When a high-precision CNC glass edge grinding machine produces acceptable dimensions at startup and then gradually shifts, establish the relationship between dimensional error and temperature before changing offsets. Heat changes the position of the wheel relative to the glass, and several parts of the machine reach operating temperature at different rates. A correction made during that transition can become an error once the machine settles.

Begin with a repeatable reference cut, stable workholding and consistent measurement conditions. Track the finished profile alongside spindle, coolant and machine temperatures. The objective is to identify which part of the tool-to-workpiece relationship moves, whether that movement stabilizes, and whether it returns after cooling. Those observations determine whether the remedy belongs in warm-up practice, cooling, mechanical maintenance or compensation.

The wheel and the glass do not share one thermal reference

The programmed toolpath assumes a known relationship between the spindle, wheel contact surface, axes and workpiece datum. That relationship changes as the spindle bearings, motor, bearing housing and supporting structure heat up. Even when the controller reports the same axis position, the grinding contact point can occupy a different physical location.

Spindle growth deserves early attention, but its dimensional effect depends on the machine layout. On a vertical spindle, axial growth primarily changes wheel height. With a profiled wheel, that height change can also alter the effective contact radius or move a chamfer boundary. On a straight peripheral grinding surface, radial movement of the spindle or supporting structure more directly changes the finished outline. The same thermal displacement therefore produces different errors with different wheel geometries.

Heat also travels through spindle mounts, slides, screws and the machine bed. Uneven heating creates bending or angular movement, rather than simple uniform expansion. A profile that remains correct near one datum but changes farther away suggests a different mechanism from an equal dimensional shift across the entire working area.

Position feedback does not necessarily capture all this movement. An encoder on a motor or screw measures motion at that component; a linear scale measures displacement at its mounting location. Neither directly measures every change between the scale, spindle nose, fixture and glass. Closed-loop positioning alone cannot establish thermal stability at the grinding contact.

Build a drift record that separates time, temperature and cutting load

Use one representative glass type and thickness, one wheel condition, a fixed program and repeatable locating surfaces. Include dimensions that reveal different error patterns: opposing edge spacing, a feature relative to the fixture datum, and, where relevant, chamfer width or profile height. A single overall dimension can conceal whether the error comes from tool radius, translation or angular movement.

Record measurements from the first cold-start cut through the production period in which the error normally appears. Use intervals short enough to capture the transition, then extend them as the readings settle. Keep measurement timing consistent after machining, since freshly ground glass and a warmed measuring fixture can introduce their own dimensional changes.

A useful record includes:

• Elapsed time, actual spindle speed and whether the machine was cutting, idling or stopped.
• Coolant supply temperature near the machine inlet, together with return or tank temperature when available. A tank reading alone can miss changes in the fluid reaching the wheel.
• Spindle housing temperature at a repeatable location, axis or bed temperature where movement is suspected, and room temperature near the machine.
• Measured profile error, wheel dressing events, offset changes, fixture resets and unusual load indications. These events often explain abrupt changes that a temperature plot cannot.

Surface temperature readings are best used as trend indicators. Infrared measurements vary with surface finish, viewing angle and reflections; a shiny housing can produce misleading values. Use a consistent measurement method and location. The measured housing temperature also lags conditions inside the bearings, so a delay between a temperature rise and dimensional drift does not rule out spindle heating.

Separate spindle heating from cutting effects with two controlled runs where practical. First observe a manufacturer-approved spindle and axis warm-up sequence without grinding. Then repeat under representative cutting conditions. Drift that appears mainly under load directs attention toward grinding heat, wheel condition, coolant delivery or load-sensitive mechanical movement. Allow comparable starting conditions before comparing the runs.

Read the error pattern before choosing a correction

These patterns guide the investigation; they do not establish a diagnosis by themselves. Wheel loading, for example, raises grinding force and heat simultaneously. A correlation between spindle temperature and size error could therefore reflect a process problem that also heats the spindle. Compare dimensional changes with load, dressing history and coolant delivery before attributing the whole error to expansion.

Inspect workholding when the error includes a translation or rotation of the profile. Vacuum pads, locating stops and supporting surfaces establish the glass datum. Slurry beneath a support, an inconsistent vacuum seal or movement of a stop can mimic thermal drift. Where vacuum fixtures include compliant components, changes in compression under clamping and grinding loads can alter seating even when the nominal locating position is unchanged.

