Spiral bevel gears carry motion between intersecting shafts with smooth, low-noise engagement and high load capacity, which is why they dominate helicopter transmission, aero-engine and automotive final drives. Predicting how a pair actually meshes — where the contact trace runs across the flank and how much the output shaft deviates from the ideal ratio — depends on a tooth contact analysis (TCA) that solves the mesh conditions of two generated flank surfaces simultaneously. No such analysis can start without an initial point: a pair of surface parameters that already lies close to the true contact position. When the gear set is assembled with a vertical offset, a horizontal offset or a shaft-angle error, that point moves, and a starting value taken from the nominal design may fall outside the convergence region of the equation solver.
The starting-value problem
The mesh condition is a system of nonlinear equations in the surface parameters of both members and their angular positions. Classical local synthesis presets a contact point by construction, so the analysis can begin there directly. Other routes to the same starting value have been proposed. Litvin and co-workers subdivided both flanks into cells and combined a minimum normal deviation criterion with a minimum distance criterion through weighting coefficients, but those coefficients are difficult to set and installation errors were not included. Tang and Lei determined the pinion point only after a point on the gear flank had been prescribed. Li and co-workers discretized the two flanks into matching cells and selected the cell of shortest distance, again without misalignment. Su and co-workers showed how the contact trace of a spiral bevel pair shifts under installation error but did not address how the starting point is obtained. Ding and co-workers developed a method for hypoid and spiral bevel gears that requires measured flank form error as input.
The procedure described below works from design and machine-tool settings alone. It treats misalignment explicitly: the two flanks are discretized, brought into a common fixed frame, rotated about their own axes until their unit normals coincide, and the pair of points with the smallest remaining separation is selected as a first estimate; a Levenberg–Marquardt refinement then turns that estimate into the initial point used by the contact analysis.
Flank geometry from the generating motion
The flanks are described through the generating process rather than by a nominal cone. The coordinate chain runs from the cutter frame Mci through the machine frame Mmi to the workpiece frame Mi, with three auxiliary frames Mni, Mei and Mfi carrying the position and motion relationships between tool, machine and blank. The position vector and unit normal of an arbitrary point on the cutting edge, expressed in the blank frame, are
$$\mathbf{r}_i(u_{ci},\theta_{ci},\varphi_i)=\mathbf{M}_{if_i}\mathbf{M}_{f_ie_i}\mathbf{M}_{e_im_i}\mathbf{M}_{m_in_i}\mathbf{M}_{n_ic_i}\,\mathbf{r}_{ci}(u_{ci},\theta_{ci})$$
$$\mathbf{n}_i(u_{ci},\theta_{ci},\varphi_i)=\mathbf{L}_{if_i}\mathbf{L}_{f_ie_i}\mathbf{L}_{e_im_i}\mathbf{L}_{m_in_i}\mathbf{L}_{n_ic_i}\,\mathbf{n}_{ci}(u_{ci},\theta_{ci})$$
where rci and nci are the position vector and normal of the cutter cone, uci and θci are the cutter surface coordinates, φi is the blank rotation angle, M denotes a 4×4 transformation matrix, L is its upper-left 3×3 submatrix, and the subscript i = 1, 2 refers to the pinion and the gear respectively. The generating motion adds the meshing condition
$$f(u_{ci},\theta_{ci},\varphi_i)=\mathbf{n}_{mi}\cdot\mathbf{v}_{oi}=0$$
in which nmi is the cutter normal in the machine frame and voi is the relative velocity between the virtual generating gear and the blank. Solving this relation for φi and substituting it back reduces each flank to a two-parameter surface:
