Spiral bevel gears carry torque between intersecting shafts through teeth whose traces curve across the face width. Because the teeth enter mesh gradually rather than along a full line, the load-bearing behaviour is governed less by the nominal profile than by three observable quantities: the size of the instantaneous contact ellipse, the direction of the contact trace relative to the root cone, and the shape of the transmission error curve over a mesh cycle. All three are set indirectly, through the machine-tool settings used to cut the pinion: cutter radius, pressure angle, radial and angular cutter position, roll ratio, axial and vertical wheel position, tilt and swivel angles.
Why Light-Load Contact Analysis Is Not Enough
Conventional tooth contact analysis (TCA) evaluates the contact pattern under light load, and it does so after the fact: the cutter settings are chosen first, then the resulting pattern is inspected. Because a small change in a single machine setting propagates into the flank curvature, the same nominal design space can produce transmission error curves ranging from nearly straight to pronounced S-shapes, and from symmetric to strongly asymmetric. The contact pattern degrades in parallel, with diagonal contact, diamond-shaped patches, and fishtail patterns appearing when the settings are pushed.
The classical remedy is local synthesis: match the normal vector and the curvatures at a single computation point. That guarantees conjugacy at that point and nothing more, since the curvature condition is not enforced over the rest of the flank. A study at Jiangxi University of Science and Technology addressed this by moving the problem upstream, treating the curvature parameters themselves as design variables and searching them globally with a genetic algorithm, with the resulting contact pattern and transmission error curve as the objectives.
Choosing Curvature Parameters as Design Variables
The pinion spiral angle $\beta_P$, pitch cone angle $\Gamma_p$ and pitch cone distance $R_p$ describe the blank and are fixed by the gear ratio and envelope; they do not shape the tooth form. What does determine the cutting cone of the pinion are the three curvature quantities at the pitch point $P$, the tangency point of the pinion and wheel pitch cones:
- $A_P$ — normal curvature along the tooth height direction;
- $B_P$ — normal curvature along the tooth length direction;
- $C_P$ — the short-range deflection (geodesic torsion) at $P$.
The design vector for each individual in the population is therefore $X_k=[A_P,\,B_P,\,C_P]^{T}$. This choice follows the principle that meshing behaviour is controlled primarily by flank curvature rather than by the absolute position of the blank.
Three Objectives That Suppress Whole Families of Defects
Instantaneous contact ellipse semi-major axis
The semi-major axis $l$ sets the size of the contact patch and suppresses patches that are too wide, too narrow, too long or too short. Its optimum is tied to the face width:
$$l_{OPT}=\frac{F\,b}{2\cos\beta}$$
where $F$ is the face-width contact coefficient, $b$ the face width and $\beta$ the spiral angle. Since a mesh cycle contains many instantaneous ellipses of differing size, the objective is evaluated on the mean over the cycle:
$$\bar l=\frac{1}{n}\sum_{k=0}^{n-1} l_k$$
with $l_k$ the optimal semi-major axis of the $k$-th ellipse and $n$ the number of ellipses in one mesh cycle.
Contact trace direction angle
The angle $\gamma$ between the contact trace and the root cone generatrix controls both the patch orientation and the overlap of successive transmission error curves. For a convex wheel flank, $\gamma<0$ leaves the error curves non-intersecting, while increasing $\gamma>0$ increases their overlap — that is, raises the contact ratio — but too large a value produces severe inner diagonal contact and demands excessive mounting accuracy. The behaviour reverses on a concave wheel flank. The admissible optimum is therefore split by flank type:
$$\gamma_{OPT}\in\begin{cases}[50^{\circ},70^{\circ}] & \text{convex wheel flank}\\[4pt] [110^{\circ},130^{\circ}] & \text{concave wheel flank}\end{cases}$$
To evaluate $\gamma$ from a computed contact pattern, the contact points are fitted with a first-order polynomial,
$$P_{ay}(P_{ax})=\sum_{i=0}^{1} c_i\,(P_{ax})^{i}$$
where $P_{ax}$ and $P_{ay}$ are the contact point coordinates and $c_i$ the fit coefficients, so that $\gamma=\arctan c_1$.
Ordinate of the transmission error curve intersection
The ordinate $\delta$ of the crossing point of successive error curves governs whether the curves bend upward, fail to overlap, or fail to intersect at all. It should be negative and as small in magnitude as possible, so that the actual ratio stays close to the theoretical one; too small a magnitude bends the curve upward. The accepted band is $\delta_{OPT}\in[-6\times10^{-5},\,-4\times10^{-5}]$ rad.
