Spiral bevel gears transmit torque between intersecting axes with high contact ratio, smooth motion, and strong load capacity, which explains their use in marine drives, precision machinery, and aerospace transmissions. Their meshing quality is commonly assessed through two geometric outputs of tooth contact analysis (TCA): the contact path, which is the locus of instantaneous contact points on the flank, and the transmission error (TE), which is the deviation of the gear pair from a constant instantaneous ratio. Small mounting errors can shift the contact path toward the toe or heel, break the symmetry of the TE curve, and increase its fluctuation amplitude. In a finished housing, the shaft angle and lateral offset are effectively fixed, but axial shim adjustments at the pinion and gear remain available. The practical question is therefore inverse: for a desired contact-path position and TE shape, what mounting adjustments should be used without recutting the flanks?
Spiral bevel gear set: the mounting position changes the contact path and transmission error without altering the generated tooth flanks.
Why mounting position changes meshing
An ideal assembly places the generated pinion and gear at the design shaft angle \(\alpha\) and at theoretical mounting settings \(E\), \(P\), and \(G\). Here \(E\) is the gear lateral offset adjustment, \(P\) is the pinion mounting-distance adjustment, and \(G\) is the gear axial adjustment. Real assemblies contain misalignment components \(\Delta\alpha\), \(\Delta E\), \(\Delta P\), and \(\Delta G\), so the actual shaft angle and mounting distances become \(\alpha+\Delta\alpha\), \(E+\Delta E\), \(P+\Delta P\), and \(G+\Delta G\). These errors enter the rigid-body coordinate transformations between the two gear axes. They do not change the cutter-generated flank geometry, but they change where the two flanks touch and how the gear rotates as contact moves across the tooth.
Classical TCA assumes ideal mounting and solves for the contact point at each pinion roll angle. An error-sensitive version, often called ETCA, embeds the misalignment in the assembly transform and solves the same contact conditions in a fixed reference frame. The contact path is then obtained as the locus of contact points, and the TE is obtained from the difference between the actual gear rotation and the rotation predicted by the nominal ratio. For a spiral bevel gear pair, a light-load contact path is normally placed near the middle of the flank, slightly toward the toe. Under load, the contact ellipse is expected to spread toward the mid-flank rather than run off the heel or toe. A TE curve that is symmetric and has a small amplitude is preferred because it reduces the risk of edge contact and impact at mesh transfer.
Tooth flank generation model
The flanks are generated by a face-milling process on a five-axis gear machine. A cutter head rotates about the cradle axis, and the straight cutting edge sweeps a generating surface. For the gear, a double-sided cutter produces convex and concave flanks; for the pinion, single-sided cutters are used with a modified roll motion. The coordinate transformation from the cutter frame to the workpiece frame has the general form
\[
\mathbf r_i=M_{i b_i}M_{b_i a_i}M_{a_i m_i}M_{m_i c_i}M_{c_i t_i}\,\mathbf r_{ti},
\]
where \(i=1\) denotes the pinion and \(i=2\) the gear. The matrices account for the cradle rotation, workpiece mounting angle, axial and vertical wheel positions, sliding base, and workpiece rotation. For the gear cutter, the straight-edge generating surface is
\[
\mathbf r_g(\mu_g,\theta_g)=
\begin{bmatrix}
(R_g\pm \mu_g\sin\alpha_g)\cos\theta_g\\
(R_g\pm \mu_g\sin\alpha_g)\sin\theta_g\\
-\mu_g\cos\alpha_g
\end{bmatrix},
\]
with cutter tip radius \(R_g\), profile angle \(\alpha_g\), and surface parameters \(\mu_g\) and \(\theta_g\). Its unit normal is
\[
\mathbf n_g(\theta_g)=
\begin{bmatrix}
\cos\alpha_g\cos\theta_g\\
\cos\alpha_g\sin\theta_g\\
\pm\sin\alpha_g
\end{bmatrix}.
