Spiral bevel gears transmit rotation between intersecting shafts through curved teeth that enter mesh gradually along the face width. Their load capacity, noise, and sensitivity to assembly error depend less on the nominal tooth outline than on the controlled mismatch between the two flanks. A flank that is too conjugate tends to concentrate contact; a flank with poorly chosen mismatch develops edge contact, a skewed contact path, or an asymmetric transmission error. Modification design is therefore a calibration problem: the tooth surface must be moved away from the conjugate state by a specified amount, and the machine settings must then be corrected so that the physical flank reproduces that target.
Ease-off methods describe the gap between a real flank and a fully conjugate reference, and five control terms are commonly used to shape that gap: spiral angle, pressure angle, lengthwise curvature, profile curvature, and twist. The weak point in many workflows is coefficient selection. If the coefficients are chosen by trial, the designer can see the resulting contact pattern but cannot directly prescribe a local mismatch. The approach described here closes that loop by expressing the five modification quantities as functions of position on the flank, solving their coefficients from prescribed mismatch values at control points, and then correcting the machine settings through a normal-deviation sensitivity model.
Conjugate Reference Flank as the Zero-Mismatch Baseline
The pinion and gear flanks generated by a machine-tool setting can be written as vector functions of two surface coordinates, the cradle rotation, and the machine-setting vector:
$$
\mathbf r_p=\mathbf f\left(u_p,\theta_p,\varphi_p,\boldsymbol{\psi}_{p,l}\right),\qquad
\mathbf r_g=\mathbf f\left(u_g,\theta_g,\varphi_g,\boldsymbol{\psi}_{g,l}\right),
$$
where \(u\) and \(\theta\) are tooth-surface parameters, \(\varphi\) is the cradle rotation, \(\boldsymbol{\psi}_{l}\) is the set of machine-tool settings, and the subscripts \(p\) and \(g\) denote pinion and gear. For a gear pair meshing under fully conjugate conditions, the workpiece rotations satisfy
$$
\frac{\phi_2}{\phi_1}=\frac{z_1}{z_2}=\frac{\omega_2}{\omega_1},
$$
with \(z\) as the tooth number and \(\omega\) as the angular velocity. Applying the corresponding coordinate transformations to the gear flank gives the position and normal of the fully conjugate reference surface:
$$
\mathbf r_e=\mathbf M_{1h}(\phi_1)\mathbf M_{hd}\mathbf M_{d2}(\phi_2)\mathbf r_g(u_2,\varphi_2),\qquad
\mathbf n_e=\mathbf L_{1h}(\phi_1)\mathbf L_{hd}\mathbf L_{d2}(\phi_2)\mathbf n_g(u_2,\varphi_2).
$$
Here, \(\mathbf M\) denotes a homogeneous coordinate transformation and \(\mathbf L\) is its rotational part. The reference surface has zero mismatch by construction. It is not a practical tooth surface, but it provides a clean datum: every modification term can be defined as a normal displacement from this conjugate baseline.
Five-Parameter Second-Order Mismatch Surface
Let \(x\) be the distance from the tooth-surface midpoint along the pitch line, and let \(y\) be the distance from the midpoint along the tooth height at mid-face. The five modification quantities are defined as linear or parabolic variations about these axes:
$$
\begin{aligned}
\Delta\delta_1 &= a_1 x, &\Delta\delta_2 &= a_2 y,\\
\Delta\delta_3 &= a_3 x^2, &\Delta\delta_4 &= a_4 y^2,\\
\Delta\delta_5 &= a_5 xy.