Glass measurement needs its own control. Compare parts at a consistent temperature and support them consistently. If the measuring instrument sits beside warm equipment or receives direct sunlight, an apparent machine trend can originate partly in the inspection setup. Repeating the measurement against a stable reference helps separate machining error from measurement drift.

Coolant temperature and delivery require separate checks

A coolant tank often warms more slowly than the spindle. Early parts may therefore experience one combination of spindle and coolant temperatures, while later parts experience another. If tank temperature continues rising after the spindle housing has stabilized, the machine can enter a second period of drift that looks like an unsuccessful warm-up.

Measure the fluid supplied to the grinding zone and observe its delivery during the actual cycle. Restricted filters, partially blocked nozzles, changing fluid level or pump problems affect heat removal. A stable tank temperature does not prove that sufficient coolant reaches the contact area. Nozzle alignment matters particularly when wheel position or profile changes place the contact outside the intended stream.

Changes in coolant concentration, where a formulated coolant is used, can alter lubrication and grinding behavior. Follow the fluid specification and verify concentration with a method suitable for that fluid. Avoid changing concentration, flow and temperature control together during diagnosis; otherwise, the cause of any improvement remains unclear.

If a cooling unit is fitted, examine its temperature cycling and sensing location. Repeated supply-temperature swings can produce a dimensional oscillation instead of a smooth drift. Setting coolant substantially below the surrounding machine temperature also changes thermal gradients. Establish the specified operating condition rather than assuming that colder coolant always improves accuracy.

Make warm-up match the operating state

A fixed startup timer is useful only when it reliably produces a repeatable thermal condition. Warm-up with the spindle at low speed and stationary axes does not reproduce a cycle that uses higher speed, repeated travel and sustained grinding. Equally, an aggressive warm-up unrelated to production can heat the machine into a state that quickly changes once normal work begins.

Use the machine’s approved warm-up procedure as the starting point, then assess its adequacy against representative machining. Include coolant circulation and relevant axis movement within the permitted procedure. Define readiness through stable reference measurements over successive checks, using a stability allowance appropriate to the part tolerance and measurement uncertainty.

Stopping behavior matters as well. A brief interruption, an extended idle period and a shutdown with coolant circulation stopped create different thermal histories. The spindle may cool while the bed remains warm, so restarting can follow a different dimensional path from the original cold start. Record recovery behavior after typical interruptions and establish when a reference cut is needed before releasing production.

Environmental changes deserve investigation when identical startup routines produce different results. Air-conditioning directed at one side of the machine, sunlight on the bed, or nearby heat-generating equipment can create uneven structural temperatures. Compare the location of the dimensional error with the exposed structure before changing axis calibration.

Compensate only for a repeatable relationship

Compensation is appropriate when the thermal error is repeatable, its relevant inputs are measurable, and the mechanical and cooling systems are functioning correctly. It is a poor substitute for unstable bearings, loose mounting surfaces, inconsistent workholding or inadequate coolant flow. Those faults change the relationship between temperature and error, making a previously effective correction unreliable.

A tool-radius adjustment fits an error that behaves like a consistent change in effective wheel radius. It does not correct every thermal displacement. A rigid shift of the profile calls for investigation of the datum or coordinate relationship, while position-dependent distortion requires a broader assessment of machine geometry. Changing radius compensation to fix one dimension can worsen another if the underlying movement has been misidentified.

Do not derive a correction from housing temperature alone without testing it across relevant operating states. The same housing reading during heating and cooling can correspond to different internal and structural temperatures. This thermal lag explains why a simple temperature-to-offset rule sometimes works during startup and fails after an interruption.

Before enabling or revising automatic compensation, establish what the controller already applies. Record the active settings and retain the original configuration. Manufacturer-specific correction functions require their documented setup procedure; stacking an external offset onto an existing correction can obscure the real movement.

Validate the resulting change from a comparable cold start, during sustained cutting and after a representative pause. Measure the profile dimensions affected by the suspected displacement, including features away from the original inspection point. The release criterion is sustained dimensional compliance under those operating conditions, with measurement uncertainty accounted for. Preserve the temperature and error record with the machine settings so that a later recurrence can be compared against a known operating state.