Spiral bevel gears are used in aerospace, automotive, and industrial drivetrains because their curved teeth provide gradual, oblique engagement, high load capacity, controllable contact area, and stable motion. Their geometric design is judged largely by two unloaded Tooth Contact Analysis (TCA) outputs: the contact pattern on the tooth flank and the transmission error along the meshing path. The contact pattern indicates where the teeth bear load and whether the pattern risks running to the tooth edges; transmission error quantifies the deviation from a constant angular ratio and is therefore linked to vibration and noise excitation.
TCA was introduced by Baxter in 1961 for hypoid gears, using a no-load contact model to study how mounting errors alter contact behavior. Since then, TCA has become a standard design tool for spiral bevel and hypoid gears. Its numerical difficulty, however, is not merely geometric. The contact equations are nonlinear, the surfaces are often defined through simulated cutting kinematics rather than simple analytical functions, and the contact path and transmission error are strongly coupled. Solving them in separate steps can produce inconsistent results, poor convergence, and excessive sensitivity to initial values.
From cutter kinematics to a numerical tooth flank
A spiral bevel gear flank is generated as the envelope of a rotating cutter surface. For a straight-sided cutter, the tool surface is a conical surface parameterized by a height coordinate s and an angular coordinate θ:
$$
\mathbf F_C(\theta,s)=
\begin{bmatrix}
\left(R_{\mathrm{tip}}+s\sin\alpha\right)\cos\theta\\
\left(R_{\mathrm{tip}}+s\sin\alpha\right)\sin\theta\\
-s\cos\alpha
\end{bmatrix}
$$
where Rtip is the cutter tip radius and α is the pressure angle. The corresponding unit normal depends on θ and α but not on s, which is advantageous for the contact solution:
$$
\mathbf N_C(\theta)=
\begin{bmatrix}
-\cos\alpha\cos\theta\\
-\cos\alpha\sin\theta\\
-\sin\alpha
\end{bmatrix}
$$
The cutter is moved relative to the workpiece by a machine-setting transformation that includes the cradle angle φC, workpiece rotation φP, cutter tilt, swivel angle, radial setting, vertical offset, bed position, and horizontal offset. Modified roll and helical motion are represented by
$$
\phi_P=R_a\left(\phi_C-C\phi_C^2-D\phi_C^3\right),
\qquad
X_{BH}=X_B+H_e\phi_C
$$
where Ra is the roll ratio, C and D are modified-roll coefficients, XB is the basic bed position, He is the helical-motion coefficient, and XBH is the effective bed position. A forming process is obtained as a limiting case by setting the roll ratio and cradle rotation to zero, so that the cutter and workpiece have no generating relative motion.
The spatial cutter surface has three parameters, but the generated flank should be a two-parameter surface. The extra parameter is removed by the meshing condition
$$
\mathbf N_O\cdot\mathbf V=0
$$
where NO is the cutter normal in the workpiece frame and V is the relative velocity between cutter and workpiece. Because s appears linearly in this condition, it can be solved explicitly as a function of θ and φC. Substitution into the cutter surface gives the numerical tooth flank in two parameters. This step is important: it converts a cutting simulation into a surface representation suitable for contact equations without requiring a closed-form analytical flank.
Initial contact with a reduced five-equation system
For two tooth surfaces in mesh, contact requires equality of position and parallelism of the normals. If each surface has two parameters and each member has one motion parameter, the complete system contains six unknowns. The position condition supplies three equations, and the normal condition appears to supply three more, but a unit normal has only two independent components. Only two normal equations are needed when the normal orientation is chosen consistently, with the outward normals of the two bodies opposing each other at contact.
A practical difficulty remains: five equations are available for six unknowns. One unknown can be eliminated by aligning the two tooth flanks near their mid-face positions and rotating the pinion and gear by equal angular amounts during the search. The two surfaces then approach or separate together, and the contact system becomes
$$
\begin{aligned}
x_P&=x_G,\qquad y_P=y_G,\qquad z_P=z_G,\\
n_{x,P}&=n_{x,G},\qquad n_{y,P}=n_{y,G}.
\end{aligned}
$$
The five unknowns are the two surface parameters of each flank and one common rotation parameter. This reduction is not merely cosmetic. It lowers the dimension of a strongly nonlinear root-finding problem and reduces the sensitivity to initial guesses, which is a common source of instability in TCA.
Contact path, blank boundary, and transmission-error truncation
Once an initial contact point is found, the contact path is generated by advancing the pinion rotation in small increments and solving for the gear rotation that restores tangency. If the pinion has PN teeth and the path is to contain Nn points, a natural step is
$$
p_{\mathrm{step}}=\frac{2\pi}{P_N N_n}.
$$
The pinion rotation at step j is ϖP,j = ϖP0 + j pstep, and the corresponding gear rotation ϖG,j is obtained from the contact equations. The resulting sequence of contact points forms the theoretical meshing path.
A theoretical flank often extends beyond the actual tooth blank, so the contact path must be truncated. The first boundary is the tooth blank itself. The addendum, dedendum, heel, and toe lines are represented in a axial-radial cross-section, and a candidate contact point is retained only if it lies inside the four line inequalities for both pinion and gear. This produces a contact path within the working flank region, but it does not yet guarantee that the path corresponds to a single tooth pair in actual mesh.
