Calibration Interval & Metrology Program for 3D Printing | Precise3D

A Printer That Has Drifted Still Produces Parts

There is no alarm that fires when a printer slips out of tolerance. A machine that has drifted two tenths of a millimetre still completes every job, still produces a part that looks correct, and still passes a visual check. The failure only appears later, in a mating part that will not seat, in a batch of fixtures that quietly need hand fitting, or in a customer complaint that arrives three weeks after the parts have shipped. This is why calibration on a printer is a different problem from calibration on a measuring instrument: the instrument is the thing being measured, and the printer is the thing that silently changes the measurement of everything downstream.

Most distributors and most print farms handle this with a fixed annual calibration card, copied from wherever they last saw one. The card is convenient and almost always wrong in both directions. It over-services the machines that hold their geometry well, burning technician hours and production time on a machine that did not need them, and it under-services the machines that move quickly, because the one variable the card assumes constant is the one that varies most between machines. The programme that works is built from measured drift, on a register, with intervals that are earned rather than assigned.

[MUST] The four things that actually move on a printer are the same four that appear in every serious calibration discussion: positional accuracy across the build volume, the squareness and flatness of the build surface, the thermal accuracy of the hotend and chamber, and the extrusion constant that ties commanded flow to delivered material. Everything else is either a consequence of these or is stable enough to spot-check rather than control.

Setting Intervals from Drift Rate, Not from Habit

[GAP] The interval should be the time in which a machine is expected to drift a meaningful fraction of its tolerance band, and it should be computed rather than guessed. The method is simple and almost nobody uses it. Take the machine's specification tolerance for a given dimension, decide what fraction of that tolerance you can afford to consume between checks, and divide by the measured drift rate. If a machine moves half a hundredth of a millimetre per hundred hours of printing on the XY axes, and you are willing to let it consume a third of a two-tenths tolerance before you intervene, the arithmetic puts the check somewhere near a thousand printing hours. That is a very different number from twelve months, and on a busy farm it is a much shorter one.

The reason this matters is that printing hours and calendar months are not the same quantity. A machine running three shifts accumulates in six weeks what a lightly used machine accumulates in a year. Interval by calendar is interval by the wrong variable.

The Four Dimensions Worth Controlling

A calibration programme that tries to control everything controls nothing, because the technician runs out of time before reaching the machines that needed attention. Four dimensions carry the quality risk and merit a formal interval. Everything else belongs in a spot-check list that a machine operator can run without a metrology technician present.

Positional accuracy across the build volume

Positional accuracy is measured with a calibrated artefact printed in a known material and measured on a coordinate measuring machine or a calibrated scanner. The artefact should span the build volume rather than sit in the centre, because the failures that matter are the ones at the corners. A printer can hold its centre tolerance for a very long time while the edges drift, and edge drift is exactly what produces the mating-part failures that appear in a customer's assembly line rather than in your inspection room. The artefact should be printed with a fixed profile that never changes, because a comparison across time only means something if the process is constant.

Build surface flatness and squareness

The build surface is the datum for everything else, so an out-of-flat bed corrupts every measurement taken on it. Flatness is checked with a straight edge and a feeler gauge across multiple axes, or with a dial indicator swept across the plate if the machine allows it. Squareness between the gantry and the bed, and between the two horizontal axes, is the parameter that most often goes unmeasured and most often explains an unexplained dimensional problem. A machine can have perfect positional accuracy in its own frame and still produce a skewed part if the frame is not square.

Thermal accuracy

[SNIPPET] Hotend thermal accuracy is checked against a calibrated thermocouple inserted at a known depth in the nozzle, not against the machine's own reading. The machine's own reading is the thing being calibrated, so trusting it is circular. The offset between the setpoint and the reference is the measurement that matters, and it should be recorded both as an absolute offset and as a direction, because an offset that grows in one direction is a sensor degrading and an offset that is stable is simply a constant to be compensated. Chamber temperature, where controlled, is checked the same way at the top and the bottom of the chamber, since a chamber that reports correctly at the sensor can still stratify.

Extrusion constant

The extrusion constant, or E-step value, ties the commanded filament length to the length actually consumed. It changes slowly with extruder gear wear and with every nozzle change, and it is the single easiest parameter for a machine operator to check, which is why it belongs in the operator spot-check rather than the metrology programme. A ten-minute extrusion test once a week catches a drifting constant long before it produces a dimensional rejection.

The Fleet Metrology Register

A drift-based programme is impossible without a register, because the whole point is comparing a machine against its own history and against its siblings. The register is a table, one row per machine per check, and it holds seven fields. Machine identifier and location. Printing hours at the time of check, not the date. The artefact or method used, with its own calibration status. The measured values for the four controlled dimensions. The specification limits in force at the time of the check. The drift rate computed since the previous check. The pass, adjust, or out-of-service decision, with the technician's initials.

The field that earns its keep is drift rate. A single measurement tells you whether a machine is in or out of tolerance today; the drift rate tells you when it will be out, which is the only information that lets you schedule the intervention instead of reacting to it. A register with drift rates turns calibration from a cost centre into a planning tool, because a machine with a known drift rate can be scheduled for its next check at the moment when its accumulated drift approaches the intervention threshold, rather than at an arbitrary calendar boundary.

When a Machine Leaves Service for Requalification

Requalification is the gate that separates adjustment from intervention. Adjustment is small: a re-levelled bed, a re-flashed extrusion constant, a thermal offset corrected in firmware. Requalification is a full four-dimension check, and it is mandatory after three events. The first is any repair that touches the motion system, the gantry, or the extruder assembly, because the geometry that the previous calibration certified no longer exists. The second is any physical relocation of the machine, because moving a printer across a workshop loads the frame in ways that relieve stresses and shift geometry. The third is a failed check, because a machine that has been adjusted back into tolerance has a new drift baseline and its interval has to be recomputed from the date of adjustment.

[MUST] The trap is treating a repair as a repair. A technician replaces a belt, re-tensions it, runs a test print, sees a part that looks good, and returns the machine to production. The test print proves the machine works; it does not prove the machine is in tolerance. Requalification is the step that converts "it works" into "it is known to be accurate", and skipping it because the test print looked fine is the single most common way a fleet accumulates machines that are working correctly but producing out-of-tolerance parts.

Building the Programme Without a Metrology Lab

A distributor or farm does not need a full metrology laboratory to run this programme. The minimum viable kit is a calibrated artefact and a calibrated reference thermometer, both on a traceable calibration cycle, plus a coordinate measuring capability that can be hired rather than owned. Many farms measure their artefacts on a service bureau's CMM once a quarter rather than buying a machine, and that is a perfectly sound arrangement as long as the artefact's own calibration is traceable and the measurement method is fixed.

What cannot be outsourced is the discipline of recording. A measurement that is not written down did not happen, because the value of the programme is in the trend rather than in any single reading. The register is the asset. The equipment is replaceable and the method can be rented, but a fleet that has two years of consistent drift records can predict its own failures, schedule its own interventions, and answer a customer's dimensional complaint with evidence instead of reassurance.

The inbound side of the same quality problem, checking what arrives before it reaches production, is covered in our incoming QC guide. The statistical process-control side, turning a stream of measurements into a control chart, is in the SPC and Cpk guide. The equipment-side problem of sensors that feed these measurements is covered in the temperature sensor guide, and the maintenance schedule that carries calibration tasks alongside mechanical ones is in the tooling maintenance guide. The end-use qualification chain that depends on a calibrated printer is set out in the end-use parts guide.

← Back to Blog