Generating motion, correction authority, and the shop-floor problem
Spiral bevel gears transmit motion between intersecting shafts through curved teeth that engage gradually along the face width. Because several teeth share the load and the contact migrates rather than jumping from tooth to tooth, the pair tolerates deflection and misalignment better than straight bevel gearing, which is why the form dominates automotive drivelines, helicopter transmissions and precision machine-tool drives.
All of the predictive work downstream — tooth contact analysis (TCA), loaded TCA, contact-pattern and transmission-error studies — assumes that the manufactured flank coincides with the theoretical one. Machine-tool geometric errors, thermal distortion and cutter wear break that assumption. The conventional remedy is proportional correction: an operator reads a deviation map from a gear measuring centre and shifts machine settings according to experience. The method works, but its quality depends entirely on how well the operator can guess which setting produces which kind of distortion.
CNC spiral bevel gear cutting and grinding machines now built in China provide the high-order motion functions needed to go beyond that: roll ratio correction, spiral motion correction and vertical motion correction. The servo axes can execute them. What is missing is the calculation chain that predicts the geometric consequence of each coefficient, so production practice rarely takes the roll ratio polynomial beyond the third order and much of the machine’s generating freedom remains unused. The question addressed here is therefore narrow and practical: when one coefficient of the roll ratio polynomial is changed, where on the flank does the surface move, in which direction, and by how much?
Tooth surface model with a high-order roll correction
Take the left-hand member of a pair as the example. In the machine frame σ = {O; i, j, k}, O is the machine centre, i and j span the machine plane and k points into the cradle. Oc is the intersection of the cutter tip plane with the cutter axis and O1 is the design crossing point of the two axes. The cutter axis has unit vector c and its projection on the machine plane is b. The setting is described by the vertical offset Em, radial cutter position S, angular cutter position q, sliding base XB, horizontal offset Xp and blank mounting angle δM, with p the workpiece axis. These values come either from a cutting calculation program or directly from the settings dialled in on the machine.
The cutting edge has two parts: a straight segment that generates the working flank and a circular arc that generates the root fillet. Because the measurement grid is contracted away from the root, the fillet falls outside the inspected region and only the straight segment is modelled. A point M on the edge is given by
$$\mathbf{r}_c = \mathbf{r}_0 + s_1\,\mathbf{t}_1 \tag{1}$$
where r0 locates the cutter tip M0, s1 is the distance from M0 to M along the edge and t1 is the unit vector of the edge direction.
The flank is conjugate to the cutter surface under the relative motion of the cradle and the blank. Introducing the high-order roll ratio correction, the workpiece rotation becomes
$$\varphi = i_0\left(\Delta q + 2c\,\Delta q^{2} + 6d\,\Delta q^{3} + 24e\,\Delta q^{4} + 120f\,\Delta q^{5} + 720g\,\Delta q^{6}\right) \tag{2}$$
with i0 the nominal roll ratio, Δq the cradle angle increment, and 2c, 6d, 24e, 120f and 720g the second- to sixth-order roll ratio correction coefficients. The instantaneous generating ratio is the derivative dφ/dΔq, so the ratio is no longer constant during the roll: every additional coefficient adds one higher-degree term to the ratio curve. Solving the meshing equation for a point M yields the cradle angle and edge parameter at which M contacts the blank, and hence the workpiece rotation φ; the position and unit normal of the generated point then follow as
$$\mathbf{r}_w = (\mathbf{p}\cdot\mathbf{r}_c)\mathbf{p} + \cos\varphi\,(\mathbf{p}\times\mathbf{r}_c)\times\mathbf{p} + \sin\varphi\,(\mathbf{p}\times\mathbf{r}_c) \tag{3}$$
$$\mathbf{n}_w = (\mathbf{p}\cdot\mathbf{n})\mathbf{p} + \cos\varphi\,(\mathbf{p}\times\mathbf{n})\times\mathbf{p} + \sin\varphi\,(\mathbf{p}\times\mathbf{n}) \tag{4}$$
Discretizing the flank into a 5 × 9 grid
Error measurement and shape analysis both need the spatial coordinates and normals of a finite set of theoretical surface points, which is also a prerequisite for digital manufacturing of the pair. The flank is therefore discretized on its rotational projection: five rows in the tooth-height direction and nine columns in the tooth-length direction, 45 points in total. Following the AGMA bevel gear standard, the grid is contracted by 10 % of the face width at both toe and heel, and by 5 % of the full tooth height at tip and root, with a lower bound of 0.6 mm. The contracted boundaries stay parallel to the original ones and the points are spaced equally between them.
In the axial section, a flank point is located by its distance x along the gear axis from O1 and its radial distance y from that axis:
$$y = \left|\mathbf{r}_w \times \mathbf{p}\right|, \qquad x = -\mathbf{r}_w\cdot\mathbf{p} \tag{5}$$
Both quantities are functions of the cradle angle and the edge parameter. For a prescribed pair (x, y) the corresponding angles are recovered by two-dimensional iteration, after which equations (2)–(4) deliver rw and nw at that point; one further coordinate transformation brings them into the measurement frame. This is what lets a three-coordinate measuring machine or a gear measuring centre probe the flank at prescribed locations with a known normal direction.
