Machine Tools Suddenly Decreases in Accuracy: 4 Diagnostic Basics

• 1
  What machine tools suddenly decreases actually looks like
• 2
  Thermal behavior is the most common hidden cause
• 3
  Mechanical causes: ballscrew, guideway, and clamping
• 4
  The five diagnostic methods in order of cost
• 5
  When the cause is not the machine at all
• 6
  Symptom, likely cause, and first check

A gradual wear curve is normal. You re-touch a tool, adjust a wear offset, and the part comes back into tolerance. The event that matters is a step change: one morning the same program, same fixture, same material produces a bore that is 20 μm oversize. That is when machine tools suddenly decreases in a way that needs diagnosis rather than another offset tweak.

The first job is to define the symptom precisely. Is the error dimensional, geometric, or surface finish? A bore that is consistently 0.02 mm small on every part points to a tool or offset issue. A bore that varies 0.015 mm part to part points to thermal growth, loose fixturing, or a ballscrew problem. The distinction changes which of the five methods you run first.

Write down three numbers before touching anything: the nominal tolerance, the measured deviation, and the repeatability across five consecutive parts. Without that baseline, every adjustment is a guess. A machine that has been running at ±0.005 mm and now holds only ±0.03 mm has a real problem. A machine that was always at ±0.03 mm has a capability mismatch, not a sudden failure.

We see this on our own 127 machines. The step change is almost never the spindle bearings. It is usually a thermal event, a clamp that moved, or a tool that chipped without breaking. The spindle is the last place to look, not the first.

• 1
  Step change vs. driftA step change within one shift points to a discrete event, not wear.
• 2
  Dimensional vs. geometricSize errors track offsets; roundness and squareness errors track mechanics.
• 3
  Repeatability firstFive consecutive parts separate a random cause from a systematic one.

Cast iron and steel grow about 10–12 μm per meter per °C. A 4,000 mm machine bed that warms 5 °C from morning to afternoon moves 200–240 μm at the extreme end if the thermal gradient is not compensated. That is 40 times larger than a ±0.005 mm tolerance. The error is real, it is geometric, and it does not show up on a cold morning check.

Spindle growth is the same story on a smaller scale. A spindle running at 12,000 rpm for two hours can grow 15–25 μm axially. If your Z-axis reference is set cold, every part after lunch is short by that amount. The fix is not a new spindle. It is a warm-up cycle, a spindle growth sensor, or a reference shift tied to spindle temperature.

Coolant temperature matters as much as ambient air. A chiller set at 20 °C and a shop at 28 °C creates a gradient across the machine column. We have seen a 0.04 mm taper appear on a 300 mm bore simply because the coolant tank was undersized and the return temperature climbed 6 °C over a shift.

The diagnostic method here is simple. Measure the same feature at 7 a.m., 11 a.m., and 3 p.m. with a warm spindle and stable coolant. If the error tracks the temperature curve, the cause is thermal. Do not chase mechanical fixes until you have ruled this out.

• 1
  Warm-up before first cutRun the spindle 30–45 minutes at production speed before setting offsets.
• 2
  Log coolant temperatureA 5 °C swing in coolant is enough to move a 300 mm bore.
• 3
  Time-stamped measurementThree measurements across a shift reveal a thermal signature.

A ballscrew that has lost preload shows up as backlash. You command 0.01 mm and the table moves 0.006 mm, then catches up. On a finishing pass this produces a step or a witness mark where the tool reverses. The error is directional and repeatable, which makes it easy to confirm with a dial indicator against the table.

Linear guideway wear is different. It shows as a change in squareness or parallelism over the stroke, not as backlash. A part that is square at the center of travel and out 0.02 mm at the ends points to rail wear or a twisted bed. This is a geometry problem, and no offset will fix it.

Clamping is the cause engineers overlook most often. A vise with 0.01 mm of lift on the movable jaw, or a hydraulic chuck with uneven jaw pressure, distorts the part during cutting and relaxes it after. The measurement on the machine looks good. The measurement on the CMM does not. If the error appears only after unclamping, the fixture is the problem.

Toolholder runout belongs here too. A worn taper or a chip on the spindle nose face gives 0.01–0.03 mm of runout at the tool tip. On a Ø6 mm end mill that is a 0.03 mm radial error on every wall. Check runout with a dial indicator on the tool shank, not the holder body.

• 1
  Backlash testCommand 0.01 mm moves and measure the actual table travel.
• 2
  Squareness across strokeCheck at center, near end, and far end of travel.
• 3
  Measure after unclampingIf the error appears only on the CMM, suspect the fixture.
• 4
  Tool tip runoutIndicate on the shank; 0.01 mm here becomes 0.01 mm on the wall.

Method one is the warm-up comparison. Run the machine to thermal steady state, cut a test part, then cut the same part cold the next morning. The difference between the two is your thermal error. It costs one hour and no tooling.

Method two is the backlash and repeatability check. Command a series of small incremental moves with a dial indicator on the table. Record the actual travel each direction. Anything above 0.005 mm of lost motion on a machine rated at ±0.005 mm needs ballscrew attention.

Method three is the circular interpolation test. Cut or trace a 100 mm circle with a ballbar or a test bar and indicator. The polar plot shows squareness, backlash, and servo mismatch in one measurement. A roundness error above 0.01 mm on a machine that used to hold 0.005 mm is a clear signal.

Method four is the thermal mapping run. Instrument the spindle, column, and coolant with thermocouples and log for four hours under load. This is the method that finds the gradient you cannot see with a hand check. Method five is the metrology cross-check: measure the same part on the machine and on a calibrated CMM. The difference isolates machine error from measurement error.

• 1
  Start with warm-up comparisonCheapest test, rules thermal in or out in one shift.
• 2
  Backlash check secondSeparates lost motion from thermal drift.
• 3
  Ballbar thirdOne circular test covers squareness, backlash, and servo.
• 4
  Thermal mapping lastOnly needed when the first three are clean.

Not every precision drop belongs to the machine. A new batch of material with different hardness changes tool wear rate and cutting temperature. A 4140 pre-hard lot at 30 HRC versus 28 HRC will push tool life down and surface finish up. The machine is fine; the process window moved.

Program changes matter too. A CAM update that adds a helical entry or changes stepover alters radial engagement and cutting force. On a thin-wall part that force deflects the wall by 0.02–0.05 mm. The fix is a process change, not a machine repair.

Measurement itself can drift. A micrometer checked out at 20 °C reads differently in a 28 °C shop. A CMM probe that has not been requalified in a week can carry 0.005 mm of stylus error. Before you condemn a machine, confirm the gauge.

Our rule on the floor: if the machine held tolerance last week with the same program and material, and the only thing that changed is time of day, suspect thermal. If material or program changed, suspect the process. If the gauge changed, suspect the gauge first.

• 1
  Same program, same materialOnly then is the machine the prime suspect.
• 2
  Gauge requalificationA CMM probe needs checking before it can judge a machine.
• 3
  Hardness lot variation2 HRC difference changes tool wear and finish.