Spiral bevel gears transmit torque between intersecting shafts through curved teeth that enter mesh gradually. This geometry gives high load capacity, smooth motion and high efficiency, which explains their use in agricultural machinery, vehicle drivelines and other compact, high-torque transmissions. The same geometry also makes contact stress sensitive to small changes in tooth-surface shape, shaft alignment and elastic deflection under load. Improving manufacturing and installation accuracy can reduce those stresses, but the cost rises quickly. Micro-modification of the tooth surface is therefore an attractive design lever: a small, deliberately introduced deviation from the conjugate surface can move the contact pattern away from the tooth boundaries and spread the load over a larger area.
Spiral bevel gears use curved teeth to transmit motion between intersecting axes; the contact pattern, not only the nominal geometry, governs the resulting contact stress.
The practical question is not whether modification helps, but which modification parameters matter and how they should be combined. For a spiral bevel gear pair with 14 pinion teeth and 25 wheel teeth, a 90° shaft angle and a 35° spiral angle, a transient dynamic finite element model was used to compute the contact stress through a complete mesh cycle. Five tooth-surface modification coefficients were then varied, their influence on the peak contact stress was ranked, and a surrogate-based optimization was used to search for a low-stress combination.
Gear set and transient meshing model
The gear pair is a right-hand pinion meshing with a left-hand wheel at a 90° shaft angle. The main geometry is listed in Table 1. Both gears are made of 20CrMnTi, with a density of 7,800 kg/m³, an elastic modulus of 207 GPa and Poisson’s ratio of 0.25.
Table 1. Main geometry of the spiral bevel gear pair.
| Parameter | Pinion | Wheel |
|---|---|---|
| Number of teeth | 14 | 25 |
| Large-end transverse module (mm) | 3.5 | 3.5 |
| Spiral angle and hand | 35°, right-hand | 35°, left-hand |
| Face width (mm) | 16 | 16 |
| Normal pressure angle (°) | 20 | 20 |
| Shaft angle (°) | 90 | |
| Pitch cone angle (°) | 29.249 | 60.751 |
| Root cone angle (°) | 26.171 | 55.544 |
| Face cone angle (°) | 34.456 | 63.829 |
| Cone distance (mm) | 50.143 | 50.143 |
| Whole tooth depth (mm) | 6.608 | 6.608 |
| Tangential modification coefficient | 0.025 | −0.025 |
| Radial modification coefficient | 0.27 | −0.27 |
The finite element model used hexahedral elements, with 70 elements along the tooth profile and 50 along the face width, giving a total of 1,308,220 elements. A mesh-convergence check preceded the contact calculations. The transient dynamic contact analysis was performed in LS-DYNA. The pinion tooth surface was defined as the contact surface and the wheel tooth surface as the target surface. Contact was treated as frictionless, consistent with the load-capacity formulation of ISO 10300-1:2023 for spiral bevel gears, in which friction is not included in the calculation.
Rotational degrees of freedom were released at the inner rings of both gears through grounded revolute joints. The pinion received a rotational speed and the wheel a resisting torque; both were ramped over 0.002 s, followed by one full pinion revolution of 0.015 s. The total simulated time was 0.017 s. The initial gear positions were adjusted so that contact began without initial penetration, which improves convergence of the contact analysis.
Baseline contact stress and the edge-contact mechanism
In the unmodified design, the contact pattern did not remain in the central part of the tooth. Under load it migrated toward the tooth tip, and during meshing-out the contact zone reached the tooth boundary. The result was edge contact: the load was carried by a narrow strip at the tip rather than by a well-distributed elliptical area. The computed peak contact stress reached 5,595 MPa at the tooth tip during the meshing-out stage, and the contact stress remained high throughout the mesh cycle.
This behavior is a geometric and elastic effect rather than a simple overload. A spiral bevel gear tooth is stiff in the central region but becomes less constrained near the tip and ends. When the contact pattern reaches those boundaries, the effective bearing area shrinks, the local compliance changes abruptly, and the same transmitted load is concentrated into a smaller region. Small alignment errors or tooth deflections can therefore produce a disproportionate rise in peak stress. The baseline result identifies the problem that modification must solve: not merely lower the average stress, but move the contact away from the tip and ends.