\end{aligned}
$$
Each coefficient follows directly from a prescribed mismatch at a known control point:
$$
a_1=\frac{\Delta\delta_1}{x},\quad
a_2=\frac{\Delta\delta_2}{y},\quad
a_3=\frac{\Delta\delta_3}{x^2},\quad
a_4=\frac{\Delta\delta_4}{y^2},\quad
a_5=\frac{\Delta\delta_5}{xy}.
$$
The total second-order mismatch is the sum of the five terms, and the target pinion flank is obtained by displacing the conjugate reference along its normal:
$$
\Sigma\delta=\sum_{i=1}^{5}\Delta\delta_i,\qquad
\mathbf r_t=\mathbf r_e+(\Sigma\delta)\mathbf n_e.
$$
| Control term | Expression | Effect on the flank |
|---|---|---|
| Spiral angle | \(a_1x\) | Linear change along the tooth trace |
| Pressure angle | \(a_2y\) | Linear change along the tooth height |
| Lengthwise curvature | \(a_3x^2\) | Parabolic crowning along the tooth trace |
| Profile curvature | \(a_4y^2\) | Parabolic crowning along the profile |
| Twist | \(a_5xy\) | Coupling between lengthwise and heightwise slopes |
This formulation replaces blind coefficient selection with a local calibration step. If a contact analysis calls for a 0.02 mm normal mismatch at a point with coordinates \((x,y)\), the corresponding coefficient amplitude is fixed by the equations above. The five terms are not strictly orthogonal in their effect on the contact path and transmission error, so the summed surface is still checked by tooth contact analysis (TCA). If the combined surface misses the contact target, the control values are adjusted and the coefficients are recalculated.
Machine-Setting Correction from Normal Deviation
The target flank and the actual pinion flank generated by the current machine settings differ by
$$
\mathbf r=\mathbf r_t-\mathbf r_p(\boldsymbol{\psi}_l).
$$
Differentiating with respect to the surface coordinates and machine settings gives
$$
\delta\mathbf r=-\left(
\frac{\partial\mathbf r_p}{\partial u_1}\delta u_1+
\frac{\partial\mathbf r_p}{\partial\theta_1}\delta\theta_1+
\sum_{i=1}^{l}\frac{\partial\mathbf r_p}{\partial\psi_i}\delta\psi_i
\right).
$$
Taking the dot product with the unit normal \(\mathbf n_p\) eliminates the two surface-coordinate terms, because \(\mathbf n_p\) is perpendicular to both \(\partial\mathbf r_p/\partial u_1\) and \(\partial\mathbf r_p/\partial\theta_1\):
$$
\delta\mathbf r\cdot\mathbf n_p=
-\sum_{i=1}^{l}\left(\frac{\partial\mathbf r_p}{\partial\psi_i}\cdot\mathbf n_p\right)\delta\psi_i.
$$
The left side is the normal tooth-surface deviation. The derivative on the right is the sensitivity of that deviation to the \(i\)-th machine parameter. In a discretized flank grid, the sensitivity is evaluated by a finite difference,
$$
\eta_i=\frac{\Delta\varepsilon}{\Delta\psi_i},
$$
where \(\Delta\varepsilon\) is the normal deviation at a grid point produced by a small perturbation \(\Delta\psi_i\). For \(k\) grid points and \(l\) machine parameters, the correction equations form the linear system
$$
\{\Delta\varepsilon_m\}=[\eta_{nm}]\{\Delta\psi_n\},\qquad m=1,\ldots,k,\quad n=1,\ldots,l.
$$
Because the number of machine parameters is much smaller than the number of grid points, the system is overdetermined and is solved in the least-squares sense:
$$
\{\Delta\psi_n\}=\left([\eta_{nm}]^{\mathsf T}[\eta_{nm}]\right)^{-1}[\eta_{nm}]^{\mathsf T}\{\Delta\varepsilon_m\}.
$$
The result is a set of machine-setting corrections that best reproduces the target normal deviation over the whole grid, rather than at a single point.
Design Workflow and Feedback
The complete design loop can be summarized as follows:
- Compute the original pinion and gear flanks from the basic gear data and the initial machine settings.
- Construct the fully conjugate reference flank for the gear pair.
- Specify the desired mismatch at control points; solve the five coefficients and build the second-order target flank.
- Sample the target and actual flanks on a grid, evaluate normal deviations, assemble the sensitivity matrix, and solve the machine-setting corrections by least squares.
- Recompute the flank, run TCA and Ease-off analysis, and check both the residual normal deviation and the contact metrics.