\]
The equation of meshing is
\[
f_g(\mu_g,\theta_g,\phi_{c2})=\mathbf N_g\cdot\mathbf V_g=0,
\]
where \(\mathbf V_g\) is the relative velocity between the cutter generating surface and the gear blank at the cutting point, and \(\phi_{c2}\) is the cradle angle. Eliminating \(\mu_g\) gives the gear flank as \(\mathbf r_2=\mathbf r_2(\theta_g,\phi_{c2})\). The pinion follows the same construction with subscript \(p\), using its own cutter radius \(R_p\), profile angle \(\alpha_p\), and modified roll coefficients. The resulting flanks are smooth generated surfaces, not approximations by discrete points; the TCA then operates on these surfaces.
Theoretical mounting settings and ETCA
The theoretical mounting settings are not taken directly from a drawing. They are solved together with the initial contact position. The initial contact is placed at the gear tooth mean point. If \(L\) is the axial distance from the pitch-cone apex to the projection of that point and \(R\) is its radial distance from the gear axis, the gear surface parameters \(\theta_g\) and \(\phi_{c2}\) satisfy
\[
x_2(\theta_g,\phi_{c2})=L,\qquad
y_2^2(\theta_g,\phi_{c2})+z_2^2(\theta_g,\phi_{c2})=R^2.
\]
The two flanks are then transformed into a common assembly frame \(S_h\). Continuous contact requires equality of position vectors and unit normal vectors:
\[
\mathbf r_h^{(1)}(\theta_p,\phi_{c1},\phi_1)=
\mathbf r_h^{(2)}(\theta_g,\phi_{c2},\phi_2),
\]
\[
\mathbf n_h^{(1)}(\theta_p,\phi_{c1},\phi_1)=
\mathbf n_h^{(2)}(\theta_g,\phi_{c2},\phi_2).
\]
These vector equations provide five independent scalar conditions. At the initial contact point the instantaneous ratio must equal the theoretical ratio \(z_2/z_1\), which supplies a sixth condition:
\[
\frac{\omega_1}{\omega_2}=\frac{z_2}{z_1}
=
\frac{\left|\mathbf n_2\cdot(\mathbf p_2\times\mathbf l_2)\right|}
{\left|\mathbf n_1\cdot(\mathbf p_1\times\mathbf l_1)\right|},
\]
where \(\mathbf p_1\) and \(\mathbf p_2\) are unit vectors along the pinion and gear axes, and \(\mathbf l_1\), \(\mathbf l_2\) locate the contact point relative to those axes. The six unknowns are \(E\), \(P\), \(\theta_p\), \(\phi_{c1}\), \(\phi_1\), and \(\phi_2\). Solving the nonlinear system gives the theoretical mounting settings and the initial pinion contact parameters. For the 40/79-tooth pair considered here, this produced \(P=-0.5751\) mm, \(E=0.4979\) mm, and \(G=0\).
Once the theoretical settings are known, the misalignment is introduced by replacing \(P\), \(E\), \(G\), and \(\alpha\) with \(P+\Delta P\), \(E+\Delta E\), \(G+\Delta G\), and \(\alpha+\Delta\alpha\) in the assembly matrices. The contact equations again give five independent scalar equations. Choosing the pinion roll angle \(\phi_1\) as the input leaves five unknowns, which are solved at each step. The contact path is the resulting locus \(\mathbf r_1(\phi_1)\), \(\mathbf r_2(\phi_1)\), and the transmission error is
\[
\Delta\phi_2(\phi_1)=
\left[\phi_2(\phi_1)-\phi_{02}\right]
-\frac{z_1}{z_2}
\left[\phi_1-\phi_{01}\right].
\]
This definition removes the nominal linear ratio and retains the periodic deviation caused by the generated flank geometry and the mounting pose.
Optimization targets
Two geometric features are extracted from each TCA result. The first is the contact-path centroid relative to the tooth-flank midpoint. If the contact points are \((x_i,y_i)\), the centroid is
\[
\bar x=\frac{1}{n}\sum_{i=1}^{n}x_i,\qquad
\bar y=\frac{1}{n}\sum_{i=1}^{n}y_i.
