The Axis That Doesn't Backlash
The Axis That Doesn't Backlash: Ballscrew, Lead Screw, Belt, or Linear Motor?
A linear axis misses position, whips at speed, or drops its load on power loss, and the fault is usually the architecture, not the motor. Ball screw vs lead screw vs belt vs rack vs linear motor, the five numbers that decide, and a worked Z-axis sizing.
Example Labs is a made-up company we use for worked examples.
Picture the Z-axis on Example Labs' sensor-hub test fixture. It lifts an 8-kilogram carriage 300 millimetres, holds it still while the fixture probes the board, and lowers it again, a thousand times a day. One afternoon the power blips.
The axis does not hold. The carriage falls the full 300 millimetres, the fixture crashes into the board, and the shift loses an hour. The motor was fine.
The controller was fine. The screw was a ball screw, and a ball screw back-drives: without a brake, gravity wins the moment the power goes away.
That is the kind of failure this article is about. A linear axis is not one part. It is three subsystems, the guide that constrains the motion to one line, the drive that turns torque into thrust, and the carriage that carries the load, and the most common sizing error is treating them as separate purchases.
The guide spacing changes the moment capacity. The screw's end fixity changes its critical speed and its buckling load. The carriage's overhang changes the rail loading.
You size them as a system, or you size them twice: once on paper, and once again after the first prototype fails.
The choice that matters most is the drive. There are five families, and each one is the right answer to a different question: ball screw, lead screw, belt, rack-and-pinion, and linear motor. This article takes them apart, gives the five numbers that decide between them, and sizes a real axis in the demo, the Example Labs Z-axis, the one that fell.
What a linear axis is, in plain language
A linear axis moves something in a straight line. A motor spins; something has to turn that spin into a push. The drive is the something, and the five families do it five different ways.
A ball screw is a screw with ball bearings rolling in the thread between the screw and the nut. The balls roll instead of sliding, so friction is low: a ball screw is 85–95% efficient, against 20–30% for a lead screw. The nut can be preloaded, oversized balls, or a double nut tensioned against itself, so there is no play at all when the direction reverses.
That is why a ball screw is the default for precision work: it is stiff, it is accurate, and the supply chain is deep. Its price is that it back-drives. Turn the screw and the nut moves; push the nut and the screw turns.
A vertical ball-screw axis falls when the power goes away, unless a brake or a counterweight holds it.
A lead screw (an Acme screw, the old-fashioned kind) is a screw with a sliding nut, metal on plastic, or metal on metal, with no balls. Friction is high, so efficiency is 20–30% and the nut wears, but the friction is also the feature: at a low lead, a lead screw is self-locking. Push the nut and nothing turns.
A vertical lead-screw axis stays where it is when the power goes away, and a 3D printer's Z-axis is the classic example, it moves rarely, it just holds position, and self-locking means no droop on power-off. Lead screws are also quiet, clean, and cheap, and polymer nuts can run dry. They are the right answer when the axis is slow, the duty is light, and holding still matters more than moving fast.
A belt drive is a toothed belt on pulleys, GT2 or HTD profile, that pulls the carriage along the rail. The belt is the fastest of the screw family and the cheapest way to cover a long stroke: a single belt can run 10 metres and beyond, because there is no rigid column to support. The price is stiffness.
A belt stretches under load, so absolute positioning accuracy is lower, and the stretch changes with tension and temperature. Belts are for long axes that need fast rapids and light payloads, not for microns.
A rack-and-pinion is a gear rack bolted along the frame with a pinion that drives along it. The stroke is effectively unlimited, the rack can be as long as the frame, and the speed and thrust are high. The price is backlash: any clearance between pinion and rack is play at the carriage, and removing it needs a split or preloaded pinion, which adds complexity and cost. Rack is for strokes measured in metres, where a screw would whip and a belt would stretch.
A linear motor skips the conversion entirely: the motor is the axis, a row of magnets on the carriage and coils in the track, pushing directly. There is no screw to whip, no belt to stretch, no backlash at all, the highest bandwidth, the highest acceleration, and the most direct control of any family. The price is everything else: the magnets, the encoder, the heat, and the cost.