Building the deviation map
Running the model with the nominal cutter and machine settings gives the theoretical flank; running it with one parameter altered gives the actual flank. Before the two can be compared, the actual surface is rotated about p until its grid midpoint M′ coincides with the midpoint M of the theoretical surface. After that alignment, the deviation at any grid point is the distance between the two surfaces measured along the normal, which is the quantity a gear measuring centre reports when it traverses the flank in its S-shaped path with the midpoint as reference. Plotting those 45 values produces the tooth surface error topology, in which solid contour lines mark regions where the actual flank lies above the theoretical one and dashed lines where it lies below.
Case gear pair and machine settings
The analysis was applied to the pinion of a pair produced in series by the duplex helical method. Basic data are listed in Table 1 and the machine and cutter settings in Table 2. The contraction of the measurement grid was set to 1.5 mm at the toe, 2.5 mm at the heel, 1 mm at the tip and 1.7 mm at the root.
| Parameter | Wheel | Pinion |
|---|---|---|
| Hand of spiral | right | left |
| Number of teeth | 17 | 17 |
| Shaft angle (°) | 90 | |
| Mean spiral angle (°) | 40 | |
| Module (mm) | 3 | |
| Pressure angle (°) | 17.5 | |
| Face width (mm) | 12 | |
| Parameter | Wheel | Pinion |
|---|---|---|
| Radial cutter position (mm) | 28.7080 | 28.6176 |
| Horizontal offset (mm) | 0 | −0.2618 |
| Vertical offset (mm) | 0 | −0.7194 |
| Sliding base (mm) | 0 | −1.7016 |
| Cutter tilt (°) | 2.58 | 7.42 |
| Cutter swivel (°) | 219.31 | 84.06 |
| Blank mounting angle (°) | 37.48 | 35.99 |
| Roll ratio | 1.402053 | 1.411378 |
| Centre roll position (°) | 81.13 | 86.87 |
| Nominal cutter diameter (mm) | 76.2 | 77.55 |
| Blade offset (mm) | 1.25 | 1.16 |
| Inside blade angle (°) | 25 | 25 |
| Outside blade angle (°) | 10 | 10 |
| First-order spiral motion coefficient | 0 | 4.9717 |
To separate the effect of each order, the second- to sixth-order coefficients were perturbed one at a time by +0.02, +0.2, +2, +20 and +200 respectively, and a dedicated analysis program computed the resulting deviation field and drew its topology for every case. Corner values in these maps reach roughly 17–18 μm, concentrated at the toe of the concave flank and the heel of the convex flank; the middle of the flank and the region near the root move far less.
Five sensitivity coefficients from a quadratic fit
Any measured or computed deviation field can be condensed into five numbers by fitting a second-order surface to it. With the origin at the grid midpoint, X along the tooth length, Y along the tooth height and Z the deviation,
$$Z = a_1X + a_2Y + a_3XY + a_4X^{2} + a_5Y^{2} \tag{6}$$
Here a1 is the spiral angle error coefficient, a2 the pressure angle error coefficient, a3 the tooth surface twist coefficient, a4 the lengthwise curvature coefficient and a5 the heightwise curvature coefficient. The first-order pair a1, a2 describes rigid tilting of the flank; the second-order triple a3, a4, a5 describes its bending. Writing the deviations of all grid points as a vector,
$$\begin{Bmatrix} Z_1 \\ Z_2 \\ \vdots \\ Z_{mn} \end{Bmatrix} = \begin{bmatrix} X_1 & Y_1 & X_1Y_1 & X_1^{2} & Y_1^{2} \\ X_2 & Y_2 & X_2Y_2 & X_2^{2} & Y_2^{2} \\ \vdots & \vdots & \vdots & \vdots & \vdots \\ X_{mn} & Y_{mn} & X_{mn}Y_{mn} & X_{mn}^{2} & Y_{mn}^{2} \end{bmatrix}\begin{Bmatrix} a_1 \\ a_2 \\ a_3 \\ a_4 \\ a_5 \end{Bmatrix}$$
which is written compactly as
$$\{Z\} = [S]\{a\} \tag{7}$$
Since there are 45 equations for five unknowns, the system is overdetermined and is solved in the least-squares sense:
$$\{a\} = \left([S]^{T}[S]\right)^{-1}[S]^{T}\{Z\} \tag{8}$$
Repeating this for each perturbed coefficient gives the sensitivity tables for the concave and convex flanks of the pinion (Tables 3 and 4). A larger magnitude means that the corresponding coefficient has a stronger influence on the shape of that flank.
