Spiral bevel gears are critical components in power transmission systems, especially in demanding applications such as heavy-duty gas turbines, due to their advantages of high load capacity, smooth operation, and good meshing conformity. One of the most important performance indicators for these gears is the contact pattern formed on the tooth flanks during operation. This pattern directly influences the transmission error, noise, stress distribution, and ultimately, the service life of the gear set. In many industrial applications, spiral bevel gears are required to operate in bidirectional rotation, meaning both the concave and convex sides of the tooth flanks serve as working surfaces. This places a stringent requirement on the quality, consistency, and overlap of the contact patterns on both flanks.
In practical manufacturing and assembly, various installation errors are inevitable. These errors, including changes in offset (ΔE), pinion axial displacement (ΔP), gear axial displacement (ΔG), and shaft angle error (ΔΣ), significantly alter the designed meshing condition. They can cause the contact pattern to shift towards the edge of the tooth, reduce its area, change its orientation, or lead to severe edge contact, thereby compromising performance. Therefore, understanding the quantitative relationship between these assembly misalignments and the resulting contact pattern characteristics is essential for making precise adjustments during the final assembly phase. This knowledge helps in achieving the desired, high-quality contact patterns for both directions of rotation without relying solely on time-consuming trial-and-error methods.
This article delves into a systematic methodology for modeling, analyzing, and correcting the tooth contact patterns of spiral bevel gears under the influence of installation errors. By employing advanced simulation techniques and establishing quantitative relationships, we provide a theoretical foundation for efficient gear assembly and performance optimization.
Mathematical Modeling of the Tooth Flank
The foundation of any contact analysis lies in an accurate mathematical model of the gear tooth surfaces. We consider Gleason-type spiral bevel gears, where the gear (wheel) is generated using a duplex spread-blade method, and the pinion is generated via a modified roll method using single-sided cutting. The manufacturing process is modeled using a series of coordinate systems representing the machine tool components.
The basic coordinate systems include the machine tool coordinate system \( S_{m_i} \), the cradle system \( S_{c_i} \), the gear blank system \( S_i \), and the cutter coordinate system \( S_{g_i} \), where the subscript \( i=1 \) denotes the pinion and \( i=2 \) denotes the gear. Key machine settings such as the radial distance \( S_i \), angular position \( q_i \), sliding base \( X_{B_i} \), vertical wheel offset \( E_{m_i} \), and machine root angle \( \gamma_{m_i} \) define the relative positions of these systems. The cradle rotation angle is \( \phi_{c_i} \) and the blank rotation angle is \( \phi_i \).
The cutter, typically a double-sided blade group, generates the tooth flank. The outer blade cuts the concave side, and the inner blade cuts the convex side. The surface of revolution formed by the cutting edge can be described in the cutter coordinate system. Let \( r_{c_i} \) be the cutter point radius (different for concave and convex), \( u_{g_i} \) a length parameter along the blade, \( \theta_{g_i} \) the rotation angle, and \( \alpha_{g_i} \) the blade pressure angle. The position vector \( \mathbf{r}_{g_i} \) and unit normal vector \( \mathbf{n}_{g_i} \) of the cutter surface are:
$$
\mathbf{r}_{g_i}(u_{g_i}, \theta_{g_i}) = \begin{bmatrix}
(r_{c_i} \pm u_{g_i} \sin \alpha_{g_i}) \cos \theta_{g_i} \\
(r_{c_i} \pm u_{g_i} \sin \alpha_{g_i}) \sin \theta_{g_i} \\
\mp u_{g_i} \cos \alpha_{g_i} \\
\end{bmatrix}
$$
$$
\mathbf{n}_{g_i}(\theta_{g_i}) = \begin{bmatrix}
\cos \alpha_{g_i} \cos \theta_{g_i} \\
\cos \alpha_{g_i} \sin \theta_{g_i} \\
\pm \sin \alpha_{g_i} \\
\end{bmatrix}
$$
where the upper sign corresponds to the concave flank and the lower sign to the convex flank.
Using a series of coordinate transformation matrices \( \mathbf{M}_{ig} \), the cutter surface is transformed into the gear blank coordinate system \( S_i \). The generated flank must satisfy the equation of meshing between the cutter and the blank. The theoretical tooth surface of the gear and its unit normal are expressed as functions of the cutter surface parameters and the machine motion parameter (cradle angle):
$$
\mathbf{r}_i = \mathbf{r}_i(\theta_{g_i}, \phi_{c_i}), \quad \mathbf{n}_i = \mathbf{L}_{ig} \cdot \mathbf{n}_{g_i}(\theta_{g_i})
$$
Here, \( \mathbf{L}_{ig} \) is the 3×3 orientation matrix extracted from \( \mathbf{M}_{ig} \). For the pinion generated with a modified roll, the ratio of cradle-to-blank rotation is not constant but a function, often a polynomial, of the cradle angle \( \phi_{c_1} \):
$$
i_1(\phi_{c_1}) = [i_1 – 2C\phi_{c_1} – 3D\phi_{c_1}^2]^{-1}
$$
where \( i_1 \) is the basic ratio, and \( C \) and \( D \) are the second and third-order modification coefficients, respectively.
