Spiral bevel gears carry power between intersecting shafts in aero-engines, vehicle axles, marine drives and heavy industrial transmissions. Their curved teeth and locally synthesized flanks let the designer dictate where the teeth touch, but a contact mark observed on a lightly loaded rolling check describes only a single point of the operating envelope. Under torque the teeth bend, the contact ellipse migrates, and a mark that sits comfortably inside the flank at no load can drift toward the heel or the tip until it runs over the edge. Edge contact concentrates stress, accelerates pitting and, in aircraft applications, can end the life of the pair long before the nominal rating is reached.
The classical design targets for a loaded contact pattern are size, orientation and position, usually assessed together with transmission error through tooth contact analysis. Optimization strategies have diversified considerably: surrogate-assisted simulated loaded tooth contact analysis to keep computation tractable; robust optimization that treats misalignment as an uncertain input; curvature-based global control of both the contact mark and the transmission error; multi-objective correction of machine settings driven by loaded contact performance; ease-off surfaces combined with kriging models; adaptive data-driven schemes for non-orthogonal duplex helical face milling; and simplified equivalent-cylinder or Hertzian formulations that avoid a full loaded contact solve. What these approaches share is a view of the contact mark as a single shape to be matched, and a treatment of the tooth boundary mainly as a constraint to be respected. The quantity that governs durability under high torque, however, is not the position of the mark at one instant but how the mark grows through the mesh cycle relative to the four boundaries of the active flank.
Reading the expansion from offset surfaces
Load sharing between two flanks in contact can be represented geometrically by offsetting each surface along its common normal by a small distance and intersecting the two offset bodies at each meshing instant. Writing the surface points and normals as pN, gN and n, the offset surfaces are
$$\mathbf{p}_C^{\,i}=\mathbf{p}_N^{\,i}+\varepsilon\,\mathbf{n}^{\,i},\qquad \mathbf{g}_C^{\,i}=\mathbf{g}_N^{\,i}+\varepsilon\,\mathbf{n}^{\,i}$$
where the shift ε stands for the local elastic approach. The intersection of the two offset surfaces is an ellipse at that instant; superposing the ellipses obtained over the whole meshing cycle reproduces the loaded contact pattern that a roll check would reveal in red lead.
To extract the ellipse boundary numerically, rays are cast from the instantaneous contact point across the transverse section,
$$\mathbf{A}_j^{n}=\left[\cos\frac{2\pi\left(j-1\right)}{n},\ \sin\frac{2\pi\left(j-1\right)}{n}\right],\qquad j=1,\ldots,n$$
and each ray is parameterized in the radial–axial plane by a length k,
$$\begin{bmatrix}R\\ L\end{bmatrix}_{j}=\begin{bmatrix}R_p\\ L_p\end{bmatrix}+k\begin{bmatrix}A_{j,R}\\ A_{j,L}\end{bmatrix}$$
The length is increased until the two offset points coincide, which locates one boundary point Cj of the ellipse. Repeating this for every candidate instant and summing the ellipses yields the loaded contact pattern of the pair.
Four parameters then describe the expansion of that pattern. Two of them measure the height of the mark above the tooth tip,
$$\delta_{ag}=\min_{j=1,\ldots,n}\frac{\left|k_{ag}L_{j,g}+R_{j,g}+b_{ag}\right|}{\sqrt{k_{ag}^{2}+1}},\qquad \delta_{ap}=\min_{j=1,\ldots,n}\frac{\left|k_{ap}L_{j,p}+R_{j,p}+b_{ap}\right|}{\sqrt{k_{ap}^{2}+1}}$$
with kag, bag and kap, bap defining the addendum lines of the gear and pinion in the same axial section. The other two measure the remaining material between the mark and the heel and toe, obtained by interpolating n+1 points between the nearest contact point and its projection on the boundary line, then accumulating the segments along the flank surface:
$$s_h=\sum_{i=0}^{n}\left|\mathbf{C}_{h,i+1}-\mathbf{C}_{h,i}\right|,\qquad s_t=\sum_{j=0}^{m}\left|\mathbf{C}_{t,j+1}-\mathbf{C}_{t,j}\right|$$
Arc length, rather than a straight radial span, is required because the mid-point of the blank does not coincide with the mid-point of the flank on a spiral bevel gear; the curvature of the tooth trace would otherwise bias the measured margins.