A linear motor is the answer only when the job genuinely needs the bandwidth and the zero-backlash directness that a screw can never give, and the budget can pay for it.
The five numbers that decide
The guides agree on the sizing order, and it is worth following in that order: life first, then critical speed, then thrust and torque, then stiffness and accuracy. Checking them in a different order is how a team sizes a screw that lives forever at a speed it can never reach.
Efficiency is the first number, because it sizes the motor. A ball screw at 90% needs a third of the torque a lead screw at 30% needs for the same thrust, the demo below shows it: 0.084 N·m against 0.251 N·m for the same 94.5 N of thrust. The missing torque does not vanish. It becomes heat in the nut, and heat in the nut is wear in the nut. A lead screw that runs all day at full load is a lead screw that gets hot, and a hot polymer nut has a PV limit, pressure times velocity, that caps how hard and how fast it can run.
Backlash is the play when the direction reverses, and it is the number that decides whether the axis repeats. A preloaded ball screw has effectively none. A lead screw's anti-backlash nut has some, and it grows as the nut wears. A belt has stretch instead of backlash, which is worse in one way: backlash is a fixed dead zone you can measure and compensate for, while stretch depends on the load, so it changes with the job. A rack's backlash is the pinion clearance, removed only by a split pinion. The trap is to treat backlash as the whole accuracy story. It is not. The gearbox piece showed the same lesson on the rotary side: a harmonic gearbox's backlash was under 0.25 mm at the end of the arm, but its wind-up under load was 10.8 mm, 43 times larger. A screw's stiffness matters the same way: the deflection under load is usually bigger than the backlash, and it is the deflection that moves the part.
Critical speed is the speed at which the screw starts to whip, the first bending resonance of the screw as a beam. It depends on three things: the root diameter of the screw, the free length between the supports, and the end fixity, how the ends are held. Fixed-fixed mounting boosts critical speed the most; fixed-free is the worst. The same screw, the same length, the same load: the mounting decides whether the axis is comfortable or marginal, and the demo shows the spread. A belt has no critical speed, which is why long fast axes use belts, but a belt has stretch instead, so the trade is never free.
Buckling is the load at which the screw, in compression, bows sideways instead of pushing straight. It follows Euler's column rule: the buckling load rises with the square of the diameter and falls with the square of the length, and the end fixity sets the constant. The guides say to target at least twice the maximum compressive load. For a light axis the margin is enormous, the demo's Z-axis has a 488× margin, and buckling only becomes the driver when the axis pushes hard: a press, a vertical lift of a heavy load, a screw in compression against a jammed carriage.
Life (L10) is the number of revolutions that 90% of screws will reach, from the ISO 3408-5 method: L10 = (Ca / (fw × Fa))³ × 10⁶ revolutions, where Ca is the basic dynamic load rating from the catalogue, Fa is the axial load, and fw is a load factor for shocks. Convert revolutions to hours by the rpm. The demo inverts the formula, it asks what Ca the axis actually needs for five years of service, and the answer is under 1 kN, which is the point: for a light axis, life does not size the screw. Stiffness and critical speed do.
Grade is the sixth number, and it is the one teams overbuy. ISO 3408 and JIS B1192 grade screws C0 to C10 by lead error, how far the actual travel deviates from the commanded travel over a length. C3 and C5 are the usual picks. A C3 screw costs more than a C5, and the difference only matters if the rest of the axis, the guide, the carriage, the encoder, the temperature, is accurate enough to see it. Buying a C3 screw for an axis with a belt-grade guide is paying for accuracy the machine cannot use.
The build: sizing the axis as a system
The sizing workflow, in order, is the one the guides describe, and it is the one the demo follows.
First, the load and the life. Work out the axial load from the moving mass, the acceleration, and any process force, then ask what dynamic load rating the life target needs. If the answer is small, life is not the driver, and the screw can be chosen on stiffness and speed instead.