$$\mathbf{r}_i=\mathbf{r}_i\left[u_{ci},\theta_{ci},\varphi_i(u_{ci},\theta_{ci})\right],\qquad \mathbf{n}_i=\mathbf{n}_i\left[u_{ci},\theta_{ci},\varphi_i(u_{ci},\theta_{ci})\right]$$
Assembly frame and installation errors
A fixed frame Ms is introduced at the assembly position, together with rotating frames M1 for the pinion and M2 for the gear, and an auxiliary frame Mt that carries the pinion offset. Three error components enter here: a vertical offset ΔV, a horizontal offset ΔH, and a shaft-angle error ΔΣ. Because the gear frame is made coincident with Ms, its flank data pass through unchanged, while the pinion flank and normal become
$$\mathbf{r}_{s1}=\mathbf{M}_{st}\mathbf{M}_{t1}\mathbf{r}_1,\qquad \mathbf{n}_{s1}=\mathbf{L}_{st}\mathbf{L}_{t1}\mathbf{n}_1,\qquad \mathbf{r}_{s2}=\mathbf{r}_2,\qquad \mathbf{n}_{s2}=\mathbf{n}_2$$
The rotation axes, given by the unit vectors i1 and i2 of the two rotating frames, transform to the fixed frame as
$$\mathbf{i}_{s1}=\mathbf{L}_{st}\mathbf{L}_{t1}\mathbf{i}_1,\qquad \mathbf{i}_{s2}=\mathbf{L}_{s2}\mathbf{i}_2$$
Simply placing the two gears in the assembly position does not bring corresponding flank points into mesh; each member must be turned about its own axis. Rotating the pinion through β1 about −is1 and the gear through β2 about is2 gives, by Rodrigues’ rotation formula,
$$\mathbf{r}_{si}^{*}=\cos\beta_i\,\mathbf{r}_{si}+\sin\beta_i\left(\mathbf{i}_{si}\times\mathbf{r}_{si}\right)+\left(1-\cos\beta_i\right)\left(\mathbf{i}_{si}\cdot\mathbf{r}_{si}\right)\mathbf{i}_{si}$$
$$\mathbf{n}_{si}^{*}=\cos\beta_i\,\mathbf{n}_{si}+\sin\beta_i\left(\mathbf{i}_{si}\times\mathbf{n}_{si}\right)+\left(1-\cos\beta_i\right)\left(\mathbf{i}_{si}\cdot\mathbf{n}_{si}\right)\mathbf{i}_{si}$$
Contact requires coincident position vectors and coincident unit normals:
$$\mathbf{r}_{s1}^{*}=\mathbf{r}_{s2}^{*},\qquad \mathbf{n}_{s1}^{*}=\mathbf{n}_{s2}^{*}$$
Since both normals are of unit length, the normal condition contributes only two independent scalar equations, so the system has five equations in six unknowns — uc1, θc1, β1, uc2, θc2, β2, with the blank rotations already eliminated by the generating condition. One parameter must therefore be fixed before the system can be solved.
Grid search on the discretized flanks
Both flanks are projected onto their axial sections and divided into m intervals along the tooth height and n intervals along the face width. The gear points are numbered from the toe at the tip to the heel at the root, the pinion points from the toe at the root to the heel at the tip, so that identically numbered points occupy roughly corresponding regions of the two flanks. For a selected pair of identically numbered points, the rotation angles β1 and β2 that align the two unit normals are computed in closed form from the axis vectors and normal vectors; after rotation the normals coincide and the residual separation of the two points is
$$d=\left|\mathbf{r}_{s1}^{*}-\mathbf{r}_{s2}^{*}\right|$$
Installation error destroys the correspondence implied by identical numbering, so the minimum of d generally lies at a neighbouring index pair. Taking the gear flank as the reference, each of its grid points is visited in turn and compared with several pinion points in the vicinity of the matching index; the pair returning the smallest d is retained. Sweeping the whole flank yields the angles β1*, β2* and the surface parameters uc1*, θc1*, uc2*, θc2* of the closest pair.
These values are estimates only, limited by the grid spacing. They serve as the starting vector for a Levenberg–Marquardt solution of the five mesh equations. Fixing β2 at its grid value leaves five unknowns for five equations, and the converged set defines the initial point on both flanks together with the angular positions at which the analysis begins.
