Why the curved tooth trace matters
A spiral bevel gear transmits motion between intersecting shafts through teeth whose longitudinal lines are arcs rather than straight generators. The curvature is a load-path decision, not a stylistic one. Because the flank enters and leaves mesh progressively, several tooth pairs share the torque at any instant, the total contact ratio rises well above that of a straight bevel pair of the same envelope, and the impact at engagement is spread over a longer arc of rotation. The result is higher torque capacity at equal size, smoother motion, and lower mesh noise — the reasons the configuration dominates final drives in mining machinery, transport, aerospace and precision instruments.
The cost of that geometry is that the flank is neither a plane involute nor a simple ruled surface. It is generated by a cutter head rotating in a cradle that represents an imaginary generating gear, so the tooth surface is defined implicitly by the machine motion. Two capabilities therefore determine how quickly a design can be evaluated: a repeatable way to rebuild the exact flank from a short list of parameters, and a contact model that resolves the real load transfer — first in the static regime, then under rotation, and finally in the modal domain where resonance risk is decided.
What follows traces that chain for a representative pair: module 6 mm, 15 and 46 teeth, 20° pressure angle, 90° shaft angle, 40 mm face width, 35° mean spiral angle, right-hand pinion and left-hand gear, with addendum and clearance coefficients of 0.85 and 0.188 following the Gleason recommendation for pinions above 12 teeth.
Geometry: the spherical involute and the base cone
Two base cones sharing an apex and a generator length roll on each other without slip. A point fixed on one cone traces, on the sphere centred at the apex, the spherical involute, which is the theoretical profile of a bevel tooth. The same motion can be read as a plane rolling on a single base cone, with every point of the plane held at the constant cone distance R from the apex.
The base cone half-angle follows from the pitch cone angle and the pressure angle. The plane containing the pitch line cuts a spherical right triangle in which the angle between base and pitch cone is 90° − α, and the sine rule gives the closure relation
$$\sin\delta_b=\sin\delta\cos\alpha .$$
For a roll parameter θ (rotation of the generating plane about the cone axis), the no-slip condition sets the in-plane angle ψ = θ sin δb, and the involute point on the sphere of radius R has components
$$x=R\left(\cos\psi\,\sin\delta_b\cos\theta-\sin\psi\,\sin\theta\right)$$
$$y=R\left(\cos\psi\,\sin\delta_b\sin\theta+\sin\psi\,\cos\theta\right)$$
$$z=R\cos\psi\,\cos\delta_b ,$$
with polar angle ω measured from the axis satisfying cos ω = cos ψ cos δb. Converting these components to spherical coordinates yields the radius, colatitude and azimuth triplets that drive the CAD curves directly.
Two refinements make the surface usable. First, the flank cannot be represented by the heel and toe involutes alone: intermediate involutes at radii r = R − 0.05nB for n = 1…19 are interpolated across the face width, which removes the ruled-surface error that a two-curve loft introduces. Second, the lengthwise curve comes from the cutter head. With cutter radius rd and mean spiral angle βm, the centre offset is
$$S=\sqrt{R^{2}+r_d^{2}-2Rr_d\sin\beta_m},$$
and the angular positions of heel and toe about the cutter centre give, through the spherical sine relation sin K = sin η / sin δ, the rotation K12 that the tooth sweeps in spherical coordinates. Between them these curves close a single-tooth boundary that can be lofted into a solid.
