Thermal Growth and Machine Accuracy: The Error Source Nobody Sees
The first part of the morning is often the one that fails inspection — not because the program is wrong, but because the machine was still growing. Every CNC machine tool is a structure made of metal, and metal grows when it is warm. A machine switched on cold and set to work immediately is a machine whose spindle, ballscrews and castings are still expanding as it cuts; the first parts are machined by a machine that is a different size from the machine that cut the parts an hour later. In precision work this single effect — thermal deformation — is routinely estimated to account for a large share of the error that remains after every other error source has been corrected. It is the error source nobody sees, because nothing looks wrong: the machine measures fine when idle, the program is verified, the tool is sharp — and the parts still drift.
This guide is the reference for that invisible error. It explains the physics of thermal growth in a machine tool, where the heat comes from, why it shows up as drifting part dimensions, and — the part that matters most — the ladder of strategies that prevent it, from choosing a thermally stable machine to running a warm-up program and controlling the shop environment. It is the dedicated treatment of a theme the spec-sheet guide raises when it reads a machine’s accuracy claims: the datasheet’s numbers are measured on a warm, stable machine, and the thermal story is what happens to accuracy when a real machine runs a real shift. The surface-finish and tolerance guide explains the 20 °C reference behind all dimensional measurement, which is the same temperature story this page tells from the machine’s side. Terms like spindle and ballscrew are in the glossary.
The physics: metal grows, and only the cut point counts
The mechanism is simple enough to hold in your head. Most structural metals expand when heated by a well-known amount per degree: a steel machine element grows on the order of eleven to twelve microns per metre of length for every degree Celsius it warms. A one-metre steel ballscrew warmed by twenty-five degrees grows by a few tenths of a millimetre — a huge amount in machining terms, hundreds of times a tight tolerance. Aluminium, common in machine castings, grows about twice as much per degree as steel; the material of the machine, the length of the element and the temperature rise together decide the expansion. The engineering formula is the straight line of a coefficient of expansion multiplied by length multiplied by temperature change — but the shop-floor version is all you need: a warm machine is a bigger machine, and the growth is in the tens of microns per degree-metre.
Two refinements make the mechanism real rather than merely interesting. First, the growth that matters is only the growth that changes the relative position of the tool tip and the workpiece at the cut — a machine that grows uniformly, like a photograph enlarging, might carry tool and part together and change little at the cutting point. What damages accuracy is non-uniform heating: the spindle warms and grows downward while the column stays cool, or one side of the bed warms faster than the other, so the structure not only grows but tilts and bends. Second, real machines are not uniform blocks — they are a spindle hanging from a column over a table, and heat sources sit at particular places (the spindle bearings, the ballscrew nuts, the axis motors), so the temperature field across the machine is never even. Uneven heat means uneven growth, and uneven growth means the tool point wanders in ways that no simple “add a correction” fully captures — orientation errors and volumetric errors that change with position and with time.
Where the heat comes from
A machine tool is a machine for making heat as much as for making parts, and the sources divide into two camps:
Heat the machine makes itself. The biggest internal sources sit exactly where the precision lives. The spindle is the dominant one: its bearings and motor generate serious heat whenever the spindle runs, and because the spindle holds the tool, its growth moves the tool tip directly — a warmed spindle grows along its axis, pushing the tool closer to the work. The ballscrews warm from friction during rapid moves and heavy cutting; a screw that runs hard for an hour is measurably hotter, and longer, than it was at start-up. The axis motors and drives, the guideways, and even the hydraulic and coolant pumps all add their share. The energy audit of a machine tool is revealing: of the power it draws, a meaningful fraction leaves as heat in the chips, the spindle and the motors rather than as cutting work.
Heat the environment imposes. The shop itself heats and cools the machine: the air temperature through the day and the seasons, a door opening in winter, sunlight falling on one side of the column, a warm machine standing next to a cold one, the coolant tank warming as the machine runs. For large machines especially, ambient temperature swings can drive deformation comparable to the machine’s own internal heat — the room is part of the machine’s thermal system whether the shop acknowledges it or not. And there is a quieter contributor: thermal memory. A machine that sat warm from yesterday’s shift and was cooled overnight carries yesterday’s temperature field into today’s start-up, so its behaviour this morning is partly a record of last night — which is exactly why warm-up and soak time matter.
The heat matters because of where it acts, not just how much there is. A non-uniform temperature field across a precision structure is the definition of the problem: some parts grow, some lag behind, and the tool point drifts as the balance shifts through the run.
