Why the Initial Point Matters in TCA
Spiral bevel gears transmit motion between intersecting shafts with curved, inclined teeth that engage gradually rather than along a full line. This geometry gives smooth operation and high load capacity, which is why the pair is common in helicopter and aircraft engine drives, automotive axles, and other compact high-torque transmissions. The design quantities that control noise and vibration are the contact path—the locus of contact points on the tooth flank as the gear rotates—and the transmission error, the deviation of the output rotation from the nominal gear ratio.
Tooth contact analysis (TCA) computes these quantities by solving the geometric conditions for two generated flanks in mesh. The equations are nonlinear, and every TCA run needs an initial point: a pair of surface parameters and rotation angles close enough to the true contact for the solver to converge. With perfect alignment, local synthesis can prescribe a contact point in advance. In an assembled gearbox, however, vertical offset, horizontal offset, and shaft-angle error displace the flanks. The prescribed point may no longer lie on the contact path, and a solver started from a nominal point can converge slowly, converge to another tooth pair, or fail altogether.
Existing strategies handle the problem in different ways. Local-synthesis-based TCA presets the contact point by construction. Other formulations search for the closest pair of discretized flank points, but may require a point to be specified on the gear flank first, may assume ideal alignment, or may need measured tooth-flank deviations as input. The approach described below works from the design and machining data alone. It discretizes the two generated flanks, locates the pair with the smallest separation under the actual installation errors, and then refines that pair by solving the contact equations with the Levenberg–Marquardt method. The result is an initial point suitable for tracing the contact path and transmission error.
Tooth Flank Generation and Surface Equations
The flanks are generated by a rotating cutter on a machine tool. For each member, the kinematic chain passes through a cutter coordinate frame, a machine frame, a workpiece frame, and auxiliary frames that describe the relative position and motion of the cutter, the generating gear, and the blank. The cutter surface is transformed into the workpiece frame while the workpiece rotates, producing the generated flank rather than a simple theoretical surface.
Let subscript i = 1 denote the pinion and i = 2 the gear. The position and unit normal of a point on the generated flank are
$$
\mathbf r_i(u_{ci},\theta_{ci},\phi_i)
= \mathbf M_{i f_i}\mathbf M_{f_i e_i}\mathbf M_{e_i m_i}\mathbf M_{m_i n_i}\mathbf M_{n_i c_i}\mathbf r_{ci}(u_{ci},\theta_{ci})
$$
$$
\mathbf n_i(u_{ci},\theta_{ci},\phi_i)
= \mathbf L_{i f_i}\mathbf L_{f_i e_i}\mathbf L_{e_i m_i}\mathbf L_{m_i n_i}\mathbf L_{n_i c_i}\mathbf n_{ci}(u_{ci},\theta_{ci})
$$
Here rci and nci are the position and normal of the cutter surface, uci and θci are the cutter surface coordinates, φi is the workpiece rotation angle, M denotes a 4×4 homogeneous transformation matrix, and L is its 3×3 rotation submatrix.
The cutter and workpiece must also satisfy the meshing condition at every instant:
$$
f(u_{ci},\theta_{ci},\phi_i)=\mathbf n_{mi}\cdot\mathbf v_{oi}=0
$$
where nmi is the cutter normal in the machine frame and voi is the relative velocity between the virtual generating gear and the workpiece. Solving this scalar equation for φi and substituting back eliminates the rotation angle:
$$
\mathbf r_i=\mathbf r_i\big[u_{ci},\theta_{ci},\phi_i(u_{ci},\theta_{ci})\big],\qquad
\mathbf n_i=\mathbf n_i\big[u_{ci},\theta_{ci},\phi_i(u_{ci},\theta_{ci})\big]
$$
The result is a two-parameter surface for each member. The surface is defined by the machine settings and cutter geometry; it is the mathematical object used in the contact search, not a measured or simply approximated tooth form.
Assembly Frame with Installation Errors
The two members are placed in a fixed frame Ms in their nominal mounting position, and the installation errors are introduced. M1 and M2 are the pinion and gear rotation frames, and Mt is an auxiliary frame used to position the pinion. The three error components are the vertical offset ΔV, the horizontal offset ΔH, and the shaft-angle error ΔΣ.
