Machine Tools Suddenly Declined: 4 Diagnostic Principles and 5 Diagnostic Methods

• 1
  Separate a step change from a slow drift
• 2
  Measure the machine before you blame the part
• 3
  Change one variable at a time
• 4
  Know the machine's real baseline
• 5
  What the numbers mean for your tolerance band
• 6
  Process causes that look like machine failure
• 7
  5 diagnostic methods for a sudden accuracy drop
• 8
  Match the symptom to the likely cause

A step change and a slow drift are different failure modes, and mixing them up wastes days. A step change appears between two consecutive parts, or after a tool change, a power interruption, or a fixture reset. Something discrete moved. A slow drift creeps across a shift: the first part is good, the twentieth is 0.03 mm out, and the next morning the machine is fine again.

Step changes point at mechanical events. Loose chuck jaws, a pulled dowel pin, a fixture clamp that released under load, a tool pulled out of its holder, or a parameter edited by mistake. Drifts point at heat, wear, or material.

Before touching a single parameter, plot the deviation against part sequence number. Mark every tool change, every pause, and every shift change on the same chart. In most shops the pattern becomes obvious within one plot, and the correct diagnostic path follows from it.

• 1
  Step changeCompare the last good part with the first bad one. Check clamping and tool retention first.
• 2
  Slow driftLog spindle and ambient temperature alongside the deviation for one full shift.
• 3
  Mixed patternDrift plus a step usually means a thermal trend with one worn element on top.

Part dimensions tell you the result, not the cause. If the part is 0.02 mm oversize on one feature, the error could come from thermal growth, tool wear, a fixture shift, or the probe itself. Run a machine check first and you cut that list down to one or two items.

The practical sequence: air-cut a warm-up cycle, then measure a known artifact. A test bar in the spindle, a ring gauge on the table, or a certified step gauge held in the vise all work. Record the reading, run a 30-minute warm-up, and measure again.

That single comparison separates thermal behavior from everything else. If the artifact reading moves with temperature, the machine structure is moving. If it stays put while the part drifts, the error lives in the process: tool, fixture, program, or material. On a machine held to ±0.005 mm, a 0.01 mm artifact shift after warm-up is enough to explain a whole batch of out-of-tolerance holes.

• 1
  Cold readingTake it before the first cut of the day, spindle at idle.
• 2
  Warm readingRepeat after 30 minutes of cycling at normal spindle speed.
• 3
  DifferenceMore than 0.01 mm means thermal compensation is the first fix.

When machine tools suddenly declined, the temptation is to re-level the machine, retune the servos, replace the tool, and rewrite the program in one afternoon. Then the parts come back good and nobody knows which change mattered. Next month the problem returns.

Change one thing, cut three parts, measure all three. If the deviation is unchanged, revert and move to the next candidate. This is slower for one afternoon and much faster over a month, because you keep a documented list of what was ruled out.

Keep the measurement method identical across tests. Same micrometer, same operator, same temperature, same feature, same datum scheme. Switching from a caliper to a CMM mid-diagnosis introduces more variation than most real faults do. A 20 °C shop and a 26 °C shop will report different numbers for the same part, so write the room temperature next to every reading.

• 1
  One changeAdjust one parameter or component per test cycle.
• 2
  Three partsCut and measure three parts to average out setup noise.
• 3
  Written recordNote the change, the room temperature, and the result.

A machine does not lose accuracy in the abstract. It loses accuracy relative to a baseline that was established when it was commissioned, after leveling, and after thermal compensation was tuned. Without that baseline number, every discussion about accuracy is opinion.

Baseline data worth keeping: backlash on each axis, squareness between X and Y, spindle runout at the gauge line, repeatability of ten rapid positioning moves, and the artifact reading from a cold start. These are cheap to collect and they age well.

Repeat the baseline check every six months, or after any crash, move, or foundation work. When a real problem appears, you compare two numbers instead of guessing. This is also the data a repair technician will ask for first, and having it ready usually removes a day of on-site diagnosis time.

• 1
  BacklashRecord per axis at commissioning and every six months.
• 2
  Spindle runoutMeasure at the gauge line with a certified test bar.
• 3
  RepeatabilityTen bidirectional moves, note the spread, not just the average.

Not every deviation is a fault. A shop at 20 °C and a shop at 26 °C will produce different measurements from the same machine, and aluminum grows about 23 μm per meter per degree Celsius. A 300 mm aluminum part that warms by 5 °C during inspection moves roughly 0.035 mm on its own.

This is why inspection temperature belongs in the drawing notes for tight work, and why a part should sit on the granite before final measurement. For parts held to ±0.005 mm, letting the part stabilize for 20 to 30 minutes is not bureaucracy. It is the difference between a real reading and a random one.

The same logic applies to the machine. Thermal compensation in the control is a model, not a measurement. It works when the machine follows its expected warm-up curve and stops working when the shop door opens, the coolant chiller fails, or the machine sits idle for two hours and then starts a finishing pass.

• 1
  AluminumAbout 23 μm/m per °C. A 300 mm part moves 0.035 mm over 5 °C.
• 2
  SteelAbout 11–12 μm/m per °C. Roughly half the aluminum figure.
• 3
  InspectionLet parts stabilize 20–30 minutes before final measurement.

Roughly half the accuracy complaints we see are not machine faults. The machine holds its geometry, the artifact checks clean, and the parts still drift. The cause sits in the process around the machine.

Common ones: a fixture clamp that deflects a thin wall, a finishing pass with too little radial engagement that rubs instead of cuts, a tool offset entered from a pre-setter that was not warmed up, or a CAM path that leaves an uneven stock allowance. Each produces a repeatable error that follows the program, not the clock.

The way to separate them is to run the same program on a different machine with the same fixture and tooling. If the deviation travels with the setup, the machine is fine. That test costs one setup and settles the argument with data instead of seniority.

• 1
  Setup travelsError follows the fixture and tool, not the machine.
• 2
  Program travelsError follows the CAM file across machines.
• 3
  Clock travelsError follows warm-up time. That is thermal.