Fully parametric CAD construction
Manual curve construction is slow and error-prone, so the whole chain from parameters to solid is worth automating. The build proceeds in the same order every time: involute and lengthwise curves, surface patches through a curve mesh, sewing of the patches into a single-tooth solid, circular patterning, boolean union with the blank, then bore, chamfer and trim operations.
| Parameter | Symbol | Value |
|---|---|---|
| Module | m | 6 mm |
| Teeth, pinion / gear | Z₁ / Z₂ | 15 / 46 |
| Pressure angle | α | 20° |
| Shaft angle | Σ | 90° |
| Face width | B | 40 mm |
| Mean spiral angle | βm | 35° |
| Addendum / clearance coefficient | ha* / c* | 0.85 / 0.188 |
| Hand of spiral | — | Pinion right, gear left |
The parameterisation was implemented with UG/OPEN GRIP, the built-in二次 development language of the CAD environment. The source is written as a plain text file, renames to the .grs extension, compiled and linked into an executable .grx module, and invoked from the menu. A dialog then accepts module, tooth counts, spiral angle, pressure angle, face width and the remaining coefficients, and regenerates the pinion, the gear and their virtual assembly. The practical gain is that a design iteration becomes a parameter edit rather than a rebuild of every curve, which is the difference between exploring a design space and checking a single point in it.
Finite element idealisation of a cyclic contact problem
Flank twist makes direct meshing of the full gear impractical, so the geometry is split in a dedicated pre-processor. Each tooth is separated from the blank, the flank is divided into a regular quad surface mesh by adding auxiliary points and lines, and the two-dimensional mesh is swept into hexahedra with a solid map operation. The single-tooth mesh is then patterned, and the duplicate nodes at the interfaces are merged with an equivalence tolerance set to 10% of the smallest element; a final element-quality check guards against excessive distortion.
The model used for static analysis is a five-tooth sector: 52,800 linear reduced-integration hexahedra (C3D8R) at 1 mm nominal size, with 61,869 nodes. Three to five teeth are sufficient when the contact zone stays inside the sector, since the teeth outside it carry no load and only reproduce the boundary stiffness. The material is steel with E = 2.1 × 10⁵ MPa, ν = 0.3 and ρ = 7.8 × 10⁻⁹ t/mm³ in an SI/mm unit system, the density being required whenever rotational inertia matters.
Because hexahedral nodes carry no rotational degree of freedom, torque and speed are applied through a reference point on the gear axis, kinematically coupled to the bore and to the cut faces. The contact pair is defined surface-to-surface with finite sliding, hard normal behaviour and a penalty tangential law with μ = 0.1, the pinion flank acting as master and the gear flank as slave.
Loading is where such models usually fail. Applying a rotational velocity to the pinion and a torque to the gear in a single step, with a real backlash present, produces either no contact at all or a normal-force spike that the applied load cannot equilibrate. The remedy is a two-step sequence. In the first step a small torque is applied to the pinion with the gear fully constrained while a gradually removed interference fit closes the backlash, so that the flanks are in firm contact at the end of the step; the second step releases the gear rotation and applies the working torque. Increments are set to 0.01 initial, 1 × 10⁻⁷ minimum and 0.1 maximum, with up to 1000 increments over 1 s of step time. Rotation senses must be consistent with the right-hand rule for the pair.
Static contact: stress distribution and its path
Under 900 N·m on the gear, the equivalent contact stress peaks at 373.9 MPa on the driven flank and 314.1 MPa on the driving flank. The contact patch is elliptical, its major axis inclined to the tooth width, and the maximum sits near the ellipse centre. Watching the patch through a mesh cycle shows it migrating from the heel region to the toe along a diagonal trace, which is the expected signature of a spiral flank: contact begins at one end of the tooth and sweeps across rather than loading the whole face simultaneously.
| Component | Max equivalent contact stress (MPa) | Max root bending stress (MPa) |
|---|---|---|
| Pinion, 15 teeth | 314.1 | 117.2 |
| Gear, 46 teeth | 373.9 | 84.3 |
Root bending stress forms an elongated elliptical zone directly beneath the contact patch and follows it from heel to toe as meshing proceeds. The asymmetry between the two members is instructive: the gear sees the higher surface stress because its flank is flatter and the contact ellipse is wider, while the pinion sees the higher root stress because its fewer, more slender teeth carry the same torque through a smaller section. Surface durability and tooth breakage are therefore governed by different members of the same pair, and a design change that helps one may not help the other.