Two feedback decisions follow from this sequence. The first is a deviation loop: if the corrected flank is not close enough to the target, the machine-setting solution is updated. The second is a performance loop: if the contact path, contact width, or transmission error is unacceptable, the prescribed mismatch quantities are changed and a new target surface is built. The method therefore separates “what mismatch is wanted” from “which machine settings produce it.”
Validation Case: An 18/64-Tooth Spiral Bevel Gear Pair
The procedure was evaluated on a spiral bevel gear pair with 18 pinion teeth and 64 gear teeth. The basic geometry and the initial machine settings are listed in Tables 1 and 2.
| Parameter | Pinion | Gear |
|---|---|---|
| Shaft angle | 90° | |
| Spiral angle | 35° | |
| Pressure angle | 20° | |
| Module | 7.16 mm | |
| Face width | 45 mm | |
| Number of teeth | 18 | 64 |
| Face cone angle | 31.65° | 64.63° |
| Pitch cone angle | 27.217° | 62.783° |
| Root cone angle | 25.36° | 58.35° |
| Root angle | 1.85° | 4.43° |
| Working tooth height | 12.33 mm | |
| Full tooth height | 13.81 mm | |
| Addendum | 8.78 mm | 3.55 mm |
| Dedendum | 5.03 mm | 10.26 mm |
| Outer diameter at heel | 144.49 mm | 253.85 mm |
| Crown distance | 121.29 mm | 61.28 mm |
| Setting | Gear convex | Pinion concave |
|---|---|---|
| Cutter tip radius | 112.70 mm | 115.44 mm |
| Profile angle | 22° | 18° |
| Cutter tilt | 0° | 3.81° |
| Cutter swivel | 203.29° | 147.61° |
| Radial cutter position | 110.0537 mm | 105.5101 mm |
| Angular cutter position | 58.29° | 58.68° |
| Vertical wheel position | 0 mm | −4.1657 mm |
| Bed position | 0 mm | −3.9308 mm |
| Blank mounting angle | 58.62° | 29.5° |
| Horizontal wheel position | 0 mm | −2.3984 mm |
| Roll ratio | 1.121524 | 2.110971 |
With the initial settings, TCA placed the contact area near mid-face. The contact path made 44.6° with the perpendicular to the root cone, and the transmission error was 9.449″. The transmission-error curve was asymmetric, which indicates a higher risk of edge contact and mesh noise under load. Those metrics defined the starting point for the modification.
Single-Parameter Checks of the Mismatch Mapping
The mapping from prescribed mismatch to coefficient was first checked one parameter at a time on the pinion concave flank. For the angular terms, a change of 1′ was prescribed at the control point; for the curvature and twist terms, the prescribed normal mismatch was 0.01 mm. The resulting normal deviations at the corresponding points are listed in Table 3.
| Applied term | Prescribed change | Normal mismatch at control point |
|---|---|---|
| Spiral angle | 1′ | 0.0052 mm |
| Pressure angle | 1′ | 0.0016 mm |
| Twist | 0.01 mm | 0.0105 mm |
| Lengthwise curvature | 0.01 mm | 0.0100 mm |
| Profile curvature | 0.01 mm | 0.0118 mm |
For the two angular terms, the normal mismatch is not numerically equal to the angular input because the comparison passes through the local surface geometry; the test confirms that a 1′ change produces a localized normal displacement of the expected order. The curvature and twist terms recover the prescribed 0.01 mm within 0.0005–0.0018 mm. The profile-curvature case shows the largest discrepancy, about 18% above the target, because the mid-profile direction of the tapered tooth is not exactly aligned with the \(y\)-axis of the modification frame. That is a coordinate approximation, not a failure of the polynomial form.
