How Gear Accuracy Affects Gearbox Noise

A gearbox is assembled, run on the test bench, and howls in one speed band. The gears come back from inspection marked DIN 7 with every item in tolerance. A qualified gear in a noisy box is a common enough situation on real projects. Accuracy grade and noise do not sit on a straight line: the grade assigns a tolerance band to each deviation, while the noise itself comes from the frequency content of the excitation, the transmission path through the housing, and the stack-up of assembly errors.

Start with the excitation. The number of teeth in mesh changes periodically through the cycle, so tooth stiffness rises and falls and produces a mesh frequency plus its harmonics, where mesh frequency equals tooth count times rpm divided by sixty. On top of that, the gear’s own deviations add a displacement excitation once per revolution and once per tooth: pitch deviation fpt, profile deviation Fα, lead deviation Fβ, and runout Fr. Together they form what is normally called transmission error. Which orders carry the harmonic energy largely decides whether the result sounds like a low hum or a sharp whine.

Two gears of the same grade are therefore not the same noise source. Take three 6-grade gears: one with a crowned profile, one with a unidirectional pressure-angle error, one with random scatter. The energy sits in different places, and in the same housing the three can behave quite differently. For noise-sensitive applications a single grade callout on the drawing is thin. Specify the control items: limits on Fα and Fβ, flank roughness, runout, and where needed a modification requirement such as tip relief or lead crowning so that load moves off the tooth ends and back toward the middle of the face width.

The next layer is machining. On the tolerance tables we work to, ground gears at ISO 4–5 carry flank roughness of Ra 0.2–0.4 μm and Rz 1.0–2.0 μm; ground or shaved gears at ISO 6–7 sit at Ra 0.4–0.8 μm; hobbed or shaped gears at ISO 8–9 run Ra 0.8–1.6 μm. Lower roughness means smaller micro-impacts at the contact, and that is grinding’s direct contribution to noise. But passing the roughness check does not mean the flank is free of waviness. Wheel dressing intervals, feed rate and chatter during grinding all leave longer-wavelength ripples, and those ripples excite frequencies that tend to land in the range the ear notices most, while Ra on the roughness tester shows nothing unusual.

Heat treatment fixes the sequence, and the order cannot be reversed. Carburizing and quenching takes a case depth of 0.5–3.0 mm against module, with 58–62 HRC at the surface and 30–42 HRC in the core, quenching at 830–860 °C and tempering at 520–650 °C. That route always distorts the part. If the gear is not ground afterwards, the profile and lead you deliver are the post-quench ones. The grinding allowance has to be set per heat-treatment route and module: too little and the distortion cannot be cut out, too much and the wheel removes a heavy layer, which invites burn. Then the datum. Grinding and inspection must locate on the same surfaces. If the drawing does not state which face is the datum, the measured lead will not match what happens inside the housing, and that gap shows up as edge contact and noise.

Assembly and housing often matter more than the gear itself. Centre distance deviation is held to roughly ±11–18 μm at ISO 5–6 and ±18–30 μm at ISO 7–8. Axis parallelism at ISO 5 is fΣδ 6 μm and fΣβ 12 μm; at ISO 7 it is 11 μm and 22 μm. Blank face runout stays within 0.02 mm, finished face runout within 0.03 mm. A bearing bore located ten-odd microns off is enough to cancel the lead advantage of a grade 5 gear. The gear is one link in an accuracy chain, not the chain.

Design choices are worth settling early as well. Helical gears have a higher contact ratio than spur gears, mesh more continuously, and generally sound better at the same accuracy level; helix angles commonly run 0–30°. At a 20° pressure angle, a minimum of 18 teeth avoids undercut. Single-stage ratios run 1:1 to 1:6, pitch line velocity up to 25 m/s, efficiency 98–99%. In the pair design, choosing relatively prime tooth counts and avoiding in-phase error stacking between the two wheels spreads both wear and excitation.

What procurement can actually enforce fits into three lines in the purchase order: inspection reports must carry measured curves, not only a grade conclusion; the scatter of Fα and Fβ within a batch has to be stated; and batch consistency needs a defined method — first-article, patrol inspection frequency, full inspection on critical parts. As a reference level, our outgoing checks for precision parts are written around Fp within 25 μm, Fα within 10 μm, Fβ within 10 μm, Fr within 15 μm and flank Ra within 0.8 μm; the drawing grade governs the actual numbers.

Two items from the industry side are worth reading together. Gear Technology’s State of the Gear Industry 2026, published 4 March 2026 with 137 survey respondents, reports that EV and precision transmission customers now ask for “noiseless gears” outright, a requirement that runs back through design, cutting and finishing. The second is Klingelnberg’s JIMTOF 2026 release (Tokyo Big Sight, 26–31 October): Quiet Bevel Grinding varies machine kinematics with a low-frequency harmonic from tooth slot to tooth slot, so each flank topography differs slightly and the excitation spreads across several frequencies instead of concentrating in one order. The result is heard as broadband, closer to a lapped set, and it does not affect cycle time or the established measurement-correction process. Both point the same way: noise is not always solved by pushing the grade higher. The distribution of energy across the spectrum is a design variable of its own.

(Sources: Gear Technology, State of the Gear Industry 2026, 4 March 2026, https://www.geartechnology.com/state-of-the-gear-industry-2026 ; Klingelnberg, JIMTOF 2026 press release, https://klingelnberg.com/en/news/news/presse )

At Geyontech, ground gears reach DIN 4–5, equivalent to ISO 4–5 and AGMA 2000-A88 grade 13–14; hobbing holds DIN 4 on single parts and DIN 5 in series. Inspection runs on a P400 and a Klingelnberg P26 measuring centre, with complex contours passed to a Zeiss CMM and flank hardness checked on a Rockwell tester. The rack range covers five series in spur and helical, ground and precision-milled, module M1–M12, standard lengths of 500 mm and 1000 mm, accuracy stated as 6h25.

Send the drawing together with the operating conditions — input speed, the orders you care about, housing structure and lubrication — and we will answer process route and inspection items line by line: https://www.geyontech.com/get-a-gear-quote/