Second, the speed. Work out the rpm the lead and the travel speed demand, then check it against the critical speed for the diameter, the free length, and the end fixity. If the margin is thin, the fix is a larger diameter, a shorter free length, or a stiffer mounting, in that order of cost.
Third, the thrust and the torque. Convert the axial load to motor torque through the lead and the efficiency, and check the motor against it, remembering that the lead screw needs three times the torque of the ball screw for the same thrust, and that the motor's own inertia and the guide friction add a small term the formula leaves out.
Fourth, the stiffness and the accuracy. Check the deflection under load against the tolerance, and the grade against the rest of the machine. A screw that is accurate to 23 microns in a machine that deflects 200 microns is a decoration.
Fifth, the holding question. Ask what happens on power loss. A ball screw back-drives: the axis needs a brake, a counterweight, or a motor with a holding brake. A lead screw self-locks: no brake, but three times the torque and three times the heat. The answer belongs in the design, not in the incident report.
The Example Labs Z-axis: the one that fell
Example Labs' sensor-hub test fixture has a vertical Z-axis: an 8-kilogram moving mass (a 5-kilogram payload from the bench arm, plus a 3-kilogram carriage), a 300-millimetre stroke, a required speed of 0.25 m/s, and an acceleration of 2 m/s². The screw is a 16-millimetre ball screw with a 5-millimetre lead, a 13-millimetre root diameter, and a 350-millimetre free length between supports, mounted fixed-supported. The arithmetic was run in Python before writing, and the assumptions are stated: the load is weight plus acceleration only (guide friction and motor inertia are left out, and they add a small term), and the efficiency figures are the guides' 90% for a ball screw and 30% for a lead screw.
Step 1, the load and the life. The axial load is the mass times gravity plus acceleration: 8 kg × (9.81 + 2.0) m/s² = 94.5 N. The axis runs 1,000 cycles a day, 250 days a year, and each cycle is 600 millimetres of travel, 300 up and 300 down, which at a 5-millimetre lead is 120 revolutions per cycle, or 30 million revolutions a year. For a five-year L10 life the screw must survive 150 million revolutions, and inverting the life formula gives the dynamic load rating it needs: (150)⅓ × 94.5 N = 502 N. Under 1 kN. Life does not size this screw, any 16-millimetre ball screw clears it by an order of magnitude, and the check that matters is stiffness and critical speed.
Step 2, the speed and the critical speed. The travel speed demands 0.25 m/s ÷ 0.005 m/rev = 3,000 rpm. The critical speed is the first bending mode of the screw as a beam, the Euler-Bernoulli frequency, with steel at 200 GPa, a 13-millimetre root diameter, and a 350-millimetre free length, and it depends on the mounting: fixed-free 4,496 rpm, supported-supported 12,621 rpm, fixed-supported 19,721 rpm, fixed-fixed 28,610 rpm. At the as-built fixed-supported mounting the margin is 6.6×, and the axis is comfortable. At fixed-free, the same screw, the same length, the same load, held at one end only, the margin is 1.5×, and the axis is marginal: 3,000 rpm is 67% of the critical speed, inside the usual 80% rule but close enough that a longer stroke or a faster rapid would break it. End fixity is not a detail. It is the difference between a comfortable axis and a marginal one.
Step 3, the thrust and the torque. The drive torque is the axial load times the lead, divided by 2π times the efficiency. For the ball screw: 94.5 N × 0.005 m ÷ (2π × 0.90) = 0.084 N·m. For a lead screw at 30% efficiency: 0.251 N·m, three times the torque for the same thrust, and three times the heat in the nut. Either motor is small; the point is the ratio, and the ratio is the reason a lead screw that runs all day gets hot.
Step 4, the buckling check. Euler's column load for the same screw, fixed-supported (K = 0.7), is 46,104 N, a 488× margin over the 94.5 N load. Buckling is not the driver for this axis. It becomes the driver when the axis pushes: a press, a heavy vertical lift, a screw in compression against a jammed carriage. The check is cheap, and the guides say to keep at least a 2× margin, so it belongs in the sizing sheet even when it passes by hundreds.
