Spiral bevel gears transmit torque between intersecting shafts and are widely used in automotive drivetrains, construction machinery, aerospace actuation, and machine tools. Their running quality is governed less by the nominal tooth form than by the controlled mismatch between the two flanks: the contact pattern, the contact path, and the transmission error. A perfectly conjugate pair has zero mismatch but is highly sensitive to manufacturing and assembly errors. A small, deliberately prescribed mismatch can shift and shape the contact pattern, suppress edge contact, and make the mesh more tolerant of alignment variation.
Tooth-surface modification can be introduced through cutter profile, cutter path, or machine settings. Ease-off topology methods describe the normal separation between a conjugate reference surface and a modified surface and are now a common way to design contact behavior. In five-coefficient topology modification, the helix angle, pressure angle, lengthwise curvature, profile curvature, and twist are varied to shape the ease-off surface. The practical difficulty is that these coefficients are usually selected by trial: the designer can see the resulting contact pattern, but the link between a physical mismatch at a specified point and the coefficient that produces it is not explicit. The design loop therefore depends heavily on experience and repeated tooth contact analysis.
The quantitative mismatch approach closes that loop. A desired normal mismatch is specified at defined check points, converted directly into five second-order coefficients, assembled into a target surface, and then converted into machine-setting corrections through a sensitivity matrix. Tooth contact analysis (TCA) provides the contact check, and a cutting test provides the physical verification.
Conjugate Baseline and Quantified Mismatch Surface
The pinion and gear tooth surfaces are represented as functions of surface parameters, cradle rotation, and machine-setting vectors:
$$ \mathbf r_p=\mathbf f_p(u_p,\theta_p,\varphi_p,\boldsymbol\psi_p),\qquad
\mathbf r_g=\mathbf f_g(u_g,\theta_g,\varphi_g,\boldsymbol\psi_g) $$
where u and θ are surface parameters, φ is the cradle rotation, and ψ is the machine-setting vector. For a gear pair meshing under the conjugate condition, the workpiece rotations satisfy
$$ \frac{\phi_2}{\phi_1}=\frac{z_1}{z_2}=\frac{\omega_2}{\omega_1} $$
where z is the tooth number and ω is the angular velocity; subscripts 1 and 2 denote the pinion and gear. The large gear surface is transformed into the pinion coordinate frame to obtain the conjugate baseline:
$$ \mathbf r_e=\mathbf M_{1h}(\phi_1)\mathbf M_{hd}\mathbf M_{d2}(\phi_2)\mathbf r_g(u_2,\varphi_2),\qquad
\mathbf n_e=\mathbf L_{1h}(\phi_1)\mathbf L_{hd}\mathbf L_{d2}(\phi_2)\mathbf n_g(u_2,\varphi_2) $$
Here M is a coordinate transformation matrix and L is its rotation part. The resulting surface re and normal ne form the reference on which the modification is defined.
Five mismatch quantities are specified relative to the pitch-line midpoint. The coordinate x runs along the pitch line from the midpoint toward the heel; y runs along the tooth height from mid-height toward the addendum. The quantities are:
- helix-angle mismatch, a linear change along the tooth line from midpoint to heel;
- pressure-angle mismatch, a linear change along the tooth height at mid-face;
- lengthwise curvature mismatch, a parabolic change along the tooth line;
- profile curvature mismatch, a parabolic change along the tooth height;
- twist, a cross-term change in slope, represented by a linear variation from the midpoint toward the heel addendum.
The second-order relation between the mismatch amounts and the surface coordinates is
$$ \Delta\delta_1=a_1x,\qquad
\Delta\delta_2=a_2y,\qquad
\Delta\delta_3=a_3x^2,\qquad
\Delta\delta_4=a_4y^2,\qquad
\Delta\delta_5=a_5xy $$
so the five coefficients follow directly from the specified mismatch at a point with known x and y:
$$ a_1=\frac{\Delta\delta_1}{x},\qquad
a_2=\frac{\Delta\delta_2}{y},\qquad
a_3=\frac{\Delta\delta_3}{x^2},\qquad
a_4=\frac{\Delta\delta_4}{y^2},\qquad
a_5=\frac{\Delta\delta_5}{xy} $$
The combined mismatch surface is the superposition of the five terms:
$$ \Sigma\delta=a_1x+a_2y+a_3x^2+a_4y^2+a_5xy $$
and the target surface is obtained by offsetting the conjugate baseline along its normal:
$$ \mathbf r_t=\mathbf r_e+(\Sigma\delta)\mathbf n_e $$
This is an ease-off surface in geometric terms: at each grid point, Σδ is the normal separation from the conjugate reference. Because the five terms are superposed, a large combination can deviate from an ideal second-order surface at points away from the check locations. The coefficient set is therefore a starting design, not a guarantee; contact analysis and coefficient adjustment remain part of the procedure.