Why it shows up as drifting parts
The signature of thermal error is unmistakable once you know to look: parts that were fine drift progressively off dimension through the morning, then stabilise as the machine reaches thermal equilibrium — and the first part after a cold start or a long break is the worst of the day. The drift is usually in one consistent direction, because the machine is growing in a consistent way, which is what distinguishes thermal drift from random error. Two machine details decide how badly it shows:
The spindle grows into the work. A spindle that warms and grows along Z moves the tool closer to the table, so a facing or boring operation cut when the spindle is warm removes slightly more than the same operation cut when it was cold. On tight axial tolerances this is the effect that makes the morning’s first part thin, the noon part right, and the part after lunch (if the machine cooled) thin again.
The ballscrew’s growth is invisible to a semi-closed loop. Here is the architectural trap that explains most “the machine used to hold tolerance” mysteries. Many machines measure position with a rotary encoder on the ballscrew motor — a semi-closed loop: the control knows where the motor turned, and assumes the slide went where the screw carried it. But the screw itself grows with heat, and a longer screw moves the slide farther than the motor’s rotation alone accounts for. In a semi-closed-loop machine, ballscrew thermal growth is invisible to the control — the machine thinks it is where it was commanded while the screw has quietly grown and carried the slide past the mark. This is why a machine can measure perfectly on a test and drift on the floor: the error is not in the positioning system’s command, but in the unmeasured physical length between the encoder and the slide. Machines with linear scales — full-closed-loop feedback, where position is read from the slide itself rather than the motor — are largely immune to this particular error, because the scale measures where the slide actually is, screw growth and all. It is one of the genuine (and genuine-cost) architectural differences the spec-sheet guide flags under feedback type, and it is the single most structural answer to thermal drift.
The strategy ladder: from machine choice to shop habit
Thermal error cannot be eliminated — a machine that cuts is a machine that heats — but it can be managed, and the management happens at several levels, from the most structural to the most habitual. Work the ladder from the top:
1. Choose a machine that manages heat well. At purchase, the thermal story is a specification worth interrogating, not a footnote: what does the builder do about the heat its own machine generates — spindle and motor cooling, coolant temperature control, structure designed so heat sources sit away from the critical geometry, castings and beds chosen for stability? Machines built for precision work manage their own heat deliberately, and the selection guide and the spec-sheet guide are where to press on this before the money moves. The same decision logic that reads accuracy claims on a warm machine applies here: the machine’s design for heat decides whether that accuracy survives a real shift.
2. Prefer full-closed-loop feedback where tolerances demand it. As above, linear scales that measure the slide directly remove the largest invisible thermal error — ballscrew growth — from the equation. If your work lives at tolerances where tens of microns matter, the scale-equipped machine is not a luxury; it is the machine that can actually hold the number.
3. Run the machine warm, and keep it warm. The cheapest and most reliable shop-floor defence is thermal consistency. A warm-up routine — running the spindle up through its speed range in steps and moving every axis through its travel for a period before critical work, typically on the order of half an hour after a long idle — brings the machine to a stable operating temperature before the first critical part is cut, so the parts are machined by the machine in the state it will hold for the rest of the shift. Shops holding tight tolerances extend the logic: they do not let the machine cool at lunch or between runs, because every cold-to-warm cycle re-runs the drift. The discipline is simple to state and widely violated: do not cut precision work with a cold machine, and do not let a warm machine cool down between precision runs.
4. Stabilise the environment. The room is part of the machine. Controlling shop temperature — keeping it stable rather than merely cool, shielding the machine from sun, doors and radiant heat, and controlling coolant temperature — removes the ambient half of the thermal problem. For the most demanding work, parts and raw material should also soak to shop temperature before they are measured or machined, because a part that arrives warm and is cut cold, or vice versa, carries its own expansion into the tolerance. The 20 °C reference behind all dimensional work, which the surface-finish guide explains, is a shop-discipline target as much as a metrology convention.
5. Compensate for what remains. Modern controls increasingly offer thermal compensation: temperature sensors at the key heat sources — spindle housing, ballscrew nuts, the structure — feed a model that estimates the current thermal state and shifts the axis offsets to correct the predicted drift in real time. When done well, compensation turns a machine that drifts through the morning into one that holds position across the run. Two honest caveats. First, compensation is only as good as its sensors and model — it needs real temperature measurements at the right places, and it is not a substitute for the machine-side and process-side discipline above; a builder’s compensation claim deserves the same sceptical reading as its accuracy claim. Second, probing is the independent backstop: measuring the part or the work offset between operations tells you where the machine actually is, thermal drift included, and correcting from a real measurement beats correcting from a prediction every time.