Worked example: a 26/47 pair
The procedure was applied to a spiral bevel pair with a left-hand pinion and a right-hand gear, 26 and 47 teeth, a 90° shaft angle and a 6.5 mm transverse module at the heel. Design data are collected in Table 1 and the machine-tool settings for the pinion concave side and the gear convex side in Table 2.
| Parameter | Pinion | Gear |
|---|---|---|
| Shaft angle (°) | 90 | |
| Transverse module at heel (mm) | 6.5 | |
| Number of teeth | 26 | 47 |
| Face width (mm) | 52.4 | |
| Mid-cone distance (mm) | 148.364 6 | |
| Pitch cone angle (°) | 28.951 1 | 61.048 9 |
| Root cone angle (°) | 27.314 4 | 58.259 2 |
| Face cone angle (°) | 31.740 8 | 62.685 6 |
| Addendum (mm) | 7.285 | 3.765 |
| Dedendum (mm) | 4.987 | 8.507 |
| Parameter | Pinion, concave side | Gear, convex side |
|---|---|---|
| Cutter tip diameter (mm) | 293.628 9 | 304.800 0 |
| Cutter point width (mm) | — | 3.048 |
| Pressure angle (°) | 18 | 20 |
| Axial wheel position (mm) | −3.009 0 | — |
| Sliding base (mm) | 1.381 6 | −0.001 5 |
| Vertical wheel position (mm) | −4.158 4 | 0 |
| Cutting roll ratio | 1.974 6 | 1.141 5 |
| Radial cutter position (mm) | 130.781 5 | 138.923 8 |
| Angular cutter position (°) | −64.227 8 | 63.976 4 |
For each of the three error components the analysis was run with the error applied in both signs, and the initial point, the contact trace and the transmission error curve were obtained in every case. The grid search converged throughout: the refined solution reproduced the mesh condition to solver tolerance from grid estimates that were themselves several orders coarser, which indicates that the discretized flanks supply a starting vector inside the convergence region even when misalignment displaces the true contact point appreciably from its nominal location.
How each error component moves the contact pattern
The computed results show a consistent directional response. With negative vertical offset, negative horizontal offset or negative shaft-angle error, the pinion contact trace migrates toward the tooth tip while the gear trace moves toward the root; with positive values of the same errors the pattern reverses, the pinion trace moving toward the root and the gear trace toward the tip. The sign convention is therefore sufficient to predict the direction in which the loaded pattern will move when a gearbox is shimmed or when a housing bore is machined off position.
The three components are not equivalent in magnitude of effect. For this pair the vertical offset produced the largest displacement of the contact trace, followed by the horizontal offset, while the shaft-angle error acted more weakly on the trace. The ranking is inverted for transmission error: the horizontal offset and the shaft-angle error perturb the angular transmission ratio more strongly than the vertical offset does. A designer concerned with pattern placement and a designer concerned with mesh-induced velocity fluctuation should therefore not treat misalignment as a single lumped disturbance — the two concerns respond to different error components.
Scope and practical considerations
The method requires only the design parameters and the machine-tool settings of the two members, together with the error components to be studied; no measured flank topography is needed, which distinguishes it from approaches built around form-error data. Its accuracy before refinement is set by the grid density, and after refinement by the quality of the grid estimate and by the choice of the fixed parameter β2; the neighbourhood window used in the search must be wide enough to bracket the displacement caused by the largest error to be analysed. Because the contact condition is solved point by point, the result describes a single contact position for an unloaded, geometrically conjugate pair, and the trace is built up from successive solutions rather than from a loaded contact model.
Within those limits, the procedure turns the initial point from a trial-and-error input into a computed quantity. That matters in practice: once the starting point is available for a given combination of vertical, horizontal and shaft-angle errors, the contact trace and transmission error curve follow directly, and the sensitivity of a gear set to misalignment can be mapped over the tolerance band before any hardware is cut.