Two design objectives follow from these definitions, both resting on the assumption that the pattern grows at comparable rates in all directions. First, the distances from the marks to the tip on the two members should be equal, δag = δap, so that the loaded zone stays near the mid-height of the tooth and neither the pinion root nor the gear tip is overloaded. Second, the clearances to the heel and the toe of the gear flank should be equal, sh = st, so that the expansion has the same room on both ends of the tooth trace. Because the two members carry conjugate marks, the correction is applied to one member while both are evaluated.
Deflecting the flank to move the contact
Once the target is expressed as four numbers, the question becomes how to change the flank so that the numbers converge. The lever used here is a controlled normal deflection superposed on the existing flank, with the machine settings re-solved afterwards.
Let the two flank surfaces be given in a common reference by R, L coordinates and normals N. The normal deviation between the mating surfaces is
$$d\left(R,L\right)=\left[\mathbf{R}_2\left(R,L\right)-\mathbf{R}_1\left(R,L\right)\right]\cdot\mathbf{N}_1\left(R,L\right)$$
which vanishes at the tangent contact point (R0, L0) and must remain non-negative elsewhere to avoid interference. Expanding about that point,
$$d\left(R_0+h,L_0+k\right)=\frac{1}{2}\left(Eh^{2}+2Fhk+Gk^{2}\right)+O\left(h^{3},k^{3}\right)$$
where E, F, G are the second derivatives of d evaluated at the contact point. The quadratic term alone is a paraboloid whose minimum sits at the origin, which is why an unmodified flank keeps its mark where the design put it. Add a small inclined plane to the deviation,
$$\delta_g\left(R,L\right)=Ah+Bk+D$$
and the modified deviation becomes
$$d_i\left(h,k\right)=\frac{1}{2}\left(Eh^{2}+2Fhk+Gk^{2}\right)+Ah+Bk+D$$
whose stationary point follows from the two linear conditions Eh + Fk + A = 0 and Fh + Gk + B = 0:
$$h^{*}=\frac{BF-AG}{EG-F^{2}},\qquad k^{*}=\frac{AF-BE}{EG-F^{2}}$$
For the minimum to remain at h = k = 0 the coefficients would have to satisfy A = B = 0, which contradicts a deflection plane with A and B not simultaneously zero and D ≠ 0. Any admissible deflection surface therefore displaces the contact point, and the direction and magnitude of that displacement are set by the coefficients of the plane — the property that makes boundary-driven regulation possible.
Constructing the deflection surface
Two elementary deflection fields are used, one across the tooth height and one across the face width. In the height direction, the same adjustment δh is prescribed at the heel addendum point (Rae, Lae) and at the toe addendum point (Rai, Lai), while the flank mid-point (Rm, Lm) is held fixed. The three conditions determine a plane,
$$\begin{bmatrix}a_h\\ b_h\\ c_h\end{bmatrix}=\begin{bmatrix}R_{ae}&L_{ae}&1\\ R_{ai}&L_{ai}&1\\ R_m&L_m&1\end{bmatrix}^{-1}\begin{bmatrix}\delta_h\\ \delta_h\\ 0\end{bmatrix}$$
so that δh(R, L) = ahR + bhL + ch. In the face-width direction the adjustment δl is applied at the heel addendum and heel dedendum points and vanishes at the mid-point, giving the analogous system for al, bl, cl.
The two fields are superposed and applied along the surface normal, so that every point of the original flank moves to
$$\mathbf{p}^{C}=\mathbf{p}+\left(a_hR+b_hL+c_h+a_lR+b_lL+c_l\right)\mathbf{n}$$
The magnitudes are deliberately small — micrometre-scale corrections rather than a redesign of the tooth form — because the regulation is meant to shift the expansion pattern without disturbing the meshing kinematics that the transmission error depends on.