The gear flank can be taken directly in Ms by making M2 coincident with the fixed frame:
$$
\mathbf r_{s2}=\mathbf r_2,\qquad \mathbf n_{s2}=\mathbf n_2
$$
The pinion flank is transformed through the auxiliary frame:
$$
\mathbf r_{s1}=\mathbf M_{st}\mathbf M_{t1}\mathbf r_1,\qquad
\mathbf n_{s1}=\mathbf L_{st}\mathbf L_{t1}\mathbf n_1
$$
Let i1 and i2 be unit vectors along the pinion and gear axes. Their expressions in the fixed frame are
$$
\mathbf i_{s1}=\mathbf L_{st}\mathbf L_{t1}\mathbf i_1,\qquad
\mathbf i_{s2}=\mathbf L_{s2}\mathbf i_2
$$
To bring a candidate pair of flank points into mesh, the pinion is rotated by β1 about −is1 and the gear by β2 about is2. Using Rodrigues’ rotation formula, the rotated position and normal become
$$
\mathbf r^*_{si}
=\mathbf r_{si}\cos\beta_i
+(\mathbf i_{si}\times\mathbf r_{si})\sin\beta_i
+\mathbf i_{si}(\mathbf i_{si}\cdot\mathbf r_{si})(1-\cos\beta_i)
$$
$$
\mathbf n^*_{si}
=\mathbf n_{si}\cos\beta_i
+(\mathbf i_{si}\times\mathbf n_{si})\sin\beta_i
+\mathbf i_{si}(\mathbf i_{si}\cdot\mathbf n_{si})(1-\cos\beta_i)
$$
At a true contact point the two surfaces coincide and their normals are parallel:
$$
\mathbf r^*_{s1}=\mathbf r^*_{s2},\qquad
\mathbf n^*_{s1}=\mathbf n^*_{s2}
$$
Because both normals are unit vectors, the normal equality contributes only two independent scalar conditions. Together with the three position equations, the contact conditions provide five independent equations. This count is important for the numerical solution described next.
Discrete Search for the Initial Guess
Solving the contact conditions from an arbitrary guess is unreliable when the initial point is unknown. The method therefore first reduces the problem to a finite search on the two flanks.
Each flank is projected onto its axial section and divided into m intervals along the tooth height and n intervals along the face width. The gear grid is numbered from the heel at the addendum to the toe at the dedendum; the pinion grid is numbered from the heel at the dedendum to the toe at the addendum. The opposite numbering reflects the conjugate orientation of the two flanks and makes it convenient to pair points at similar positions.
For a candidate pair, the pinion and gear are rotated about their axes until their unit normals coincide; the required rotation angles follow from the vector geometry of the two normals. After this alignment, the separation of the two points is
$$
d=\left\|\mathbf r^*_{s1}-\mathbf r^*_{s2}\right\|
$$
The pair with the smallest d on the grid is the discrete approximation of the contact point. At exact contact the distance would be zero; on a finite grid, the closest pair is the best available approximation.
Because installation errors break exact index correspondence, the search is not limited to points with the same number. The gear grid is used as the reference. For each gear point, several pinion points with nearby indices are tested, and the distance is evaluated for each candidate. The search continues until all gear points have been visited. The output is the grid point with the global minimum distance, together with its rotation angles β1*, β2* and surface parameters uc1*, θc1*, uc2*, θc2*.
The discrete minimum is only an initial guess. The surface parameters are continuous, and the exact contact point generally lies between grid points. The discrete values are therefore passed to a local solver. The unknowns are uc1, θc1, β1, uc2, θc2, and β2. Since the contact conditions supply only five independent equations, one unknown must be prescribed. The gear rotation β2 is a natural choice: it defines the roll position at which contact is evaluated, and TCA sweeps this quantity in any case. Fixing β2 = β2* leaves five unknowns and five equations, which are solved by the Levenberg–Marquardt method.
The procedure can be summarized as follows:
- Generate the pinion and gear flanks from the machine settings and cutter geometry.
- Transform both flanks into the assembly frame with the specified ΔV, ΔH, and ΔΣ.
- Mesh the flanks and find the minimum-distance pair on the discrete grid.
- Fix β2 at the discrete value and solve the five remaining contact equations.
- Use the converged point as the seed for the first TCA step, then sweep the roll angle to trace the contact path and transmission error.