Loaded contact ratio
Geometric overlap is only part of the story. With no load the pair behaves as a line contact and the effective contact ratio collapses toward unity regardless of how many teeth the design places in the zone; as torque rises, elastic deformation spreads the contact and closes the gaps to neighbouring teeth. The classical combination of transverse and overlap components,
$$\varepsilon_\gamma=\sqrt{\varepsilon_\alpha^{2}+\varepsilon_\beta^{2}},$$
describes the unloaded geometry. The loaded value is better extracted from the transient contact-force history: if a tooth pair carries normal force from time t₁ to t₃ and the neighbouring pair engages at t₂, then
$$\varepsilon_\gamma=\frac{T’}{\Delta t}=\frac{t_3-t_1}{t_2-t_1}.$$
| Gear torque (N·m) | Loaded contact ratio εγ |
|---|---|
| 100 | 2.3 |
| 500 | 2.6 |
| 900 | 2.9 |
At 900 N·m the measured intervals were Δt = 0.20 s and T′ = 0.58 s. The monotonic rise from 2.3 to 2.9 across a ninefold torque increase is a deformation effect: larger load flattens the flanks further, widening the window in which adjacent teeth share the normal force. The seven-tooth sector used for this measurement exists precisely because the loaded zone grows beyond the five-tooth sector employed for stress extraction.
Transmission error
Transmission error arises from three sources that cannot be separated experimentally without care: the quasi-conjugate nature of the generated pair, elastic deflection of teeth and rims under load, and manufacturing deviation. Denoting the actual gear rotation A₂ and the pinion rotation A₁,
$$\Delta A=A_2-\frac{A_1Z_1}{Z_2},$$
plotted against A₂, gives the error curve whose smoothness is a direct indicator of how quietly the pair will run. For this gear set the curve amplitude decreases as torque increases from 100 to 900 N·m. That direction is not accidental: deflection under load compensates part of the geometric mismatch built into the conjugate approximation, so the pair runs measurably smoother when it is working than when it is idling. A design optimised for the unloaded transmission error alone would therefore be optimised for the wrong operating point.
Dynamic behaviour under rotation
Static analysis cannot describe a process in which the load path changes every millisecond. The explicit solver used here integrates the dynamic equilibrium
$$\mathbf{M}\ddot{\mathbf{u}}=\mathbf{P}-\mathbf{I}$$
with the central difference scheme, evaluating accelerations at the start of each increment and updating velocity at the midpoint and displacement at the end:
$$\dot{\mathbf{u}}^{(t+\Delta t/2)}=\dot{\mathbf{u}}^{(t-\Delta t/2)}+\frac{\Delta t^{(t+\Delta t)}+\Delta t^{(t)}}{2}\,\ddot{\mathbf{u}}^{(t)},$$
$$\mathbf{u}^{(t+\Delta t)}=\mathbf{u}^{(t)}+\Delta t^{(t+\Delta t)}\,\dot{\mathbf{u}}^{(t+\Delta t/2)}.$$
No global stiffness matrix is formed and no iteration is required, which suits a problem dominated by changing contact status; the stable increment depends only on the highest natural frequency of the model. The full gear is modelled for this stage, since mass distribution now matters, with a refined mesh confined to the meshing zone and tied to a coarser mesh elsewhere — accuracy where the contact lives, economy where it does not. The baseline case is a pinion speed of 100 rad/s ramped over 0.001 s against a 500 N·m resisting torque on the gear, reaching a steady 32.6 rad/s on the gear side, consistent with the 15/46 ratio.
Damping
Rayleigh damping is used, with the element damping matrix assembled from mass and stiffness contributions:
$$\mathbf{C}=\alpha_R\mathbf{M}+\beta_R\mathbf{K}.$$
The mass-proportional term is the one that governs low-order oscillation, which is the regime relevant to a meshing gear pair. Three values of αR were compared (Table 4). Increasing αR shortens the settling time from 0.015 s to 0.006 s and reduces the amplitude of the speed oscillation at engagement, but it does not change the final steady speed. It does, however, raise the force transmitted through the mesh: the steady contact force climbs from 24 to 52 kN across the same range. A damping coefficient chosen purely to suppress start-up oscillation will thus increase the sustained load on the flanks, which is a trade-off worth making explicitly rather than by default.