Five tooth-surface modification parameters
The modification was applied to the pinion because its smaller tooth count and greater freedom in machining adjustment make it more efficient to modify than the wheel. A second-order surface was used to approximate the tooth-surface deviation, and the topology was decomposed into five independent directions. Each direction has a clear geometric role in controlling the contact pattern.
Table 2. Modification coefficients, allowable ranges and geometric roles.
| Modification | Symbol | Range (μm) | Primary geometric role |
|---|---|---|---|
| Spiral angle correction | a1 | −20 to +20 | Shifts the contact pattern along the face and changes the effective spiral overlap |
| Pressure angle correction | a2 | 0 to 55 | Alters profile curvature and the position of profile contact |
| Lengthwise crowning | a3 | 0 to 20 | Relieves the tooth ends and reduces end loading |
| Profile crowning | a4 | 0 to 55 | Relieves the tip and root and controls profile contact |
| Torsional twist | a5 | −55 to +5 | Superposes a diagonal ease-off component across the face |
The design vector is \(A=\{a_1,a_2,a_3,a_4,a_5\}\). The optimization problem minimizes the maximum contact stress \(f_1(A)=\sigma_{\max}(A)\) over the five-dimensional box:
$$
\begin{aligned}
A &= \{a_1,a_2,a_3,a_4,a_5\},\\
\min_{A}\quad & f_1(A)=\sigma_{\max}(A),\\
\text{s.t.}\quad
& -20 \le a_1 \le 20,\\
& 0 \le a_2 \le 55,\\
& 0 \le a_3 \le 20,\\
& 0 \le a_4 \le 55,\\
& -55 \le a_5 \le 5,
\end{aligned}
$$
All bounds are in micrometres. The asymmetric bounds reflect the fact that the sign and magnitude of a correction are not equally useful in both directions: a spiral angle correction can move the pattern either way, whereas the pressure angle and crowning terms are defined as non-negative relief in their chosen directions.
A design of experiments (DOE) was used to sample the space. An optimal Latin hypercube design was generated in ISIGHT. For \(n\) design variables, the minimum number of points is estimated by
$$
N=\frac{(n+1)(n+2)}{2}.
$$
For five variables this gives \(N=21\). Thirty points were actually used so that the later surrogate model would have more information than the bare minimum. Each design point defined a modified tooth surface, and each surface was evaluated with the same transient contact model.
Sensitivity: profile crowning dominates
The peak contact stress from each design point was extracted, and a sensitivity analysis ranked the five coefficients by their contribution to the maximum stress. The results are summarized in Table 3.
Table 3. Sensitivity of the maximum contact stress to the five modification parameters.
| Rank | Parameter | Sensitivity magnitude (%) |
|---|---|---|
| 1 | Profile crowning, a4 | 42.57 |
| 2 | Torsional twist, a5 | 22.07 |
| 3 | Pressure angle correction, a2 | 15.29 |
| 4 | Spiral angle correction, a1 | 14.48 |
| 5 | Lengthwise crowning, a3 | 5.59 |
Profile crowning is the dominant single parameter, accounting for 42.57% of the sensitivity. This is consistent with the baseline failure mode: the peak stress occurs at the tooth tip, and profile crowning directly controls the tip relief and the shape of the profile contact zone. Torsional twist is second at 22.07%, reflecting its ability to rotate the contact pattern diagonally across the face and to redistribute load between the central and end regions. Pressure angle correction and spiral angle correction follow at 15.29% and 14.48%, respectively. Lengthwise crowning has the smallest sensitivity at 5.59%, but it is not negligible: the baseline edge contact is concentrated at the tip, so a parameter that mainly relieves the tooth ends has less leverage than the profile-related parameters.