Tooth Contact Analysis (TCA) Considering Installation Errors
Boundary Definition and Contact Point Computation
To analyze the contact pattern within the usable tooth area, the boundaries of the tooth flank must be defined. The four corner points A, B, C, D of the flank projection are calculated based on gear blank geometry: pitch cone angle \( \delta_i \), face width \( b_i \), and addendum \( h_{a_i} \) and dedendum \( h_{f_i} \). The coordinates in the projection plane \( X_iO_iY_i \) are:
$$
\begin{aligned}
&X_A = R_i \cos \delta_i – h_{a_i} \sin \delta_i, \quad &Y_A = R_i \sin \delta_i – h_{a_i} \cos \delta_i \\
&X_B = R_i \cos \delta_i + h_{f_i} \sin \delta_i, \quad &Y_B = R_i \sin \delta_i – h_{f_i} \cos \delta_i \\
&X_C = X_B – b_i \cos \delta_{f_i} / \cos(\delta_i – \delta_{f_i}), \quad &Y_C = Y_B – b_i \sin \delta_{f_i} / \cos(\delta_i – \delta_{f_i}) \\
&X_D = X_A – b_i \cos \delta_{a_i} / \cos(\delta_{a_i} – \delta_i), \quad &Y_D = Y_A – b_i \sin \delta_{a_i} / \cos(\delta_{a_i} – \delta_i)
\end{aligned}
$$
where \( R_i \) is the outer cone distance, \( \delta_{a_i} \) is the face angle, and \( \delta_{f_i} \) is the root angle. The equations of the four boundary lines (AB, BC, CD, DA) can be derived from these points. Any contact point \( M \) on the flank with coordinates \( (X_M, Y_M) \) must lie within this closed boundary polygon.
Error Tooth Contact Analysis (ETCA) Algorithm
Installation errors disrupt the theoretically aligned position of the gear pair. Key assembly errors considered are: offset error \( \Delta E \), pinion axial error \( \Delta P \), gear axial error \( \Delta G \), and shaft angle error \( \Delta \Sigma \). These errors modify the transformation matrices between the gear and pinion coordinate systems in the fixed assembly housing.
The position and normal vectors of both gear and pinion flanks are transformed into a common global housing coordinate system \( S_h \), considering their respective rotational angles \( \phi_2 \) and \( \phi_1 \):
$$
\mathbf{r}_h^{(1)} = \mathbf{M}_{h1} \mathbf{M}_{d1} \mathbf{r}_1, \quad \mathbf{n}_h^{(1)} = \mathbf{L}_{h1} \mathbf{L}_{d1} \mathbf{n}_1 \\
\mathbf{r}_h^{(2)} = \mathbf{M}_{h2} \mathbf{M}_{d2} \mathbf{r}_2, \quad \mathbf{n}_h^{(2)} = \mathbf{L}_{h2} \mathbf{L}_{d2} \mathbf{n}_2
$$
The matrices \( \mathbf{M}_{h1}, \mathbf{M}_{h2}, \mathbf{M}_{d1}, \mathbf{M}_{d2} \) incorporate the assembly errors and gear rotations. For instance, the matrix positioning the gear in the housing is:
$$
\mathbf{M}_{h1} = \begin{bmatrix}
\cos(\Sigma + \Delta \Sigma) & 0 & \sin(\Sigma + \Delta \Sigma) & (G+\Delta G)\cos(\Sigma+\Delta \Sigma) \\
0 & 1 & 0 & E + \Delta E \\
-\sin(\Sigma + \Delta \Sigma) & 0 & \cos(\Sigma + \Delta \Sigma) & -(G+\Delta G)\sin(\Sigma+\Delta \Sigma) \\
0 & 0 & 0 & 1
\end{bmatrix}
$$
The fundamental ETCA equations require that at a point of contact, the position vectors and the unit normals of the two flanks coincide in the housing system:
$$
\mathbf{r}_h^{(1)}(\theta_{g1}, \phi_{c1}, \phi_1) = \mathbf{r}_h^{(2)}(\theta_{g2}, \phi_{c2}, \phi_2) \\
\mathbf{n}_h^{(1)}(\theta_{g1}, \phi_{c1}, \phi_1) = \mathbf{n}_h^{(2)}(\theta_{g2}, \phi_{c2}, \phi_2)
$$
This system contains five independent scalar equations. By prescribing the pinion rotation angle \( \phi_1 \) as an input, the remaining five unknowns \( (\theta_{g1}, \phi_{c1}, \theta_{g2}, \phi_{c2}, \phi_2) \) can be solved numerically. The transmission error (TE) is then computed as the deviation from perfect conjugate motion:
$$
\Delta \phi_2(\phi_1) = (\phi_2(\phi_1) – \phi_{20}) – \frac{z_1}{z_2} (\phi_1 – \phi_{10})
$$
where \( z_1, z_2 \) are the tooth numbers and \( \phi_{10}, \phi_{20} \) are initial reference angles.