6. Let the process absorb the last microns. For the tightest work, the finishing pass itself is the final answer to drift: take the critical finishing cuts when the machine is in its stable, warm state, rather than when it is still climbing to equilibrium — the machinist’s instinct to do the important dimension last, on a warm machine, is the thermal strategy in its oldest form.
How to see the error for yourself
Before spending on scales or compensation, confirm the diagnosis — thermal drift has a signature you can check in an afternoon. The test is simple: set up a light finishing operation on a feature with a tight tolerance, and cut the same feature repeatedly from a cold start, measuring each part as you go. Thermal error shows as parts that drift in a consistent direction through the warm-up period and then stabilise — first part worst, later parts consistent once the machine reaches equilibrium. The same experiment run after a proper warm-up, or with the machine left running through the test, shows the drift shrink toward nothing. That single comparison — cold-start drift versus warm-machine stability — tells you more about your machine’s thermal behaviour than any brochure, and it is the same discipline the spec-sheet guide applies when it reads a builder’s thermal claims: measure the machine as it actually runs, not as it is advertised. For spindle thermal behaviour specifically, the international test standard ISO 230-3 defines how thermal effects on machine tools are measured, and the machine’s compliance with it is worth asking about when a builder claims thermal stability.
Frequently asked questions
Why do my parts drift off tolerance through the morning even though the machine is accurate when I test it? Because the machine is growing as it warms, and the accuracy test was done on a warm, stable machine while the morning’s parts were cut by a machine still expanding to its operating state. The spindle grows along its axis, moving the tool closer to the work, and in a semi-closed-loop machine the ballscrew’s thermal growth is invisible to the control — the slide goes further than the motor’s rotation commands. The signature is parts drifting in one consistent direction until the machine reaches thermal equilibrium.
What is the difference between semi-closed-loop and full-closed-loop control for thermal error? Where the machine measures its position. A semi-closed loop measures rotation at the motor and infers the slide’s position from the ballscrew — so when the screw grows with heat, the machine does not know the slide has moved further than commanded. A full-closed-loop machine measures position from a linear scale on the slide itself, so it knows where the slide actually is, screw growth and all, and corrects for it. For tight tolerances, full-closed-loop feedback removes the largest invisible thermal error.
How long should I warm up my CNC machine? Long enough to bring it to its stable operating temperature before critical work — typically on the order of half an hour after a machine has sat idle for several hours or overnight. A proper routine runs the spindle up through its speed range in steps and moves every axis through its travel, rather than just idling the spindle. The deeper principle matters more than the exact minutes: cut precision work only when the machine is warm and stable, and do not let it cool down between precision runs, because every cold-to-warm cycle re-runs the drift.
Does thermal compensation fix the problem? It reduces it. Modern controls can estimate the machine’s thermal state from temperature sensors and shift the axis offsets to correct predicted drift in real time — genuinely useful, but only as good as its sensors and model, and not a substitute for the fundamentals: a thermally managed machine, full-closed-loop feedback where tolerances demand it, a warm-up discipline, and a stable environment. Probing — measuring where the machine actually is — remains the independent backstop that catches what compensation predicts imperfectly.
Can I machine precise parts on a machine that is not thermally stable? Yes, if you manage the process around it: keep the machine warm and run it in its stable state, do the critical finishing cuts when it has reached equilibrium, probe and correct the offsets from real measurements, and hold the environment steady. The machine’s thermal stability sets how much of this discipline you need, not whether you need it — which is why the thermal story belongs in the machine-buying decision, where it is cheapest to solve.
Bottom line
Thermal deformation is the error source nobody sees because nothing looks wrong: the machine measures fine when idle, yet the parts drift. The mechanism is ordinary physics — metal grows with heat, unevenly, and only the growth at the cut point counts — and the heat comes from both the machine’s own spindle, ballscrews and motors and from the room around it. The consequences show up as the classic signature: parts drifting in one direction until the machine reaches equilibrium, first part worst. And the defence is a ladder, not a single fix: choose a machine designed to manage its own heat; prefer full-closed-loop feedback that measures the slide rather than inferring it from a growing screw; warm the machine up and keep it warm; stabilise the shop environment and let parts soak; compensate with real sensors and confirm with probing; and do the critical cuts in the machine’s stable state. Work that ladder and the machine that drifted through the morning becomes the machine that holds its number all day — which is what the accuracy claim on the datasheet always promised, and what thermal discipline finally delivers.
This guide is part of the CNC Media guides library — the thermal-error reference of the accuracy topic, deliberately free of prices and of any single builder’s or compensation vendor’s product to promote.