Step 5, the holding question, the one that failed. A ball screw back-drives, so on power loss the 8-kilogram carriage falls. The holding torque the brake must supply is the weight times the lead over 2π times the back-driving efficiency: 8 kg × 9.81 m/s² × 0.005 m ÷ (2π × 0.90) = 0.069 N·m. A brake rated above that holds the axis; without it, the carriage falls the full 300 millimetres. A lead screw would have held on its own, self-locking at a 5-millimetre lead, and that is the genuine trade: the ball screw needs a brake and a 0.084 N·m motor, the lead screw needs no brake and a 0.251 N·m motor that runs three times hotter. Example Labs' axis had the ball screw and no brake. The fix is a brake, not a different screw.
The belt alternative, for comparison. The same axis on a belt has no critical speed and no buckling check at all, the belt does not whip and does not buckle, and at 0.25 m/s it would be comfortable. What it loses is the stiffness: the belt stretches under the 94.5 N load, and the stretch changes with the load, so the positioning error changes with the job. For a test fixture that probes a board to a tenth of a millimetre, that is the wrong trade. For a pick-and-place axis that runs 2 metres at 2 m/s, it is the right one.
The traps
Sizing the screw alone. The guide, the drive, and the carriage are one system. A screw sized in isolation is a screw that meets its own datasheet and misses the machine: the guide spacing sets the moment capacity, the carriage overhang sets the rail loading, and the screw's end fixity sets its critical speed. Size them together, or size them twice.
Forgetting the end fixity. The same screw at the same length has a critical speed of 4,496 rpm held at one end and 28,610 rpm held at both ends, a 6.4× spread. A catalogue that lists a screw's critical speed without the mounting is not giving a number; it is giving a hint. The mounting belongs on the sizing sheet, next to the diameter.
Forgetting the brake. A ball screw back-drives, and a vertical ball-screw axis without a brake is a drop waiting for a power blip. The brake is not an accessory. It is part of the axis, and the holding torque is a line in the sizing sheet, next to the drive torque.
Buying grade C3 when C5 will do. The grade is the lead error, and the lead error only matters if the rest of the machine is accurate enough to see it. A C3 screw in a machine whose guide deflects ten times further is money spent on a number nobody can measure. Match the grade to the machine, not to the catalogue's top line.
Treating backlash as the whole accuracy story. Backlash is a fixed dead zone. Deflection under load is usually bigger, and it changes with the load. The gearbox piece proved it on the rotary side, 10.8 mm of wind-up against 0.25 mm of backlash, and the same ratio holds here: check the stiffness before you celebrate the zero-backlash nut.
The decision test: five questions
A team choosing a linear drive can answer five questions in an afternoon. If it cannot answer three of them, it has not chosen a drive, it has chosen a catalogue page.
- What is the axial load, and what does the life target actually need? If the required Ca is under 1 kN, life is not the driver, and the screw should be chosen on stiffness and speed.
- What rpm does the travel speed demand, and what is the critical speed at the as-built mounting? If the mounting is fixed-free and the margin is under 2×, the axis is marginal, change the mounting or the diameter, not the motor.
- What happens on power loss? If the answer is "it falls", the axis needs a brake, a counterweight, or a self-locking lead screw, and the choice belongs in the design, not the incident report.
- What is the deflection under load, and what is the tolerance? If the deflection is larger than the tolerance, the screw is the wrong size no matter what its grade says.
- What is the stroke, and what is the speed? If the stroke is measured in metres and the speed in metres per second, the answer is a belt or a rack, and the screw was the wrong family from the start.
Objections
"A ball screw is always the right answer." It is the right answer when the job needs precision, stiffness, and thrust at a moderate stroke. It is the wrong answer when the stroke is long and the speed is high (a belt), when the stroke is measured in metres (a rack), when the axis must hold still on power loss without a brake (a lead screw), and when the bandwidth genuinely needs direct drive (a linear motor). The default axis for most machines in 2026 is profile rails plus a ground ball screw, and a default is a starting point, not a decision.
