at once.
- 4Write it downThe next shift needs your findings, not your memory of them.
Start with the alarm log, then confirm power and signal
Pull the alarm history before touching anything. On most controls the log shows the first-out alarm, not the cascade that followed it. A drive overload on the Z axis will trigger three or four downstream alarms about position deviation and emergency stop; the first entry is the one that matters.
Write down the alarm number, the axis or spindle named, and the machine coordinates at the moment of the fault. If the same alarm appears at the same coordinate three times, you have a repeatable fault and a strong clue. If the coordinate moves each time, suspect a loose connection or a thermal effect rather than a hard failure.
Next, confirm the control is actually talking to the drive. Check the 24 V DC logic supply at the terminal strip with a meter, not by eye. Anything below 23 V under load causes intermittent signal faults that look like software problems. Then read the drive status LEDs or the diagnostic screen: ready, enabled, fault, or no communication.
No communication is an electrical question. Check the fiber optic or bus cable seating, the termination resistor, and the shield ground at both ends. A shield grounded at one end only is correct for most signal cables; grounding both ends invites a ground loop that shows up as random alarms during spindle acceleration.
- 1First-out alarm onlyDownstream alarms are usually consequences.
- 224 V under loadBelow 23 V DC, treat the supply as faulty.
- 3Shield grounded one endTwo-end grounding creates ground loops.
Check parameters and compare them against a known backup
Parameter drift is quiet. Nobody changes a servo gain on purpose, but a battery failure, a control swap or a power surge during a storm can reset values to default. The symptom is a machine that runs but leaves a poor surface finish, overshoots on rapid moves, or alarms only at high feed rates.
Compare the live parameter set against your dated backup, axis by axis. Focus on servo gain, position loop gain, acceleration and deceleration time constants, backlash compensation, and pitch error compensation. A single digit change in backlash compensation shows up as a step in the part, not as an alarm.
If the machine has been sitting idle for weeks, re-check the absolute position data and the battery voltage on the encoder. A weak battery lets position data shift while the machine is powered down, and the first move after startup will be wrong by a fixed offset. That offset is not a mechanical problem, and no amount of gib adjustment will remove it.
Only after the parameters match the backup should you touch the mechanics. Editing compensation values to hide a worn ball screw guarantees the error returns when the wear increases.
- 1Compare, do not assumeDiff the live set against a dated backup.
- 2Encoder battery voltageReplace before it drops below the manufacturer's threshold.
- 3Do not mask wearCompensation hides a failing ball screw for a few weeks.
Run a functional program to test interpolation and cycles
Once power, signal and parameters check out, prove the control's motion functions with a short test program rather than a production job. Write one that exercises linear interpolation on all three axes, circular interpolation in at least two planes, a spindle speed change through the gear range, and one canned cycle.
Run the program in single block with rapid override at 25 percent for the first pass. Watch the position display against a dial indicator on a known surface. A deviation that grows with distance points to pitch error or a slipping coupling. A deviation that appears only on reversal points to backlash.
Add a thread cutting pass if the machine has a spindle encoder. Thread pitch error is one of the clearest indicators that the encoder feedback is losing counts, and it shows up long before a general position alarm.
Keep the test program short, under two minutes, and store it under a fixed program number. It becomes your repeatable health check, and it costs almost nothing to run between jobs.
- 1Single block first25 percent rapid override on the first pass.
- 2Error grows with distanceSuspect pitch error or a slipping coupling.
- 3Error on reversal onlySuspect backlash in the drive train.
- 4Thread pitch checkCatches encoder count loss early.
Measure backlash and spindle runout, then judge mechanical wear
Backlash measurement is simple and it settles most arguments. Mount a dial indicator with 0.001 mm resolution against a flat surface on the table or turret, move the axis in one direction, zero the indicator, then reverse. Read the lost motion. On a production machining center, keep X and Y backlash under 0.010 mm and Z under 0.015 mm.
If backlash is out of range, check the coupling, the thrust bearing preload and the ball screw nut before you replace anything. A loose coupling clamp bolt produces backlash that varies with load, while a worn nut produces consistent backlash. The two feel similar at the indicator but need different repairs.
Spindle runout tells you whether the problem is in the spindle or in the toolholder. Measure taper runout with a test bar, typically under 0.005 mm for a healthy spindle. Then measure at the tool tip. If taper runout is good and tip runout is not, the problem is the holder, the collet or chips on the taper.
Judgment call: if backlash and runout are within range, the fault is electrical or in the parameters. If they are out of range, fix the mechanics first, then re-run the functional program. Do not chase both at the same time.
- 1X and Y backlashKeep under 0.010 mm on a production machine.
- 2Z backlashKeep under 0.015 mm.
- 3Taper runoutUnder 0.005 mm with a clean test bar.
- 4Taper good, tip badThe holder or collet is the problem.
Seven-step CNC system checking method
Work through these in order. Stop at the first step that fails and fix that before continuing.
