Spiral bevel gears transmit torque between intersecting shafts through curved teeth that engage gradually along the face width. The curved tooth trace gives these drives high load capacity, smooth motion, and a compact layout, which explains their use in aerospace, transport, and industrial machinery. The same geometry also makes them sensitive: the tooth surface is a spatial surface, and micron-scale changes in its topology can shift the contact path, redistribute load, and alter the vibration signature of the whole drive.
The central design variable is transmission error, the difference between the actual output rotation and the rotation that perfect conjugate motion would produce. Under light load, its geometric component is fixed by the tooth-surface design. Under torque, elastic deformation adds a loaded component. A large peak-to-peak loaded transmission error means that the output angular velocity fluctuates during each mesh cycle, which directly excites gear whine and dynamic bearing loads. Edge contact—where the contact ellipse runs off the tooth boundary—is especially harmful because it produces local stress spikes and a strongly varying compliance. A 2020 finite element study examined how two deliberately different geometric transmission errors affect contact stress, loaded transmission error, and vibration of a 16/27-tooth spiral bevel pair at 200 N·m.
Transmission Error as a Design Variable
Geometric transmission error is usually shaped as a smooth function of pinion roll angle, with the contact path and contact ellipse placed by the tooth-surface design. In the comparison, the contact path angle was held at 70° and the ratio of the instantaneous contact ellipse semi-major axis to face width at 0.3, while the geometric transmission error was set to 15 arcsec in one design and 30 arcsec in the other. Holding the contact pattern descriptors constant isolates the effect of transmission-error magnitude from other contact-pattern changes.
Loaded tooth contact analysis supplies the bridge between geometry and system behavior. It solves the contact problem under a prescribed torque and returns the load distribution, contact stress, and loaded transmission error. The peak-to-peak amplitude of the loaded transmission error is a practical indicator of smoothness: a larger amplitude corresponds to stronger velocity fluctuation and, generally, higher vibration. For a 16-tooth pinion, one mesh cycle covers 22.5° of pinion rotation, so the contact stress and transmission-error histories are evaluated over that interval.
A Tooth-Surface-Based Finite Element Model
The finite element model was generated directly from the tooth-surface equations rather than from a converted CAD solid. The wheel tooth surface is written as $r_2(u_2,v_2)$. Grid nodes are laid out in the rotation projection plane, and the position equation is solved for the position vector and normal vector at each node on the surface. This avoids the two geometry conversions of a CAD-based workflow, both of which can lose micron-level tooth-surface detail and degrade assembly accuracy.
The working tooth surface was divided into 25 nodes along the face width and 9 nodes along the working tooth depth. The root transition curve, generated by a cutter tip radius of 0.6 mm, has a high curvature and was refined with 6 additional nodes along its height. Tooth thickness was set by angular positioning: for the convex surface, a thickness reference point was taken at the midpoint of the pitch line, and the surface was rotated until the $y$ coordinate of that point vanished. The rotation angle $\phi$ for the required tooth thickness follows
$$ \phi = \frac{\pi}{z} – \frac{j_n}{z m_t} $$
where $z$ is the number of teeth, $j_n$ the normal backlash, and $m_t$ the large-end transverse module. The same procedure was applied to the concave surface with its own reference point and rotation angle.
Tooth contact analysis then provided the wheel and pinion rotation angles needed to bring the two surfaces into mesh. The node and element data were written in the Abaqus input-file format and assembled directly at the correct meshing position. All elements were eight-node linear bricks (C3D8). Because only a few teeth carry load at any instant, a five-tooth sector was used for each member to reduce computational cost while retaining the contacting region.
| Parameter | Pinion | Wheel |
|---|---|---|
| Number of teeth | 16 | 27 |
| Module (mm) | 4.25 | 4.25 |
| Face width (mm) | 17 | 17 |
| Pressure angle (°) | 20 | 20 |
| Shaft angle (°) | 90 | 90 |
| Spiral angle (°) | 35 | 35 |
The contact analysis used a density of $7.9 \times 10^{-9}$ t/mm³, an elastic modulus of $2.1 \times 10^5$ MPa, and a Poisson ratio of 0.3 for both members. The tooth flanks, root transition regions, and tip transition regions of adjacent teeth were grouped into one contact surface per member, with the pinion as the master surface and the wheel as the slave surface. The friction coefficient was 0.06. Reference points on the shaft axes were coupled to the inner rings and sections through rigid constraints. The implicit static analysis used two steps: 1 s to establish stable contact and 5 s to apply the load. The pinion was fully constrained, while the wheel was free to rotate about its own axis under a 200 N·m torque applied through an amplitude ramp.
Two Contact Patterns under the Same Torque
The two tooth-surface designs produced different contact patterns under the same 200 N·m torque. At the instant of entry into mesh and the instant before exit, the contact ellipse was truncated by the tooth boundary, producing edge contact in both designs. The 15 arcsec design showed the more severe form: the edge contact was more pronounced, and the contact stress rose sharply at the beginning and end of the mesh cycle. The 30 arcsec design also experienced edge contact, but to a lesser degree, and its contact pattern remained farther from the tooth edges during the loaded portion of the cycle.