Recovering machine settings with Levenberg–Marquardt
A deflected flank is only useful if a grinding machine can produce it. The blank surface is therefore discretized into an (n+1)×(m+1) grid, with normalized interpolation matrices X of size (n+1)×2 and Y of size 2×(m+1) mapping the four blank boundary curves into the radial and axial coordinate fields,
$$\mathbf{R}=\mathbf{X}\,\mathbf{R}_0\,\mathbf{Y}^{\mathrm{T}},\qquad \mathbf{L}=\mathbf{X}\,\mathbf{L}_0\,\mathbf{Y}^{\mathrm{T}}$$
The target flank supplies a vector of normal deviations f at the grid nodes. The machine setting vector Xk = [α, β, r0, q, i, δm, Em, Xb, Xp, γ, σ] collects the cradle angle, eccentricity, cutter radius, tilt, swivel, vertical and horizontal offsets, sliding base, angular cutter position, ratio of roll and the second-order curvature coefficients. The correction solves
$$\min\ \left\|f\left(X_k+\Delta\right)\right\|,\qquad F\left(x\right)=0,\quad i=1,\ldots,n\times m$$
with the Levenberg–Marquardt update
$$X_{k+1}=X_k-\left(J_k^{\mathrm{T}}J_k+\lambda I\right)^{-1}J_k^{\mathrm{T}}F\left(X_k\right)$$
where Jk is the Jacobian of the residual vector with entries ∂Fi/∂xj and λ is the damping factor that keeps the step well behaved when the system is ill-conditioned. The converged set is written into the numerical control of the machine tool, closing the loop between evaluation, correction and manufacturing.
Case study: an aero-engine face-milled gear set
The procedure was applied to a face-milled pair intended for aero-engine duty, with a 12-tooth right-hand pinion driving a 49-tooth left-hand gear. Blank data are listed in Table 1 and the grinding wheel and machine settings used for the reference flank in Table 2.
Table 1. Blank design parameters of the aero-engine spiral bevel gear pair.
| Parameter | Pinion | Gear |
|---|---|---|
| Number of teeth | 12 | 49 |
| Hand | Right | Left |
| Face width (mm) | 41.05 | 37.00 |
| Outer cone distance (mm) | 109.12 | 114.20 |
| Pitch cone angle (°) | 17.5833 | 71.8167 |
| Face cone angle (°) | 21.3667 | 72.6167 |
| Root cone angle (°) | 16.8167 | 67.9333 |
| Addendum (mm) | 6.98 | 1.44 |
| Dedendum (mm) | 2.72 | 8.13 |
Table 2. Wheel parameters and machine settings of the reference flank.
| Parameter | Pinion concave | Gear convex |
|---|---|---|
| Cutter diameter (mm) | 144.272 | 150.100 |
| Pressure angle (°) | 16.79 | 22.00 |
| Fillet radius (mm) | 1.0133 | 1.0133 |
| Radial cutter position (mm) | 86.9701 | 87.6797 |
| Cutter tilt (°) | 0 | 0 |
| Cutter swivel (°) | 18.79 | 201.02 |
| Vertical offset (mm) | 21.7923 | −3.1061 |
| Installation angle (°) | 16.81 | 67.93 |
| Horizontal offset (mm) | −1.9346 | 4.3773 |
| Sliding base (mm) | −0.59334 | −1.9455 |
| Angular cutter position (°) | 64.83 | 47.73 |
| Ratio of roll | 4.2235 | 1.0116 |
| Second-order coefficient 2C | 0.06305 | 0 |
| Second-order coefficient 6D | −0.106 | 0 |
Baseline tooth contact analysis placed the pinion mark near the tip and the gear mark near the root, with evaluation parameters δag = 0.5 mm, δap = 1.6 mm, sh = 9.3 mm and st = 11.5 mm; the peak-to-peak transmission error was 66 μrad. The mismatch in both pairs of numbers is the signature of an off-centre pattern that has little room left at the toe.
Sensitivity analysis identified the radial cutter position and the cutter pressure angle as the two settings with the strongest and most selective influence on the four evaluation parameters, and these were chosen as the regulation variables to keep the inverse problem small and well conditioned. A height-direction adjustment δh = −0.0015 mm and a face-width adjustment δl = 0.0025 mm were prescribed, the deflection surface was constructed, and the machine settings of the pinion concave flank were re-solved by the Levenberg–Marquardt procedure.