Quantitative Analysis of Installation Error Impact on Contact Patterns
To systematically study the effects, we parameterize the contact pattern. The pattern is approximated as an ellipse, and its key characteristics are defined as: midpoint coordinates \( (x, y) \), area \( s \), and the inclination angle \( k \) of its major axis relative to the pitch cone direction. A numerical example based on a real industrial spiral bevel gear pair for a gas turbine drive is used. The basic gear data and machine settings are summarized below.
| Parameter (Unit) | Pinion | Gear |
|---|---|---|
| Number of Teeth | 28 | 33 |
| Hand of Spiral | Left | Right |
| Shaft Angle (°) | 90 | |
| Module at Outer End (mm) | 6.1 | |
| Face Width (mm) | 39.6 | |
| Mid Spiral Angle (°) | 35 | |
| Normal Pressure Angle (°) | 20 | |
We analyze the gear flank contact patterns for both forward (convex side active) and reverse (concave side active) rotation. In practice, adjusting the pinion’s position (\( \Delta E, \Delta P, \Delta \Sigma \)) is more common and economical than moving the gear.
Impact of Offset Error \( \Delta E \) and Pinion Axial Error \( \Delta P \)
The variations of the pattern parameters within the range \( \Delta E, \Delta P \in [-0.3 \, \text{mm}, +0.3 \, \text{mm}] \) were studied. The results for forward and reverse rotation are distinct. The following equations represent fitted curves for the convex (F) and concave (R) flanks’ pattern midpoint coordinates.
For the convex flank (Forward):
$$
\begin{aligned}
x_F(\Delta E) &= 6.928 \times 10^6 – 6.928 \times 10^6 \cos(3.325 \times 10^{-5} \Delta E) + 2.826 \times 10^4 \sin(3.325 \times 10^{-5} \Delta E) \\
x_F(\Delta P) &= 5.5 \times 10^6 – 5.5 \times 10^6 \cos(-4.409 \times 10^{-5} \Delta P) + 7501 \sin(-4.409 \times 10^{-5} \Delta P) \\
y_F(\Delta E) &= -0.2727 + 0.09012 \cos(4.763 \Delta E) + 1.387 \sin(4.763 \Delta E) \\
y_F(\Delta P) &= -1.881 + 1.665 \cos(2.208 \Delta P) – 3.352 \sin(2.208 \Delta P)
\end{aligned}
$$
For the concave flank (Reverse):
$$
\begin{aligned}
x_R(\Delta E) &= -1.378 \times 10^7 + 1.378 \times 10^7 \cos(-5.718 \times 10^{-4} \Delta E) – 5.039 \times 10^4 \sin(-5.718 \times 10^{-4} \Delta E) \\
x_R(\Delta P) &= -1.238 \times 10^6 + 1.238 \times 10^6 \cos(-2.099 \times 10^{-3} \Delta P) + 2250 \sin(-2.099 \times 10^{-3} \Delta P) \\
y_R(\Delta E) &= 0.1513 + 0.3859 \cos(4.331 \Delta E) – 1.74 \sin(4.331 \Delta E) \\
y_R(\Delta P) &= -4.528 \times 10^7 + 4.528 \times 10^7 \cos(4.362 \times 10^{-4} \Delta P) + 8933 \sin(4.362 \times 10^{-4} \Delta P)
\end{aligned}
$$
The general trends are summarized in the table below:
| Error | Direction | Midpoint X (Toe/Heel) | Midpoint Y (Root/Top) | Area \( s \) Trend | Angle \( k \) Trend |
|---|---|---|---|---|---|
| \( \Delta E \) Increase | Forward (Conv.) | Moves to Heel | Moves to Top | Increases then decreases | Increases |
| Reverse (Conc.) | Moves to Toe | Moves to Root | Decreases | ||
| \( \Delta P \) Increase | Forward (Conv.) | Moves to Toe | Moves to Root | Increases then decreases | Decreases |
| Reverse (Conc.) | Minor Change | Moves to Top | Increases |
Impact of Shaft Angle Error \( \Delta \Sigma \)
The shaft angle error \( \Delta \Sigma \) was varied within \( [-10′, +10′] \). Its effect is also significant, though generally less pronounced on pattern area compared to \( \Delta E \) and \( \Delta P \). The fitted trends are:
$$
\begin{aligned}
x_F(\Delta \Sigma) &= 19.71 + 0.1205 \cos(0.1451 \Delta \Sigma) – 4.395 \sin(0.1451 \Delta \Sigma) \\
x_R(\Delta \Sigma) &= 20.32 – 0.1731 \cos(0.1569 \Delta \Sigma) + 1.881 \sin(0.1569 \Delta \Sigma) \\