Inertial load
Inertial load scales with mass, so the gear mass was varied in the ratio 1 : 2 : 4 by adjusting density while holding the geometry fixed. Peak contact force rose from 246 to 427 to 713 kN, and the steady value from 38 to 71 to 136 kN — close to proportional to mass, as expected for a fixed kinematic excitation. Settling time lengthens and the oscillation frequency falls as mass increases. Heavy rotating components therefore amplify mesh force even when the transmitted torque is unchanged.
Friction
Tangential behaviour follows the classical Coulomb model, with a critical shear stress proportional to contact pressure:
$$\tau_{crit}=\mu p .$$
Raising μ from 0.05 to 0.15 lowered the peak contact force from 261 to 246 to 232 kN while leaving the steady force unchanged at 38 kN. Friction dissipates kinetic energy during the initial impact and shortens the transient, so a slightly rougher, more dissipative contact behaves better at start-up without altering the established load path.
Pinion speed
Speed is the single most influential factor in the sweep. Raising the pinion speed from 40 to 100 to 250 rad/s increased peak contact force from 128 to 246 to 508 kN and the steady force from 18 to 38 to 89 kN, while settling time grew from 0.006 to 0.011 s. Higher speed means higher impact velocity at engagement, stronger inertial resistance to the resulting acceleration, and a longer interval before the transient decays.
| Factor | Setting | Peak contact force (kN) | Steady contact force (kN) | Settling time (s) |
|---|---|---|---|---|
| Mass-proportional damping αR | 400 | 242 | 24 | 0.015 |
| 700 | 246 | 38 | 0.009 | |
| 1000 | 249 | 52 | 0.006 | |
| Gear mass ratio | 1 | 246 | 38 | — |
| 2 | 427 | 71 | — | |
| 4 | 713 | 136 | — | |
| Friction coefficient μ | 0.05 | 261 | 38 | — |
| 0.10 | 246 | 38 | — | |
| 0.15 | 232 | 38 | — | |
| Pinion speed (rad/s) | 40 | 128 | 18 | 0.006 |
| 100 | 246 | 38 | 0.008 | |
| 250 | 508 | 89 | 0.011 |
Which factor dominates: speed or torque
A three-by-three matrix of pinion speeds (40, 100, 250 rad/s) and gear torques (100, 500, 900 N·m) separates the two influences. Peak contact force rises with torque at every speed, but the sensitivity collapses as speed increases: a ninefold torque increase adds 13.8% to the peak force at 40 rad/s, only 3.2% at 100 rad/s and 0.6% at 250 rad/s.
| Gear torque | 40 rad/s | 100 rad/s | 250 rad/s |
|---|---|---|---|
| 100 N·m | 119.3 kN | 241.1 kN | 506.4 kN |
| 500 N·m | 128.0 kN (+7.30%) | 246.0 kN (+2.03%) | 508.1 kN (+0.33%) |
| 900 N·m | 135.7 kN (+6.02%) | 248.8 kN (+1.14%) | 509.4 kN (+0.25%) |
The mechanism is a change of regime rather than a change of magnitude. At low speed the transmitted torque controls the mesh force; at high speed the inertial term dominates and the torque becomes a secondary perturbation. Any dynamic model of a spiral bevel gear therefore needs the operating speed fixed before its load sensitivity is interpreted, and a design validated at low speed says little about behaviour at 250 rad/s.