| Coefficient perturbed | a1 (spiral angle) | a2 (pressure angle) | a3 (twist) | a4 (lengthwise curvature) | a5 (heightwise curvature) |
|---|---|---|---|---|---|
| Second order | −0.001459 | −0.000085 | 0.030401 | −0.032993 | 0.001665 |
| Third order | 0.000669 | −0.000398 | −0.024025 | 0.026346 | −0.000377 |
| Fourth order | −0.000855 | 0.000232 | 0.023961 | −0.025666 | 0.000699 |
| Fifth order | 0.000385 | −0.000179 | −0.012491 | 0.013369 | −0.000246 |
| Sixth order | −0.000227 | 0.000088 | 0.006956 | −0.007385 | 0.000159 |
| Coefficient perturbed | a1 (spiral angle) | a2 (pressure angle) | a3 (twist) | a4 (lengthwise curvature) | a5 (heightwise curvature) |
|---|---|---|---|---|---|
| Second order | 0.000101 | 0.001265 | 0.023800 | −0.024071 | −0.001349 |
| Third order | 0.000372 | 0.000912 | 0.009852 | −0.010666 | −0.001245 |
| Fourth order | 0.000245 | 0.000797 | 0.011255 | −0.011419 | −0.001034 |
| Fifth order | 0.000167 | 0.000395 | 0.000395 | −0.004645 | −0.000555 |
| Sixth order | 0.000075 | 0.000192 | 0.002383 | −0.002416 | −0.000265 |
How each order acts on the flank
Location of the influence. All five coefficients act most strongly in the same two areas: the tip at the toe of the concave flank and the tip at the heel of the convex flank. The middle of the flank and the fillet region respond weakly. Practically, this means the roll correction polynomial is a tool for the ends of the tooth and the top of the profile, not for the centre of the contact pattern.
Magnitude and decay. On the concave flank the twist and lengthwise curvature terms dominate the response: a4 runs from −0.0330 at second order through +0.0263, −0.0257 and +0.0134 to −0.0074 at sixth order, with a3 tracking it at slightly smaller magnitude and opposite sign in every row. The tilt coefficients are an order of magnitude smaller, a1 lying between 0.00023 and 0.00146 and a2 between 0.00008 and 0.00040, while a5 stays near 0.0017 and below. On the convex flank the same pair dominates, a3 falling from +0.0238 to +0.0024 and a4 from −0.0241 to −0.0024. The magnitude roughly halves with each order step, and the sign alternates, so successive even orders push the surface one way and successive odd orders the other.
Tilt versus bending. On the concave flank the spiral-angle coefficient a1 exceeds the pressure-angle coefficient a2 in every row, most clearly at second order where the two differ by more than an order of magnitude. On the convex flank the relation reverses: a2 is consistently larger than a1, from 0.00127 against 0.00010 at second order to 0.00019 against 0.00008 at sixth, and the gap widens as the order rises. Heightwise curvature a5 is negative throughout on the convex flank, whereas on the concave flank it follows the alternating sign of a3.
Direction at toe and heel. At the heel, concave and convex flanks move in the same sense. At the toe the behaviour splits by parity: the second, fourth and sixth orders produce the same pattern, with the actual concave surface below the theoretical one and the actual convex surface above it, while the third and fifth orders reverse both.
Grinding trial and measurement check
The predicted maps were checked on an H350GH CNC spiral bevel gear grinding machine. Test gears were ground with each coefficient perturbation applied in turn, using the settings of Tables 1 and 2, and then measured on a gear measuring centre over the same 5 × 9 grid. The measured topologies reproduced both the direction of the shape change and the deviaton values at corresponding points within a few micrometres; the residual differences are attributable to random and measurement error. Only the second-order case (+0.02) is reproduced here for reasons of space, but it exhibits the same tilt-and-twist pattern and the same concentration at the toe of the concave flank predicted by the model.
What the results mean for correction practice
The five coefficients give the shop floor a map from a machine setting to a symptom on the flank. A lengthwise curvature error visible along the whole face width is addressed with a4-dominated correction, a twist between toe and heel with a3, a pressure angle error with a2 and a spiral angle error with a1. The alternating sign of successive orders also means that a pair of adjacent orders can be combined to cancel an unwanted tilt while reinforcing a desired bend, which is exactly the freedom that a three-term roll ratio polynomial does not provide.
The decay of roughly a factor of two per order, together with the tenfold larger coefficient values needed to make the higher orders visible in the topology plots, sets the practical limit: fourth- and fifth-order terms are fine-adjustment tools with limited authority, useful for trimming the extreme ends of the flank rather than for correcting gross distortion. Making that authority usable requires the calculation software to keep pace with the machine, so that the roll ratio polynomial can be driven to its full order in production rather than being truncated at the third term. Within that framework the model and the sensitivity tables provide the quantitative basis for adjusting flank geometry and for correcting profile error on the shop floor, and the grinding trials confirm that the predicted deviation fields are the ones the machine actually produces.