All five sensitivities exceed 5%. No single coefficient can therefore be treated as irrelevant, and the useful design question is how the five should be combined. A one-factor-at-a-time adjustment would miss the interaction between profile shape, lengthwise shape and diagonal twist.
Surrogate model and NSGA-II optimization
The thirty design points were used to construct a Kriging surrogate model in ISIGHT, with a Gaussian correlation function, relating the five modification coefficients to the maximum contact stress. The fitted model achieved a coefficient of determination \(R^2=0.933\) and an average relative error of 0.051, or 5.1%. That accuracy is sufficient for search and screening, but the final candidate was still verified with the finite element model rather than accepted from the surrogate alone.
The optimization was carried out with the non-dominated sorting genetic algorithm with elitism, NSGA-II. The population size was 40, the number of generations was 400, the crossover probability was 0.9, the crossover distribution index was 10, and the mutation distribution index was 20. The search minimized the peak contact stress over the five-dimensional box defined above. Although NSGA-II is normally associated with multi-objective problems, here it served as the evolutionary search engine for a single stress objective; the result is a stress-minimizing combination of the five coefficients rather than a trade-off curve among competing objectives.
Optimized contact pattern and stress reduction
The optimized modification was re-evaluated with the transient contact model. The resulting contact pattern was no longer at the tooth boundary. The contact area was an ellipse located in the middle of the tooth surface, and the maximum contact stress occurred at the centre of that ellipse. The peak stress fell to 1,120 MPa.
Table 4. Baseline and optimized contact-stress results.
| Metric | Baseline | Optimized |
|---|---|---|
| Maximum contact stress | 5,595 MPa | 1,120 MPa |
| Location of maximum stress | Tooth tip | Centre of contact ellipse |
| Contact pattern | Edge contact at the tip during meshing-out | Centred elliptical area at mid-tooth |
| Reduction relative to baseline | — | 4,475 MPa, or 80.0% |
The reduction is approximately a factor of five. The stress histories of the baseline and optimized designs both fluctuate periodically, with a period matching the gear mesh frequency. The optimized design is lower throughout the cycle, and the difference is especially clear in the mid-to-late meshing period, where the baseline design was dominated by tip edge contact. This is the expected signature of a geometry change that removes a boundary concentration rather than merely shifting the same peak to another location.
The result also explains why the optimization did not simply maximize profile crowning. The five coefficients interact: profile crowning relieves the tip, but excessive crowning can shorten the contact ellipse and reduce the load-sharing area; torsional twist can compensate for the diagonal migration of the contact pattern; spiral angle and pressure angle corrections position the pattern along the face and profile. The optimizer found a combination that keeps the ellipse centred and sufficiently large, rather than a single extreme value.
Design implications and limits of the result
For this gear pair, the dominant route to lower contact stress is control of the profile contact boundary. Profile crowning and torsional twist together account for nearly two-thirds of the sensitivity, so they deserve the most attention in tolerance allocation and in any subsequent design refinement. Lengthwise crowning remains useful for end relief, but it cannot by itself correct a tip-dominated edge contact. The five-parameter representation is also more informative than a single crowning value because it separates the lengthwise, profile and diagonal components of the surface deviation.
The quantitative result—5,595 MPa reduced to 1,120 MPa—belongs to the specific geometry, material, load case and frictionless contact model described here. It is not a universal modification recipe. The model does not include lubrication, thermal effects, manufacturing scatter or a range of mounting misalignments, and the optimization used one load level. A design intended for series production should therefore treat the optimized coefficients as a starting point and examine how the contact pattern moves under torque variation, misalignment and tooth-to-tooth manufacturing deviation. Robust optimization over those sources of variation would be the natural next step.
Within those limits, the mechanism is clear. Edge contact concentrates load into a small boundary region and produces a stress peak that is much higher than the nominal contact stress. A five-parameter tooth-surface modification can move the contact ellipse back to the middle of the tooth, enlarge the effective bearing area and reduce the peak stress by about 80%. The sensitivity ranking shows where the design effort should go, and the surrogate-based optimization shows how the interacting coefficients can be combined to achieve a centred contact pattern.