Numerical Example
The example is a right-hand gear meshing with a left-hand pinion. The basic design is listed in Table 1 and the machine settings in Table 2. The shaft angle is 90°, the outer transverse module is 6.5 mm, the tooth numbers are 26 and 47, the face width is 52.4 mm, and the midpoint cone distance is 148.3646 mm.
| Parameter | Pinion | Gear |
|---|---|---|
| Shaft angle (°) | 90 | |
| Outer transverse module (mm) | 6.5 | |
| Number of teeth | 26 | 47 |
| Face width (mm) | 52.4 | |
| Midpoint cone distance (mm) | 148.3646 | |
| Pitch cone angle (°) | 28.9511 | 61.0489 |
| Root cone angle (°) | 27.3144 | 58.2592 |
| Face cone angle (°) | 31.7408 | 62.6856 |
| Addendum (mm) | 7.285 | 3.765 |
| Dedendum (mm) | 4.987 | 8.507 |
| Parameter | Pinion concave flank | Gear convex flank |
|---|---|---|
| Cutter tip diameter (mm) | 293.6289 | 304.8000 |
| Cutter top distance (mm) | — | 3.048 |
| Pressure angle (°) | 18 | 20 |
| Axial wheel position (mm) | −3.0090 | — |
| Bed position (mm) | 1.3816 | −0.0015 |
| Vertical wheel position (mm) | −4.1584 | 0 |
| Cutting roll ratio | 1.9746 | 1.1415 |
| Radial cutter position (mm) | 130.7815 | 138.9238 |
| Angular cutter position (°) | −64.2278 | 63.9764 |
These values define the generated flanks; no measured tooth-flank deviation is used. The method is applied separately to negative and positive values of the vertical offset, the horizontal offset, and the shaft-angle error. In each case it returns an initial point and the corresponding contact path and transmission-error curve.
Three trends are consistent across the cases. First, the sign of the error controls the direction of the shift. With the assembly frame used in the example, a negative vertical, horizontal, or shaft-angle error moves the pinion contact trace toward the addendum and the gear trace toward the dedendum. A positive error reverses the direction: the pinion trace moves toward the dedendum and the gear trace toward the addendum. Second, the vertical offset produces a larger displacement of the contact path than the horizontal offset. Third, the ranking is reversed for transmission error: the horizontal offset produces the larger change in transmission error, while the vertical offset produces the smaller change. The shaft-angle error likewise shifts the contact path and alters the transmission error.
These are geometric, unloaded results. They show how the initial point and the contact path move when the mounting is perturbed; they do not predict contact stress, tooth deflection, or loaded transmission error.
Interpreting the Error Sensitivity
In the fixed frame, each error component changes a different element of the relative axis position. The vertical offset displaces the pinion axis in the vertical plane, the horizontal offset displaces it in the horizontal plane, and the shaft-angle error changes the angle between the axes. Because the contact point is determined by the intersection of the two generated surfaces, these changes do not produce identical shifts. The pinion and gear traces move in opposite directions because they are images of the same contact point on conjugate flanks: a shift toward the addendum on one member corresponds to a shift toward the dedendum on the other.
The different sensitivities follow from the way each error enters the assembly transformation. The vertical offset changes the effective center distance more directly in this mounting, which is consistent with its stronger effect on the contact path. The horizontal offset and the shaft-angle error alter the angular relation between the two members more strongly, which is consistent with their larger influence on transmission error. These rankings are specific to the example and to the sign conventions of the assembly frame; they should be checked for a different mounting rather than assumed.
For assembly and tolerance analysis, the practical value of the method is that the initial contact point can be computed before any TCA run. It needs only the design parameters, the machine settings, and the assumed installation errors. It is therefore suitable for batch studies of shim thickness, axis alignment, and shaft-angle tolerance, where many error combinations must be evaluated. It also avoids the awkward step of guessing a point on the gear flank and hoping that the solver converges.
Scope and Limitations
The grid resolution controls the quality of the initial guess. A coarse grid can place the seed in the wrong tooth region, while a fine grid increases the search cost. The final point is also sensitive to the prescribed β2, because β2 selects the roll position; in a full TCA the contact point moves as β2 is swept. The contact equations are solved for perfectly generated surfaces, so measured tooth-flank deviations and elastic deformation are outside the scope. The method is a geometric front end to TCA, not a replacement for loaded contact analysis.
Within those limits, the procedure provides a repeatable way to start TCA of spiral bevel gears with installation errors. It converts an unbounded nonlinear root-finding problem into a bounded discrete search followed by a local correction, and it returns the contact path and transmission error without requiring a pre-specified contact point.