Table 3. Contact pattern expansion parameters before and after regulation.
| Parameter | Before | After |
|---|---|---|
| δag (mm) | 0.5 | 1.1 |
| δap (mm) | 1.6 | 1.1 |
| sh (mm) | 9.3 | 10.7 |
| st (mm) | 11.5 | 10.7 |
| Peak transmission error (μrad) | 66 | 66 |
Both symmetry targets were met to within the resolution of the evaluation grid: the marks on the two members now sit at the same height, and the remaining material at the heel and toe of the gear flank is balanced. The transmission error stayed at 66 μrad, confirming that the micrometre-scale deflection corrected the pattern without degrading the meshing kinematics.
Loaded contact pressure before and after regulation
Finite element models of the flank before and after regulation were built from the respective machine settings and solved in Abaqus to obtain the loaded contact pressure distribution.
Table 4. Peak contact pressure before and after regulation.
| Member | Before (MPa) | After (MPa) | Location after |
|---|---|---|---|
| Pinion | 1978.3 | 1351.1 | Central flank region |
| Gear | 1773.9 | 1500.5 | Central flank region |
Before regulation the pinion pressure peaked at 1978.3 MPa close to the root, the gear pressure reached 1773.9 MPa near the tip, and the pressure did not fall to zero at the tooth boundaries — the pattern was crowding the edges of the active surface. Pressure traces taken along a line in the face-width direction showed a secondary rise near the heel instead of a monotonic decay, and traces in the tooth-height direction showed concentrations at the pinion root and the gear tip, consistent with the computed fields.
After regulation the peak pressures dropped to 1351.1 MPa on the pinion and 1500.5 MPa on the gear, both located in the central region of the flank, while the pressure at the root and tip boundaries fell to 0 MPa. The maximum contact pressure was reduced by 31.7 %, and the edge concentrations disappeared. The face-width trace now rises and falls smoothly from toe to heel, and the height-direction trace is nearly symmetric about the mid-height of the tooth.
Grinding and roll-check validation
The corrected settings were loaded into the numerical control of a Phoenix 600G grinder and the flanks were ground, after which the pair was roll-checked at two load levels to compare the marked contact areas with the predicted ones.
At 10 Nm, close to the unloaded condition, the reference pair showed a mark centred along the face width but displaced toward the tip in the height direction. The regulated pair showed a mark shifted slightly toward the toe but noticeably more uniform in the height direction, matching the small δh and δl corrections that were prescribed.
At 1500 Nm the difference became decisive. The reference mark migrated to the heel and ran past the boundary of the flank, leaving roughly one quarter of the tooth surface out of contact. The regulated mark stayed in the middle of the flank, with essentially equal margins to the heel and the toe, and grew until it covered almost the entire tooth surface. The measured areas at both load levels were larger after regulation than before, in agreement with the finite element predictions.
What the regulation buys, and what it assumes
Treating the loaded contact pattern as four boundary distances rather than a single shape converts an inspection judgement into a solvable inverse problem. Because the objective is expressed in the same coordinates in which the flank is manufactured, a micrometre-scale normal deflection can be translated directly into corrected machine settings, and the resulting pattern is centred by construction rather than by trial and error on the roll-check stand.
The practical payoff is largest where the design margin is thinnest. Under high-speed and heavy-load operation the regulated flank develops a contact area that is as large as the geometry allows — in the 1500 Nm test, essentially the whole tooth surface — and that area is uniformly loaded, so the peak contact stress falls even though the transmitted torque is unchanged. A lower peak stress at equal torque permits a smaller gear for a given fatigue life, which is the route to a lighter transmission without sacrificing durability. Conversely, for a gear of fixed size, the reduced stress translates into longer life or a higher rating.
The method rests on two assumptions that define its range of validity. The first is that the pattern expands at comparable rates toward all four boundaries, which justifies the symmetric targets δag = δap and sh = st; strongly anisotropic expansion, as might occur with a markedly asymmetric misalignment, would call for weighted targets. The second is that a small number of sensitive machine settings can absorb the required correction, which held for this pair with the radial cutter position and the cutter pressure angle but should be re-examined for each new tooth form. Within those limits, the approach gives designers a quantitative handle on where the contact pattern will be at high load, before any hardware is cut.