"A lead screw is just a cheap ball screw." It is a different machine. It is quieter, cleaner, self-locking, and cheaper, and it is three times less efficient, it wears faster, and it gets hot. For a slow axis that holds position, a printer's Z, a focus stage, a clamp, it is often the better machine. For an axis that runs all day, it is a heater with a screw attached.
"Our axis is slow, so critical speed doesn't matter." Critical speed is a property of the screw and the mounting, not of the job. A slow axis with a long, thin, poorly supported screw can still whip, and a screw that whips once, at a rapid move nobody planned for, is a screw that bends. The check is cheap. Run it.
"We'll just buy the stiffest screw." Stiffness is not free. A larger screw is heavier, needs a larger motor, and costs more, and the stiffness that matters is the stiffness of the whole axis, including the guide and the carriage. A stiff screw in a flexible carriage is a stiff screw bolted to a spring. Size the system.
"The belt is fine, we'll compensate in software." Software can compensate for a fixed dead zone. It cannot compensate for stretch that changes with the load, because the load changes with the job. A belt axis can be accurate at one load and wrong at another, and the error appears in the part, not in the log.
Sources
- JLCMC, "Ball Screw Linear Actuators: Guide to Selection and Purchase"
What we used
Ball screw 85–95% efficient vs lead screw 20–30%; zero-backlash via oversized balls or double-nut preload; lead screws anti-backlash polymer nuts, more wear-dependent; ball screws back-drivable (not self-locking), lead screws self-locking at low leads; sizing order life → critical speed → thrust → stiffness/accuracy; L10 = (Ca / (fw × Fa))³ × 10⁶ revolutions; critical speed from root diameter, free length, end fixity (fixed-fixed best, fixed-free worst); buckling Euler's column check, target ≥2× margin; belt for long-stroke speed, screw for precision and thrust
- JLCMC, "What Is a Ball Screw in a CNC Machine?"
What we used
Ball screw ~90% efficient vs Acme/lead screws 20–40%; preload by oversized balls, tensioned double-nut, or lead-shift single nut; too much preload adds torque and wear; ISO 3408 / JIS B1192 grades C0–C10, C3 and C5 the usual picks; fixed end + supported (floating) end is the usual mounting; roller screw more load and stiffness in a smaller package, ball screw more efficient and cooler for long high-speed cycles; Z on a lead screw holds position on power-off
- Robo2U, "Linear Motion Systems: Rails, Ball Screws & Linear Motors"
What we used
The three subsystems (guide, drive, carriage) and the sizing error of treating them separately; default axis for most machines in 2026 is profile rails + ground ball screw; belt when the stroke is long and speed matters more than micron accuracy; rack when the stroke is measured in metres; linear motor only when the bandwidth, acceleration, and zero-backlash directness justify the magnets, encoder, and heat
- KKM Solutions, "Belt, Screw or Rack: Choosing a Linear Drive" (August 2026)
What we used
Belts suit long strokes (10 m and beyond) at high speed and moderate thrust; screws deliver the highest accuracy and stiffness with high thrust; rack offers effectively unlimited stroke with high speed and thrust; belt stretch reduces stiffness and absolute accuracy; screw critical speed caps usable velocity; rack needs a split or preloaded pinion to remove backlash
- Example Labs figures from the cluster: the bench arm (Link 1 0.425 m / 4 kg, Link 2 0.425 m / 3 kg, payload 5 kg at 0.85 m; shoulder requirement 126.8 N·m
What we used
#29); the machined 6061-T6 hub lid (#27); the sheet-metal hub tray (5052-H32, T 2.0 mm, flat 172.80 mm, #30); the field sensor hub (100 × 60 × 40 mm, #7)
- Critical-speed and buckling figures in the demo are the article's own calculation from the Euler-Bernoulli beam first mode and Euler's column formula, with the inputs stated (steel E = 200 GPa, root diameter 13 mm, free length 350 mm, end-fixity factors β = 1.875 / π / 3.927 / 4.730 and K = 2.0 / 1.0 / 0.7 / 0.5)
What we used
Standard structural mechanics, not catalogue data