The difference is visible in the stress distribution on the wheel convex flank at zero pinion rotation. The 15 arcsec model concentrated stress closer to the boundary, while the 30 arcsec model spread the loaded region more toward the interior of the flank. The maximum contact stress at each instant was extracted to form the contact stress history over the 22.5° mesh cycle. Both histories show the characteristic peaks at mesh entry and exit caused by edge contact, but the peaks are more severe for the smaller geometric transmission error. Increasing the geometric transmission error from 15 to 30 arcsec therefore reduces, but does not eliminate, edge contact.
Loaded Transmission Error
Loaded transmission error is the difference between the actual wheel rotation and the theoretical rotation when the pinion turns at a uniform speed under load. It is dominated by elastic deformation of the teeth and the contact zone, so it responds directly to changes in contact pattern and edge loading. The computed mean values and peak-to-peak amplitudes are summarized in Table 2.
| Geometric transmission error (arcsec) | Mean loaded transmission error (arcsec) | Peak-to-peak amplitude (arcsec) | Relative amplitude |
|---|---|---|---|
| 15 | 201.66 | 52.82 | 1.44 |
| 30 | 188.53 | 36.72 | 1.00 |
The 15 arcsec design had both the higher mean loaded transmission error and the larger fluctuation. Its peak-to-peak amplitude of 52.82 arcsec was about 44% larger than the 36.72 arcsec of the 30 arcsec design; expressed the other way, increasing the geometric transmission error to 30 arcsec reduced the amplitude by about 30% relative to the 15 arcsec case. Because loaded transmission error is mainly an elastic quantity, this result is consistent with the contact stress observation: the more severe edge contact of the 15 arcsec design produces a larger variation in compliance through the mesh cycle, and therefore a larger variation in output rotation.
Vibration Response
The vibration analysis used modal superposition. The contact force history from the dynamic analysis was decomposed into axial, radial, and tangential components and applied along the contact path as the excitation. The response was evaluated at a point on the pinion shaft bore, with the pinion axis as $X$, the wheel axis as $Z$, and $Y$ completing the orthogonal frame. Root-mean-square values of displacement, velocity, and acceleration at that point are listed in Tables 3–5.
| Geometric transmission error (arcsec) | X (μm) | Y (μm) | Z (μm) | Total (μm) |
|---|---|---|---|---|
| 15 | 12.740 | 9.512 | 5.402 | 16.792 |
| 30 | 10.968 | 7.853 | 5.051 | 14.404 |
| Geometric transmission error (arcsec) | X (mm/s) | Y (mm/s) | Z (mm/s) | Total (mm/s) |
|---|---|---|---|---|
| 15 | 19.574 | 23.721 | 2.023 | 30.820 |
| 30 | 18.905 | 21.421 | 2.209 | 28.655 |
| Geometric transmission error (arcsec) | X (m/s2) | Y (m/s2) | Z (m/s2) | Total (m/s2) |
|---|---|---|---|---|
| 15 | 70.881 | 71.454 | 17.332 | 102.128 |
| 30 | 72.846 | 68.623 | 16.337 | 101.403 |
The displacement response follows the loaded transmission error trend most clearly. The total RMS displacement fell from 16.792 μm to 14.404 μm, about 14% lower for the 30 arcsec design. The total RMS velocity fell from 30.820 mm/s to 28.655 mm/s, about 7% lower. These reductions are consistent with the smaller loaded transmission error amplitude: a more stable output rotation means less motion at the shaft bore.
The acceleration result is less uniform. The total RMS acceleration was almost unchanged, falling only from 102.128 to 101.403 m/s2, and the $X$ component was slightly higher for the 30 arcsec design (72.846 versus 70.881 m/s2). The $Y$ and $Z$ components were lower, by about 4% and 6%, respectively. The acceleration response therefore does not simply inherit the displacement trend in every direction. At this response point, the larger geometric transmission error clearly improves displacement and velocity, while its acceleration benefit is marginal and direction-dependent.
What the Comparison Supports
Within the tested window, increasing the geometric transmission error from 15 to 30 arcsec acted as a useful tuning direction. It reduced the severity of edge contact at mesh entry and exit, lowered the peak-to-peak loaded transmission error, and reduced the RMS displacement and velocity at the pinion shaft bore. The mechanism is consistent across the three result types: less edge loading means a more stable contact compliance through the mesh cycle, which appears as a smaller fluctuation in loaded transmission error and as lower motion at the response point.
The result should not be read as a rule to maximize geometric transmission error. Only two values were compared, at one torque, with one contact path angle and one contact ellipse ratio, and at one response point. Contact ratio, tooth strength, and behavior at other loads were not part of the comparison. The acceleration data also show that a smoother loaded transmission error does not guarantee lower acceleration in every direction. A design optimum would require sweeping the geometric transmission error over a wider range, checking other loads and contact-pattern parameters, and confirming the vibration ranking against measurement.
The modeling chain itself is transferable: tooth-surface equations generate the mesh directly, tooth contact analysis sets the assembly position, loaded contact analysis returns stress and transmission error, and modal superposition converts the contact force history into a vibration estimate. That chain preserves tooth-surface detail that a CAD conversion can lose, but its predictions remain dependent on mesh density, friction definition, and boundary constraints. For spiral bevel gears, the practical conclusion is that geometric transmission error is not merely a quality metric to be minimized; within a bounded range it is a design lever that can trade edge contact against transmission-error fluctuation and vibration.