Modal behaviour
Free vibration with the bore constrained follows from
$$\left(\mathbf{M}\ddot{\mathbf{X}}\right)+\left(\mathbf{K}\right)\mathbf{X}=\mathbf{0},\qquad\left[\mathbf{K}-\omega^{2}\mathbf{M}\right]\left\{\mathbf{X}\right\}=\mathbf{0},$$
solved with the Lanczos algorithm for the six lowest eigenvalues. The full gear is used, meshed freely with C3D4 tetrahedra; splitting the wheel would alter exactly the rim stiffness that governs these modes. Only linear behaviour is admissible, so any nonlinear definitions are linearised, and no external load is applied — the bore faces are simply fixed.
| Order | Pinion frequency (Hz) | Gear frequency (Hz) |
|---|---|---|
| 1 | 27,256 | 12,262 |
| 2 | 29,566 | 12,309 |
| 3 | 29,576 | 12,311 |
| 4 | 31,758 | 12,464 |
| 5 | 31,768 | 12,467 |
| 6 | 32,200 | 12,262* |
*The sixth eigenvalue coincides with the first at the precision reported.
Frequencies increase monotonically with mode order for both members, and the pinion sits roughly 2.4 times higher than the gear — a direct consequence of its smaller, stiffer rim. The mode shapes fall into a small set of recognisable families: circumferential (ring) modes, torsional modes in which the two rotational degrees of freedom dominate, diametral bending modes with one or two nodal diameters, and umbrella modes in which the rim cone deflects axially.
| Order | Pinion | Gear |
|---|---|---|
| 1 | Circumferential (Z rotation dominant) | Umbrella |
| 2, 3 | Torsional (X and Y rotation dominant) | Torsional |
| 4, 5 | Second-order diametral bending | Second-order diametral bending |
| 6 | Umbrella plus circumferential | Third-order diametral bending |
The participation factors confirm the classification: for the pinion’s first mode the Z-rotation component is −0.945 while the translational components are of order 10⁻³ to 10⁻⁵; modes two and three are dominated by X and Y rotations at magnitudes near 1.2 to 1.5. Diametral bending is the family that matters most for durability, since it repeatedly flexes the rim and root fillet.
A resonance check needs the excitation frequency, not just the natural frequencies. At 100 rad/s the pinion turns at 955 rpm, giving a mesh frequency of 15 × 955/60 ≈ 239 Hz; at 250 rad/s it reaches roughly 597 Hz. Both are two orders of magnitude below the lowest computed natural frequency, so direct mesh-harmonic resonance is not a concern in this operating range for the bore-fixed configuration. The caveat is that a fixed bore is a stiff idealisation: bearing and housing compliance will lower the assembled natural frequencies considerably, and impact or parametric excitation from tooth engagement can still reach high-order modes. Adjusting pressure angle, spiral angle or root fillet geometry shifts these frequencies, and keeping the assembled natural frequencies away from the mesh harmonics is the practical design rule.
Design implications
- Spherical involute geometry parameterised through the base cone angle and a roll parameter gives an exact, editable definition of the flank; interpolating intermediate involutes across the face width removes the lofting error that end-curve-only models carry.
- Automating the build through a compiled parametric module converts an iteration from hours of curve work into a parameter edit, which is what makes load-case exploration feasible.
- Static contact stress and root bending stress peak on opposite members of the pair (373.9 MPa surface on the gear, 117.2 MPa root on the pinion), so surface and bending criteria must be checked separately.
- Loaded contact ratio rises with torque (2.3 at 100 N·m to 2.9 at 900 N·m), and transmission error amplitude falls over the same range; both effects come from elastic deformation compensating geometric mismatch, so unloaded predictions understate real performance.
- In dynamic operation, damping and friction both shorten the start-up transient while damping also raises the sustained mesh force; gear mass scales peak and steady force almost proportionally; pinion speed is the dominant variable above roughly 100 rad/s.
- Dominant mode families are torsional and diametral bending at 12.3–12.5 kHz for the gear and 27.3–32.2 kHz for the pinion, far above the mesh harmonics of the modelled speed range but sensitive to boundary compliance in the assembled system.