y_F(\Delta \Sigma) &= -1.954 \times 10^7 + 1.954 \times 10^7 \cos(6.707 \times 10^{-6} \Delta \Sigma) + 1.897 \times 10^4 \sin(6.707 \times 10^{-6} \Delta \Sigma) \\
y_R(\Delta \Sigma) &= -0.6735 + 0.1591 \cos(0.1896 \Delta \Sigma) – 1.015 \times 10^4 \sin(0.1896 \Delta \Sigma)
\end{aligned}
$$
An increase in \( \Delta \Sigma \) moves the convex flank pattern towards the toe and root, while moving the concave flank pattern towards the heel and top. The pattern area for both flanks shows a peak near zero error. The angle \( k \) decreases for the convex flank and increases for the concave flank as \( \Delta \Sigma \) increases.
Optimization of Installation Errors and Experimental Validation
Optimization Model for Pattern Correction
Given the complex influence of errors, an optimization approach is used to find the best installation adjustments. Let the target (ideal) pattern characteristics be \( \lambda_0 = [x_0, y_0, s_0, k_0] \). The actual pattern parameters \( \lambda = [x, y, s, k] \) are functions of the installation errors \( \mathbf{e} = [\Delta E, \Delta P, \Delta \Sigma] \). The goal is to minimize the deviation of the pattern’s major axis inclination while keeping the pattern centered and of adequate size. This is formulated as:
$$
\min f(\mathbf{e}) = \min | \, k(\mathbf{e}) – k_0 \, |
$$
Subject to constraints ensuring the pattern is centered within the flank and has a sufficient area:
$$
\begin{aligned}
\Delta x &= \left| \frac{\sum x_j}{n} – x_m \right| \leq \frac{b}{4} \\
\Delta y &= \left| \frac{\sum y_j}{n} – y_m \right| \leq \frac{h}{4} \\
& s_{\min} \leq s(\mathbf{e}) \leq s_0
\end{aligned}
$$
where \( (x_m, y_m) \) is the midpoint of the usable flank area, \( (x_j, y_j) \) are contact points, \( b \) is the face width, \( h \) is the mid-face height, and \( s_{\min} \) is a minimum acceptable area.
Optimization Results and Verification
For the example gear pair, the initial simulation with zero assumed installation errors yielded unsatisfactory patterns. The convex flank pattern was too close to the heel and top edge with an area of 388.9 mm² and a large negative inclination (\(k = -8.86^\circ\)). The concave flank pattern was also edge-prone with an area of 342.7 mm² and a positive inclination (\(k = 11.02^\circ\)). The transmission error showed a significant “jump” of 21.49 arc-seconds at the pitch point.
Applying the optimization model yielded optimal adjustment values: \( \Delta E = 0.0113 \, \text{mm}, \Delta P = 0.0095 \, \text{mm}, \Delta \Sigma = 0′ \). The corrected patterns showed remarkable improvement. Both patterns moved towards the center of the flank. The convex pattern parameters became \([x=17.06, y=0.06, s=332.83, k=-3.98^\circ]\) and the concave pattern became \([x=16.63, y=-0.08, s=360.41, k=2.81^\circ]\). The patterns were more elliptical, centrally located, and the transmission error jump was reduced to 16.12 arc-seconds, indicating smoother meshing.
The optimized results were validated using a physical rolling test on a gear testing machine. The pinion teeth were coated with marking compound, and the gear pair was assembled according to the optimized error values and run under light load. The contact patterns obtained from the test at four different gear positions (90° apart) were consistent and closely matched the simulated patterns in terms of location (centered slightly towards the toe) and shape (elongated ellipse). This confirmed the accuracy of the ETCA simulation and the effectiveness of the optimization-based correction methodology for achieving high-quality, consistent, and overlapping contact patterns on both the concave and convex flanks of spiral bevel gears.