The final combination for the pinion concave flank used a −3.5′ spiral-angle term, a 3′ pressure-angle term, 0.04 mm lengthwise curvature, 0.025 mm profile curvature, and 0.025 mm twist. The corresponding coefficients are listed in Table 4.
| Term | Modification | Coefficient |
|---|---|---|
| Spiral angle | −3.5′ | −0.001 |
| Pressure angle | 3′ | 0.00087266 |
| Lengthwise curvature | 0.04 mm | 0.000079012 |
| Profile curvature | 0.025 mm | 0.001 |
| Twist | 0.025 mm | 0.00015793 |
The target surface was then converted into machine-setting corrections. The least-squares solution is given in Table 5. The largest linear corrections are the cutter tip radius (+1.775 mm), the horizontal wheel position (−2.3687 mm), and the radial cutter position (−0.9568 mm); the roll ratio changes by −0.01963.
| Setting | Correction |
|---|---|
| Cutter tip radius | +1.775 mm |
| Profile angle | 0° |
| Cutter tilt | −0.19° |
| Cutter swivel | −3.98° |
| Radial cutter position | −0.9568 mm |
| Angular cutter position | +0.17° |
| Vertical wheel position | +0.6264 mm |
| Bed position | +0.377 mm |
| Blank mounting angle | 0° |
| Horizontal wheel position | −2.3687 mm |
| Roll ratio | −0.01963 |
Contact and Ease-off Response After Correction
The modified Ease-off topology separates from the reference flank around the boundary, which suppresses tip and edge contact. The predicted contact zone runs from the heel top to the toe root, producing a diagonal pattern along the face. After modification, the normal mismatch increases at the toe top and heel root in the tooth-height direction, so the contact zone becomes narrower in profile and longer along the face. The four boundary points also develop a larger mismatch spread, which is consistent with the stronger contact diagonal.
The TCA results follow the same trend. The contact path angle rises from 44.6° to 64°, the contact zone lengthens along the face and narrows in the profile, and the transmission error changes from 9.449″ to 13.47″. The larger transmission error is a deliberate tradeoff: the modified design does not minimize unloaded transmission error, but it makes the curve more symmetric and relieves the boundary. The symmetric error reduces impact at mesh entry and exit, while the diagonal pattern distributes contact over a longer face path.
| Metric | Initial | Modified |
|---|---|---|
| Contact path angle to root-cone perpendicular | 44.6° | 64° |
| Transmission error | 9.449″ | 13.47″ |
| Transmission-error symmetry | Asymmetric | Improved |
| Contact length along face | Shorter | Longer |
| Contact width in profile | Wider | Narrower |
| Edge-contact tendency | Higher | Reduced by boundary relief |
Cutting and Rolling Test
The corrected settings were transferred to a Gleason 600 milling machine. The cut parts were mounted at the standard mounting distance and checked by a rolling test. The rolled contact pattern before modification was concentrated near mid-face; after modification, the pattern showed the predicted diagonal from heel top to toe root and remained inside the flank boundary. The position and orientation of the rolled pattern agreed with the TCA and Ease-off predictions.
That agreement is the practical test of the method. A numerical flank correction is only useful if the machine can reproduce it. In this case, the chain from prescribed mismatch to coefficients, from coefficients to target normal deviation, and from normal deviation to machine-setting corrections produced a physical contact pattern consistent with the calculated one. The rolling test does not replace the TCA metrics, but it confirms that the corrected settings act on the real tooth surface in the intended direction and magnitude.
Scope, Limitations, and Transferability
The validation was limited to single-face milling of the pinion concave flank. In duplex cutting, one set of machine settings generates both flanks, so a correction intended for one side also changes the other. The coefficient-to-machine-setting mapping then becomes a coupled two-flank problem, and the target must be posed as a joint objective rather than a single-flank deviation. That is a different optimization problem from the one demonstrated here.
The broader value of the approach lies in converting coefficient selection into a calculable calibration step. A designer can specify a local mismatch, obtain the second-order surface coefficients directly, and solve the machine corrections from a normal-deviation sensitivity matrix. The same structure can be re-derived for other gear flank parameterizations, but the control-point coordinates, polynomial basis, and sensitivity matrix must be rebuilt for each geometry. For the 18/64-tooth spiral bevel gear pair, the result was a controlled diagonal contact pattern, a more symmetric transmission error, and a cutting test that matched the predicted flank behavior.