Transmission error provides the second truncation. Taking the first computed point as a reference, the raw transmission error is
$$
T’_{e,j}=
\left|\varpi_{G,j}-\varpi_{G,n}\right|
–
\left|i\left(\varpi_{P,j}-\varpi_{P,n}\right)\right|,
$$
where i is the nominal transmission ratio. Because the reference point need not coincide with the ideal-ratio point, the curve is shifted so that its maximum is zero:
$$
T_{e,j}=T’_{e,j}-\max_j T’_{e,j}.
$$
The transmission-error curve of a single tooth pair is approximately parabolic. Curves for adjacent tooth pairs have the same shape but are shifted along the pinion-rotation axis by one pitch, 2π/PN. The intersection of adjacent curves marks the changeover between tooth pairs. Within a finite blank boundary, the intersection also defines the usable end of the single-tooth contact path. If the curves do not intersect inside the flank region, the calculated contact runs to an edge; under load, the pattern would tend to expand further toward that edge.
This truncation strategy avoids solving a simultaneous two-tooth contact system containing ten equations. Instead, it uses the already computed single-tooth transmission-error curves and a segment-intersection test. With a sufficiently small rotation step, the lower endpoint of the intersecting segment provides a practical cutoff for the contact path.
Contact pattern without second derivatives of the numerical flank
Conventional contact-pattern calculations require the principal curvatures and principal directions at each contact point. For a numerical flank generated through cutter kinematics, first and second partial derivatives can be cumbersome and numerically noisy. An alternative is to construct the instantaneous contact region directly from the separation between the two flanks.
At a contact point with position P and unit normal N, two orthogonal unit vectors a and b are built in the tangent plane. A family of rays is generated at uniformly spaced angles around the contact point. Each ray is projected onto the pinion and gear flanks, and the separation is evaluated along the normal:
$$
d_j=\left|\left(\mathbf P_{P,j}-\mathbf P_{G,j}\right)\cdot\mathbf N_{P,j}\right|.
$$
The boundary of the instantaneous contact region is reached where this separation exceeds a prescribed threshold. In the numerical example, the threshold is d = 0.003 mm. The long axis of the contact ellipse is then approximated by the two boundary points with the greatest separation. Repeating the procedure along the contact path produces the full contact pattern. Because the method uses only point coordinates and normals, it avoids differentiating a numerically defined surface twice.
Numerical example
The procedure was applied to a spiral bevel gear pair with a 31-tooth pinion and a 38-tooth gear. The face width is 32 mm for both members, and the outer cone distance is 208.81 mm. The nominal cutter radii are 301.8666 mm for the pinion and 304.8 mm for the gear. The cutter pressure angles are 17.6° on the outer side and 22.6° on the inner side for the pinion, and 20° and 25°, respectively, for the gear.
| Parameter | Pinion | Gear |
|---|---|---|
| Number of teeth | 31 | 38 |
| Pitch cone angle (°) | 23.1667 | 28.8333 |
| Face cone angle (°) | 24.2667 | 29.6 |
| Root cone angle (°) | 22.4 | 27.7333 |
| Addendum (mm) | 5.09 | 3.44 |
| Dedendum (mm) | 4.65 | 6.30 |
| Face width (mm) | 32 | 32 |
| Outer cone distance (mm) | 208.81 | 208.81 |
| Nominal cutter radius (mm) | 301.8666 | 304.8 |
| Cutter pressure angle (°) | 17.6 outer / 22.6 inner | 20 outer / 25 inner |
The collaborative TCA calculation gives a transmission-error amplitude of 108.6 μrad. Calculation with the commercial Gleason CAGE software for the same gear set gives 110.3 μrad, a difference of 1.7 μrad, or about 1.5%. The contact pattern position and trend are also consistent between the two calculations. This agreement supports both the contact-path solution and the numerical flank representation used in the separation-based pattern calculation.
| Indicator | Collaborative TCA | Gleason CAGE |
|---|---|---|
| Transmission-error amplitude | 108.6 μrad | 110.3 μrad |
| Contact-pattern position and trend | Consistent with CAGE | Reference result |
| Pattern boundary criterion | Normal separation ≤ 0.003 mm | Default mid-face contact setting |
Interpretation and limits
The main mechanical point is that contact pattern and transmission error should not be treated as independent post-processing outputs. The contact path supplies the points at which gear rotation is evaluated; the transmission-error curves determine where the single-tooth contact path should be cut off; and the same contact points supply the local tangent frames used to construct the pattern. Coupling these steps improves consistency and avoids some of the numerical difficulties associated with separate curvature-based or multi-tooth simultaneous solutions.
The approach also has clear limits. It is an unloaded geometric TCA: tooth deflection, load sharing, contact pressure, and misalignment-induced elastic deformation are outside its scope. The pattern threshold of 0.003 mm is a geometric separation criterion, not a stress-based contact width. The numerical comparison covers one gear pair and one set of machine settings, so it demonstrates feasibility and consistency with commercial software rather than broad statistical validation. Edge-contact prediction from the absence of transmission-error intersections should also be treated as a design warning, because loaded contact can alter the pattern near the tooth boundaries.
Within those limits, the method provides a coherent way to obtain the two primary TCA indicators from one numerical flank model. It reduces the contact problem from six unknowns to five, truncates the path through transmission-error intersections rather than a larger simultaneous contact system, and constructs the contact pattern from normal separation rather than second derivatives. These choices address the accuracy, efficiency, and stability problems that arise when strongly nonlinear contact equations are solved separately for pattern and transmission error.