Fitting the transmission error curve with a fourth-order polynomial in the wheel rotation angle $\psi_g$,
$$\Delta\psi(\psi_g)=\sum_{i=0}^{4} c_i\,\psi_g^{\,i}$$
gives the intersection ordinate directly as
$$\delta=c_0+c_2\left(\frac{\pi}{N_g}\right)^{2}+c_4\left(\frac{\pi}{N_g}\right)^{4}$$
with $N_g$ the normal vector at the contact point on the wheel.
Objective Function, Feasibility Constraint and Genetic Search
The three targets are combined in a minimax form so that no single objective can be traded away against the others:
$$\min f(l,\gamma,\delta)=\max\left\{\left|\frac{l-l_{OPT}}{l_{OPT}}\right|,\Delta\varepsilon\right\}+\max\left\{\left|\frac{\gamma-\gamma_{OPT}}{\gamma_{OPT}}\right|,\Delta\varepsilon\right\}+\max\left\{\left|\frac{\delta-\delta_{OPT}}{\delta_{OPT}}\right|,\Delta\varepsilon\right\}$$
where $\Delta\varepsilon$ is the required accuracy. When $f(l,\gamma,\delta)=3\Delta\varepsilon$, every objective has individually reached its tolerance band.
Contact points must lie inside both flanks. Writing $\Delta S^{g}$ and $\Delta S^{P}$ for the axial-section areas of wheel and pinion, the four triangular sub-areas bounded by the contact point $M$ and the tip, toe, root and heel lines must close the total area:
$$\begin{cases}\Delta S^{g}_{top}+\Delta S^{g}_{toe}+\Delta S^{g}_{root}+\Delta S^{g}_{heel}=\Delta S^{g}\\[4pt] \Delta S^{P}_{top}+\Delta S^{P}_{toe}+\Delta S^{P}_{root}+\Delta S^{P}_{heel}=\Delta S^{P}\end{cases}$$
If $M$ falls outside the wheel flank, the sum of the four sub-areas exceeds $\Delta S^{g}$; the same test applies to the pinion. Individuals violating the test are discarded before evaluation.
The search uses binary coding of the control variables, a fitness function $fitvalue=-\lg f(l,\gamma,\delta)$, roulette-wheel selection with
$$P_i=\frac{fitvalue_i}{\sum_{i=1}^{n} fitvalue_i}$$
and the following settings: population $m=30$, crossover probability $P_c=0.6$, mutation probability $P_m=0.09$, with 250 generations as the termination criterion. Each individual requires a full pass through the pipeline — pitch cone parameters of the pinion generating gear, all pinion machine settings, pinion flank generation, assembly with the wheel flank, TCA, feasibility filtering, feature extraction of the contact patch and error curve, then evaluation of $f$. Convergence was reported as well behaved at these settings, which the authors read as evidence that the objective set, the control parameters and the search operator are mutually consistent.
Optimization Outcome
The converged solution reached $l_{OPT}=3.0036$ mm, $\gamma_{OPT}=59.9693^{\circ}$ and $\delta_{OPT}=-5.00077\times10^{-5}$ rad at a computation accuracy $\Delta\varepsilon=0.001$, corresponding to curvature parameters $A_P=-0.0026$, $B_P=0.031988$ and $C_P=-0.01723$. The resulting corrections to the pinion machine settings are listed below; convex and concave flanks are corrected in opposite directions for most quantities, which reflects the asymmetric curvature requirement of the two flanks.