Machine-Setting Correction from Surface Deviation
The target surface must be reproduced by the generating machine. Let ψ be the vector of machine settings to be corrected. The deviation between the target and the original pinion surface is
$$ \mathbf r=\mathbf r_t-\mathbf r_p(\boldsymbol\psi) $$
Differentiating with respect to the machine settings and projecting onto the original surface normal gives
$$ \delta\mathbf r\cdot\mathbf n_p=
-\sum_j\left(\frac{\partial\mathbf r_p}{\partial\psi_j}\delta\psi_j\right)\cdot\mathbf n_p $$
The terms involving the surface tangents ∂rp/∂u1 and ∂rp/∂θ1 vanish because the unit normal is perpendicular to both tangents:
$$ \mathbf n_p=
\frac{(\partial\mathbf r_p/\partial u_1)\times(\partial\mathbf r_p/\partial\theta_1)}
{\left\|(\partial\mathbf r_p/\partial u_1)\times(\partial\mathbf r_p/\partial\theta_1)\right\|} $$
The left-hand side of the projected equation is the normal surface deviation. The right-hand side contains the sensitivity of that deviation to each machine setting. For a finite change, the sensitivity coefficient for setting j can be evaluated by a difference:
$$ \eta_j=\frac{\Delta\varepsilon}{\Delta\psi_j} $$
With the tooth surface divided into k grid points and q machine settings, the linearized correction model is
$$ \{\Delta\varepsilon_m\}=[\eta_{mj}]\{\Delta\psi_j\},\qquad
m=1,\ldots,k,\quad j=1,\ldots,q $$
Because the number of grid points is much larger than the number of machine settings, the system is overdetermined. It is solved in the least-squares sense:
$$ \{\Delta\psi_j\}=
\left([\eta_{mj}]^{\mathsf T}[\eta_{mj}]\right)^{-1}
[\eta_{mj}]^{\mathsf T}\{\Delta\varepsilon_m\} $$
The design procedure has two feedback levels. The inner level corrects the machine settings until the residual deviation between the generated surface and the target surface is acceptably small. The outer level evaluates the contact path, contact pattern, and transmission error by TCA. If those contact metrics do not meet the requirement, the mismatch amounts are adjusted, a new second-order surface is built, and the correction is repeated. This is how the method avoids blind coefficient selection without pretending that a single linear correction is exact.
Numerical Example and Single-Parameter Checks
The procedure was applied to a spiral bevel gear pair with a 90° shaft angle, 35° spiral angle, 20° pressure angle, 7.16 mm module, and 45 mm face width. The pinion has 18 teeth and the gear has 35 teeth. The main geometry is listed in Table 1, and the generating machine settings for the gear convex flank and pinion concave flank are listed in Table 2.
| Parameter | Pinion | Gear |
|---|---|---|
| Shaft angle (°) | 90 | |
| Spiral angle (°) | 35 | |
| Pressure angle (°) | 20 | |
| Module (mm) | 7.16 | |
| Face width (mm) | 45 | |
| Number of teeth | 18 | 35 |
| 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 (mm) | 12.33 | |
| Full tooth height (mm) | 13.81 | |
| Addendum (mm) | 8.78 | 3.55 |
| Dedendum (mm) | 5.03 | 10.26 |
| Outer diameter at heel (mm) | 144.49 | 253.85 |
| Crown distance (mm) | 121.29 | 61.28 |
| Machine setting | Gear convex flank | Pinion concave flank |
|---|---|---|
| Cutter tip radius (mm) | 112.70 | 115.44 |
| Profile angle (°) | 22 | 18 |
| Cutter tilt (°) | 0 | 3.81 |
| Cutter swivel (°) | 203.29 | 147.61 |
| Radial setting (mm) | 110.0537 | 105.5101 |
| Angular setting (°) | 58.29 | 58.68 |
| Vertical offset (mm) | 0 | -4.1657 |
| Sliding base (mm) | 0 | -3.9308 |
| Workpiece mounting angle (°) | 58.62 | 29.5 |
| Horizontal offset (mm) | 0.1198 | -2.3984 |
| Roll ratio | 1.121524 | 2.110971 |
Before modification, TCA placed the contact zone near the middle of the tooth. The contact path made an angle of 44.6° with the root-cone normal, and the transmission error was 9.449 arcsec. The transmission-error curve was asymmetric, which indicated a risk of edge contact and uneven loading. The initial design therefore needed a controlled mismatch rather than a larger conjugate contact area.