\]
With face width \(b\), mid-tooth height \(h_m\), and midpoint coordinates \((x_m,y_m)\), the position coefficients are
\[
\lambda_x=\frac{\bar x-x_m}{b},\qquad
\lambda_y=\frac{\bar y-y_m}{h_m}.
\]
These coefficients describe whether the contact path lies toward the toe or heel and toward the addendum or dedendum. The second feature is the TE curve. Its amplitude at the mesh transfer point is \(E_{T\max}\), and its end droop deviation is
\[
\Delta E_T=\left|E_{Ta}-E_{Tb}\right|,
\]
where \(E_{Ta}\) and \(E_{Tb}\) are the TE values at the two ends of a single-tooth meshing cycle. A small \(E_{T\max}\) limits the angular fluctuation, while a small \(\Delta E_T\) means the two ends of the curve are nearly at the same level, giving a more symmetric TE curve.
The four residuals are written as
\[
(f_{\rm auto})_1=\bar x-x_m-\Delta_1,\qquad
(f_{\rm auto})_2=\bar y-y_m-\Delta_2,
\]
\[
(f_{\rm auto})_3=E_{T\max}-\Delta_3,\qquad
(f_{\rm auto})_4=\Delta E_T-\Delta_4,
\]
where \(\Delta_1=\lambda_x b\), \(\Delta_2=\lambda_y h_m\), and \(\Delta_3\), \(\Delta_4\) are the prescribed TE targets. Because the flank equations are nonlinear, solving all four residuals simultaneously as independent objectives is not reliable. A nonlinear programming formulation is used instead: one residual is minimized, and the other three become inequality constraints. With \(\Delta P\) and \(\Delta G\) as design variables,
\[
\min_{\Delta P,\Delta G}(f_{\rm auto})_4
\]
\[
\text{s.t.}\qquad
\bar x-x_m\le\Delta_1,\qquad
\bar y-y_m\le\Delta_2,\qquad
E_{T\max}\le\Delta_3.
\]
The shaft angle and lateral offset are held fixed because the housing has already been machined. The optimization therefore answers a constrained assembly question: what axial shim adjustments at the pinion and gear best reproduce the prescribed contact position and TE symmetry?
Case study: a face-milled 40/79 pair
The procedure was applied to a face-milled spiral bevel gear pair with a 40-tooth left-hand pinion and a 79-tooth right-hand gear. The main design data are listed in Table 1, and the machine settings used to generate the flanks are listed in Table 2. The gear was generated with a double-sided cutter; the pinion was generated with single-sided cutters and a modified roll coefficient.
| Design parameter | Pinion | Gear |
|---|---|---|
| Number of teeth \(z\) | 40 | 79 |
| Module \(m\) (mm) | 3 | 3 |
| Face width \(b\) (mm) | 30 | 30 |
| Pressure angle \(\alpha_0\) (°) | 20 | 20 |
| Shaft angle \(\alpha\) (°) | 90 | 90 |
| Whole tooth depth \(h\) (mm) | 5.664 | 5.664 |
| Pitch cone angle \(\delta\) (°) | 26.5651 | 63.4349 |
| Hand | Left | Right |
| Machine setting | Pinion | Gear |
|---|---|---|
| Cutter profile angle \(\alpha_p,\alpha_g\) (°) | 19.925 | 20.075 |
| Cutter-head radius \(R\) (mm) | 226.6202 | 227.0462 |
| Radial setting \(s\) (mm) | 107.7669 | 109.0728 |
| Cradle angle \(q\) (°) | 298.2101 | 59.9098 |
| Axial wheel position \(X_P\) (mm) | -1.567 | 0 |
| Vertical wheel position \(E_m\) (mm) | 0 | 0 |
| Sliding base \(X_b\) (mm) | 0.6109 | 0 |
| Workpiece mounting angle \(\gamma\) (°) | 25.6196 | 61.7213 |
| Roll ratio \(i_{pg}\) | 2.194 | 1.118 |
| First-order modified roll \(C\) | -0.05 | 0 |
| Second-order modified roll \(D\) | 0 | 0 |
Before optimization, the contact path lay near the heel and addendum side of the gear flank. Its centroid was \((15.59,\,2.54)\) mm, while the flank midpoint was \((15.00,\,2.52)\) mm. The path was therefore about 0.59 mm on the heel side of the midpoint. The geometric TE amplitude was \(E_{T\max}=12.0025\) arcsec, and the end droop deviation was \(\Delta E_T=5.3211\) arcsec, indicating a noticeably asymmetric curve.