- 1Capture the fault stateRecord the first-out alarm number, the named axis or spindle, and the machine coordinates. Photograph the diagnostic screen. Ask the operator what the machine was doing in the ten seconds before the fault.
- 2Verify the 24 V DC and drive statusMeasure at the terminal strip under load. Treat anything under 23 V DC as a fault. Read the drive LEDs or diagnostic page and note ready, enabled, fault or no communication.
- 3Check communication and groundingReseat fiber optic and bus connectors. Confirm the termination resistor is fitted at the end of the bus. Confirm signal cable shields are grounded at one end only.
- 4Diff parameters against the backupCompare servo gain, position loop gain, accel and decel time constants, backlash compensation and pitch error compensation, axis by axis. Restore any value that differs without a written reason.
- 5Run the functional test programSingle block, 25 percent rapid override, one pass. Check linear moves, circular interpolation in two planes, a spindle speed change and one canned cycle. Add a thread pass if a spindle encoder is fitted.
- 6Measure backlash and runoutDial indicator on a flat surface, reverse direction, read lost motion. Target under 0.010 mm on X and Y, under 0.015 mm on Z. Then measure taper runout and tool tip runout separately.
- 7Decide and documentIf mechanics are in range, the fault is electrical or parameter related. If not, repair the drive train first and re-run the test program. Write the finding, the fix and the new baseline numbers into the machine log.
Symptoms, likely causes and where to check first
Use this as a quick pointer after you have read the alarm log.
| Symptom | Likely cause | Check first |
|---|---|---|
| Position alarm at the same coordinate | Mechanical binding or a hard stop | Backlash and coupling on that axis |
| Position alarm at random coordinates | Loose connection or ground loop | Cable seating and shield grounding |
| Poor finish, no alarm | Parameter drift or worn spindle bearing | Servo gain and taper runout |
| Dimension shifts after power down | Weak encoder battery | Absolute position data and battery voltage |
| Overshoot on rapid moves | Accel or decel time constant changed | Compare parameters to backup |
| Step in the part at one position | Backlash compensation value wrong | Backlash compensation per axis |
| Thread pitch error | Encoder losing counts | Spindle encoder and thread test pass |
| Alarm only at high feed rate | Drive overload or mechanical drag | Servo load display and way lubrication |
Measure first, swap last
A CNC system checking method only pays off if you follow the order: alarm log, power and signal, parameters, functional test, then mechanics. If backlash and runout are in range, the fault is electrical. If they are out, fix the drive train before you touch another parameter.
Questions engineers ask about CNC system checks
How often should the full checking method be run?
Run the short functional program and read backlash monthly on machines that run two or three shifts. Run the full seven-step sequence after any control repair, a power event, or a machine move.
If the machine sits idle for more than four weeks, run steps one through five before releasing it to production. Encoder batteries and absolute position data are the usual casualties of a long stop.
Can a parameter error really look like a mechanical fault?
Yes, and it is one of the most common misdiagnoses. A wrong backlash compensation value produces a visible step in the part that looks exactly like a worn ball screw nut.
The difference is that parameter drift appears suddenly after a battery change or control swap, while mechanical wear develops over weeks. The machine log usually tells you which one you are looking at.
When is a board swap justified?
Only when you already have measurement evidence pointing at that board. For example, the drive reports a fault, the motor and cables test good, and the same fault follows the board when you move it to an identical axis.
Swapping a board to see what happens is how good machines end up with two faults. It also risks damaging the replacement board if the original fault is in the wiring.
What backlash numbers are acceptable on a production machine?
Keep X and Y under 0.010 mm and Z under 0.015 mm for general machining center work. For finishing work at ±0.005 mm tolerance, tighter figures are needed on the axes that set the critical dimension.
Check the backlash on the axis that holds your tightest tolerance first. Not every axis needs the same number, and chasing the last micron on a non-critical axis wastes downtime.
How do we tell a spindle problem from a toolholder problem?
Measure taper runout with a clean test bar, then measure at the tool tip. Taper runout under 0.005 mm with a bad tip reading means the holder, collet or taper cleanliness is at fault.
If both readings are out, the spindle itself needs attention. Clean the taper and re-measure before you conclude anything, because a single chip can double the tip reading.
Do we need to stop production to run these checks?
No. Steps one through four can be done between jobs with the machine powered and the fixture still mounted. Steps five through seven need the spindle stopped and the table clear.
The functional program takes under two minutes, so it fits into a tool change window. Backlash measurement takes about ten minutes per axis with practice.
Need machining support while a machine is down?
Send us the drawing and we will run a free DFM analysis within 12 hours, with production able to start in 24 hours and no minimum order quantity. Uploads stay secure and confidential, and an NDA is available on request.
12-hour quote100% inspection±0.005 mm toleranceNo minimum order quantity
More CNC machining know-how
We publish setup notes, tooling trials and inspection data from the factory floor.