To isolate the mapping from each mismatch quantity to the surface, the pinion concave flank was modified one quantity at a time. Angle-type modifications were applied as 1 arcmin; curvature and twist modifications were applied as 0.01 mm. The normal mismatch at the corresponding check point was then computed from the surface geometry. The results are summarized in Table 3.
| Modification | Applied quantity | Computed normal mismatch at check point (mm) |
|---|---|---|
| Helix angle | 1 arcmin | 0.0052 |
| Pressure angle | 1 arcmin | 0.0016 |
| Lengthwise curvature | 0.01 mm | 0.0100 |
| Profile curvature | 0.01 mm | 0.0118 |
| Twist | 0.01 mm | 0.0105 |
The curvature and twist checks returned normal offsets close to the specified values: 0.0100 mm for lengthwise curvature and 0.0105 mm for twist. The helix and pressure-angle checks produced smaller normal offsets because the input was angular rather than a direct normal displacement. The profile-curvature result was 0.0118 mm, about 18% above the 0.01 mm nominal value. This larger discrepancy comes from the tapered tooth form: the mid-height direction of the tooth is not exactly aligned with the Y axis of the modification coordinate frame. The check confirms that the coefficient equations are usable at the reference point while showing where coordinate alignment matters.
The final combined modification used a helix-angle change of -3.5 arcmin, a pressure-angle change of 3 arcmin, a lengthwise curvature of 0.04 mm, a profile curvature of 0.025 mm, and a twist of 0.025 mm. The corresponding coefficients are listed in Table 4. The machine-setting corrections obtained from the least-squares model are listed in Table 5.
| Quantity | Modification amount | Coefficient |
|---|---|---|
| Helix angle | -3.5 arcmin | -0.001 |
| Pressure angle | 3 arcmin | 0.00087266 |
| Lengthwise curvature | 0.04 mm | 0.000079012 |
| Profile curvature | 0.025 mm | 0.001 |
| Twist | 0.025 mm | 0.00015793 |
| Machine setting | Correction |
|---|---|
| Cutter tip radius (mm) | 1.775 |
| Profile angle (°) | 0 |
| Cutter tilt (°) | -0.19 |
| Cutter swivel (°) | -3.98 |
| Radial setting (mm) | -0.9568 |
| Angular setting (°) | 0.17 |
| Vertical offset (mm) | 0.6264 |
| Sliding base (mm) | 0.377 |
| Workpiece mounting angle (°) | 0 |
| Horizontal offset (mm) | -2.3687 |
| Roll ratio | -0.01963 |
The ease-off topology after modification separated from the conjugate reference at all boundaries, which suppresses edge contact at the addendum. It also indicated that the contact zone would run diagonally from the heel addendum toward the toe root. The change was not uniformly beneficial in every metric: at the toe addendum and heel root, the heightwise mismatch increased, so the contact zone became narrower in the tooth-height direction while becoming longer in the lengthwise direction. The four boundary mismatch values also became more different, which increased the contact diagonal.
TCA of the corrected machine settings matched those geometric predictions. The contact path angle to the root-cone normal rose from 44.6° to 64°, the contact zone became longer lengthwise and narrower heightwise, and the contact diagonal increased. The transmission error changed from 9.449 to 13.47 arcsec, but its curve became more symmetric. The larger transmission-error amplitude is a trade-off against the improved symmetry and reduced edge-contact risk; it should not be read as an unconditional improvement in every mesh metric.
| Metric | Initial | After modification |
|---|---|---|
| Contact path angle to root-cone normal (°) | 44.6 | 64.0 |
| Transmission error (arcsec) | 9.449 | 13.47 |
| Contact-zone extent in lengthwise direction | Shorter | Longer |
| Contact-zone extent in tooth-height direction | Wider | Narrower |
| Transmission-error symmetry | Poor | Improved |
| Edge-contact risk | Present | Suppressed |
Cutting Test and Engineering Interpretation
The corrected settings were used to cut a pair on a Gleason 600 milling machine. The workpieces were mounted at the standard mounting distance and checked by rolling inspection. The contact pattern after modification followed the predicted diagonal: it was longer along the face and narrower along the tooth height, with a larger contact diagonal than the original pattern. The physical contact zone was consistent with the TCA and ease-off results, which verifies that the machine-setting corrections act in the intended direction on the real tooth surface.
The practical value of the method is that it converts coefficient selection into a pointwise mismatch specification. A designer can start from a contact requirement, choose the normal mismatch at the pitch-line midpoint and at the boundaries, calculate the five coefficients, and then solve for machine-setting corrections. The TCA loop is still required, but it becomes a verification and fine-adjustment step rather than a blind search over five coefficients.
The scope of the demonstration is a single-side cutting process for one gear pair. In double-side machining, both flanks are produced with one set of machine settings, so a correction intended for one flank also changes the other. The coupling is not captured by the single-flank sensitivity model presented here, and the extension to double-side quantitative mismatch modification requires a coupled correction strategy. Within the single-side case, the method provides a direct link among specified normal mismatch, second-order surface coefficients, machine-setting corrections, and the resulting contact pattern.
The same structure is applicable to other gear surfaces when a conjugate baseline and a machine-setting sensitivity matrix can be formulated. The essential requirement is not the specific five-coefficient parameterization but the separation of the problem into three measurable steps: prescribe the normal mismatch, map it to a target surface, and correct the generating settings until the target is reproduced within the required tolerance.