The optimization used \(\Delta P\) and \(\Delta G\) as variables, with position targets \(\lambda_x=\lambda_y=-0.01\). The TE amplitude was constrained on an arcsecond scale, and the end droop deviation was minimized. The resulting adjustments were
\[
\Delta P=0.0722\ \text{mm},\qquad
\Delta G=-0.0795\ \text{mm}.
\]
With the theoretical setting \(P=-0.5751\) mm, the adjusted pinion mounting distance becomes \(P+\Delta P=-0.5029\) mm. The adjusted gear axial setting is \(G+\Delta G=-0.0795\) mm. The lateral offset remains \(E=0.4979\) mm and the shaft angle remains \(90^\circ\).
| Quantity | Before | After |
|---|---|---|
| Contact centroid \(\bar x\) (mm) | 15.59 | 13.83 |
| Contact centroid \(\bar y\) (mm) | 2.54 | 2.47 |
| TE amplitude \(E_{T\max}\) (arcsec) | 12.0025 | 11.3376 |
| End droop deviation \(\Delta E_T\) (arcsec) | 5.3211 | 0.009 |
| Pinion offset \(\Delta P\) (mm) | 0 | 0.0722 |
| Gear axial offset \(\Delta G\) (mm) | 0 | -0.0795 |
After optimization, the contact centroid moved to \((13.83,\,2.47)\) mm, which is 1.17 mm on the toe side of the flank midpoint. The path thus shifted from the heel side to the small-end side. This is the intended direction for a light-load contact pattern: a path placed too close to the heel can run toward the edge under load, whereas a path biased toward the toe leaves room for the contact ellipse to spread back toward the mid-flank. The TE amplitude decreased to 11.3376 arcsec, a reduction of about 5.5%, and the end droop deviation fell to 0.009 arcsec, making the two ends of the TE curve nearly equal. The optimized geometric TE is therefore much more symmetric, with a smaller fluctuation amplitude.
Interpretation and limits
The result shows that small axial shim changes, on the order of 0.07–0.08 mm, can produce a measurable shift in the contact path and a large change in TE symmetry. This makes installation adjustment a useful complement to flank modification: the gear set can be generated with a fixed cutter design, and the final mesh can be trimmed at assembly by adjusting \(P\) and \(G\). The inverse mapping from prescribed contact and TE features to mounting offsets is the main practical value of the method. It can support shim selection on an assembly line and can be used to check whether a housing tolerance leaves enough adjustment range.
The calculation is geometric and quasi-static. It captures the rigid-body effect of misalignment and the generated flank curvature, but it does not include elastic deflection under torque, thermal distortion, or manufacturing scatter in the flanks. The claim that the loaded contact area will expand toward the mid-flank is a design expectation based on the unloaded path position; it should be confirmed with a loaded tooth contact analysis or a roll test. The TE amplitude reduction is modest, while the symmetry improvement is large, so the dominant benefit in this case is the removal of end-to-end droop rather than a major reduction in peak TE. The method is also limited to misalignments that can be compensated by the available adjustments: if the shaft angle or lateral offset deviates substantially, axial shims alone may not restore the desired contact, and housing correction or flank modification may be required.
For spiral bevel gear drives, the contact path and transmission error are coupled outputs of the same meshing process. Adjusting the mounting position changes both at once, which is why a constrained optimization is more appropriate than tuning one feature independently. The 40/79 case illustrates the principle: a small change in pinion mounting distance and gear axial position moved the contact path toward the toe and nearly equalized the two ends of the TE curve, while leaving the generated tooth flanks unchanged.