| Setting | Unit | Pinion convex flank | Pinion concave flank |
|---|---|---|---|
| Cutter radius $\Delta r_{0P}$ | mm | −0.8849 | 0.8385 |
| Cutter pressure angle $\Delta\alpha’_{P}$ | deg | 2.1312 | −1.9527 |
| Point width $\Delta U_{P}$ | mm | — | — |
| Radial cutter position $\Delta V_{P}$ | mm | 2.0619 | −2.1729 |
| Angular cutter position $\Delta q_{P}$ | deg | −0.5621 | 0.4975 |
| Machine roll ratio $\Delta i_{P}$ | — | 0.1129 | −0.0109 |
| Vertical wheel position $\Delta Y_{BP}$ | mm | 0.2037 | −0.0826 |
| Axial wheel position $\Delta Y_{P}$ | mm | 2.65 | −2.6435 |
| Perpendicular wheel offset $\Delta F_{P}$ | mm | 2.54 | −2.54 |
| Blank mounting angle $\Delta\Gamma_{MP}$ | deg | 0.1731 | — |
| Cutter tilt angle $\Delta i$ | deg | 0.5234 | −0.1831 |
| Cutter swivel angle $\Delta j$ | deg | 0.2893 | −0.57 |
Finite Element Verification Under Torque
The optimized geometry was rebuilt as a finite element model of the spiral bevel gear pair used in a rotary forging machine, with a contact ratio of $\varepsilon=1.6134$ and simplified to two tooth pairs in mesh. Mapped surface meshing was applied to the whole model in ANSYS, with local refinement on one contacting pair, and the mesh behaviour was evaluated at wheel torques of 100, 200 and 400 N·m.
| Wheel torque $T_g$ (N·m) | Peak von Mises stress at patch centre (MPa) | Stress reached toward the root (MPa) | Patch area and trace length | Trace direction angle $\gamma$ |
|---|---|---|---|---|
| 100 | 241.9 | 2.016 | Reference | Reference |
| 200 | 318.3 | 7.959 | Markedly increased | Reduced |
| 400 | 447.9 (upper contour level) | 14.93 | Further increased | Further reduced |
In every load case the stress field spreads from the centre of the flank outward, with the gradient directed toward the root rather than toward the toe. The patch remains essentially elliptical and centred on the convex flank at the position predicted by the zero-load optimization, which is the central validation argument: the light-load optimum survives the introduction of load, at least within this torque range. The growth pattern is useful in practice. Because the patch expands toward the root as torque rises, the assembly should be positioned so that the light-load contact sits near the toe, leaving the heel and root portions of the flank available as load reserve and thereby using the full face width.
Transmission Error and Instantaneous Ratio Under Load
The theoretical error curve oscillates periodically between a maximum of $-4.64\times10^{-11}$ rad and a minimum of $-5.743\times10^{-5}$ rad, a fluctuation of $5.743\times10^{-5}$ rad. Loading changes both the amplitude and the topology of the curve:
- 100 N·m — maximum $3.055\times10^{-8}$ rad, minimum $-2.351\times10^{-5}$ rad, fluctuation $2.3541\times10^{-5}$ rad; three wave regions per motion cycle, indicating repeated switching between single and double tooth contact.
- 200 N·m — maximum $-2.973\times10^{-8}$ rad, minimum $-3.279\times10^{-5}$ rad, fluctuation $3.260\times10^{-5}$ rad; again three wave regions.
- 400 N·m — maximum $-2.053\times10^{-7}$ rad, minimum $-5.431\times10^{-5}$ rad, fluctuation $5.4105\times10^{-5}$ rad; the wave regions disappear, and the transition becomes smooth because the mesh state has shifted to double or triple tooth contact, redistributing load among teeth and changing the effective working rotation and contact position.
The instantaneous ratio $\Delta i=\Delta\psi_P/\Delta\psi_g$ fluctuates about the theoretical value of 2.083, with a band of about 0.006 at 100 and 200 N·m widening to about 0.009 at 400 N·m. The optimized pair therefore retains its designed meshing behaviour at part load, but the ratio fluctuation grows with torque — a stability penalty that any downstream vibration or noise assessment should take into account.
Scope and Design Implications
The method converts an open-ended machine-setting trial into a constrained optimization on three curvature variables, and it does so without requiring the analyst to guess which of a dozen machine parameters to move. Its value lies in the coupling of the two outputs that are usually tuned separately: the contact ellipse size, the trace direction angle and the error curve intersection ordinate are optimized simultaneously, so a patch that looks acceptable cannot hide a non-intersecting error curve, and vice versa.
The boundaries of the result should be stated as plainly as the result itself. The optimization is performed on unloaded geometry and verified at three torque levels up to 400 N·m; the contact ratio of 1.6134 and the two-tooth-pair simplification bound the finite element evidence; and the setting corrections are specific to the cutter head and machine kinematics of the gear pair studied, so they transfer as a methodology rather than as a table of numbers. Within those limits, the agreement between the predicted zero-load patch and the loaded finite element contact pattern supports treating curvature-space optimization as a practical front end to conventional TCA for spiral bevel gears.